You are on page 1of 170

2016

Pumping Station for the Future


Re-design using multi-objective robust simulation optimization

Emiel van Druten


Title : Pumping Station for the Future
Re-design using multi-objective robust simulation optimization

Author : E.J. van Druten


Telephone : +31 68682935
E-mail : emiel.van.druten@witteveenbos.com
Student ID : 4077369
Published : October 2016

University : Delft University of Technology


Faculty : Faculty of Civil Engineering
Department : Hydraulic Engineering

TU Delft : Prof. Ir. T. Vellinga


supervisors Dr.ir. J.S. Timmermans
Ir. E.C. van Berchum

Witteveen+Bos : Ir. M.F.E. Wauben


supervisors Ir. G.A. van Dorp

Type : Master of Science Thesis


Summary

Uncertainties about climate change and the energy market constitute a major challenge
in designing a pumping station. To tackle this challenge, a robust design is needed, that
performs satisfactory under a wide variety of future conditions. Multi-Objective Robust
Simulation Optimization (MORSO) was applied, to create a robust re-design for pumping
station Vissering in the Dutch Noordoostpolder.
The Pumping Station Simulation & Testing (PSST) model, was developed and calibrated
to reproduce the behavior of pumping station Vissering. The PSST model was used to
simulate design choices, input uncertainties and the resulting performance, as shown in
Figure 1.

Figure 1: Simulation framework

The input uncertainties were characterized by climate scenarios and energy scenarios.
Design choices in the categories pumps, motors, control and innovation, resulted in nu-
merous design alternatives. The performance of the design was expressed by performance
metrics in the categories water quantity & water safety, water quality & ecology, energy
performance and costs.
Combining the future scenarios and design alternatives led to a large set of simulations,
that was analyzed using sensitivity analysis. This ranked the sensitivity of the perfor-
mance, for the design choices and future scenarios, resulting in Figure 2. The future
scenarios caused the same order of magnitude sensitivity as the design choices. As such,
this insight confirmed the importance of including uncertainties about the future in a
pumping station design.

v
vi Summary

In addition, sensitivity analysis revealed several design choices that created a trade-off be-
tween different performance metrics. The resulting spread in performance, could be used
to investigate these trade-offs. The pump control for instance, created a trade-off between
the energy bill and sustainability. On the one hand, pump control based on the availabil-
ity of local wind and solar power showed robust behavior for CO2 emissions and use of
local energy. That is a good average performance (µ) within a narrow bandwidth (σ), as
shown in Figure 3. On the other hand, pump control based on hourly APX electricity
prices was extremely sensitive for the energy scenarios, resulting in poor robustness.

Figure 2: Sensitivity of the performance Figure 3: Robustness of the pump


control

By considering multiple performance objectives, instead of just a single cost objective,


many trade-offs and interactions between design choices and future scenarios were re-
vealed. Traditional design strategies and the other robustness frameworks would not
have come to these insights. In the end MORSO enabled us to design a robust pumping
station and create a starting point for an adaptive operational strategy.
Preface

When I first visited the TU Delft during an open day in 2009 my first question was which
Bachelor to choose if I wanted to learn about sustainable energy. Mechanical Engineering
was the answer and so I started there the following year. During my Masters I decided to
follow my other passion, which is for water, by delving into Hydraulic Engineering. For
my thesis I wanted to combine my passions for water and energy. I called Witteveen+Bos
and together with Marcel Wauben and Bert van Dorp the topic Pumping Station for the
Future was formulated.
I would like to thank Marcel for always making time for me when I had questions about
pumping stations and helping me out with all other problems I encountered.
I would also like to thank Bert and my other colleagues of the energy and sustainability
group for their support and for offering me a job only two months into the project.
Working part-time for Witteveen+Bos besides my thesis research was both exciting and
challenging. I am motivated to start my full-time job at Witteveen+Bos next month. I
hope I can help shape our sustainable energy system for the future and integrate the role
of water in the energy system.
I would like to express my gratitude for the members of my committee at the TU Delft.
To professor Tiedo Vellinga, for trusting me with this unconventional research topic. To
Erik van Berchum, for helping me out with all sorts of questions. To Jos Timmermans,
for his enthusiasm and help to communicate my research to a broad public. Furthermore,
I thank Jan Kwakkel of the TU Delft for helping me with the robustness methodology.
A big thanks to the people at water board Zuiderzeeland that helped me to set up the
case study about pumping station Vissering. Your experience helped me gain insights
about robust design choices and I hope my insights can contribute in making Vissering
the most sustainable pumping station of the Netherlands.
Since many of the subjects I was investigating were new and innovative, little literature
was yet available. Therefore, I gathered much information by interviewing many experts
from different companies and institutions, as listed in Figure 4. I want to thank all experts
for having me over and sharing your experience with me.

vii
viii Preface

I want to mention DNV GL in special and thank them for sharing their energy market
model results with me. This saved me from a futile attempt to model the entire energy
market myself.

Figure 4: List of companies and institutions that contributed to this thesis

I want finish by thanking my family for their support and love, not only during this thesis,
but also during my entire education. Without their support, I would not have made it
this far. I want to thank my sister Tasja for spending a lot of time in helping me with
graphics and multimedia. I want to thank by dad for reading and checking my thesis.
The biggest thanks I owe my mum for giving me support and advice whenever I needed
it, especially those times I did not want to hear it but definitely needed it.

Emiel van Druten


Delft, October 2016
Contents

Summary v

Preface vii

List of Figures xiv

List of Tables xv

Nomenclature xvii

1 Introduction 1
1.1 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Research Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Research Question . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Method and Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Introduction to pumping station design . . . . . . . . . . . . . . . . . . . 11
1.6 Case study: pumping station Vissering . . . . . . . . . . . . . . . . . . . . 17

2 Input Uncertainties 19
2.1 Energy scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Climate scenarios and water system . . . . . . . . . . . . . . . . . . . . . 28

3 Design Choices 35
3.1 Pump choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2 Motor choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.3 Control choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4 Innovations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

ix
x Contents

4 Performance 51
4.1 Water quantity and water safety . . . . . . . . . . . . . . . . . . . . . . . 52
4.2 Water quality and ecology . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.3 Energy performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.4 Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

5 PSST Model 61
5.1 Relationships in System . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.2 Simulating the model in Vensim (manual) . . . . . . . . . . . . . . . . . . 62
5.3 Simulating the model via Python (automated) . . . . . . . . . . . . . . . 63

6 Insight via learning 65


6.1 Expressing robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
6.2 Sensitivity analysis and Finding trade-offs . . . . . . . . . . . . . . . . . . 68
6.3 Determining a robust design . . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.4 Adaptive operational strategy . . . . . . . . . . . . . . . . . . . . . . . . . 88

7 Conclusions and Recommendations 91


7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
7.2 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

A Expert interviews 101


A.1 Interview with Bart Dekens . . . . . . . . . . . . . . . . . . . . . . . . . . 102
A.2 Interview with Jules van Haaren . . . . . . . . . . . . . . . . . . . . . . . 105
A.3 Interview with Barry Scholten . . . . . . . . . . . . . . . . . . . . . . . . . 107
A.4 Interview with Paulien Hoogvorst . . . . . . . . . . . . . . . . . . . . . . . 111
A.5 Interview with Davı́d Brakkenhoff . . . . . . . . . . . . . . . . . . . . . . 112
A.6 Interview with Jan Kwakkel & Jos Timmermans . . . . . . . . . . . . . . 115
A.7 Interview with Ivo Pothof . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
A.8 Interview with A. Koffeman & C. Bakker . . . . . . . . . . . . . . . . . . 120
A.9 Interview with Friso Roest & Marcel Wauben . . . . . . . . . . . . . . . . 123
A.10 Interview with Ernst Moerman & Coen Rood . . . . . . . . . . . . . . . . 125
A.11 Interview with Ronald van Nooyen . . . . . . . . . . . . . . . . . . . . . . 127
A.12 Interview with Kaz Vermeer . . . . . . . . . . . . . . . . . . . . . . . . . . 129
A.13 Interview with Marijke Visser . . . . . . . . . . . . . . . . . . . . . . . . . 132
A.14 Interview with Thomas van Egmond . . . . . . . . . . . . . . . . . . . . . 134
A.15 Interview with Rudolf de Vetten . . . . . . . . . . . . . . . . . . . . . . . 136
A.16 Interview with Wouter Zomer . . . . . . . . . . . . . . . . . . . . . . . . . 138

B DNV GL energy scenario graphs 141


B.1 Merit order plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
B.2 Price versus CO2 intensity plots . . . . . . . . . . . . . . . . . . . . . . . 142

C Python simulation code 143


List of Figures

1 Simulation framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
2 Sensitivity of the performance . . . . . . . . . . . . . . . . . . . . . . . . . vi
3 Robustness of the pump control . . . . . . . . . . . . . . . . . . . . . . . . vi
4 List of companies and institutions that contributed to this thesis . . . . . viii

1.1 PSST model logo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3


1.2 Stock and flow example in Vensim . . . . . . . . . . . . . . . . . . . . . . 4
1.3 XLRM framework (top) and implementation(bottom) . . . . . . . . . . . 5
1.4 Robustness taxonomy. Source: Herman et al. (2015) . . . . . . . . . . . . 7
1.5 Comparison of robustness frameworks. Source: Herman et al. (2015) . . . 8
1.6 Slot machine analogy of the simulation approach with explanation . . . . 10
1.7 Research process and outline of the report . . . . . . . . . . . . . . . . . . 11
1.8 Pumping station functional specification . . . . . . . . . . . . . . . . . . . 12
1.9 Pumping station types. Source: Kunst et al. (2010) . . . . . . . . . . . . . 13
1.10 Pump type history. Source: NGS (2016) . . . . . . . . . . . . . . . . . . . 13
1.11 Water wheel. Source: NGS (2016) . . . . . . . . . . . . . . . . . . . . . . 14
1.12 Screw pump. Source: NGS (2016) . . . . . . . . . . . . . . . . . . . . . . 14
1.13 Centrifugal pump. Source: NGS (2016) . . . . . . . . . . . . . . . . . . . 14
1.14 Axial pump. Source: NGS (2016) . . . . . . . . . . . . . . . . . . . . . . . 14
1.15 Motor type history. Source: NGS (2016) . . . . . . . . . . . . . . . . . . . 15
1.16 Diesel engine at pumping station Vissering . . . . . . . . . . . . . . . . . 15
1.17 Electric motors at pumping station Verdoold . . . . . . . . . . . . . . . . 15
1.18 Verdoold renovation. Source: HHSK (2014) . . . . . . . . . . . . . . . . . 16
1.19 Verdoold design. Source: HHSK (2014) . . . . . . . . . . . . . . . . . . . 16
1.20 Map of the NOP. Source: Zuiderzeeland (2015) . . . . . . . . . . . . . . . 17

xi
xii List of Figures

1.21 Vissering gas engine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18


1.22 Vissering engine shafts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.1 Electricity price trend NL [EUR/MWh]. Source: Schoots et al. (2016) . . 20


2.2 Projection of the installed renewable capacity. Source: Schoots et al. (2016) 22
2.3 Extrapolation of the installed renewable capacity . . . . . . . . . . . . . . 22
2.4 Cost assessment for electricity storage. Source: Kempener and Vivero (2015) 22
2.5 Price duration curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.6 Average APX price during the day . . . . . . . . . . . . . . . . . . . . . . 24
2.7 2030 BaU - merit order. Source: Buck et al. (2014) . . . . . . . . . . . . . 25
2.8 2030 vision 4 - merit order. Source: DNV GL (2015) . . . . . . . . . . . . 25
2.9 CO2 duration curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.10 Average CO2 intensity during the day . . . . . . . . . . . . . . . . . . . . 26
2.11 2030 BaU - Price vs CO2. Source: Buck et al. (2014) . . . . . . . . . . . . 27
2.12 2030 vision 4 - Price CO2. Source: DNV GL (2015) . . . . . . . . . . . . 27
2.13 Energy scenario model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.14 Rainfall regimes. Source: Wolters et al. (2015) . . . . . . . . . . . . . . . 29
2.15 KNMI scenarios. Source: Klein Tank et al. (2014) . . . . . . . . . . . . . 29
2.16 FLEX model. Source: Fenicia et al. (2006) . . . . . . . . . . . . . . . . . . 30
2.17 HBV model. Source: Bergström and Singh (1995) . . . . . . . . . . . . . 30
2.18 Water system model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.19 NOP water balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.20 Water balance measured and modeled . . . . . . . . . . . . . . . . . . . . 33
2.21 Rainfall discharge calibration . . . . . . . . . . . . . . . . . . . . . . . . . 34

3.1 Archimedean screw pumps at Kinderdijk. Source: NGS (2016) . . . . . . 36


3.2 Bulb pump placement at pumping station IJmuiden. Source: NGS (2016) 36
3.3 (L ⇒ R) Centrifugal pump, mixed-flow pump and axial pump. Source:
Pumpfundamentals (2016) . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4 Pump model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.5 Pump inflow CFD analysis. Source: Xylem (2016) . . . . . . . . . . . . . 40
3.6 Trash rack cleaner. Source: Bosker (2013) . . . . . . . . . . . . . . . . . . 40
3.7 Motor model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.8 Control model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.9 Normal fixed on/off control . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.10 Rainfall event 01-01-2012 . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.11 Weather scenarios control . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.12 Day night control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.13 APX control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.14 Decision support systems model . . . . . . . . . . . . . . . . . . . . . . . . 45
List of Figures xiii

3.15 Solar-dike impression. Source: Kaia (2014) . . . . . . . . . . . . . . . . . 46


3.16 Solar-dike planning at Buma. Source: Zuiderzeeland (2015) . . . . . . . . 46
3.17 Power curve of the E-92. Source: Enercon (2016) . . . . . . . . . . . . . . 47
3.18 E-92 wind turbine and older wind mills. Source: Smith (2014) . . . . . . . 47
3.19 Local energy supply model . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.20 Local energy production over a week . . . . . . . . . . . . . . . . . . . . . 48
3.21 Local energy production 2015 . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.22 Local energy production 2015 averaged . . . . . . . . . . . . . . . . . . . . 48
3.23 ATES summer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.24 ATES winter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4.1 Open water evaporation and steady inflow . . . . . . . . . . . . . . . . . . 53


4.2 Water quantity and safety model . . . . . . . . . . . . . . . . . . . . . . . 53
4.3 Water quality and ecology model . . . . . . . . . . . . . . . . . . . . . . . 55
4.4 Trias Energetica. Source: RVO (2015) . . . . . . . . . . . . . . . . . . . . 56
4.5 Energy performance model . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.6 Cost model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

5.1 XLRM simulation framework . . . . . . . . . . . . . . . . . . . . . . . . . 64

6.1 Visual explanation of the robustness metric . . . . . . . . . . . . . . . . . 67


6.2 First loop feature scoring results . . . . . . . . . . . . . . . . . . . . . . . 69
6.3 Fish damage, Energy used and Investment grouped by Fish safe solution . 70
6.4 Water level exceedance and Energy used grouped by Climate scenario . . 72
6.5 Water level exceedance and Energy used grouped by Number of pumps . 73
6.6 Water level exceedance and Energy used grouped by Weather alarm . . . 73
6.7 Water level exceedance and Energy used grouped by Motor choice . . . . 74
6.8 Water level exceedance and Energy used grouped by Pump inflow improve-
ment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.9 Energy bill and Operational cost by Percentage energy neutral . . . . . . 75
6.10 Energy bill and Operational cost by Energy scenario . . . . . . . . . . . . 75
6.11 APX control savings for energy scenarios . . . . . . . . . . . . . . . . . . . 75
6.12 Energy bill and Local energy use by Pump control . . . . . . . . . . . . . 76
6.13 Second loop feature scoring results . . . . . . . . . . . . . . . . . . . . . . 77
6.14 Energy bill and Local energy used grouped by Energy scenario . . . . . . 78
6.15 Energy bill and Local energy used grouped by Pump control . . . . . . . . 79
6.16 Energy bill and Local energy used grouped by Pump control . . . . . . . . 80
6.17 Third loop feature scoring of uncertainties . . . . . . . . . . . . . . . . . . 82
6.18 Scatter plots of most influential uncertainties . . . . . . . . . . . . . . . . 83
6.19 Final robust design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.20 PRIM method explained. Source: Greeven (2015) . . . . . . . . . . . . . . 85
xiv List of Figures

6.21 PRIM results from the first insight loop . . . . . . . . . . . . . . . . . . . 86


6.22 PRIM results from the second insight loop . . . . . . . . . . . . . . . . . . 86
6.23 Final design performance compared to all designs . . . . . . . . . . . . . . 87
6.24 Decision three for third motor . . . . . . . . . . . . . . . . . . . . . . . . . 89

A.1 Hydrological model at 08-03-2016 . . . . . . . . . . . . . . . . . . . . . . . 112


A.2 FLEX model. Source: Fenicia et al. (2006) . . . . . . . . . . . . . . . . . . 113
A.3 Hydrological model at 22-04-2016 . . . . . . . . . . . . . . . . . . . . . . . 114
A.4 Suggested research process . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
A.5 Vissering gas engine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
A.6 Vissering engine shafts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
A.7 Water system model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
A.8 SGvdT goals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
A.9 Gemaal van de Toekomst. . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

B.1 2015 merit. Source: APX (2016) . . . . . . . . . . . . . . . . . . . . . . . 141


B.2 2030 High CO2 price merit. Source: Buck et al. (2014) . . . . . . . . . . . 141
B.3 2030 sustainable merit. Source: Buck et al. (2014) . . . . . . . . . . . . . 141
B.4 2030 vision 3 merit. Source: DNV GL (2015) . . . . . . . . . . . . . . . . 141
B.5 2015 Price vs CO2 intensity. Source: Entsoe (2016) . . . . . . . . . . . . . 142
B.6 2030 High CO2 Price vs CO2 intensity. Source: Buck et al. (2014) . . . . 142
B.7 2030 sustainable Price vs CO2 intensity. Source: Buck et al. (2014) . . . . 142
B.8 2030 vision 3 Price vs CO2 intensity. Source: DNV GL (2015) . . . . . . 142
List of Tables

2.1 Scenario overview. Sources: APX (2016), Buck et al. (2014), DNV GL (2015) 23
2.2 climate overview 2011-2015. Source: Klein Tank et al. (2014) . . . . . . . 28
2.3 climate overview 2030. Source: Klein Tank et al. (2014) . . . . . . . . . . 29
2.4 climate overview 2050. Source: Klein Tank et al. (2014) . . . . . . . . . . 30

3.1 Weather scenarios setting . . . . . . . . . . . . . . . . . . . . . . . . . . . 44


3.2 Day night control setting . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

4.1 Parameters influencing fish safety. Source: Kunst et al. (2010) . . . . . . . 55


4.2 Estimations for social cost of a ton CO2 . . . . . . . . . . . . . . . . . . . 58

5.1 Numerical error due to time-step . . . . . . . . . . . . . . . . . . . . . . . 63


5.2 Simulation time for different time-steps . . . . . . . . . . . . . . . . . . . 63

6.1 Feature scoring for water level exceedance . . . . . . . . . . . . . . . . . . 69


6.2 Feature scoring for local energy used fish damage . . . . . . . . . . . . . . 69
6.3 Operating hours for number of pumps is 3 . . . . . . . . . . . . . . . . . . 73
6.4 Operating hours for number of pumps is 2 . . . . . . . . . . . . . . . . . . 73
6.5 List of the 39 sampled uncertainties . . . . . . . . . . . . . . . . . . . . . 81
6.6 Investment cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.7 Thresholds for PRIM analysis . . . . . . . . . . . . . . . . . . . . . . . . . 85

A.1 Overview of interviews . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101


A.2 Pros and cons of APX and imbalance market . . . . . . . . . . . . . . . . 129

xv
Nomenclature

Latin Symbols
A blade area [m2 ]
C specific heat of water (= 4.182 at 20 ◦ C) [kJ/(kg ∗ K)]
Cp power coefficient [−]
g the gravity constant (= 9.81) [m/s2 ]
∆H head or differential height [m]
P power [kW ]
Qsun solar radiation [W/m2 ]
Q capacity or discharge [m3 /h]
∆T temperature difference [K]
v speed [m/s]

Greek Symbols
η efficiency [−]
µ average performance [−]
ρair density of air (= 1.2 at 20 ◦ C) [kg/m3 ]
ρwater density of water (= 1000) [kg/m3 ]
σ standard deviation [−]

xvii
xviii Nomenclature

Abbreviations
APX Amsterdam Power Exchange
ATES Aquifer Thermal Energy Storage
BaU Business as Usual
CBS Central Bureau for Statistics
CFD Computational Fluid Dynamic
CFL Courant Friedrichs Lewy
COP Coefficient of Performance
DNO Distribution Network Operator
DSS Decision Support System
ECN Energy research Center Netherlands
EMA Exploratory Modeling and Analysis
FEWS Flood Early Warning System
GvO Guaranty of Origin
GWW Soil, Road and Hydraulic engineering
HHNK Hoogheemraadschap Hollands Noorderkwartier
LCC Life Cycle Costing
MORO Multi-Objective Robust Optimization
MORSO Multi-Objective Robust Simulation Optimization
MPC Model Predictive Control
NAP Normaal Amsterdams Peil
NEV National Energy Vision
NOP Noordoostpolder
PRIM Patient Rule Induction Method
PSST Pumping Station Simulation and Testing model
PV Photo-voltaic
SD System Dynamics
SGvdT Stichting Gemaal van de Toekomst
SOW States Of the World
UvW Unie van Waterschappen
WB Water Board
XLRM eXternal factors, policy Levers, Relationships and performance Metrics
Chapter 1

Introduction

1
2 Introduction

1.1 Problem Statement

The Dutch started around the 10th century to actively manage the water. This was
both for keeping high water out, as for keeping the land behind the dikes dry. Human
induced deforestation, peat excavation and de-watering caused massive land subsidence.
Today still 26 % of the Netherlands is situated below sea level in polders PBL (2016).
This forced people to build dikes and mounds. Sluices where constructed to be able to
discharge water at low tide. Wind mills were introduced in the 17th century, enabling
active pumping of water. Later, windmills were replaced by pumping stations driven by
steam engines. From there the pumping stations gradually developed into the modern
pumping station as we know it today.

In the Netherlands water boards are responsible for these pumping stations. These water
boards are also responsible for water quantity and water quality. For centuries, the Dutch
water boards have managed to do this very well. Every 20 years a pumping station requires
replacement of the mechanical and electrical components. Although renovating pumping
stations is nothing new for the water boards, due to uncertainties in climate change and
energy market, this re-design task has become a more complex challenge.

In pumping station (re-)design many design choices have to be made, choosing between
different available pumps, motors and controls. Moreover, there are several new innova-
tions that can possibly be implemented. There are for instance decision support systems
that help operators to determine the optimal pumping strategy. Making a pumping sta-
tion energy neutral, by installing wind mills and solar panels is another recent innovation.

The performance of the design can be expressed by different performance metrics, for
which the design should be optimized. The performance of the design was expressed by
performance metrics in the categories costs, energy, water quantity & safety and water
quality & ecology.

Traditional design methods for designing a pumping station start with determining the
most likely future scenario. Next design choices are made based on that specific scenario.
In fact, the real future will never exactly match the scenario that was predicted. The
designed pumping station is than no longer optimal for the current scenario and might
even show unpredicted or undesired behavior.

There is a need for a new design method that generates a design that is more robust
pumping for different future scenarios. A promising new method to tackle such com-
plex re-design challenges is Multi-Objective Robust Simulation Optimization (MORSO)
Hamarat et al. (2014). MORSO helps in finding a robust design, by simulating differ-
ent designs in different future scenario. The spread in the performance metrics over the
future scenarios is used to quantify robustness. Visual analysis of the simulation results
will be used to determine robustness and provide insights into the trade-offs between the
performance metrics.

This research is a first attempt to apply the MORSO method to a civil engineering work.
A case study will be performed for the re-design of a pumping station, serving as a proof
of concept, that will result in a robust pumping station design.
1.2 Research Objectives 3

1.2 Research Objectives

Goal 1: Pumping Station Simulation & Testing (PSST) model


The first goal of this research is to build a model that can reproduce the behavior of a
polder and pumping station. It should also be possible to test different design alterna-
tives in different states of the world. Therefore, the model is named Pumping Station
Simulation & Testing (PSST) model.
Goal 2: Insight
The second goal is to use PSST model to gain insight in the performance of pumping
station designs choices under uncertainty characterized by the future scenarios. It is
important to understand the trade-offs between performance metrics that arise from a
design choice.
Goal 3: Robust pumping station
The final goal is determining a robust pumping station design, that is finding a design or
set of design alternatives that have good performance across all plausible future scenarios.

Figure 1.1: PSST model logo

1.3 Research Question

On the one hand, this research is about applying a design method that is new in the field of
Civil Engineering. On the other hand, this research is about developing a robust pumping
station for the future. Both aspects are reflected in the following research question:
”What is the value of applying the multi-objective robust simulation opti-
mization framework in designing a robust pumping station for the future?”
In order to answer this question, thorough knowledge and understanding of pumping
stations is needed. Model simulations help creating insights from the PSST model, but a
clear analyzing approach is needed to find all sensitivities and trade-offs. Finally, a clear
definition of robustness is needed to use simulation optimization to find a robust pumping
station design.
4 Introduction

1.4 Method and Outline

In order to achieve the final goal of designing a robust pumping station for the future,
the first step is to build a model that can reproduce the behavior of a pumping station.
The section start with explaining the model building approach. In order to find a robust
pumping station, first a definition of robustness is required. The robustness frameworks
in existing literature, and their relation to this research, are discussed afterwards. The
section concludes with the outline of the report.

1.4.1 Model building approach (SD, Vensim, XLRM framework)

To develop the PSST model three choices concerning the modeling approach need to be
made. A modeling technique, a modeling software application and a modeling framework
for structuring the model should be chosen. In this research the choice was to use the
System Dynamics (SD) modeling technique Forrester (1997), in Vensim software Ventana
Systems (2016), in accordance with the XLRM framework Lempert et al. (2003).
SD is an often-applied modeling technique in policy analysis and design. ”SD is a method
to describe, model, simulate and analyze dynamically complex systems. SD facilitates the
(re)design of systems and design of control structures” Wolstenholme (1990). From this
definition is seems that SD is a fit for purpose match for this research. In SD modeling,
dynamic system behavior is thought to arise due to flows that accumulate in stocks, as
demonstrated in Figure 1.2. An additional advantage of the SD modeling technique is
that it is a graphical method of modeling. SD modeling is graphical in the sense that
stocks, flows, constant and auxiliary variables can be drawn to build a diagram of the
system. This diagram allows the model builder to discuss the model with experts and
stakeholders. The SD technique applied in this research conform the description in the
learning e-book Pruyt (2013).

Stock
flow

constant auxiliary variable


Figure 1.2: Stock and flow example in Vensim

To build a SD model a software application is needed to build the model in. The SD
learning e-book Pruyt (2013) makes use of Vensim software, which is one of the most
commonly applied SD software packages SD Society (2016). Many (comparable) software
applications are available that accommodate SD modeling, so for convenience the same
package as in the learning e-book has been chosen. Vensim translates the relations in
model into a set of differential equations that describe the system. When simulating the
model, Vensim will solve these differential equations over a number of time steps.
The XLRM framework, describes how information about the system should be organized.
This structure makes it possible use the model later for simulations of different design
1.4 Method and Outline 5

alternatives in different future scenarios. XLRM sorts all system elements into one of four
categories: X, L, R, and M, which refer to:

• External factors (X) ”are factors outside the control of decision-makers that may
nonetheless prove important in determining the success of their strategies.” Lempert
et al. (2003). In this research (X) represents the Input uncertainties.

• Policy levers (L) ”are near-term actions that, in various combinations, comprise
the alternative strategies decision-makers want to explore.” Lempert et al. (2003).
The levers (L) are the design choices.

• Relationships in System (R) ”describe the ways in which the factors relate
to one another and so govern how the future may evolve over time based on the
decision-makers’ choices of levers and the manifestation of the external factors.”
Lempert et al. (2003). The relationships (R) between (X), (L) and (M) are linked
together in the PSST model.

• Performance Metrics (M) ”are the performance standards that decision-makers


and other interested stakeholders would use to rank the desirability of various sce-
narios.” Lempert et al. (2003). The same definition is used in this research.

Figure 1.3: XLRM framework (top) and implementation(bottom)

A modeling technique, a software application and a framework to organize the information


don’t make a model yet. Off course first the information should be collected before it can
be translated into a model. Information gathering will be done via literature study and
expert interviews (Appendix A) on pumping station design and other relevant elements.
This information will be used for the development of different sub-models that linked
together will form the PSST model.
6 Introduction

1.4.2 Defining robustness

When we try to make long term plans for the future we run into the problem that the
future is uncertain. In the light of decision making in a future that is uncertain Walker
et al. (2013) says the following: Given that one cannot predict, which actions available
today are likely to serve best in the future? In Walker et al. (2013) four approaches for
sustainable planning are presented:

1. Resistance: Plan for the worst possible case or future situation.

2. Resilience: Whatever happens in the future, make sure that the system can quickly
recover.

3. Static robustness: A plan that performs satisfactory under a wide variety of


future conditions.

4. Dynamic robustness (or flexibility): A plan that leaves options open and can
adapt to changing future conditions Rosenhead et al. (1972).

In case of a pumping station a resistant plan would deliver an over-dimensioned pumping


station that is too expensive. A resilient plan would suggest that quite some flooding and
damage would be accepted, but the system would be able to recover. It is conceivable
that a lot of stakeholders would not accept such a plan.
A robust plan on the other hand, would deliver a plan that is neither too expensive,
nor allow a lot of damage. Consequently, robustness is the approach of choice for this
research. This plan may be either static as it is fixed, or dynamic as the plan can change
over time. About every 20 years a pumping station requires replacement of the mechanical
and electrical components. Thus, this research considers a static robust plan for every
20 year renovation cycle. Within this period an adaptive operational strategy can be
formulated that dynamically robust.
Kwakkel et al. (2016) distinguishes two definitions of robustness:

• Reducing the uncertainty about the expected consequences of a given policy. So, no
matter how the future plays out, the policy performance falls in a narrow bandwidth.

• Minimizing the undesirable outcomes. So, no matter how the future unfolds, policy
performance will be satisfactory.

Although these definitions may seem quite similar, they can result in two completely
different designs. The distinction between the definitions lays in the fact that the first
definition is focused on performance in within a narrow bandwidth, while the second
definition in only cares about the outcomes meeting a certain threshold. For instance,
a water board does not like an energy bill that is higher than expected, but they do
not mind if the energy bill is lower than expected. In the first definition of robustness
good and bad deviations from the expected outcome are treated equally, while in the
second definition only the bad deviations are considered. This suggest that the second
definition applies better, but to be sure both definitions will be tested. In Chapter 6 both
1.4 Method and Outline 7

definitions will be used in a visual analysis of the simulation results to quantify robustness.
The flexibility definition of robustness Rosenhead et al. (1972) will be incorporated in
operational strategy of design.
Now that we defined robustness there is the distinction between single-objective optimiza-
tion and multi-objective optimization left to be made. In the single objective optimization,
multiple objectives are combined into one objective. In the multi-objective optimization,
one aims at finding the trade-offs between the performance metrics, leaving discussions
about which performance metric is more important to the decision makers.

1.4.3 Relation to robustness frameworks in literature

In the paper ”How should robustness be defined for water systems planning under change?”
Herman et al. (2015) different robustness frameworks are compared. Figure 1.4 shows the
robustness taxonomy that Herman et al. (2015) constructed to facilitate the comparison.
The robustness taxonomy takes the XLRM framework to a higher level by specifying
how the different categories can be substantiated and how this leads an expression for
robustness.

Figure 1.4: Robustness taxonomy. Source: Herman et al. (2015)


8 Introduction

The XLRM categories come back in the robustness framework in the following way:

1. Identifying alternatives, corresponding to the way the design alternatives or


policy levers (L) are identified and operationalized.

2. Sampling states the world, describing the way uncertain external factors are
found and sampled.

3. Quantification of robustness metrics, describing how robust is defined for a


design alternative (L), by looking at the resulting performance metrics (M) over the
different states of the world (X).

4. Identification of key uncertainties or robustness controls. This can be done


using different forms of sensitivity analysis.

Figure 1.5: Comparison of robustness frameworks. Source: Herman et al. (2015)

Figure 1.5 shows the robustness frameworks that were compared in Herman et al. (2015)
and their components classification within the robustness taxonomy. Below the different
frameworks will be described.
Info-gap: ”An information gap is defined as the disparity between what is known and
what needs to be known in order to make a reliable and responsible decision. Info-gap
assumes that the design alternatives are known prior to the analysis. It then proceeds to
evaluate how large the uncertainty should become before a given action fails to meet its
1.4 Method and Outline 9

specified performance requirements.” Walker et al. (2013). Info-gap is not focused finding
a robust design out of all possible design alternatives and is thereby not suited for this
research.
Decision Scaling and Robust Decision Making(RDM) are powerful frameworks
used to produce a static robust plan. Both Decision Scaling and RDM consist of about
the following steps:

1. Organize the problem within the XLRM framework and build the model.

2. Model policy solutions.

3. Consider policy solutions effectiveness and its vulnerabilities. Iterate step 2 and 3.

4. Plot expected outcomes of all policy solutions and choose the most robust solutions.

There are some small differences in the way both framework operationalize these steps.
The major difference between Decision Scaling and RDM however is that first-mentioned
is focused only at climate uncertainties, while RDM is a more generic framework.
Many-Objective Robust Decision Making (MORDM) Kasprzyk et al. (2013)
MORDM takes the RDM framework as a starting point and try to combine it with
evolutionary algorithms to improve approach and make them less labor-intensive for the
analyst. The evolutionary algorithms are used to find a small set of promising design
alternatives for a reference state of the world. Next these designs are stress-tested over
all plausible states of the word. MORDM used interactive visual analytics of the results
to quantify the robustness of the design alternatives.
Multi-Objective Robust Simulation Optimization (MORSO / MORO / RO)
Hamarat et al. (2014) takes the use potential of computer simulations even further by
combining the simulation of design alternatives and states of the world into a single
optimization. MORSO is also a framework that builds further on the RDM framework.
The advantage of MORSO over MORDM is thus that MORDM is not able to take in-
teractions between the design alternative and states of the world into account when it
is selecting the design alternatives of interest. Two disadvantages of MORSO on the
other hand are that is more computationally intensive and it gives limited insight as the
robustness quantification consist of a statistical calculation instead of visual analytics.
This research applies the MORSO framework, but also introduces some features of
MORDM. The simulation approach from MORSO was used, as it makes sense to simulate
all design alternatives over the future states of the world and not only the design that
appear to perform well for a reference scenario. Insight is gathered from these simulations
by using sensitivity analysis of the ranges and ranking type. This helps identifying trade-
offs that arise from the design choices. In MORSO the robustness is normally quantified
statistically, but in this research the choice was made to do this by means of visual
analysis comparable to that of MORDM. By combining MORSO and MORDM in this
way maximum insight is gained. The cover of this chapter shows a fragment from the
online introduction video van Druten (2016) about this research. In this video the analogy
of a slot machine was used to represent the simulation approach.
10 Introduction

Figure 1.6: Slot machine analogy of the simulation approach with explanation

Figure 1.6 explains how the slot machine analogy should be interpreted. In the video the
input uncertainty column and design choice column are randomly rotated. The central
row represents one simulation. The future scenarios together form the States of the World
(SOW). The design options together form a design alternative. One simulation calculates
the performance of one design alternative in one SOW. The performance is summarized
by four performance groups, that each contain one or more performance metrics.
1.5 Introduction to pumping station design 11

1.4.4 Outline of the report

1. This introductory chapter will conclude with a brief history of pumping station
design and an introduction of the case study.
Next the main-matter will start with explaining how the model was build conform
the XLRM framework, as can be seen in Figure 1.7.
2. Chapter 2 starts with specifying the input states of the world.
3. In Chapter 3 all design choices and new innovations that form the design alternatives
are introduced.
4. The performance metrics are presented in Chapter 4.
5. The model description is finalized in Chapter 5 with relationships that link the
previous elements together in the PSST model.
6. Chapter 6 describes the insight via learning loop that performs a sensitivity analysis
to find trade-offs. This is an iterative loop, that will repeat a couple of times, to
come to a robust design.
7. The report will conclude with the conclusions and recommendations in Chapter 7.

Figure 1.7: Research process and outline of the report

1.5 Introduction to pumping station design


A pumping station always two important parts, the pump that raises the water to a
higher level, and the motor that drives the pump. The history of pump and motor types
is introduced in this section. But first this section start with defining the function of a
pumping station and the possible pumping station types.
12 Introduction

1.5.1 Pumping station functions

Rijkswaterstaat defines the function of a pumping station as follows:


”A Hydraulic structure that regulates water levels. A pumping station pumps
or lifts water from a low point to a higher point.” van Beijen (2016)
Another definition of the function of a pumping station is:
”A pumping station serves to bring water from a low lever to a higher level.
The need for it is in a water surplus at the low side (discharge), or a water
demand at the high side (supply).” Cultuurtechnische Vereniging (1988)
The main and sub functions of a certain object can be expressed in a functional specifi-
cation. To establish this functional specification, different experts and stakeholders were
interviewed, see Appendix A.8, A.9, A.10, A.13 and A.14. This yielded the functional
specification shown in Figure 1.8. The main function of regulating water levels can be
needed because of a water surplus (water safety), or a water demand (water quantity).
Facilitating fish migrating can be considered a sub function of a pumping station. Pump-
ing stations can also have a function in water quality, for instance in flushing salt intrusion
through the water system.
Pumping station
functions

Regulating water
Fish migration Water quality
levels

Water quantity Water safety

Figure 1.8: Pumping station functional specification

1.5.2 Pumping station types

Pumping stations can be split up into closed system sewage pumping stations and open
water pumping stations, as seen in Figure 1.9. An open water pumping station can either
have the function of water level regulation, or water transport. This research considers
drainage canal pumping stations and the larger polder pumping stations which are both
in the water level regulating category. They are most critical for the functioning of the
water system and improving them has the biggest impact.
Drainage canal pumping stations and polder pumping stations are not different in their
function, but in the area they pump water from. Polder pumping stations pump water
from the polders to the higher drainage canal. The drainage canal, a term loosely trans-
lated from the Dutch word boezem, can be considered the aorta of the water system.
At the end of the drainage canal large drainage pumping stations pump water from the
drainage canal into water bodies that directly convey to the sea.
1.5 Introduction to pumping station design 13

Pumping stations

Closed systems Open water


sewage systems

Water transport

Water level Transport pumping Circulation puming Flushing pumping


regulation station station station

Drainage canal Polder pumping Lifting pumping High water Vacuum soil
pumping station station station pumping station drainage

Figure 1.9: Pumping station types. Source: Kunst et al. (2010)

1.5.3 History of pump types

From the name pumping station it is no surprise that the pump is the most important
part of the pumping station. The pump is the part that raise the water to a higher level.

Figure 1.10: Pump type history. Source: NGS (2016)

There are two broad categories of pumps NGS (2016), lifting types that raise the water
by lifting it up, and the pumping types that use an impeller to create a pressure difference
to propel the water upward.
The first category contains the water wheel (scheprad) and the Archimedean screw pump
(vijzel). The water wheel, Figure 1.11, is the oldest existing pump type, but has not
applied any-more since the start 20th century. The second ’lifting’ pump type is the
Archimedean screw pump, Figure 1.12, which has been around since 1860 but was not
applied a lot until a revival in the 1950ties that continues until the present day.
14 Introduction

Figure 1.11: Water wheel. Figure 1.12: Screw pump.


Source: NGS (2016) Source: NGS (2016)

The first pump with an impeller is the centrifugal pump (centrifugaalpomp), Figure 1.13,
which was most applied in late 19th century and beginning of the 20th century. Around
1915 NGS (2016) the axial pump (schoefpomp) is introduced, Figure 1.14, and is mainly
applied in small and mid-size pumping stations. The axial pump is still applied today,
in vertical position as well as horizontal. A special variant of this type is the horizontal
bulb pump where motor is integrated in the pump, which is a pump for high capacities
at low head.

Figure 1.13: Centrifugal pump. Figure 1.14: Axial pump.


Source: NGS (2016) Source: NGS (2016)

Around 1930 a crossover of the centrifugal and the axial pump is introduced called the
mixed-flow pump (schoef-centrifugaalpomp). This type increases in popularity when it
is applied in concrete later in the 20th century. Because of this material and shape this
pump type is known as the concrete volute pump. This pump type is still applied a lot
today, especially in larger pump stations.
1.5 Introduction to pumping station design 15

1.5.4 History of motor types

The motor or drive system is the component that drives the pump. In Figure 1.15 it can
be seen that windmills are the first system actively driving a pump. Wind mills improved
the ability of the Dutch to discharge water, but this method was very depended on the
weather. Until the introduction of the steam engine around 1850 windmills were the most
important driving system (motor) for pumping stations.

Figure 1.15: Motor type history. Source: NGS (2016)

After 1900 new motors for pumping stations are introduced. Between 1910 and 1960 NGS
(2016) diesel engines are often applied. The diesel engine at pumping station Vissering is
shown in Figure 1.16. In 1910 the electrical motor is introduced an is now the standard
in pump station design. The two electrical induction motors of pumping station Verdoold
are shown in Figure 1.17. There was also a short period that dual-fuel and natural gas
engines were applied. Diesel-electric hybrid motors are also applied sporadically.

Figure 1.16: Diesel engine at pumping Figure 1.17: Electric motors at pumping
station Vissering station Verdoold
16 Introduction

1.5.5 Verdoold: from historical to modern pumping station

Pumping station Verdoold is one of the most modern pumping stations in the Nether-
lands since its renovation in 2014. Pumping station Verdoold also tells a story about
pumping station history in the Netherlands. A site visit was made to Verdoold, to learn
about is history and the innovative renovation. Pumping station Verdoold is located in
Gouderak along the Hollandse IJssel, and managed by Hoogheemraadschap Schieland en
de Krimpenerwaard.

This location has been used for discharging water since the 1371. First with a discharge
sluice, and in 1866 a water wheel with a steam engine was added to this sluice. Not
much later in 1880 a new steam engine was installed with two centrifugal pumps. This
pumping station was so powerful that is could also serve the neighboring polder. Moreover,
all remaining windmills and discharge sluices could be put out of use. In 1920 the steam
engines were replaced by suction gas motors. In 1950 these were replaced by electrical
motors.

In 2014 the Verdoold has been fully renovated. The old building was renovated, Figure
1.18 and completely new installation, shown in Figure 1.19 was placed inside. This inno-
vative design included two fish safe axial pumps with electrical motors. Pumping station
operator Jan Spek explained that in the fish breading season water is let into the polder so
fish can migrate into the polder. Fish migration thus has an influence in pump operation.

Figure 1.18: Verdoold renovation.


Figure 1.19: Verdoold design.
Source: HHSK (2014)
Source: HHSK (2014)

The shape and positioning of the pump inflow was optimized with a Computational Fluid
Dynamic (CFD) analysis. This improved the overall efficiency of the pumping station.

The electrical motors are water cooled and the rest heat is stored in an aquifer thermal
energy storage system in the ground. This heat can later be used for heating the building.

Pumping station Verdoold was introduced as an example of an innovative pumping station


renovation. Now that the context of the innovative renovation of Verdoold gas been given,
we can move to the actual case study of this research, namely the planned renovation of
pumping station Vissering.
1.6 Case study: pumping station Vissering 17

1.6 Case study: pumping station Vissering

The PSST model will be tested in a case study for pumping station Vissering, which drains
the Noordoostpolder (NOP). Water Board (WB) Zuiderzeeland which manages pumping
station Vissering was interested in cooperating in the case study, find the interview in
Appendix A.8. Pumping station Vissering will be re-designed and renovated in 2018. This
makes Vissering an interesting case study, since the recommendations of this research
might really be applied in the re-design.
The NOP is quite different from other polders in the Netherlands. The NOP, with a
size of 48 km2 was namely reclaimed from the IJsselmeer (the former Zuiderzee) between
1937 and 1942. The NOP is one big polder and has no smaller polders discharging into a
drainage canal, as is the case in most of the Netherlands. There is a separation between
a high, middle and lower area in the NOP. Vissering is situated in the lower area near the
former island of Urk, as can be seen in Figure 1.20. Apart from Vissering the NOP has
two more pumping stations: Buma in the north and Smeenge in the east.

Figure 1.20: Map of the NOP. Source: Zuiderzeeland (2015)

WB Zuiderzeeland was formed by merger of WB Noordoostpolder and WB Fleverwaard in


the year 2000 canonnoordoostpolder.nl (2016). Before the merger, WB Noordoostpolder
was only responsible for water quantity and safety. WB West-Overijssel was namely
responsible for the water quality in the NOP. After the merger, WB Zuiderzeeland took
over the water quality task.
Because the NOP has a small wet area percentage (about 1%) the WB had to deal with
water level exceedance a few times in the past decades. The water board has worked hard
to increase the water storage capacity. This does however not mean that all problems
are solved. Climate change, land subsidence and a possible water level raise of the IJs-
selmeer pose a new challenge for the water board (canonnoordoostpolder.nl (2016) and
Zuiderzeeland (2010)).
Pumping station Vissering was opened in 1940 NGS (2016). Vissering has three vertical
centrifugal pumps in a concrete volute casing. Two pumps have a capacity of 800 m3 /s
18 Introduction

and a third one of 720 m3 /s. The head is 5.5 m. In 1946 the original steam engines
were replaced by three diesel engines. In 1983 one of the diesel engines was replaced for a
newer diesel engine. In 1989 the pump impellers were replaced. The last major renovation
included the replacement of the older two diesel engines by gas engines.
The building of pumping station Vissering is shown of the cover of this report and Figure
1.22 shows Vissering form inside. The two yellow colored shafts are driven by the gas
engines (Figure 1.21), while the green shaft in the middle is diesel driven (Figure 1.16).

Figure 1.21: Vissering gas engine Figure 1.22: Vissering engine shafts

When there is no need to discharge, the gas engines can drive a generator that produces
electricity that is fed into the grid. The rest heat produced by gas engines is currently
delivered to companies nearby via a piping system.
In the renovation in 2018 WB Zuiderzeeland is planning to electrify the pumping station.
This means replacing the existing engines by electrical motors. The goal of the project is
to make Vissering the most sustainable pumping station of the Netherlands.
Chapter 2

Input Uncertainties

19
20 Input Uncertainties

The cover of this chapter neatly summarizes the chapter about uncertainties about the
future forming the input States Of the World (SOW). The key factors were assumed to be
energy and climate. They are characterized by prespecified scenarios. The energy market
is changing due to the energy transition. This chapter start with section 2.1 Energy
scenarios, followed by section 2.2 Climate scenarios and water system.

2.1 Energy scenarios

Climate change is putting pressure on society to shift from fossil fuels to a renewable
energy supply. This energy transition has extensive consequences for the energy market.
How the energy market will develop in the future is uncertain, but there are quite some
plausible scenarios. Energy scenarios from DNV GL DNV GL (2015) were used in this
research. DNV GL is a leading company in energy transition advice and they were willing
to share their knowledge to support this research. This section starts with a general
description of energy market and the energy transition and then the future scenarios are
from DNV GL are introduced.

2.1.1 Energy market

The electricity price that the end consumer pays is build up out of three mayor elements:
the bare energy price, grid costs and taxes. For small consumers, the taxes and grid
costs are the largest part of the energy bill. For large consumers, like water boards, the
distribution and taxes are relatively small compared to the bare energy price.
The bare energy price is determined at the Amsterdam Power Exchange (APX), which
changes every hour based on supply and demand. Already the APX prices are changing.
The average APX price has gradually decreased since 2008, see Figure 2.1. In the autumn
of 2015 this decrease accelerated, but since the spring of 2016 there has been some recovery.
Figure 2.1 suggest that the price slowly keeps increasing the coming years. There are many
factors playing a role in this price development, but mainly the oil, gas and coal prices
are of large influence.

Figure 2.1: Electricity price trend NL [EUR/MWh]. Source: Schoots et al. (2016)
2.1 Energy scenarios 21

2.1.2 Energy transition

The energy transition can be summarized by three trends DNV GL (2015):

1. An increase in decentralized (sustainable) electricity generation.

2. Integration of European energy markets and grids.

3. A strong increase in installed sustainable capacity.

The introduction of solar and wind power creates new dynamics in the energy system
Rooijers et al. (2014). Electricity prices are currently demand driven, but as more wind
and solar enters the market they may become supply driven. Nowadays the daily pattern
has a morning and evening price peak, see Figure 2.6. In the future, this pattern might
change due to fluctuating supply. On a sunny and windy day, a surplus in supply will
drive prices down.
Prices can even fall negative. Sunday 8 may 2016 prices dropped to a staggering -130
e/MWh Quartz (2016) in Germany and stayed negative for several hours. Germany
had to pay neighboring countries to take their electricity. This caused prices in the
Netherlands to also become negative. Similar events happen on a regular, say monthly
basis APX (2016). Negative prices caused by a surplus in supply may become more
frequent as installed renewable capacity increases. These moments with negative prices
are really interesting for the water boards, as they can earn money by turning on the
pumping stations.
The installed renewable capacity in 2015 and two predictions for 2030 are given by fol-
lowing reference cases:

• The 2015 the installed renewable capacity in the Netherlands was 1.5 GW solar plus
3.4 GW wind Schoots et al. (2016).

• The 2030 projected installed renewable capacity is 12 GW solar and 17 GW wind


Schoots et al. (2016), see Figure 2.2.

• A 2030 projection based on a self-conducted trend fit, shown in Figure 2.3, shows 9
GW solar and 19 GW wind.

– For onshore wind and solar historical data was used from Windstats.nl (2016)
and Schoots and Hammingh (2015).
– For offshore wind the planned project by the government until 2023 were used
from Noordzeeloket (2016).

The difference between Figure 2.2 and 2.3 is mainly they the self-conducted trend fit
has a higher estimation for offshore wind. The difference was larger when Figure 2.2
was still showing a figure from the National Energy Vision 2015 (NEV2015) Schoots and
Hammingh (2015), but when it was updated to NEV2016 Schoots et al. (2016) both
predictions were closer together.
22 Input Uncertainties

Figure 2.2: Projection of the installed


renewable capacity. Source: Figure 2.3: Extrapolation of the in-
Schoots et al. (2016) stalled renewable capacity

For these 2030 scenarios, the installed renewable power is the same order of magnitude
as the total electricity demand Rooijers et al. (2014). On a sunny and windy day, nearly
no residual (conventional) load is required to meet demand. A surplus of energy must
stored, exported or otherwise curtailed, meaning solar and wind generation is turned off.
Energy storage can be active, when energy is stored in for instance a battery. Storage
can also be passive, when energy is stored within a business process, by means of flexible
demand. Pumping station are a good example of a process that can be flexible. In
section 3.4.1 we will investigate the possibility of using a pumping station as a flexible
demand. The cost of energy storage is currently about 100 e/MWh, but is decreasing
due to technological developments.
As can be seen in Figure 2.4 the cost of storage will reach the same order of magnitude
as the bare energy price in the following years, making storage commercially interesting.
Active energy storage will only be applied on large scale if price fluctuations with a
magnitude large than the storage costs occur on a regular basis. How cheap storage will
become, and how quickly it can be scaled up and installed remains a question.

Figure 2.4: Cost assessment for electricity storage. Source: Kempener and Vivero (2015)
2.1 Energy scenarios 23

2.1.3 DNV GL scenarios - price

To assess the effect of using a pumping station as flexible demand on the APX market,
future electricity prices scenarios on an hourly basis are needed. Large and complex energy
market models are needed to simulate these scenarios. Only a small number of companies
possess such a model, which made in a challenge to acquire the data. Fortunately, DNV
GL was willing to share their 2030 energy price datasets DNV GL (2015). The dataset
also includes also scenarios from an earlier research DNL GL together with CE Delft
Buck et al. (2014). The scenarios and their assumed renewable capacity and CO2 price
are summarized in Table 2.1.

2015 BaU high CO2 sust- vision 3 vision 4


price ainable
Solar + Wind [GW] 1.5+3.4 20 20 30 8+12 9+19
CO2 price [e/ton] e5 e25 e71 e25 e25 e25
average price [e/MWh] e40 e60 e74 e53 e39 e24

Table 2.1: Scenario overview. Sources: APX (2016), Buck et al. (2014), DNV GL (2015)

The reference dataset consists of historical APX prices from 2011 to 2015 APX (2016), of
which only 2015 is depicted in the table and graphs. Note from Table 2.1 that the scenario
with the highest installed renewable capacity is vision 4. However this ”extreme” scenario
has a comparable amount of installed renewable capacity as the previously described
reference cases.
Another way of summarizing the price scenarios is with a price duration curve, see Figure
2.5. This graph shows all the prices that occur over the year sorted from high to low.
What stands out is that in vision 4 has an energy price of 0 e/MWh for nearly 1/3 of
the year. Vision 3 and the sustainable scenario also contain some zero priced hours.

Figure 2.5: Price duration curves

The scenarios 2015 APX, BaU and high CO2 price on the other hand show a quite flat
price duration curve, meaning that there is little volatility. An important, but debatable,
assumption by DNV GL is curtailment when prices threaten to become negative. This
24 Input Uncertainties

means shutting down wind mills and solar panels to prevent negative prices. As was
previously stated, currently negative prices do occur. With the current policies renewable
suppliers can still make a profit, when prices are negative, as they still receive SDE subsidy
when prices are negative. Only when prices are negative for a period longer than 6 hours
there will be no SDE compensation over these hours Kamp (2015).
DNV GL (2015) concluded the following about the price duration curve: ”The volatility in
2030 increases in most of the scenarios, and as expected the volatility is higher with more
renewable capacity. The scenarios with a large installed renewable show many hours with
low prices and even prices at 0 e/MWh, especially in vision 4. No significant increase in
price peaks is seen. In general, the 2030 scenarios contain either a lot of low price peaks
(vision 3 and 4), or a higher average energy price (other scenarios).”
Comparing the DNV GL (2015) to Frontier Economics (2015), a study commissioned by
the Dutch Ministry of Economic Affairs, quite some differences can be observed. Frontier
Economics (2015) presents only a base case scenario with a 2030 average price of 50
e/MWh. Interesting is that Frontier Economics (2015) is showing both negative prices
and high peak prices ranging up-to 2386 e/MWh, which were not seen in the DNV GL
(2015) scenarios.
Another important aspect is the price development over the day, which is visualized in
Figure 2.6. All scenarios show about the same pattern, that is a cheap night, a morning
peak and an evening peak.

Figure 2.6: Average APX price during the day

DNV GL (2015) concluded the following about the daily price development: ”The scenar-
ios show an earlier price peak in the morning. Moreover, the ’day plateau’ is replaced by
a midday price dip.” Both changes can be attributed to solar power, as it only starts pro-
ducing power after sunrise and peaks during midday. Lastly, DNV GL (2015) concludes
about price fluctuations: ”Most fluctuations are in the order of 20 e/MWh and mostly
occur within a time frame of 6 hours. The frequency of large price fluctuations is low.”
The merit order is the relation between price and (residual) load. As load increases,
also power plants with higher marginal cost will be fired up. Renewables have near zero
marginal costs, and are thus at the bottom of the merit order. The merit order graphs
for Business as Usual (BaU) and vision 4 are shown in respectively Figure 2.7 and 2.8.
2.1 Energy scenarios 25

Figure 2.7: 2030 BaU - merit order. Figure 2.8: 2030 vision 4 - merit order.
Source: Buck et al. (2014) Source: DNV GL (2015)

Vision 4 has a steep merit order compared to BaU, this corresponds with the price duration
curves of Figure 2.5. The graphs of the other scenarios are shown in Appendix B. Vision
3 and 4 show a lot of scattered points, while the other scenarios show a clear merit order.
This indicated that a different model was used to simulate vision 3 and 4. In this model
the merit order effect is apparently not so well incorporated.

2.1.4 DNV GL scenarios - CO2

The DNV GL (2015) scenarios also contains an hourly energy mix and CO2 intensity
[kg/MWh] of this mix, but report does not go into detail about energy mix and CO2
intensity. Currently the CO2 emission of a pumping station are calculated by multiplying
the yearly energy use by the yearly average CO2 intensity of the energy supply. An
interesting experiment would be to evaluate on the CO2 intensity on an hourly instead
of yearly basis. By strategically turning pumping stations on and off, at certain hours,
water boards could potentially reduce national CO2 emissions and stimulate the energy
transition to renewable energy. This would however require real-time monitoring of CO2
emissions.

Figure 2.9: CO2 duration curves

Figure 2.9 visualizes the CO2 duration curves. It can be seen that the CO2 intensity is less
volatile than the prices. The 2015 scenario seems to have the lowest CO2 intensity, but this
26 Input Uncertainties

is in fact incorrect, as the average should have been 355 kg/MWh co2emissiefactoren.nl
(2016). This error is caused by the fact that the source used for this data Entsoe (2016)
overestimates the renewable fraction in the energy mix.

Ideally one wants cheap and clean electricity. From the daily price development in Figure
2.6 we have seen that prices are lowest during the night. Let us check if these cheap
night hours also result in clean energy, by looking at the daily development of the CO2
in Figure 2.10.

Figure 2.10: Average CO2 intensity during the day

In fact, the opposite can be seen: CO2 intensity is high during the night and has a
minimum at midday. In the 2015 case, although being structurally too low, it can be
seen that the CO2 intensity has a comparable curve to that of the price. The morning
and afternoon peak are probably caused by the firing up of extra fossil supply to meet
demand. The graph suggests that currently, pump control based on day and night tariff
is both cheap and clean, but in the future this will no longer be clean. The graphs show
that in the future scenarios 12:00 to 15:00 is a more appropriate time to run the pumping
station.

Demand is not the same every day and with a weather dependent supply the daily averaged
graphs do not depict the full story. We have seen that sun and wind can create a surplus
of energy driving the price down to zero. It may seem logical that in a zero-price situation
there is only renewable supply, as renewables have near zero marginal cost. However, it
might be the case that coal-fired plants cannot shut down quick enough, and will still be
running at a reduced output level.

To investigate this, the relation between price and CO2 intensity was plotted for BaU
in Figure 2.11 and vision 4 in Figure 2.12. BaU shows some near zero prices and these
all have a low CO2 intensity. For increasing price, also the CO2 intensity increases, thus
there is some sort of CO2 merit order. From the averaged daily pattern, we concluded
that cheap was not clean. From the price CO2 intensity however, one concludes that
cheap is in fact clean energy.
2.1 Energy scenarios 27

Figure 2.11: 2030 BaU - Price vs CO2. Figure 2.12: 2030 vision 4 - Price CO2.
Source: Buck et al. (2014) Source: DNV GL (2015)

Vision 4 (and 3) however shows no correlation between price and CO2 intensity at all.
There are even zero prices that show very high CO2 intensity. Perhaps vision 3 and 4
assume a coal base load that is not quickly shut down.
The other scenarios can be found in Appendix B. In short the 2015 data showed a slight
CO2 merit order. The sustainable scenario shows the clearest CO2 merit.

2.1.5 Energy scenarios model implementation

Figure 2.13 shows the model implementation of the energy scenarios. The APX price
real-time CO2 intensity are loaded to the model via price lookup and CO2 lookup from
Excel. By referring to another column in the lookup, another scenario can be loaded from
the Excel file.

CO2 lookup MW to kW

realtime CO2 realtime CO2 tax and network


<Time>
intensity MW intensity kW costs

price lookup SDE solar SDE Wind

SDE solar SDE wind


<Time> APX price APX year average
compensation compensation

Figure 2.13: Energy scenario model

The tax and network costs are added to the model based on information of the WB
Zuiderzeeland. SDE solar and SDE wind represent subsidies for wind and solar. These
variables are defined here, but will be used later in calculating the energy bill.
28 Input Uncertainties

2.2 Climate scenarios and water system

Eight climate scenarios were used to describe the current and future climate. This section
starts with a scenario consisting of historical measurements. Next the three KNMI 2030
estimate scenarios are introduced, followed by the four KNMI 2050 scenarios. The section
ends with model implementation of the climate scenarios and the water system.

2.2.1 KNMI 2011-2015 measurements and area reduction factor

A historical hourly dataset of the KNMI measuring station number273 in Marknesse for
the period 2011-2015 was downloaded from KNMI (2016). This was used as a reference
scenario for the current climate in the Noordoostpolder (NOP). The data consist of hourly
precipitation and daily potential transpiration (Makking). Table 2.2 compares the 2011-
2015 Marknesse to the 1981-2010 Dutch average. There is some deviation from the Dutch
average, which could be partly due to natural variation and partly due to the spacial
variation between Marknesse and the Dutch average. The 2011-2015 reference scenario
could be characterized as dry on average, but including at least one very extreme ten-day
rainfall sum.

1981-2010 2011-2015 natural variation


Temperature 10.1 ◦C +0.3 ◦C ±0.16 ◦C
precipitation 851mm - 5% ±4.2%
potential evaporation 559mm +5% ±1.9%
10-day rainfall sum 89mm +20% ±11%

Table 2.2: climate overview 2011-2015. Source: Klein Tank et al. (2014)

Since Marknesse is the only measuring station in the NOP this single measuring point is
taken as an estimation for the entire NOP. For light rainfall this estimation is quite accu-
rate, but for intense local showers this can result in an overestimation of the precipitation.
An area reduction factor can be applied to compensate for the fact that rainfall data in
one single point is used to represent the rainfall over a larger area. This area reduction
factor is applied to intense rainfall events above a certain threshold.
Meteobase.nl Wolters et al. (2015) provides a universal area reduction calculation tool.
This tool was used to derive a reduced precipitation dataset. The area was set to 48 km2
equal to the size of the NOP. In the reduced dataset the maximum 24-hour precipitation
was decreased with 11.4%, but the yearly average precipitation was only reduced by 1.7%.

2.2.2 KNMI 2030 scenario

For the future scenarios, the KNMI ’14 climate scenarios Klein Tank et al. (2014) were
used. For the purpose of model testing by water boards these scenarios were transformed
into hourly datasets that can be downloaded from meteobase.nl Wolters et al. (2015).
The Netherlands is divided in precipitation regimes, which are areas that are general (G),
dry (L), wet (H) and very wet (H+). The NOP is situated in the average precipitation
2.2 Climate scenarios and water system 29

regime (G) as can be seen in Figure 2.14, thus the climate scenario for the general regime
was used. Wolters et al. (2015) uses normalized historical data from 1906 to 2014 that
is rescaled to represent a future climate. This means that the dataset has a stationary
climate. Only the first five years of the dataset were used in the model, in order to have
the same length as the 2011-2015 reference case.
The three 2030 scenarios are actually a single base scenario with a lower, center and upper
estimate. Table 2.3 shows the difference between the three estimates and their deviation
from the 1981-2010 Dutch average. The estimates are quite comparable, with only a
significant difference in the 1/10 year 10-day rainfall sum.

1981-2010 2030 lower 2030 center 2030 upper


Temperature 10.1 ◦C +1.0 ◦C +1.0 ◦C +1.0 ◦C
precipitation 851mm +5% +5% +5%
potential evaporation 559mm +2% +2% +2%
10-day rainfall sum 89mm +7% +8% +9%

Table 2.3: climate overview 2030. Source: Klein Tank et al. (2014)

2.2.3 KNMI 2050 scenario

The 2050 scenarios present four scenarios, based on different assumption, thereby making
them more interesting than the 2030 scenarios. The assumptions considered are the global
temperature rise (G ⇒ W), as well as the change in air circulation patterns (L ⇑ H). This
is shown in Figure 2.15.

Figure 2.15: KNMI scenarios. Source:


Figure 2.14: Rainfall regimes. Source:
Klein Tank et al. (2014)
Wolters et al. (2015)
30 Input Uncertainties

All four scenarios also contained a lower, center and upper estimate. As the estimates
were close together only the center estimates were used. The four scenarios GL, GH,
WL and WH and their deviation from the 1981-2010 Dutch average are shown in Table
2.3. The range of variation between the scenarios is quite large. Assuming that climate
change does not reverse, it is also interesting to mention that the 2050 GL scenario is less
extreme than the all three 2030 estimates. The 2030 and 2050 scenarios both contained
a single dataset for potential transpiration.

1981-2010 2050 GL 2050 GH 2050 WL 2050 WH


Temperature 10.1 ◦C +1.0 ◦C +1.4 ◦C +2.0 ◦C +2.3 ◦C
precipitation 851mm +4% +2.5% +5.5% +5%
potential evaporation 559mm +3% +5% +4% +7%
10-day rainfall sum 89mm +6% +10% +12% +17%

Table 2.4: climate overview 2050. Source: Klein Tank et al. (2014)

2.2.4 Climate scenarios and water system model implementation

Water level management is the main function of a pumping station. Changes in the
water level are mainly caused by precipitation, for which scenarios were described it the
previous sections. There is a time delay between precipitation falling and the water level
rising caused by the water system. The water system can be modeled using a hydrological
reservoir model or stocks and flows model. Hydrological models that are often used are
the FLEX hydrological model, see Figure 2.16 and the HBV model, see Figure 2.17. These
models were used as bases for defining the water system model.

Figure 2.16: FLEX model. Source: Figure 2.17: HBV model. Source:
Fenicia et al. (2006) Bergström and Singh
(1995)

The FLEX and HBV model are intended for sloping landscapes, where the natural gra-
dient causes water to discharge into a river. In a flat polder below sea level however a
pumping station is needed to discharge water out of the polder. Evaporation from open
water and transpiration from plants combine to the term evapotranspiration, which is
second outflow term.
2.2 Climate scenarios and water system 31

Apart from precipitation, a polder has two more inflow terms not present in the FLEX
and HBV model. These two terms are inlet of water and seepage. Inlet of water from
outside the polder is done by means of a movable weir or a siphon constructions. This
is done in the dry season form April until September for water quality reasons. The
head difference over the outer polder dike causes seepage. This seepage slowly moves
underneath the dike. 75% of seepage enters into the ditches directly behind the dike and
the other 25% enters into the groundwater.
How the FLEX and HBV models were combined with the polder specific terms can be
seen in Figure 2.18. For simplicity only stocks and flows are shown, but not the auxiliary
variables. The full model diagram there including auxiliary variables can be found in the
interview with Marijke Visser A.13, hydrologist of WB Zuiderzeeland. In the interview
the model set up was verified by Marijke Visser. This was also done by Davı́d Brakkenhoff
A.5 geo-hydrologist at Witteveen+Bos.

water level differential


IJselmeer height
evaporation
surface water

Ditches Drainage Cumulative


Interception canal discharge
rain lookup precipitation surface ditch discharge
runoff runoff
<Time> infiltra inlet
tion
preferential
recharge seep discharge volume
Unsaturated
actual soil ditches
transpiration percol
ation responsible for
evap lookup potential area
<Time> evapotranspiration Groundwater
drainage
seep responsible for
groundwater NOP area
percentage of NOP

groundwater flow

Figure 2.18: Water system model

The climate scenario datasets for precipitation and potential evapotranspiration are en-
tered via respectively the rain lookup and evap lookup. The water level in the polder
is represented by value of the drainage canal stock. Later we will see that the control
can turn on the pumps that will discharge water to the IJsselmeer. The IJsselmeer is as-
sumed to be so large that Vissering has no influence on this water level. Thus, cumulative
discharge and water level IJsselmeer are not linked together.
The differential height depends on the IJsselmeer water level and the drainage canal water
level, as can be seen in 3.4. As there are no tides at the IJsselmeer the differential height
does not change much.
32 Input Uncertainties

The IJsselmeer has a winter level of -0.40 m (NAP) and a summer level between -0.10
and -0.30 m NAP. The maximum allowed drainage canal water level at Vissering is -5.50
m NAP and the minimum allowed water level is -5.85 m NAP. This means the larges
possible differential height is 5.75 m (+5%) and the smallest 5.10 m (-7%). A +5% and
-7% deviation from the average differential height can be assumed relatively small. In the
future, the IJsselmeer level may be raised by 50 cm to adapt to climate change. This the
effect will be investigated later in section 6.2.3 sensitivity analysis for uncertainties.

The difference between discharge and discharge volume is that their units are respectively
mm/hour and m3 /hour. This transformation of units is required to connect the water
model with the pump model. In this transformation includes a multiplication with the
area responsible for.

Vissering is together with the two other pumping stations Buma and Smeenge responsible
for draining the NOP area of 48 km2 . Vissering will thus be responsible for a percentage
of the NOP. Vissering has currently 40% of installed capacity and was responsible for
27% of the total discharge in 2014 Zuiderzeeland (2015). The first number is important
for extreme rainfall and the second for normal operation. Under normal operation Buma
currently has the preference over Vissering, because Vissering is less efficient and has
high fuel costs. After renovation Vissering will most likely be more efficient and have
the preference. Together with Marijke Visser A.13 is was concluded that responsibility of
50% would be reasonable for Vissering after renovation.

This assumption contains some simplifications of reality. Buma and Smeenge and the
middle and high area op the NOP were included in the model. By making Vissering re-
sponsible for 50% of the NOP while it has only 40% of installed capacity we underestimate
the ability of the other pumping stations to help in a heavy rainfall event.

To investigate the effect of this assumption the responsibility will be variated between
30% and 70% in section 6.2.3 sensitivity analysis for uncertainties. Note this percentage
is here seen as an uncertainty, but could also be considered a design choice as the water
board can decide which pumping station has the preference.

2.2.5 Water balance check

To check if the model replicated the NOP well enough there are two checks available. The
water balance check and the representation of the rainfall runoff relation. We start with
the water balance. The water balance for the NOP, originating from the water system
check Visser et al. (2012), is shown in Figure 2.19.

Five different years are shown, including the wet year of 1998 and the dry year of 2003.
There is a whopping difference in total volume of 40% between these years. Every year the
inflow and outflow are equal. A hydrological year is measured form April 1rst to March
31th , because normally around this period the system is saturated with water resulting
in a good reference point.
2.2 Climate scenarios and water system 33

Figure 2.19: NOP water balance

In order to make the PSST model not more complicated than necessary only the inflow
terms with a contribution of at least 2% to the total balance were included in the model.
Note that the term inflow from regional water represents inflow from the high area into
the low area and is actually (delayed) precipitation.

Figure 2.20: Water balance measured and modeled

A comparison of average water balance between measured and PPST model results is
shown in Figure 2.20. The average measured results come from Visser et al. (2012) and
the PPST results are averaged over a 2011 to 2015 simulation. The differences are small
and therefore it can be concluded that the PPST model properly represents the water
balance of the NOP.
34 Input Uncertainties

2.2.6 Rainfall discharge relation calibration

A second way to check if the PSST model represents the NOP is by looking at time delay
between precipitation falling and the runoff in the drainage canal.
In an interview Ronald Nooyen A.11 explained: ”The hydrological model is not only of
importance for the rainfall runoff delay, but works also as a low-pass filter. Sudden peaks
in rainfall will be dampened a bit, and that is satisfying for the control.”
Since runoff is not measured, we have used discharge data of first three months of 2002
Visser et al. (2012) to calibrated the model as shown in 2.21.

Figure 2.21: Rainfall discharge calibration

The PSST model has the same order of magnitude deviation from the measured discharge
as the SOBEK model of WB Zuiderzeeland. Do not pay much attention to the first weeks,
as the PSST model shows start up effects there. The SOBEK model shows a to slow decay
after a rainfall event, while the PSST model shows a too quick decay.
Note also that input rainfall SOBEK for was not exactly the same as the input rainfall
PSST, since the PSST model only used the rainfall data of measuring station Marknesse.
Here the model was calibrated by hand, through changing auxiliary variables, until the
results would closely match the measured data. A future recommendation is to calibrate
the model using optimization algorithms for data fitting. This might not only deliver a
well calibrated model, but might even help improving the existing SOBEK water models
the water boards have.
Chapter 3

Design Choices

35
36 Design Choices

This chapter about the design choices that together form the design alternatives is again
summarized on chapter cover. In the search for a robust design this research actually
tests all possible design alternatives. If that is not computationally feasible, enough the
alternatives will be sampled to properly represent all possibilities. This sampling can
be called a design of experiments. There is thus no need to reduce the set of design
alternatives by means of optimization. This chapter will handle all design choices in the
categories pump, motor, control and innovations.

3.1 Pump choice

In a pumping station renovation is not always necessary to replace the pumps as they
have a longer lifetime than the motors. It might also be possible to replace only the pump
impellers. If the pumps have not reached the end of their lifetime a change in boundary
conditions might still be a reason to replace and adjust them.

3.1.1 Pump selection

If the pumps have to be replaced, a selection for new pumps has to be made. The pump
choice mainly depends on the expected working range of capacity Q (⇒ discharge) and
head H (⇒ differential height). The pump characteristics, which are the relations between
Q, H, pump efficiency ηpump and pump power Ppump , help explaining the behavior of the
pump in this working range. For the different pump types are available in different ranges
capacity and head.
The subsection 1.5.3 about the history of pump types showed that the pump types still
widely applied nowadays are the Archimedean screw pumps in the lifting category and in
the pumping category we find the axial pump, mixed flow pump and centrifugal pump.
Archimedean screw pumps have the advantage of a high efficiency, little wear and
good accessibility for maintenance. This pump type is available in the range from 0.015
to 11 m3 /s Spaans Babcock (2016). Pumping station Kinderdijk, shown in Figure 3.1
has three Archimedean screw pumps of 8.8 m3 /s.

Figure 3.1: Archimedean screw pumps Figure 3.2: Bulb pump placement at
at Kinderdijk. pumping station IJmuiden.
Source: NGS (2016) Source: NGS (2016)
3.1 Pump choice 37

Screw pumps can handle very limited water level fluctuation, on both the in and outside
of the polder. Screw pumps can thus not be applied in a situation where the outside water
level in influenced by tides. There are some tricks to make it possible for an Archimedean
screw pump to handle water level fluctuations, but they are expensive and undermine the
simplicity of the Archimedean screw pump concept.

Figure 3.3: (L ⇒ R) Centrifugal pump, mixed-flow pump and axial pump.


Source: Pumpfundamentals (2016)

Axial pumps have a relatively high efficiency is for low heads situations and can head
fluctuations well, but for large heads the efficiency significantly drops. The vertical axial
pump image shown right in Figure 3.3 is the BVOP pump Bosman (2016) and is available
in the range of 0.2 to 6 m3 /s. Axial pumps are suited for small to midrange capacities
and not too large differential heights. The axial pump is characterized a relatively high
rotational speed resulting in a smaller size, but also higher hydraulic losses due to tur-
bulence. It can be applied in vertical position as well as horizontal. The vertical type
is cheaper and more compact. The horizontal type might be applied when there are no
integration capabilities for the vertical type, because for instance no change to a dike can
be made.
A special variant of the axial type is the horizontal bulb pump where the motor is inte-
grated in the pump and applied under water. This pump is applied for low head situations
where a very high capacity is required. Bulb pumps were also applied in the larges pump-
ing station of Europe, which is pumping station IJmuiden, shown in Figure 3.2. These
bulb pumps have a capacity of 50 m3 /s, that is 3.75 times that of the pumps in Vissering
which are also considered to be very large.
Mixed-flow pumps are a crossover of the centrifugal and the axial pump type. The
mixed-flow pump image shown in the middle of Figure 3.3 is the Beveron pump Bosman
(2016) and is available in the range of 3 to 30 m3 /s. This type has a high efficiency due
to a relatively low rotational speed. The mixed-flow pump characteristics are combining
the characteristics of the centrifugal and axial pump. Where the centrifugal pump has
38 Design Choices

increasing power with capacity and the axial pump decreasing power with capacity, the
mixed-flow pump first gradually increased, then has a maximum power point after which
the power decreases. The largest advantage of this pump type is has maximum efficiency
and maximum power at the same working point.
Centrifugal pumps have the advantage of a high efficiency for high heads. They are
thus mainly applied in high head situations and a varying range of capacities. Their
rotational speed is slightly higher than that of the mixed-flow type. The centrifugal
pump image shown left in Figure 3.3 is the LNN pump Flowserve (2016), which available
in the range of 0.2 to 6 m3 /s. In concrete, capacities can even be larger. A disadvantage of
the centrifugal pump that maximum efficiency point is somewhere halfway the maximum
capacity (check Figure 3.3. There is thus always a trade-off between running efficiently
or at high capacity and power. Centrifugal pumps are also applied in Vissering, where
the pumps are performed in a concrete housing instead of steal.
The vertical centrifugal pumps at Vissering can be characterized as a high head
(5.5 m) and high capacity (13.3 m3 /s) pumps. The foundation and concrete volute pump
housings originate from the opening in 1940. In 1989 the pump impellers were replaced.
The existing pumps are functioning well and there is no need to replace them. A 90-
degree bevel gear transmits the power from the engines to the vertical pump shaft. The
two gas powered pumps have a capacity of 13.3 m3 /s and a third diesel powered one does
12 m3 /s. All three pumps are actually the same, but gas engines rotate at 1500 rpm
reduced by a gear ratio of 12 to 120 rpm, while the diesel engine has a lower rotational
speed of 951 rpm reduced by a gear ratio of 8 to 117 rpm Zuiderzeeland (2015).
If the pumps had to be replaced it can conclude based on the available capacity ranges
that only the bulb, centrifugal, mixed-flow type remain. However, the bulb type is not
possible since it is only applicable for small differential heights. The centrifugal and
mixed-flow pump are thus the only technically feasible pumps for Vissering. Based on
the indicative pump curves in Figure 3.3 it is not possible to make a well-grounded pump
choice for Vissering.

3.1.2 Model implementation of pumps

Figure 3.4 that shows how the pumps were implemented in the PSST model.

<discharge original pump pump efficientcy incl fish


volume> efficiency safety and pump inflow

pump1 capacity
<differential pump inflow
hydraulic power pump2 capacity <fish safe switch>
height> improvement switch
pump3 capacity

pump inflow
specific pumping
number of pumps improvement
energy
investment

Figure 3.4: Pump model


3.1 Pump choice 39

To check if Vissering could also function with just two instead of three pumps, for instance
during maintenance, the design choice number of pumps was added. The capacity after
renovation was assumed to be 13.3 m3 /s for all three pumps, equal to the capacity of the
existing gas powered pumps. If the design choice number of pumps is set to two, then
pump3 capacity is set to zero.
The hydraulic power Phydr describes the amount of potential energy added to the water
over time, by lifting it to a higher level. Phydr is a function of the <discharge volume>,
<differential height> (from the water system model) and the specific pumping energy:
ρwater ∗ g
Phydr = Q ∗ H ∗ = Q ∗ H ∗ 0.002725 [kW ] (3.1)
3.6 ∗ 106
Due to hydraulic losses in the inflow, inside the pump and the outflow and mechanical
losses in the pump and gears the hydraulic power Phydr is lower than to amount of power
that delivered to the pump Ppump , namely:

Phydr Phydr
Ppump = = [kW ] (3.2)
ηpump ηinf low ∗ ηinside pump ∗ ηoutf ow ∗ ηmech

3.1.3 Current pump efficiency and possible improvements

For Vissering neither ηinside pump nor the combined ηpump are exactly known. Having the
pump curves of the existing pumps would have allowed to find ηinside pump , unfortunately
these were not available. Based on a comparison with other (electric) pumping stations
in Zuiderzeeland Zuiderzeeland (2015), an estimation of ηpump of 80% was assumed.
Centrifugal pumps have the option to run at a higher capacity with a lower efficiency, but
it was assumed that the pumps are only operated at the most efficient point. In section
2.2.4 it was concluded that fluctuations in differential height were small (+5% and -7%).
The deviation from 80% can be assumed small, especially since the efficiency curve is flat
near the maximum efficiency point as can be seen in Figure 3.3. These considerations
make it defensible to assume a constant efficiency. Investigating the effect of pump curves
is thereby a recommendation for further research. When the pump curves are available,
they can be added to the PSST model via a lookup table with little effort.
ηinf low starts already in the drainage canal itself. Turbulence in the drainage canal and
pump inflow are the first hydraulic loss. By dredging the canal, or changing the shape of
the inflow this hydraulic loss can be decreased. To determine the optimal inflow shape
computational fluid dynamics analysis can be used. Such a CFD analysis is shown in
Figure 3.5. Another possibility is to improve the inflow by means of a coating, creating a
smoother surface.
Another hydraulic loss that occurs even before water enters the pump is caused by the
trash rack. The trash rack itself has some resistance, but as more trash and floating
plant material is collected on the rack the hydraulic loss becomes larges. A trash rack
cleaner, see Figure 3.6, removes the trash when the hydraulic loss becomes too large.
This hydraulic loss is usually measured by the water level difference between both sides
of the trash rack. The trash rack cleaning will start cleaning when the difference be-
comes more than circa 5 cm. For Vissering with a head of 5.5 m this corresponds to an
40 Design Choices

efficiency loss of 1%. For pumping stations with a smaller head this percentage would
significantly increase. Operating the trash rack cleaner more often can thus save energy.
The cleaner also uses energy itself, so there is off course an optimum cleaning threshold.

Figure 3.5: Pump inflow CFD analysis. Figure 3.6: Trash rack cleaner.
Source: Xylem (2016) Source: Bosker (2013)
Inside the pump itself the shape of the impellers influences ηinside pump . It is an option to
change the impellers of a pump for a newer type to make them more efficient.
At the outflow side ηoutf ow can be influenced by the outflow tube, shut-off valve or one-
way valve, but their contribution to the total hydraulic loss is small. ηmech can be caused
mechanical losses in the pump and by the gears. The gears can be seen as a separate
part, but here gear losses were included in the pump efficiency.
All losses described above together form the pump efficiency ηpump . Improvement option
were mentioned to increase this ηpump . For the pumping stations Buma, Colijn and Lovink
the improvement options were investigated in Tauw (2015). In the end, all improvement
options come down to a pump efficiency increase at a certain investment cost.
For Vissering it is not yet clear what the possible improvement options cost and what
efficiency improvement they will yield. Without a pump curve or measurement this is
also very difficult to determine.
In the Tauw (2015) the pump inflow improvement was one of the most promising im-
provement options with a 1.5% ηpump increase of the pump efficiency original (80%) at
an investment of e30,000 per pump. This option was adopted in the PSST model, as can
be seen in Figure 3.1.
It is not clear if the current pumps at Vissering are fish safe. If not it might be needed
to change the impellers or even the entire pumps, modeled by <fish safe switch>. These
adaptations will result in pump efficiency incl. fish safety and pump inflow. The topic of
fish safety is further explained in Section 4.2.

3.2 Motor choice

The motor power is directly linked on the power of the chosen pump, and is therefore not
in the model as a separate variable. The rotational speed of the motor and pump have
to be matched. As the efficient rotational speed of motor is often higher than the pump
3.2 Motor choice 41

a gear is often applied. We have seen that hydraulic and mechanical losses in the pump
leaded to ηpump . Similarly heat losses in the motor lead to ηmotor .
This motor efficiency is the only motor characteristic that was modeled, as can be seen in
Figure 3.7. Motors do, just like pumps, have a working range in which they can operate
efficiently. For motors this range is much larger than for pumps and thus working ranges
of motors were no focus of this research.
Electrical induction motors have been the standard motor at pumping stations for last
decades. The induction motors at Zuiderzeeland pumping stations Colijn and Lovink
are currently at 92% efficiency. New induction motors can nowadays be purchased up to
95% efficiency. Apart from traditional induction motors and the more efficient permanent
magnet motors.
Permanent magnet motors, which are applied in pumping stations since a few years, even
go up to 96% efficiency. The pump choice between induction and permanent magnet
motor was added to the model, with respectively 95% and 96% efficiency. These numbers
are based on the motor selection for pumping station Verdoold Wauben (2011) and the
interview at Zuiderzeeland Appendix A.8.

<number of
pumps>

motor investment

motor choice
motor efficiency
switch

Figure 3.7: Motor model

The estimated motor investment cost for the induction motors is e2,000,000 per motor.
The more efficient permanent magnet motor is actually cheaper. In Wauben (2011) the
permanent magnet motors were 6.8% cheaper. Zuiderzeeland expects that the permanent
magnet motor can be up to 25% cheaper. To be conservative 6.8% saving when choosing
permanent magnet motors was used in the model.
A motor choice for a permanent magnet instead of induction motors leads to a better
efficiency at lower cost, making it seem obvious to choose for this option. With the
considered model definition, the permanent magnet motor definitely will show better
performance. Still it will be interesting to see how large the impact of this decision is
compared to the other design choices.
Not mentioned yet is that permanent magnet motors are smaller, lighter, have a broader
efficient working range and require less maintenance than induction motors. Based on
these advantages permanent magnet motors were advised in pumping station Verdoold.
But as this motor type was relatively new, the water board was not sure about its long-
term performance. Therefore, the less efficient traditional induction motors were applied.
This shows that a water board is a complex organization, where decision makers might
promote innovation and sustainability, while another department might be more conser-
vative.
42 Design Choices

3.3 Control choice

This section describes how the pumps are controlled. As was assumed that the pumps
only run at the efficient working point, pump control comes down to determining when
the pumps are turned on and off. As there are three pumps there is also the question how
many pumps to turn on at one give time.
In a polder there is a target water level, with some allowable range around this level,
resulting in a minimum and maximum allowed water level. In heavy rainfall events
the task of the pumping station is to prevent exceedance of the maximum water level.
Allowable ranges around the target water level for polders are in the order 10 cm to 50
cm, depending on polder specific factors as soil type and agricultural use. Peat soils for
instance allow small water level ranges as they are prone to land subsidence. At Vissering
the maximum allowed water level is -5.50 m NAP and the minimum allowed water level
is -5.85 m NAP.
The simplest implementation of pump control is by means of on/off water levels. Pump
control can however become very complex by programming a model predictive control
(MPC). In a MPC the polder behavior of the following day is predicted and a control
strategy is optimized. MPC was discussed in the interviews with A.1 Bart Dekens, A.7
Ivo Pothof and A.11 Ronald van Nooyen. MPC is very computational intensive and would
no longer allow us to run the model in numerous future scenarios. Thus, the choice was
Model: D:\OneDrive\Documenten\VENSIM\Pumping Station for the Future\sub systems\control.mdl View: View 1
made to keep the pump control simple by using on/off levels. These on of levels can be
fixed, or can flexibly adapt to the external conditions.

<Drainage canal
water level>

control1
on switch1 off switch1

<TIME STEP>
on level1 off level1

<smart factor>

scenario switch

<dry normal wet


and vey wet>

Figure 3.8: Control model


3.3 Control choice 43

3.3.1 Fixed on/off water levels

Figure 3.8 shows how the control was implemented in the model. The stock variable
control1 can be either 1 or 0, determining if pump1 is on or off. On switch1 is a trigger
that turns the pump on when <Drainage canal water level> reaches on level1. This is
done by a nifty formulation in Vensim. On switch1 namely sends a pulse with magnitude
1 and duration equal to <Time step>, to control1, thereby raising its value by exactly 1.
When the pump1 is on it will only be turned off by off switch1 after the water level
reaches off level1. By control based fixed on/off water levels is meant that on level1 and
off level1 do not change in time. Control2 and control3 were similarly defined for pump2
and pump3, but with higher on/off levels.
At pumping station Vissering the pump control is also based on on/off levels. An example
with on/off level to test the PSST model, has an on level at -5.70 m NAP and an off level
at -5.75 m NAP. This 5 cm difference can be considered a narrow on/off margin, keeping
the water level closely around the target level.
Figure 3.9 shows the above described control would behave for the heavy rainfall event
that occurred on 01-01-2012 shown in Figure 3.10. The time series shown consists of four
days from 30-12-2011 to 02-01-2012. Figure 3.9 shows that the water level exceeds the
maximum water level, despite all three pumps are discharging.
Selected Variables Selected Variables
.0006 m/Hour .007
-5 m
.006

.005
.0003 m/Hour
-5.5 m .004
m/Hour

.003

0 m/Hour .002
-6 m
8718 8730 8742 8754 8766 8778 8790 8802 8814 .001
Time (Hour)
discharge : no scenarios m/Hour
Drainage canal water level : no scenarios m 0
maximal water level : no scenarios m 8718 8730 8742 8754 8766 8778 8790 8802 8814
minimal water level : no scenarios m Time (Hour)
off level1 : no scenarios m
on level1 : no scenarios m ditch runoff : normal precipitation : normal

Figure 3.9: Normal fixed on/off control Figure 3.10: Rainfall event 01-01-2012

3.3.2 Weather scenarios and alarm

In order to prevent exceedance of the maximum allowed water level and to regulate water
quantity WB Zuiderzeeland works with the weather scenarios that change the on and off
level via the scenario switch.
The scenarios dry, normal and wet are based on current groundwater level inside the
polder. In wet periods the target level is lower to increase the water storage capacity,
thereby improving the water safety. The scenario very wet is initiated by a rainfall alarm,
based on the 24-hour rainfall prediction by MeteoGroup (2016). The alarm threshold is
40 mm/day in the dry scenario, 30 mm/day in the normal scenario and 20 mm/day in
the wet scenario. The reason for the different alarm thresholds is that the system can
take up more water when it is dry.
44 Design Choices

The corresponding on/off levels for the weather scenarios are shown in Table 3.1 and
the control behavior in Figure 3.11. Just after time 8754, scenario very wet is initiated
because of a MeteoGroup rainfall alert. Setting both the on and off level to the minimum
level of -5.85 m NAP. In this way, it is possible to pre-pump and prevent an exceedance
of the maximum allowed water level.
Selected Variables
.0006 m/Hour
-5 m

On level ⇒ On level
.0003 m/Hour
Off level ⇒ Of level
-5.5 m
Scenario ⇓ [cm NAP] [cm NAP]
Very dry -570 -575
0 m/Hour
-6 m Normal -575 -580
8718 8730 8742 8754 8766 8778 8790 8802 8814
Time (Hour) Wet -580 -585
discharge : normal m/Hour
Drainage canal water level : normal m
maximal water level : normal m
Very wet -585 -585
minimal water level : normal m
off level1 : normal m
on level1 : normal m
Table 3.1: Weather scenarios setting
Figure 3.11: Weather scenarios control

3.4 Innovations

In this section innovations are presented that can possibly improve the pumping station,
but are not necessary for its functioning. The innovations that are considered are decision
support systems (DSS), local energy supply and aquifer thermal energy storage (ATES).

3.4.1 Decision support systems (DSS)

The control system presented in Section 3.3 is focused only on the main pumping station
function of preventing water level exceedance. Decision support systems (DSS) can be
incorporated into the control, to optimize the control for other performance metrics. An
example of a DSS optimization that is already implemented at many water boards is
optimizing the energy use by including wind set-up or an outside water level that is
influenced by tides. The DSS will suggest pumping when the differential height is small.

On level ⇒ day night


Off level ⇒ day night
Scenario ⇓ [cm NAP] [cm NAP] [cm NAP] [cm NAP]
Very dry -555 -560 -570 -575
Normal -560 -565 -575 -580
Wet -565 -570 -580 -585
Very wet -585 -585 -585 -585

Table 3.2: Day night control setting

Many water boards use a day and night energy tariff. Table 3.2 shows how WB Zuiderzee-
land has implemented different on/off levels for day and night.
3.4 Innovations 45

Figure 3.12 shows the behavior of the day night control. The on/off levels move like a
pulse-train, causing the discharge to be mainly at night. The weather scenarios and alarm
are still in operation, as can be seen at time 8754 when pre-pumping starts.
Selected Variables Selected Variables
.0006 m/Hour .0006 m/Hour
-5 m -5 m

.0003 m/Hour .0003 m/Hour


-5.5 m -5.5 m

0 m/Hour 0 m/Hour
-6 m -6 m
8718 8730 8742 8754 8766 8778 8790 8802 8814 8718 8730 8742 8754 8766 8778 8790 8802 8814
Time (Hour) Time (Hour)
discharge : day night control m/Hour discharge : smart m/Hour
Drainage canal water level : day night control m Drainage canal water level : smart m
maximal water level : day night control m maximal water level : smart m
minimal water level : day night control m minimal water level : smart m
off level1 : day night control m off level1 : smart m
on level1 : day night control m on level1 : smart m

Figure 3.12: Day night control Figure 3.13: APX control

A new idea for DSS energy bill optimization is using hourly fluctuating APX energy prices.
A consortium of STOWA, WB Zuiderzeeland, University of Eindhoven and Deltares is
currently researching this topic in a project called ”Smart Pumping” Zuiderzeeland (2016).
The goal of the project is to establish a more sustainable pump control.

This research tries to extend ”Smart Pumping” imitative of APX based control with the
option to optimize for real-time CO2 intensity and the availability of local energy supply.
In the model, see Figure 3.14, this was done by a smart switch that determines to what
external inputs the smart factor adapts. The external inputs <APX price> and <real-
time CO2 intensity> go through some mathematical operations to transform them to a
proper signal for the smart factor.

Figure 3.13 shows how the control behaves when the smart factor is linked to the APX
price. The pump behavior is quite comparable to that the day night control. The only
notable difference is that the pump duration at the first night is shorter and with two
pumps running instead of one, to profit from the lowest energy prices.

smooting time smart


cut fraction
variable
<APX price>
smart variable smart variable smart variable
smart variable
smooth relative pricing relative pricing cut
<realtime CO2
intensity kW>

smart switch smart control


investment
<solar plus wind
local factor> smart factor
<night off peak>

Figure 3.14: Decision support systems model


46 Design Choices

3.4.2 Local energy supply

In line with the energy transition to renewable energy, water board are striving to generate
their own energy in a sustainable way. Photo-voltaic (PV) solar panels and wind turbines
are proven and widely applied renewable energy technologies. Installed near a pumping
station solar and wind power can supply the pumping station with locally generated
energy. PV panels transform solar energy Qsun into electrical energy Psolar with an
efficiency ηpv of about 15% in the following way:

Psolar = A ∗ Qsun ∗ ηpv [W ] (3.3)

The rated power of a PV panel is the power it produces when Qsun = 1000 W/m2 . In the
Netherlands, the sun does not often reach this intensity. The power a PV panel produces
is thus usually lower than the rated power.
In Zuiderzeeland (2015) the potential of solar rooftops, solar meadows and solar-dikes was
calculated. These measures would make the water board respectively 1%, 10% and 30%
energy neutral. Zuiderzeeland calculated that two solar-dikes would generate 6.6 GWh
per year at an investment cost of e7 to e9 million. Vissering uses about 4 GWh a year,
so a solar-dike is definitely an option to make Vissering energy neutral.

Figure 3.15: Solar-dike impression. Figure 3.16: Solar-dike planning at Buma.


Source: Kaia (2014) Source: Zuiderzeeland (2015)

Wind power involves the harvesting of energy by turbine blades. Energy is extracted
from the mass of air blowing at a speed v through the blade area A. Equation 3.4 (Grogg
(2005)) is used to calculate the power that that the wind turbine will generate:

1
Pwind = Cp ∗ ∗ ρair ∗ A ∗ v 3 [W ] (3.4)
2
Were the power coefficient Cp is determined by the design of the turbine and has a
theoretical maximum of Cp = 0,593 called Betz limit Betz (1920)). However, because of
mechanical and aerodynamic inefficiencies all wind turbines are less efficient than this.
The Cp for commercial wind turbines of typically about 0.45 Beggs (2010).
In fact, the power coefficient also variates with the wind speed. This can be seen in Figure
3.17. Filling in Cp equation 3.4 leads to the calculated power curve.
3.4 Innovations 47

Figure 3.17 shows the power curve for the Enercon E-92 wind turbine. This wind turbine
is with a rotor diameter of 92 m quite large compared to a traditional wind mill or an
older type wind turbine as can be seen in Figure 3.18. The power curve shows that the
E-92 produces a constant 2.35 MW for wind speeds above 15 m/s. This 2.35 MW is the
rated power of this wind turbine.

Figure 3.18: E-92 wind turbine and


Figure 3.17: Power curve of the E-92. older wind mills.
Source: Enercon (2016) Source: Smith (2014)

Figure 3.19 shows the model implementation of local energy supply by wind and solar. By
entering the previously described calculations for the solar power local and wind power
local, we were able to simulate the energy yields. It was found that an installed capacity
of 2.5 MW wind or 4.0 MW solar, was able to make Vissering energy neutral on year basis
(4 GWh/year). Percentage energy neutral and mix solar wind are the design choices we
have to make. When the pumps are on the local power will be directly used, and otherwise
sold to the grid.

Solar power
solar radiation produced
solar radiation solar power local
lookup
<Time>
installed solar
capacity local Solar power used
solar used
<electric power>
percentage energy cost of solar
neutral solar plus wind
investment
cost of wind
mix solar wind

installed wind
capacity local Wind power
U10 onshore power curve local produced
wind power local

U10 onshore <Time> U100 over U10


lookup Wind power used
wind used
<electric power>

Figure 3.19: Local energy supply model


48 Design Choices

Now that the local energy supply is modeled, we can check how it behaves. Figure 3.20
shows the power production over a week for 100% wind, 100% solar and a 50/50 mix.
The first two days are windless with a lot of sun and the last days are cloudy with a lot of
wind. The 50/50 mix however shows a steady energy production on all days of the week.

Figure 3.20: Local energy production over a week

Apart from daily fluctuation in availability one can also expect some seasonality. When
looking at the energy production over the course of a full year, shown in Figure 3.21, a
lot of spikes are observed. For solar these are caused by the days as the sun goes down
every night.

Figure 3.21: Local energy production Figure 3.22: Local energy production
2015 2015 averaged

Figure 3.22 shows the same power productions, but that averaged over a month. It quickly
becomes clear from this graph that the solar production has a sinusoidal shape that is
maximal in summer. Wind production on the other hand is higher in the winter. The
50/50 mix combines both effects and thus has better availability throughout the year.
Nevertheless, around time 42071 there is a period of some weeks with very little sun and
very little wind.

3.4.3 Aquifer thermal energy storage (ATES)

Vissering is currently delivering the rest heat produced by gas engines to companies
nearby via a piping system. This is especially done in winter, when there is a high heat
3.4 Innovations 49

demand. When the pumping station is electrified there will no longer be so much rest
heat available.

Figure 3.23: ATES summer Figure 3.24: ATES winter

In summer, large volumes of warm water pass through a pumping station. If Zuiderzeeland
can extract the thermal energy from surface water, they could keep operating the heat
grid. In order to insulate and conserve the heat for the winter it can be stored in the
ground in an ATES system, as depicted in Figure 3.23. Figure 3.24 depicts how in winter
the heat is be pumped up again and delivered to a building.
The result of extracting energy from surface water in summer, is cooling of the surface
water. This has a positive effect on (urban) water quality and the urban heat island
effect. As Vissering is not located in a city this is not of importance.
WB Zuiderzeeland informed the companies connect to the heat grid about the renovation
and found out that the companies were actually interested in cooling instead. The compa-
nies were mainly fishery related, and needed cooling for large refrigerates. Consequently,
Zuiderzeeland is considering using the ATES system for cooling instead of heating. The
cool water passing the pumping station in winter can be stored in an aquifer thermal
energy storage, and pumped back up for cooling in summer.
To investigate the potential of this concept it makes sense to compare the thermal energy
that can be extracted to the energy needed for pumping. With the specific heat of water,
C the thermal power Ptherm can easily be calculated:
C ∗ Q ∗ ∆T ∗ ρwater
Ptherm = = 1.16 ∗ q ∗ ∆T [kW t] (3.5)
3600
The ratio thermal power that can be extracted over the hydraulic power (Equation 3.1)
needed for pumping can thus be expressed as:
Ptherm 1.16 ∗ ∆T ∆T
= = 426 ∗ [kW t/kW ] (3.6)
Phydr 0.002725 ∗ ∆H ∆H

A reasonable number for ∆T over the heat exchanger is of 3 ◦ C NVOE (2011). This
implicates that potential thermal energy that can be extracted from surface water at
50 Design Choices

pumping station Vissering (∆H = 5.5 m) is 232 times larger than the energy needed for
pumping. In the interview with Barry Scholten of IF Technology (Appendix A.3) the
concept is explained more thoroughly.
WB Zuiderzeeland explained that they see this option as a separate installation from the
pumping station. Zuiderzeeland plans to let a third party operate the installation, but
only if this has no negative impacts on the pumping station. This led to the choice not
to model ATES in the PPST model.
Chapter 4

Performance

51
52 Performance

The cover of this chapter shows the four performance groups and the performance metrics
within these groups. Water quantity and water safety is the main function of a pumping
station and is thus treated first. Next Water quality and ecology is threated as it is
a sub functions the pumping station. Energy performance is largely determining how
sustainable the pumping station is and is thus treated next. The chapter finishes with
the cost performance group.
The cover also shows the six Witteveen+Bos sustainable design criteria. These criteria
were used as a guideline for drafting the performance metrics. The sustainable design
criteria, including the way they guided this research, are listed below:

1. Nature and Climate: Design hand in hand with nature and climate robust.
Reflected in CO2 emission (4.3) and Water quality and ecology (4.2).

2. Chain: Strive to circular economy and assess the impact of the entire chain.
Reflected in the analysis of the CO2 emission (4.3) of the energy supply.

3. Multifunctional: Integrating functions and creating social values.


All three considered Innovations (3.4) are examples that either integrate of create
new functions.

4. Well-being: Physical needs, safety, security, social contact, appreciation and recog-
nition and self-development are aspects of this design process.
Water safety (4.1) gives people a safe and secure feeling. Water quality and ecology
(4.2) determine how people experience the environment they live in.

5. Participation: Include stakeholders in the design criterion.


To capture the view of different stakeholder a lot of interviews were performed, see
Appendix A.

6. Trias-principle: Three-step guide for energy performance.


This principle is used as a main guideline for the Energy use performance group
and explained in Trias Energetica (4.4).

4.1 Water quantity and water safety

The main function of regulating water levels can be needed because of a water surplus
(water safety), or a water demand (water quantity).

4.1.1 Water quantity

Water demand is barely a problem in the NOP as the water board has the possibility to
let in fresh water from outside the polder. The water level can fall below the minimum
water level when in a dry period outflow terms from open water are larger than the sum
of continues inflows to the open water.
4.1 Water quantity and water safety 53

The inflow terms, shown over a five-year period in Figure 4.1, are water inlet in summer
(≈ 0.9 mm/day), groundwater flow (≈ 0.4 mm/day) and seepage into ditches (≈ 0.4
mm/day). These inflow terms add up 1.7 mm/day.
Selected Variables
.06

mm/Hour .048

.036

.024

.012

0
1 8766 17530 26295 35059 43824
Time (Hour)
evaporation surface water : test
inlet : test
groundwater flow : test
seep ditches : test

Figure 4.1: Open water evaporation and steady inflow

The outflow term is open water evaporation, which in summer is 3 to 5 mm/day. One
should however not forget that the inflow terms are expressed over the entire polder, while
the open water evaporation only occurs over the wet area percentage. Taking that into
account it can be seen from Figure 4.1 that the evaporation from surface water is much
smaller than the steady inflow terms. From Figure 4.1 one would conclude that water
level never drops below the minimum water level.
However, one (internal) flow that also impacts the water level, namely extraction for
irrigation, was not included in the model. The NOP namely has a lot of agricultural area,
and especially in dry periods, water is extracted from open water to irrigate the land.
This water might then leave the polder via transpiration in plants. The purpose of water
inlet also is to compensate for extractions from open water.
Nevertheless, the point remains that water demand is not causing a problem in the NOP,
as there is possibility to let in water.

4.1.2 Water safety

During heavy rainfall events the task of the pumping station is to prevent exceedance of
the maximum water level. At Vissering the maximum allowed water level is -5.50 m NAP.

direct flooding indirect


cost ratio flooding cost

Cumulative
water level flooding cost direct flooding cost
<Drainage canal water level exceedance
water level> exceedance

<maximal water cost of flooding


level>

Figure 4.2: Water quantity and safety model


54 Performance

Figure 4.2 shows that water level exceedance [mm] can be integrated over time to the
cumulative water level exceedance [mm*hour] and can be linked to flooding cost. The
cost of flooding was estimated at 5 e/(mm*Hour).
These flooding costs might be direct when the water board pays to compensate for the
damage. Here a 50/50 ratio between direct and indirect costs was assumed. The flood-
ing costs estimate may be significantly improved using waterschadeschatter.nl (2014).
Additional information of the water board is required to do this calculation.

4.2 Water quality and ecology

The dike isolates the polder from the surrounding waters. As a consequence, water quality
and ecology inside and outside the polder develop independently. The pumping stations
present one of the view connections where exchange of water and fish between the polder
and surrounding waters can take place.

4.2.1 Water quality

The water quality in the polder can be influenced by chemical substances, fertilizers and
pesticides. Water inlet from outside the polder can also cause problems for water quality
as the inlet of extra nutrients can cause excessive algae growth. As dry periods, when
water inlet happens, are also warm periods this algae growth problem is reinforced.
Precipitation can both cool down and flush water through the system with fresh water,
easing the water quality problems. When climate change leads to more extreme dry
periods this may become a serious problem for which no solutions are yet developed
Zuiderzeeland (2010). As water quality is a whole different dimension as water quantity
for now it was left out of the model.

4.2.2 Ecology (fish safety)

Fish are the only species known to be impacted by pumping stations. Fish safety has
been a topic of discussion over the past decade. Research by Kunst et al. (2010) showed
that a lot of pumping stations were not fish safe. Thus, pump manufactures have brought
several fish safe pumps to the market. Fish safety can be split up into the components:

1. Passing factor: measure for fish migration through the pumping station.

2. Damage factor: measure for fish getting damaged or not surviving an attempt to
pass the pumping station.

3. Fish safety factor = passing factor * damage factor.

The passing factor and damage factor can be combined to the fish safety factor. In the
model this combined fish safety factor was used, as can be seen in Figure 4.3.
4.2 Water quality and ecology 55

pump efficiency <number of fish safety


loss pumps> investment

fish safe switch Cumulative fish damage cost


fish damage
fish damage

potential fish cost of fish


fish safety factor weight per fish
migration rate damage

Figure 4.3: Water quality and ecology model

The fish safety factor multiplied by the potential fish migration passing factor leads to
fish damage or fish not surviving an attempt to pass the pumping station.
To determine if fish safety measures are needed at a certain pumping station the fish
safety factor of the existing pumps have to be determined. The (indicative) influence
of pump specifications on the fish safety factor are shown in Table 4.1. The research of
Esch and Rommens (2010) developed a fish collision model to calculate the damage factor
based on pump specifications. This model indicates the chance that a fish hits the pump
impellers. Esch and Rommens (2010) concluded that axial type pumps generally result
in high fish damage, other pump also cause types fish damage, but for larger pump fish
damage is small.

Pump Impeller Differential Rotational Operating


capacity diameter height speed hours
Fish unsafe low small high high large
⇓ ⇓ ⇓ ⇓ ⇓
Fish safe high large low low small

Table 4.1: Parameters influencing fish safety. Source: Kunst et al. (2010)

With the collision model and with measurement the fish safety of a pumping station can
be determined. If fish safety is not up to standard, measures are needed. There are several
measures to improve fish safety. These measures will induce investment costs and might
have impact on the pump efficiency.
Due to a lack of time and resources a full fish safety assessment for the Vissering was
not included. Instead the fish safety assessment of pumping station Krimpenerwaard
Nieuwkamer and Klinge (2016) was adopted as a showcase for the model, but that is
not representative for Vissering. The two most promising measures in this case were
changing the pump impeller and replacing the entire pump by new fish safe pumps. Both
measures, along with doing nothing, were including in the fish safe switch in the model.
The considered measures cause a pump efficiency loss of respectively 2% and 4%.
How this relates to Vissering, and if fish safe pump are always less efficient, is unknown.
The theory of Esch and Rommens (2010) suggest that Vissering is most likely already
56 Performance

quite fish safe, as it has very big pumps. Fish measurements at pumping station Colijn
of Zuiderzeeland, with similar pumps, also showed little fish damage.
To make the model case specific for Vissering the fish safety factor of the current impellers
could be determined with the collision model of Esch and Rommens (2010). This should
be backed by local fish measurement at Vissering, that can also help determining the
potential fish migration rate. In case it is concluded that Vissering is not fish safe,
improvement measures will have to be determined and added to the model.
Nieuwkamer and Klinge (2016) judged the measures by comparing the investment costs
and higher energy use to the prevented fish damage cost, or rather missed catches due to
fish mortality. This was done by including a cost of fish damage that was based on the
market price of fish (e12/kg). It is also good to note that fish safety is not just ”nice to
have”, but is regulated by law. The legal framework for fish safety consist of Ministers
van de Benelux (1996), Water Framework Directive (2000) and EU Regulation (2007).

4.3 Energy performance

The climate change problem has caused sustainability to become a more important topic
for water boards. This year the Union of Water Boards (UvW) signed a Green Deal
Energy UvW et al. (2016) committing to 40% self-generated renewable energy supply by
2020 and the ambition to ultimately become energy neutral.

4.3.1 Trias Energetica

When dealing with energy related problems Trias Energetica RVO (2015) is an often
applied principle. It is also one of the Witteveen+Bos sustainable design criteria. The
principle is summarized in Figure 4.4. Water board Zuiderzeeland is currently investi-
gating how to reduce energy use and increase local sustainable generation in the project
Energy and Pumping stations Zuiderzeeland (2015).

Figure 4.4: Trias Energetica. Source: RVO (2015)


4.3 Energy performance 57

Zuiderzeeland interpreted the Trias principle in the following way:

1. Minimize the demand for energy: Less and more efficient pumping.
2a. Use energy from waste streams: Producing biogas from sewage sludge.
2b. Use sustainable energy: Self-generated sustainable energy.
3. If necessary use fossil fuels as clean and energy efficient as possible:
Electrification of all motors and purchase of sustainable energy.

Zuiderzeeland has consequently set the following goals in line with the Trias principle:

1. Total energy use: 30% reduction in 2020 compared to 2005.


2. Self-generated share of sustainable energy: 35% to 45% in 2035.
3. CO2 emitted: 30% reduction in 2020 compared to 1990.

4.3.2 Energy performance model implementation

Energy bill
cash flow

<tax and network


costs>
Energy
purchased
<APX price> cash out

<solar plus wind


local used>

<SDE solar
<solar power compensation> Energy sold
local> cash in
<wind power
<SDE wind
local>
compensation>
<solar plus wind
local>
<motor efficiency> social cost per ton
Energy used
total efficiency CO2
<pump efficientcy incl electric power
fish safety measure> <hydraulic power>

<realtime CO2 averaged CO2


intensity kW> intensity
Social cost of
CO2 emitted
CO2 switch CO2
CO2 intensity CO2 emmison

<local energy used Local energy


directly> used

Figure 4.5: Energy performance model


58 Performance

The energy performance model, as can be seen in Figure 4.5. The model contains four
performance metrics that are shown in the circles. Three of them represent the Trias
Energetica and the fourth one is Energy bill. Minimizing the Energy used is the first
Trias principle. This can be achieved by either improving total efficiency, or by reducing
required <hydraulic power>. The latter is difficult to achieve as it requires pumping less
or reducing the differential height, which are both no feasible options. Minimizing energy
use mainly comes down to improving the motor efficiency or the pump efficiency.
Currently the total efficiency of Vissering is 31% Zuiderzeeland (2015). The most efficient
pumping station of Zuiderzeeland is currently Colijn with 79%. So, there is a lot of
room for improvement, mainly due to the heat loss of the diesel and gas engines. The
current situation is estimated to have an 80% pump efficiency and a 39% motor efficiency.
Changing to electrical motors improves the motor efficiency to at least 95%, and thereby
the total efficiency will at least be 76%.
The use of sustainable energy is the next Trias principle. Zuiderzeeland is currently
buying Guaranty of Origin (GvO) certificates of Scandinavian water power, so that the
energy supply can be called sustainable. For imported GvO certificates from abroad it is
known that they do not contribute to making the Dutch energy supply more sustainable,
see co2emissiefactoren.nl (2016) and Appendix A.15.
Zuiderzeeland is recognizing this fact, and will consequently start producing sustainable
energy themselves. In section 3.4.2 it was explained that is possible to make Vissering
energy neutral with wind and solar. Ideally one wants to use the locally generated energy
directly at the pumping station. This is reflected in the Local energy used performance.
The last Trias principle is about the use of fossil fuels and corresponding CO2 emissions.
The main CO2 emission of Vissering is currently directly produced by use of fossil fuels
in the gas and diesel motors. When the motors are electrified there are no longer direct
CO2 emissions. Instead there are indirect CO2 emissions related to the electricity sup-
ply. It was already concluded that the traditional way of reducing the CO2 intensity by
purchasing GvO certificates is not desirable.
Then two alternatives remain to reduce the CO2 intensity. The first one is buying elec-
tricity when the national real-time CO2 intensity is low. The second one is maximizing
the Local energy used, as wind and solar have a near zero CO2 emission.
A way of expressing the impact of CO2 emissions is by looking at the social cost and
climate damage caused by the CO2 emission. Comparing the climate damage caused to
for instance the energy bill gives an insight to decision makers how large the impact of
CO2 emissions is. There is a large difference in the estimations for what the social cost
of a ton CO2 is, as can be seen in Table 4.2.

Cost per ton CO2 Party Citation


$ 37 (≈ e30) US Whitehouse Shelanski (2013)
$ 83 (≈ e70) UK Government Bell and Callan (2011)
e90 to e100 Dutch energy experts Klimaatplein (2015)
$ 220 (≈ e200) Stanford University Moore and Diaz (2015)

Table 4.2: Estimations for social cost of a ton CO2


4.4 Costs 59

Note however that all mentioned prices are far higher than the current price of CO2 of
about e5 per ton of the Emission Trading System. In this research a price of e100 per
ton was used, in line with the Dutch energy experts. With this price and an average
CO2 intensity of 0.355 ton/MWh co2emissiefactoren.nl (2016) the CO2 impact of a ton
is e35.50 per MWh. Note that this in the same order of magnitude as the 2015 average
electricity price of e40.
The Energy bill has the largest share in the operational costs of a pumping station. The
energy scenarios range from e24 to e74 per MWh, so this could result in quite some
impact on the energy bill. If the water boards start purchasing their energy on the APX
market, the energy price is no longer constant over time and thus the energy related cash
flow will not be linear with the electric power.
There is not only energy purchased, but also energy sold to the APX market when there
is local production that cannot be used by the pumping station. SDE subsidy comes in
for every self-generated MWh of wind and solar. The SDE for solar is slightly higher. No
tax and network costs have to be payed over energy that is self-generated and used in the
pumping station. This is one more reason for maximizing the use of local energy.

4.4 Costs

A pumping station renovation requires some investment costs. The idea is that the
components are improved thereby reducing the operational costs that grow over time.
Apart from the direct operational costs directly payed by water board there are also
indirect costs caused by the pumping station.

<pump inflow
improvement
investment>

<motor
investment>

<smart control Investment costs


investment>

<solar plus wind


investment>

<fish safety
investment> Operational
Operational costs costs incl
indirect costs

<direct flooding <indirect flooding <fish damage <Social cost of


<Energy bill>
cost> cost> cost> CO2>

Figure 4.6: Cost model


60 Performance

4.4.1 Investment costs

In pumping station re-design all parts that are replaced or renovated have an initial
investment cost. In case of hardware parts those are mainly the purchase of the parts,
but also some engineering and building costs. In case of the smart control these are mainly
software development costs.
All parts have a certain physical lifetime, when the part breaks down and an economic
lifetime, when it becomes cheaper to replace the part. Cheaper in that sense means that a
new part can reduce the operational costs by so much that it can earn back the investment
costs within its own lifetime.
After the lifetime of a part a new investment is required to replace it. This means a part
only has to be robust for future changes within its own lifetime.

4.4.2 Operational costs

For a pumping station the main operational cost is the energy bill. If there is a lot of
water level exceedance the direct flooding costs will also become important.
All parts require maintenance or need to be handled by an operator. It was mentioned
that permanent magnet motors are less maintenance intensive than induction motors.
Investing in a more expensive part that requires little maintenance might pay off in the
long run. In this research the effect of maintenance was not quantified, but in a more
detailed engineering phase they should not be overlooked.

4.4.3 Operational costs including indirect costs

It is also interesting to look at indirect costs, such as the social cost of CO2, damage
costs from crop loss due to water level exceedance and fish damage. The flooding costs
are direct when the water board pays to compensate for the damage. When damages
are not claimed, they are indirect cost. The water board does not pay for indirect costs,
but nevertheless they should be minimized as, the pumping station should not impact its
surroundings more than necessary.
Chapter 5

PSST Model

Maintenance
and operational
costs

control

energy scenarios Energy use


and energy supply

motors

climate scenarios Water quantity


and water system and safety

pumps

Water quality
and ecology

61
62 PSST Model

In this chapter the model description will be finalized by adding relationships that link
the previously described elements together in the PSST model. Next will be explained
how the model can be simulated manually in Vensim (one simulation).

5.1 Relationships in System

Relationships in System (R) ”describe the ways in which the factors relate to one another
and so govern how the future may evolve over time based on the decision-makers’ choices
of levers and the manifestation of the external factors.” Lempert et al. (2003). The cover
of this chapter shows these relationships between input scenarios, design choices and
performance metrics. Together this forms the complete PSST model.

5.2 Simulating the model in Vensim (manual)

This section describes the Vensim settings with which the model was simulated. For the
integration type Euler was chosen. The more advanced Runge-Kutta integration method
caused errors in Vensim, so the simulation would not be finished. Runge-Kutta Stewart
(2010) uses a weighted average at four points near the current point. The errors with
Runge-Kutta are most likely caused by model values abruptly changing, like a pump being
switched on and off. The Euler integration method Stewart (2010) used only one single
point, and can thus better handle abrupt changes.
Hours were chosen as the units of time. The input datasets were also given in hours.
Next the calculation time-step had to be chosen. The larger the time-step the faster the
simulation, which is desirable. A large time-step will however introduce numerical errors.
The Courant-Friedrichs-Lewy (CFL) Courant et al. (1928) condition is normally used in
modeling to estimate how small the time-step ∆t should be for convergence. The CFL
condition looks as follows:
u ∗ ∆t
C= ≤ Cmax [−] (5.1)
∆x

A SD model is however non-spatial, thus the term ∆x had no meaning in a SD model.


Alternatively SD Society (2016) advised to use a time-step equal to one-eighth of the
smallest time constant in the model. The smallest time constant in the model was the
runoff time, which is equal to one hour. Input data is also entered on an hourly scale.
This suggest that a time-step of one-eighth (0.125) of an hour is required.
To be sure if the resulting numerical error was tested for seven time-steps, subsequently
halving in length. This was done for four performance metrics, as shown in Table 5.1. The
previously suggested time-step of 0.125 hour still caused an error of circa 2% at flooding
costs. A time-step of 0.03125 hours (112.5 seconds) seemed to be a better option with an
acceptable numerical error of 0.2%. The reason that the flooding costs are causing the
largest numerical errors is probably caused by the fact that water level exceedance is the
quickest process in the model.
5.3 Simulating the model via Python (automated) 63

Numerical error on performance metric:


Time step [hour] Energy bill flooding costs local energy used CO2 emitted
0.015625 0.0% 0.0% 0.0% 0.0%
0.03125 0.1% 0.1% 0.2% 0.0%
0.0625 0.1% 0.5% 1.1% 0.0%
0.125 0.4% 2.2% 1.7% 0.1%
0.25 0.8% 13.6% 3.7% 0.2%
0.5 1.3% 8.9% 7.8% 0.3%
1 1.5% 3.1% 15.6% 0.5%

Table 5.1: Numerical error due to time-step

Next is was tested if the suggested time-step of 0.03125 had an acceptable simulation
time. Remember that a single simulation will cover 5 years, or 43824 hours and we want
to do a lot of these simulations.
The simulation time for the seven different time-steps was measured with a stopwatch.
The subsequent time-steps were halving in size, and thereby doubling the number of steps.
This suggested that the simulation time would double as the time-step was halved. At
first the measurement did not seem to double. After taking into account a fixed time for
loading the initial data, a doubling simulation time was found. The time-step of 0.03125
had a loading+simulation time of 15 seconds, which was maybe not as fast as we hoped,
but nevertheless acceptable.

Simulation run time [s]


Time step [hour] measured loading simulating load+sim
0.015625 28 2.5 25.6 28.1
0.03125 15 2.5 12.8 15.3
0.0625 8 2.5 5.6 8.1
0.125 5 2.5 3.2 5.7
0.25 4 2.5 1.6 4.1
0.5 3.5 2.5 0.8 3.3
1 3 2.5 0.4 2.9

Table 5.2: Simulation time for different time-steps

5.3 Simulating the model via Python (automated)

In Vensim manual simulation can be performed with the PSST model. That is, by clicking
on the simulate button and waiting 15 seconds a single simulation of one design alternative
in one state of the world will be executed. Simulating numerous design alternatives
in different SOW required many simulations this was not possible within the Vensim
software.
64 PSST Model

Python was used therefore to simulate and analyze the model. Python executes simula-
tions by loading the ingoing input uncertainties and design choices to the PSST model,
then Vensim runs a simulation and Python extracts and stores the outgoing performance.
In analogy in the XLRM simulation framework shown in Figure 5.1 this means that
Vensim is the PSST model box, but Python regulates the arrows.

Figure 5.1: XLRM simulation framework

Most of the Python scripts were already available from the Exploratory Modeling and
Analysis (EMA) Workbench Kwakkel (2016). EMA is a research methodology that uses
computational experiments to analyze complex and uncertain systems Bankes (1993).
That is, exploratory modeling aims at offering computational decision support for decision
making under deep uncertainty and Robust decision making Kwakkel (2016).
The workbench provided a standardized format to perform certain simulation and analysis
algorithms for any Vensim model. The uncertainties and design choices related to the
PSST model off course had to be added to the script. The Python simulation code is
found in Appendix C.
The total simulation time could be reduced by running parallel simulations the calculation
cluster at the TU Delft with 24 parallel cores. There are 24000 simulation to be performed,
as we will see in the next chapter. Consider the Vensim simulation time of 15 seconds,
then the total simulation time will be:

Simulating time = 15 ∗ 24000/24 = 15000 [s] = 4.2 [hour] (5.2)


Chapter 6

Insight via learning

65
66 Insight via learning

In the previous chapters, we have seen that the PSST model was built in Vensim. We
have seen that we were able manually simulate the model and successfully calibrate it for
pumping station Vissering in the NOP. Thereby the first goal of this research is completed.
The second goal of this research was to gain insights in the performance of pumping
station designs choices under uncertainty characterized by the input scenarios for climate
and energy. A large set of simulations is used to gain insight via learning. The term
learning indicates that it is iterative loop that will repeated a couple of times to come to
a robust design.

6.1 Expressing robustness

First the robustness metric, that quantifies robustness, is defined. Next it will be explained
why we test robustness multiple objectives instead of a single objective.

6.1.1 Robustness metric

A robustness metric determines how a definition of robustness is operationalized. In the


Introduction 1.4.2 two definitions of robustness were introduced:

• Reducing the uncertainty about the expected consequences of a given policy. So, no
matter how the future plays out, the policy performance falls in a narrow bandwidth.

• Minimizing the undesirable outcomes. So, no matter how the future unfolds, policy
performance will be satisfactory.

Kwakkel et al. (2016) distinguishes tree families of robustness metrics:

• The regret based metrics try to minimize regret by comparing a result to a


baseline or best case result.

• The satisficing metrics maximize the number of future scenarios for which a
design meets a performance threshold. Satisficing is a combination of words satisfy
and suffice.

• The statistical metrics look at the distribution of the performance metrics.

In this research the focus is on the statistical metrics, as it is easy to visually asses
them from a graph. In Kwakkel et al. (2016) five statistical based robustness metrics
are presented. The best metric expressed robustness as a combination of a good average
performance (µ) and a narrow bandwidth (σ). This metric fits well our first robustness
definition. The mathematical expression of this metric, that should be minimized to
increase robustness, looks like:

fi (L) = [µi , σi ] (f or minimization objectives) (6.1)

fi (L) = [−µi , σi ] (f or maximizing objectives) (6.2)


6.1 Expressing robustness 67

Figure 6.1 gives a visual explanation of the robustness metric, for the performance metric
local energy used, which is a objective to be maximized. From equation 6.2 then follows a
robust solution has a small −µ, which is the same as high µ, and small σ. Consequently,
in Figure 6.1 local pump control is more robust than APX pump control.

Figure 6.1: Visual explanation of the robustness metric

A disadvantage of this metric is that it looks at the standard deviation to both sides,
while the water board might only be interested in the undesirable deviations away from
the objective direction. The water board might for instance not care about some future
scenarios leading to a lower energy bill, but they would care if scenarios lead to a much
higher energy bill. To take that into account, the skewness of the performance distribution
must be assessed. In Figure 6.1 local control shows a bell curve, while APX shows an
irregular distribution, but no clear skewness can be recognized.
The regret based metric can also be operationalized from Figure 6.1, as can for instance
be said that normal control deviates 0.8/1.2 from the best case, which is local control.

6.1.2 Multi-objective VS single-objective

Traditional design strategies for hydraulic engineering problems tend to aggregate all
performance metrics into one single composite objective. Weighting factors to reduce all
performance metrics to a single metric, used in for instance multiple criteria analysis and
costbenefit analysis, are an example of that.
The Arrow’s Paradox Arrow (1950) implies that aggregate objective functions are inade-
quately able to optimize a design. Kasprzyk et al. (2016) showed that a multi-objective
function, that explicitly disaggregated the different performance metrics, improved the
ability to find diverse trade-offs and innovative solutions. To test if Kasprzyk et al.
(2016) is right we will test both a single- and a multi-objective optimization function.
For the single-objective function it is most convenient to make a cost function that ex-
presses all performance metrics into euro and adds them together. In section 4.4 the
68 Insight via learning

investment and operational cost were already introduced. A single-objective function


could be the investment cost plus operational costs times the lifetime:

minimize Fcost (L) = [finvest + lif etime ∗ foperational ] (6.3)

A multi-objective optimization function could look like:

minimize Fmulti (L) = [finvest , fenergy used , −flocal energy , fCO2 , ff ish damage ] (6.4)

Note that in 6.4 the objectives are comma separated, indication they are assessed sepa-
rately. The minus sign at local energy indicates that is a maximizing objective.

6.2 Sensitivity analysis and Finding trade-offs

Sensitivity analysis with feature scoring was used to rank the sensitivity of the perfor-
mance metrics for the design choices and future scenarios. This analysis searches for
design choices that created a trade-off between performance metrics.

6.2.1 First insight loop - 500 designs in 48 SOW

Section 5.3 explained how to set up an experiment of multiple simulations in Python


using the EMA Workbench Kwakkel (2016). This section explains an experiment that
was performed and shows how sensitivity analysis was used to find trade-offs.
In chapter 2 Input uncertainties six energy scenarios and eight climate scenarios were
presented. This leads to the following number of possible future states of the world:

N r.SOW = 6 ∗ 8 = 48 [−] (6.5)

Out of the eight considered design choices only mix solar wind is of the Real type, meaning
every value between 0% and 100% was possible. The other seven design choices are of
the Categorical type, meaning they have a limited number of possibilities. Three choices
are binary, while the fish safe solution has three options and the pump control has six
options. These Categorical choices lead to the following possible design alternatives:

N r. design alternatives = 25 ∗ 3 ∗ 6 = 576 [−] (6.6)

The remaining choice of mix solar wind has infinite options, making the total number of
design alternatives also infinite. To get a sense of the possible outcomes Latin Hyper-
cube sampling was used to choose 500 design alternatives to simulate. Latin Hypercube
sampling is comparable to Monte Carlo sampling, only Latin Hypercube is not random
but structured. For the Categorical type, it will structurally sample all the possibili-
ties, and thus never two times the same design. For the Real type (mix solar wind) it
first divides the range in ten equal parts of and then samples randomly from each tenth,
thereby guaranteeing better sampling coverage over the full range. The 500 sampled
design alternatives lead to the following number of simulations:

N r. simulations = 48 ∗ 500 = 24000 [−] (6.7)


6.2 Sensitivity analysis and Finding trade-offs 69

Storing the performance for all time steps, in a five-year simulation, required too much
storage capacity. Since this research only looks at performance metrics that (slowly)
accumulate, storing only the end values sufficed.
To learn from the results, the sensitivity of the performance metrics for the design choices
and future scenarios, is interesting. This can be investigated with sensitivity analysis
with the ranking type. The algorithm used, is called feature scoring with extra trees
Kwakkel (2016). This algorithm checks per factor (design choice or scenario) how large
the influences on the performance is. Feature scoring makes a ranking of the most sensitive
factors, for a certain performance metric. The feature scoring rankings, for water level
exceedance and fish damage, are shown in respectively Table 6.1 and 6.2.

Rank Factor Sensitivity Rank Factor Sensitivity


1 number of pumps 32,6% 1 fish safe solution 100,0%
2 weather alarm 28,5% 2 Energy scenario 0,0%
3 Climate scenario 23,5% 3 Climate scenario 0,0%
4 pump control 13,2% 4 mix solar wind 0,0%
5 Energy scenario 0,4% 5 pump control 0,0%
6 % energy neutral 0,4% 6 % energy neutral 0,0%
7 mix solar wind 0,2% 7 pump inflow 0,0%
8 pump inflow 0,1% 8 motor choice 0,0%
9 fish safe solution 0,0% 9 number of pumps 0,0%
10 motor choice 0,0% 10 weather alarm 0,0%

Table 6.1: Feature scoring for Table 6.2: Feature scoring for local
water level exceedance energy used fish damage

Figure 6.2: First loop feature scoring results


70 Insight via learning

A similar feature scoring was performed for all nine performance metrics. Instead of
showing all nine tables a more effective way of presenting the feature scoring results is in
a parallel coordinate plot, shown in Figure 6.2. The first striking thing is that there are
two dominant design choices. The percentage energy neutral is dominating the energy
performance and operational costs.

The fish safe solution only factor influencing fish damage, as already indicated by
Table 6.2 by the fact that it scored 100% sensitivity. When looking better at the fish safe
solution, it can be seen that the not only dominates fish damage, but also impacts energy
use and investment cost. This indicates that there might be a trade-off created by the
fish safe solution. Insight into this potential trade-off can be gained by looking at the
spread of performance outcomes in a scatter plot, as shown in Figure 6.3.

Figure 6.3: Fish damage, Energy used and Investment grouped by Fish safe solution
6.2 Sensitivity analysis and Finding trade-offs 71

The left top shows no kernel density, as there are just three possible results, rather than a
spread in results. The original impellers have a very high fish damage, while both change
impeller and new fish safe pumps can reduce this significantly.
The investment costs on the right show that the new fish pumps have very high investment
cost compared to doing nothing and only changing the impellers. The change of impellers
and new fish safe pump also decrease the pumps efficiency by respectively 2% and 4%,
which is reflected in the energy use shown in the middle.
From these results one would conclude that fish safe switch creates a trade-off between
costs (investment), sustainability (energy use) and ecology (fish). The change impellers
option is the near best solution for these three performance metrics. It can be concluded
that the change of impellers provides the most balanced trade-off.
Note that this conclusion uses the regret based robustness metric, since we tried to min-
imize regret by comparing a result to best case result.
Keep in mind that the fish safety design choices were not parametrized specifically for
Vissering, but they represent the situation at the pumping station Krimpenerwaard
Nieuwkamer and Klinge (2016). Thereby the conclusion about the fish safe solution
can NOT be linked to Vissering. The analysis however has shown to provide a proper
framework to a robust fish safe solution. The framework enables a fish safety assessment
for Vissering once more information about is known.
Another startling observation is that there appeared to be designs with near zero invest-
ment cost, while all design would at least include two electrical motors. This finding
was investigated, and it was found that there was a mistake in the way the price of the
permanent magnet motor was defined in the PSST model. It was namely programmed as
6.8% of the induction motor price, instead of 6.8% cheaper. As the investment costs are
a simple sum of different parts, and have no further influence for the rest of the model,
we decided not to do simulations all over again.
72 Insight via learning

The most important performance group for the water board is water quantity & water
safety. Water quantity is related to the cumulative discharge and consequently the energy
use, while the water safety is linked to water level exceedance. The feature score graph
in Figure 6.2 shows that the climate scenarios are important for both energy use and
water level exceedance. Figure 6.4 investigates this.

Figure 6.4: Water level exceedance and Energy used grouped by Climate scenario

In section A.7 we learned that the total precipitation was the most extreme in 2050 WL
+5.5%, followed by 2050 WH and the 2030 scenarios all +5%. The total precipitation of
2011-2015 was much lower with -20%, clearly seen in energy used. The somewhat trivial
conclusion is that energy use is a directly linked to total precipitation.
Water level exceedance is caused by extreme rainfall events. In section A.7 we learned
that the 10-day extreme rainfall (1/10 year) was most extreme in 2011-2015 with +20%
followed by 2050 WH +17% and 2050 WL +12%. Figure 6.4 shown that 2050 WH and
2050 WL indeed cause the highest water level exceedance. 2011-2015 however shows low
water level exceedance, although it had the 10-day extreme rainfall. This suggest that the
highest 10-day extreme rainfall, is maybe not the right measure for water level exceedance,
and we should for instance focus on the 24-hour extreme rainfall.
6.2 Sensitivity analysis and Finding trade-offs 73

The ranking in Figure 6.2 shows that for water level exceedance the number of pumps
(Figure 6.5) and the weather alarm (Figure 6.6) are even a bit more influential than
the climate scenario. When number of pumps is 3, there is no water level exceedance.
This does not mean that having just 2 pumps always leads to high water level exceedance,
because the weather scenarios (alarm) can significantly reduce the water level exceedance
to no more than 50,000 mm/hour in five years. With a water level exceedance cost of e5
that comes down to e250,000 in five-year period.

Figure 6.5: Water level exceedance and Figure 6.6: Water level exceedance and
Energy used grouped by Energy used grouped by
Number of pumps Weather alarm

To get a feeling for the number of operating hours of the pumps, a manual simulation in
Vensim was done with the weather alarm on. Table 6.3 shows the operating hours for 3
pumps, resulting in zero water level exceedance. Table 6.4 shows the operating hours for
2 pumps, resulting in 17,000 mm/hour water level exceedance.

Pump Operating % of year


Pump Operating % of year
Pump 1 3627 hours 41.4%
Pump 1 3643 hours 41.6%
Pump 2 483 hours 5.5%
Pump 2 580 hours 6.6%
Pump 3 113 hours 1.3%
Table 6.4: Operating hours for
Table 6.3: Operating hours for
number of pumps is 2
number of pumps is 3
The numbers show that two pumps can suffice. A third pump, that is used circa 113
hours a year can completely prevent water level exceedance. Thus, we suggest to have
three pumps, but not to place a third electrical motor for only 113 hours a year.
Instead we advice to keep one of the gas engines. There is no doubt that the weather
scenarios and alarm should remain, as it prevents water level exceedance.
74 Insight via learning

After climate, the motor choice is the second design choice impacting the energy use.
Figure 6.7 shows the permanent magnet motor is lower energy use and has no impact
on water level exceedance. It was already known that the permanent magnet motor had
lower investment costs. There is thus no trade-off here, as the permanent magnet
motor is clearly the better option.
The pump inflow improvement is the last design choice impacting energy use. Figure
6.8 shows no impact on water level exceedance and a small improvement in energy use.
There is thus trade-off between investment costs and energy use here. As the investment
costs of the inflow improvement are small, just e30,000 per pump, it makes sense to
choose for the improved inflow. It is however not a design choice that is very relevant
for the final robustness of pumping station.

Figure 6.7: Water level exceedance and Figure 6.8: Water level exceedance and
Energy used grouped by Energy used grouped by
Motor choice Pump inflow improvement

The percentage energy neutral is dominating both the operational costs and the
energy performance metrics, except for energy use. In Figure 6.9 the operational costs
show close resemblance with the energy bill, explaining the linear increase in the scatter
plot. With a single cost objective, Figure 6.9 would be the only graph available, and all
trade-offs that were previously found, would not have been found.
Without local energy production, the pumping station is costing money. The 100% energy
neutral pumping station on the other hand, is earning money. The difference in energy
bill is about e2 million in five years. The investment, which is between e3.6 and e4.8
million, depending on the solar wind mix, is earned back within at least decade. Moreover,
the 100% energy neutral pumping station was the most effective measure for bringing
CO2 emissions down, making it a robust solution.
6.2 Sensitivity analysis and Finding trade-offs 75

The second impacting factor on operational costs and energy bill is the energy scenario.
Figure 6.10 shows the average energy bill comes in the same sequence as average energy
price mentioned in Table 2.1. The only remarkable observation is that vision 3 is closest
to vision 4 in the energy neutral case, whilst it closest to 2011-2015 for the 0% energy
neutral case.

Figure 6.9: Energy bill and Operational


Figure 6.10: Energy bill and Operational
cost by Percentage energy
cost by Energy scenario
neutral

In Figure 6.12 (next page) the effect of the pump control can be seen. The energy bill
(left middle) for both 0% and 100% energy neutral is lowest for the APX energy price
control and highest for the normal control with fixed on/off levels. The kernel density
lines are however close together, making it is difficult to clearly see the differences.
A comparison of APX and normal control for the 0% energy neutral pumping station
is shown in Figure 6.11. The APX control realized 19% savings compared to normal
control in the period 2011-2015. The performance of the APX control, appeared to be
very sensitive for the energy scenario. The lowest savings from the APX control were 6%,
for high CO2 price scenario, while it yielded 45% saving for the vision 4 scenario. It can
be concluded that savings increase with the price volatility of the scenario.

Figure 6.11: APX control savings for energy scenarios


76 Insight via learning

At the right of Figure 6.12, the local energy used results are much more clear. For 0%
energy neutral, the use of local energy used is off course zero. For the 100% energy neutral
pumping station the use of local energy gets as high as 55%, by using the pump control
bases on the availability of local energy. APX and day night control score poor for local
energy use. These two mainly pump at night, when it is off course not possible to directly
use solar power.
This can also be seen from the graph, by moving vertically up from the day night control
to the top plot, where the effect of the mix solar wind is shown. For the day night
control the use local energy is lowest when 100% solar is installed. For other controls an
optimum use of local energy is seen around a 50/50 mix. These results are in line with
the earlier insight considering the weekly and yearly power production in section 3.4.2.

Figure 6.12: Energy bill and Local energy use by Pump control

The energy profits are slightly higher for solar, but this comes from the fact that the SDE
subsidy was higher for solar. However, this is compensated by the fact that the investment
costs for solar were also 33% higher. Based on local energy use and CO2 emissions (not
shown here) it was concluded that a solar wind mix of 50% is optimal.
6.2 Sensitivity analysis and Finding trade-offs 77

6.2.2 Second insight loop - Reduced choices - 120 designs in 48 scenarios

In the first insight loop seven of the eight design choices could be decided upon, but
for the pump control it was difficult to come to a solid conclusion, as the results were
overshadowed by the choice for energy neutrality. In this simulation experiment, for
second insight loop, the percentage energy neutral is fixed at 100%. Besides that, the mix
solar wind was set to 50% and the number of pumps was set to 2.
For this experiment 8-3 = 5 design choices were considered. The remaining five design
choices, all of the Categorical type, resulted the following number of designs:

N r. designalternatives = 23 ∗ 3 ∗ 5 = 120 [−] (6.8)

It was possible to run a full factorial sampling, meaning all 120 designs in all 48 future
states of the world were simulated. The number of simulation required was thus:

N r. simulations = 48 ∗ 120 = 5760 [−] (6.9)

This took one hour to simulate and store the results. Again feature scoring was applied to
analyze the results, resulting in Figure 6.13. First thing that stands out from this graph is
that now the energy scenarios dominate the energy bill and thereby the operational costs.
The Energy scenarios have also some impact on local energy used and CO2 emitted, but
the pump control is more important.

Figure 6.13: Second loop feature scoring results


78 Insight via learning

Figure 6.14 shows the effect of the energy scenarios on energy bill, local energy used
and CO2 emitted. It is interesting to see that Gaussian kernel density plots show one bell
shaped peak per scenario for the energy bill, while in CO2 emitted and local energy used
three peaks are seen for every scenario.

Figure 6.14: Energy bill and Local energy used grouped by Energy scenario

Now moving to Figure 6.15, which shows that the pump control has large impact CO2
emitted and local energy used. It can now be seen that the three peaks previously observed
in Figure 6.14 are caused by the pump control. From Figure 6.15 an interesting trade-off
between costs (energy bill) and sustainability (CO2 emitted and local energy use) can be
observed.

On the one hand, pump control based on the availability of local wind and solar power
shows robust behavior for CO2 emitted and use of local energy, that is a good average
performance (µ) within a narrow bandwidth (σ), as can be seen in Figure 3.

On the other hand, APX control and APX combined with local control, were extremely
sensitive for the energy scenarios, resulting in a large bandwidth and thus poor robustness.
6.2 Sensitivity analysis and Finding trade-offs 79

Figure 6.15: Energy bill and Local energy used grouped by Pump control

Normal and day night control, although showing a narrow bandwidth, are less robust
than local control, since their average performance is less. For CO2 emitted and local
energy use the local control is thus clearly the best option.
For the energy bill the differences remain small, the scenario simply have more impact
than the control. All controls have about the same bandwidth, but APX control seems to
yield a slightly lower average energy bill compared to local control. To get a more insight
into how the energy bill distribution was caused, it makes sense to look at the build-up
of the energy bill:

Energy bill = Energy purchased − Energy sold (6.10)

Figure 6.16 is showing these three elements. All controls are quite evenly distributed
over energy Bill, but they are very different in energy purchased and energy sold. For
local control, little money is spent to energy being purchased, probably following from
the robust use of local energy we previously observed. This is however also the reason
local control makes less money by selling energy back to the grid.
80 Insight via learning

Day night control and to a lesser extent APX control, lead to purchasing a lot of energy,
but also selling back a lot to the grid. When we add op these two effect they balance each
other, resulting in small differences in the energy bill.

Figure 6.16: Energy bill and Local energy used grouped by Pump control

Since for local control less energy is being purchased and less energy is being sold, this
control is less sensitive for changes in subsidy and tax policies. Because of this fact it
can now be concluded that local control is most robust for all energy performance
metrics.
An important remark that should be made here is that the energy scenarios and climate
scenarios are not weather consistent. This inconsistency in the solar intensity and wind
speed causes a mismatch between local energy production and national production of
sustainable energy and the energy mix. When weather in the energy and climate scenarios
would have been consistent, probably both the APX and the local control would improve.
However, that is speculation and therefore we recommend including weather consistency
for following research. This does require bringing together parties that make climate
scenarios, like KNMI, and parties that make energy scenarios, like DNV GL.
6.2 Sensitivity analysis and Finding trade-offs 81

6.2.3 Third insight loop - Uncertainties - 10 designs in 1000 scenarios

So far the input uncertainties have only been characterized by the prespecified energy
and climate scenarios, as they were assumed to be the key uncertainty factors. Off course
those are not the only things that can change in the future. It was already mentioned
that the IJsselmeer level may be raised by 50 cm to adapt to climate change. Moreover,
many assumptions were made to parameterize the model. The assumed percentage of the
NOP for which Vissering would become responsible, was for instance already discussed.
In this insight loop, the sensitivity analysis was performed for 39 uncertainties in the
assumptions. These 39 assumptions were sampled with a very large range, as shown in
Table 6.5.

Variable From To Units


Beta 0.05 0.1 -
rho 0.85 0.95 -
Lp 0.5 0.7 -
wet area percentage 0.01 0.02 -
unsaturated soil max 40 800 mm
percolation max 0.00416 0.0313 mm
runoff time 0.024 96 hour
groundwater flow time 480 1920 hour
ditch runoff time 3 12 hour
drainage capacity 0.417 0.542 mm/hour
drainage depth -1200 -800 mm
seep total 0.0188 0.0227 mm/hour
inlet yearly total 0.0186 0.0203 mm/hour
delay time prediction 12 24 hour
capacity per pump 12 13.33 mm/hour
pump η original 0.7 0.85 -
η increase inflow improvement 0.01 0.03 -
inflow improvement investment 20000 40000 EUR
water level IJsselmeer -400 -100 mm NAP
responsible for % of the NOP 0.3 0.7 -
induction motor η 0.94 0.96 -
PM motor η improvement 0.01 0.02 -
induction motor investment 1.50E+06 2.20E+06 EUR
PM motor investment saving 0.05 0.25 -
smoothing time smart variable 48 168 hour
cut fraction 0.2 0.4 -
tax and network costs 0.01381 0.03471 EUR
U100 over U10 1 1.4 m/s
SDE solar 0.09 0.128 EUR/kWh
SDE wind 0.076 0.082 EUR/kWh
cost of solar 900 1200 EUR/kW
cost of wind 1000 1500 EUR/kW
social cost of CO2 0.033 0.2 EUR/kg
cost of flooding 1 10 EUR/(mm/hour)
direct flooding cost ratio 0.3 0.7 -
potential fish migration rate 0.167 2 fish/hour
weight per fish 0.5 1 kg
cost of fish damage 12 15 EUR/kg

Table 6.5: List of the 39 sampled uncertainties

For this simulation experiment we sampled 10 designs and 1000 scenarios, the latter
one containing the uncertainty assumptions. Again feature scoring of the results was
performed. The performance was not sensitive for most of the assumptions, however 4
out of the 39 assumptions were large enough to be visible in Figure 6.17 (dotted lines).
82 Insight via learning

Figure 6.17: Third loop feature scoring of uncertainties

It is no surprise fish damage is sensitive for the potential fish migration rate. It
was sampled for the extremely large range from 0.167 to 2 fish/hour. Still, the fish safe
solution was of more importance for the result. With some measurements the uncertainty
about the potential fish migration rate can be significantly reduced.
The responsible for percentage of total NOP area factor is the most important
factor on CO2 emitted, energy use and water level exceedance. Even more important
than the design choices. The responsibility was variated between 30% and 70% of the
NOP, thereby scaling the amount of water to be discharged by more than a factor two. The
left-hand side of Figure 6.18 shows the effect of the responsibility factor. The energy use
scales linearly with the responsibility factor, and the water level exceedance shows steep
growth. Both these factors combine via energy bill and flooding costs to the operational
costs incl. indirect costs.
A responsibility of 50% was chosen in the first and second insight loop simulations, in-
dicated by the dotted line in the figure. As Vissering has 40% of installed capacity a
higher value than 40% underestimates the ability of the other pumping stations to help
in a heavy rainfall event.
In section 2.2.4 is was already discussed that this factor can be seen as an uncertainty, but
could also be considered a design choice as the water board can decide which pumping
station has the preference. Moreover, for the entire NOP, the amount of water to be
discharged remains the same. Discharging more water via Vissering will namely result
in lower energy use at the other pumping station. It is recommend to model the entire
NOP, including pumping station Buma and Smeenge, to get more grip on this factor.
6.2 Sensitivity analysis and Finding trade-offs 83

The unsaturated soil max dictates how much water can be taken up by the soil. When
this value is low more water will pass through the quick surface runoff, impacting the water
level exceedance. For very low values that it even impacts the energy use, because then
evapotranspiration will be limited. Actually, the water board has much more certainty
about this factor, than the sampled range. Nevertheless, it is good to know that the
unsaturated soil max is a very important factor in the hydrological model.

Figure 6.18: Scatter plots of most influential uncertainties

The final assumption was with some influence was the wet area percentage. This factor
is a measure for the amount of water storage capacity in the polder. This determines the
speed with which the water level rises. It was modeled between 1% and 2%, and shows a
bit higher water level exceedance for lower values. In the first and second insight loop, the
value was set at 2%, since it showed a good fit in the water system calibration. Later we
found out that the real value for the NOP is actually close to 1%. Fortunately, this error
was not very influential. Especially since it is compensated by the fact that we assumed
the responsibility at 50%, even though Vissering has only 40% installed capacity in the
NOP.
84 Insight via learning

The potential raise of IJsselmeer level with 50 cm showed little effect, only 0.2% sensitivity
at energy use. However, if pump and motor curve are added to the model, this sensitivity
will become larger. Is advisable not to neglect this factor and include it in the pump and
motor choice.
All in all, it can be concluded that most of the assumptions had little effect on the end
performance. The assumptions that did show sensitivity, were either known with far more
certainty than sampled or well controllable. It is thus justifiable that climate change and
energy market change were assumed as the key uncertainty factors. Nevertheless, this
sensitivity analysis for uncertainties has shown what the most critical assumptions in the
PSST model are.

6.3 Determining a robust design

Bases on the sensitivity analysis with feature scoring, we were able to find the trade-offs
that the design choices caused. We have not given preference to one of the performance
metrics, but have strived to achieve balanced trade-offs. This way, we were able to identify
the most robust design option for all design choices. This resulting robust design is shown
in Figure 6.19.

Most expensive design


3x induction motor e6,000,000
100% solar e4,800,000
3x new fish safe pumps e4,860,000
3x inflow improvement e90,000
new pump control e200,000
Investment costs e15,950,000

Cheapest design
2x PM motor e3,728,000
0% energy neutral e0
original impellers e0
no inflow improvement e0
Original pump control e0
Investment costs e3,728,000

Final design
2x PM motor e3,728,000
50% solar / 50% wind e4,187,500
2x impeller change e400,000
2x inflow improvement e60,000
new pump control e200,000
Investment costs e8,575,500

Figure 6.19: Final robust design Table 6.6: Investment cost

The robust design was found using visual analysis of the results. There are however also
algorithms that can help determining a robust design. PRIM is one of these algorithms.
PRIM will be used to check if it comes to the same conclusions as we did by means of
visual analytics.
6.3 Determining a robust design 85

6.3.1 Ranges by PRIM

Patient Rule Induction Method (PRIM) is powerful algorithm that finds input ranges
responsible for meeting a set of performance thresholds. The PRIM method is visually
explained in Figure 6.20 and contains the following steps:

1. Classification. Define a set of performance thresholds.


2. PRIM. The algorithm will:
(a) Identify runs of interest that meet the performance thresholds (red dots).
(b) Peel off layers form the input space with a low density of runs of interest.
(c) Find ranges of input space with high density of runs of interest (white box).
3. Interpretation. Investigate the design consequences.

Figure 6.20: PRIM method explained. Source: Greeven (2015)

Table 6.7 shows the chosen of performance thresholds used for the PRIM analysis.

Performance metric Threshold Units Based on


Investment costs < 10,000,000 [e] Budget by WB Zuiderzeeland
Energy used < 18,000,000 [kWh] > Below average of all simulation
CO2 emitted < 2,500,000 [kg] > Lowest 1/3 of all simulations
Water level exceedance < 10,000 [mm*Hour] Corresponds to e10,000/year
Cumulative Fish damage < 5,000 [fish] Assume regulated by law

Table 6.7: Thresholds for PRIM analysis


86 Insight via learning

The thresholds were based on the goals in Zuiderzeeland (2010), Zuiderzeeland (2015)
and the interviews with WB Zuiderzeeland. A threshold for local energy use was left out,
in order not choose for the energy neutral pumping station in advance of the analysis.
Figure 6.21 and 6.22 show the resulting input ranges for respectively the first and second
insight loop simulation experiments. These input ranges represent the design options
responsible for meeting the set of performance thresholds. The resulting design options
from PRIM are consistent with the visual analysis.

Figure 6.21: PRIM results from the first insight loop

In the first insight loop 6.21, not only local control, as was our most robust choice, but
also APX combined with local control and CO2 control meet the performance thresholds
with a high concentration. In the second loop the choice for energy neutral was fixed and
consequently does not come back in Figure 6.22.

Figure 6.22: PRIM results from the second insight loop

In both Figure 6.21 and 6.22 the new fish safe pumps also meet the performance thresholds
with a high concentration. Since the change of impellers showed a better balanced trade-
off, we chose for that option.
6.3 Determining a robust design 87

PRIM only shows the design choices that were critical for meeting performance threshold,
thus not all eight design choices are shown. It is no surprising that the design choices
with high sensitivity in the feature scoring, are now also critical in the PRIM analyses.
As such PRIM has affirmed the previously established robust design.
A final remark on PRIM is that the resulting robust design from the PRIM analysis
changed, when the performance thresholds were changed. It would be thus be interesting
to have an interactive PRIM session with decision makers and stakeholders. By inter-
actively changing these threshold, one can learn the consequences of choosing objectives
and threshold for the final design.

6.3.2 Performance of the final design

The spread of performance, over 48 future states of the world and thereby the robustness
of the final design, is shown in red in Figure 6.23. The final design can be compared to
the spread of results from the original 500 design alternatives.

Figure 6.23: Final design performance compared to all designs


88 Insight via learning

For the final design, the average outcomes have all moved to the desired direction, except
for the investment cost. The investment cost however included a mistake in the model
resulting in some too cheap designs. As an alternative, table 6.6 shows the investment
costs for the most expensive, cheapest and final design. The final design of e8,575,500 is
closer to the cheapest design, as to the most expensive design. It is also below the budget
of e10 million, although engineering and installing costs are not included yet.

In the final design, bandwidth has decreased for all performance metrics, but that is off
course trivial. For the energy bill and closely related operational cost, the bandwidth is
still quite large due to the energy scenario. Fortunately the costs are negative because of
the choice for a 100% energy neutral design.

The final design shows low energy use. The small peak at 1.3 is caused by the 2011-2015
climate scenario that had less rainfall.

The choice for local control results in robust performance on local energy use and CO2
emissions. The final model design shows still some water level exceedance. In the real
design however, the third pump with gas engine will be kept in use as a backup that
reduces the water level exceedance to near zero.

6.4 Adaptive operational strategy

So far we have been able to determine a static robust design. Until here we have not used
the dynamic robustness definition from Rosenhead et al. (1972):

• Flexibility by leaving options open to be able to adapt to future conditions.

Using this definition requires some human interpretation. We have seen that with just
two pumps the maximum water level exceedance is still very limited. In order to prevent
maximum water level exceedance, the third pump is necessary about 113 hours a year.
We suggest just two pumps are electrified and the third gas engine is kept in service.

This leaves the option open to have the gas-powered pump as a backup in case of a
power cut or when very high electricity prices are experienced. If this start happening
on a regular basis there is option add biogas tanks, to make the gas engine run more
environmentally friendly. The biogas could come from the sewage sludge gasification
plants owned by the water board.

It may seem implausible that gas becomes cheaper than electricity, but in France for in-
stance electricity prices have quadrupled between April and October 2016 due to technical
problems at a number of nuclear power plants Bloomberg (2016). Even if such problems
might not occur here, the European market is becoming more interconnected, whereby
such evens also have an effect here in the Netherlands.

Figure 6.24 gives a decision three for third motor, that can be consulted as the future
enfolds. Only in a very specific case is there reason to place a third electrical motor.
6.4 Adaptive operational strategy 89

Does gas become


structurally cheaper
than electricity?

No Yes

Do heavy rainfall
events become more
frequent?

No Yes

Do price volitility and


power cuts increase?

No Yes

Keep gas engine and


Keep gas engine Electrify third pump Keep gas engine
enable biogas

Figure 6.24: Decision three for third motor

Electrifying just two pumps saves e2.1 million, consequently leaving budget for other
sustainable design choices, for instance making the pumping station already 50% energy
neutral local with wind and solar power.
When it comes to the pump control we have seen that both the APX control and the local
control have their advantages. The performance of the APX control was not statically
robust, as it was sensitive for the energy scenarios. As the investment cost of the smart
control is relatively low, it would be advisable to implement both the APX as the local
control. In this way, the water board is flexible in choosing to APX or local control, or a
combination of both, as the future enfolds.
When to switch from one control to the other is a difficult question. This question is
also relevant for the decision three in Figure 6.24. An essential part of such adaptive
strategies, is specifying under exactly what conditions to adapt to a different strategy.
The questions in the decision three should thus be formulated as an adaptation tipping
point, which is a trigger that specifies the exact conditions under which a prespecified
action to change the strategy is to be taken Haasnoot et al. (2013).
The formulation of an adaptive operational strategies, or dynamic adaptive policy path-
ways Haasnoot et al. (2013), are a whole study on its own. The decision three in Figure
6.24 is thus a starting point for a future study, into an adaptive operational strategy
for Vissering. By performing this study Vissering will not only have a statically robust
design, but also a dynamically robust operational strategy to adapt itself as the future
enfolds.
Chapter 7

Conclusions and Recommendations

91
92 Conclusions and Recommendations

7.1 Conclusions

Uncertainties about climate change and the energy market constitute a major challenge
in designing a pumping station. To tackle this challenge, a robust design is needed, that
performs satisfactory under a wide variety of future conditions. Multi-Objective Robust
Simulation Optimization (MORSO) was applied, to create a robust re-design for pumping
station Vissering. The main question was: ”What is the value of applying MORSO in
designing a robust pumping station for the future?”
The first goal was to build a model that could reproduce the behavior of pumping station
Vissering and the Noordoostpolder. A fast System Dynamics model, named Pumping
Station Simulation & Testing (PSST) model, was developed and calibrated with historical
measurements, thereby accomplishing the first goal of the research.
The second goal was to use the PSST model to gain insight in effect of design choices
and input uncertainties on the performance. The input uncertainties were characterized
by eight climate scenarios and six energy scenarios, leading to 48 plausible future states
of the world (SOW). Eight design choices in the categories pumps, motors, control and
innovation, resulted in numerous design alternatives. The performance of the design was
expressed by performance metrics in the categories water quantity & water safety, water
quality & ecology, energy performance and costs.
Combining the future SOW and design alternatives led to a large set of simulations, that
was analyzed using sensitivity analysis with feature scoring. This ranked the sensitivity
of the performance, for the design choices and future scenarios. The future scenarios
caused the same order of magnitude sensitivity as the design choices. As such, this insight
confirmed the importance of including uncertainties about the future in a pumping station
design.
In addition, feature scoring revealed several design choices that created a trade-off between
different performance metrics. The resulting spread in performance, our measure for
robustness, could be used to investigate these trade-offs. The pump control for instance,
created a trade-off between the energy bill and sustainability.
On the one hand, pump control based on the availability of local wind and solar power
showed robust behavior for CO2 emissions and use of local energy. That is a good average
performance (µ) within a narrow bandwidth (σ). On the other hand, pump control based
on hourly APX electricity prices was extremely sensitive for the energy scenarios, resulting
in poor robustness.
By considering multiple performance objectives, instead of just a single cost objective,
many trade-offs and interactions between design choices and future scenarios were re-
vealed. These were only revealed because numerous design alternatives were simulated in
different future scenarios. Traditional design strategies and the other robustness frame-
works would not have come to these insights. At this point we can say that the MORSO
method succeeded in second goal of creating insights.
The third and final goal was to determine a robust pumping station design. By means of
visual analysis of the resulting performance spread, the most robust design options could
be identified. In this way, the research conducted was able to establish the final robust
design, shown on the cover of this chapter.
7.2 Recommendations 93

PRIM analysis confirmed that this set of design choices was indeed responsible for meeting
the performance thresholds, thereby concluding final goal was accomplished.
MORSO proved to be a key factor in creating the acquired insights. As such, the main
question can convincingly be answered:
”Yes, MORSO was of added value in designing a robust pumping station, as
the method was able to establish a robust design and created many insights, as
a result of simulating numerous design choices in different future scenarios.”

7.2 Recommendations

The research has shown that the MORSO framework was of added value creating insight in
the robust pumping station for the future. There are also some general recommendations
of how the model and method can be improved for future use. Furthermore, there are
some recommendation of how to improve the analysis for the specific case of Vissering.

7.2.1 General recommendations

• Adaptive operational strategy


A starting point was made for a future study, into an adaptive operational strategy
for Vissering. The formulation of an adaptive operational strategies, or dynamic
adaptive policy pathways Haasnoot et al. (2013), are a whole study on its own. By
performing this study Vissering will not only have a statically robust design, but
also a dynamically robust operational strategy to adapt itself as the future enfolds.

• Model predictive control


In this thesis pump control was based on adapting target water levels. This simple
and quick way of pump control, created new insights in the potential of pump
control strategy based on APX, CO2 intensity and local energy availability. A model
predictive control is more complex and computationally intensive, but can most
likely produce a better pumping strategy. There are ongoing research initiatives
on this topic by the Smart Pumping initiative Kuipers et al. (2016), Peeeks Power
and Rijkswaterstaat (2016). In meetings with these parties the broad insights of
this thesis were explained, so that they can be incorporated in a complex model
predictive control algorithm.

• Weather consistency in climate and energy scenarios


The energy scenarios and climate scenarios are not weather consistent. This in-
consistency in the solar intensity and wind speed causes a mismatch between local
energy production and national production of sustainable energy. We recommend
including weather consistent scenarios for following research. This does require
bringing together parties that make climate scenarios, like KNMI, and parties that
make energy scenarios, like DNV GL.
94 Conclusions and Recommendations

7.2.2 Case study specific recommendations

• Calibrate water system using optimization


The water system was in this thesis calibrated to measured data iteratively by hand.
A future recommendation is to use optimization algorithms instead, to calibrate the
model. Because the PSST model uses a simple and fast SD model it is computa-
tionally feasible to use these optimization algorithms. This might not only deliver
a well calibrated model, but might even help improving the existing water model
(SOBEK) the water boards uses.

• Include the entire NOP in the model


It is recommended to model the entire NOP including pumping station Buma and
Smeenge to get more grip on the interactions between the three pumping stations.
This will also allow to investigate the consequences of having preference for a certain
pumping station in the pumping strategy. This will also improve insight in how to
prevent water level exceedance in the NOP in case of extreme rainfall events.

• Study into water level exceedance


Only five years of climate data was used. That is not enough to make solid state-
ments about extreme rainfall events and consequentially water level exceedance.
Simulating longer climate datasets and extreme value analysis, are needed before
firm conclusions can be made about water level exceedance. Moreover, a raw es-
timate for the flooding costs was used. The water board might already have a
reasonable estimate, otherwise the current estimate may be significantly improved
using waterschadeschatter.nl (2014).

• Fish safety assessment


Due to a lack of data, the fish safety assessment of pumping station Krimpenerwaard
Nieuwkamer and Klinge (2016) was adopted. To make the model case specific for
Vissering the fish safety factor of the current impellers could be determined with
the collision model of Esch and Rommens (2010). This should be backed by local
fish measurement at Vissering, that can also help determining the potential fish
migration rate. In case it is concluded that Vissering is not fish safe, improvement
measures will have to be determined and added to the model.

• Add pump curves


By adding pump curves to the PSST model the effect of water level fluctuations on
pump performance can better be assessed.

• Interactive PRIM session with stakeholders


The resulting robust design from the PRIM analysis changed, when the performance
objectives were changed. It would be thus be interesting to have an interactive
PRIM session with decision makers and stakeholders. By interactively changing
these objectives, one can learn the consequences of choosing objectives and threshold
for the final design.
Bibliography

Water Framework Directive (2000). Directive. EC of the European Parliament and of the
Council of.

APX (2016). Apx power nl day ahead. https://www.apxgroup.com/market-results/


apx-power-nl/dashboard/.

Arrow, K. J. (1950). A difficulty in the concept of social welfare. The Journal of Political
Economy, pages 328–346.

Bankes, S. (1993). Exploratory modeling for policy analysis. Operations research,


41(3):435–449.

Beggs, C. (2010). Energy: management, supply and conservation. Routledge.

Bell, R. G. and Callan, D. (2011). More than meets the eye: the social cost of carbon in
us climate policy. Environmental Law.

Bergström, S. and Singh, V. (1995). The hbv model. Computer models of watershed
hydrology., pages 443–476.

Betz, A. (1920). Das maximum der theoretisch möglichen ausnützung des windes durch
windmotoren. Zeitschrift für das gesamte Turbinenwesen, 26(307-309):8.

Biesboer, F. (2016). Rioolwaterwarmte. De Ingenieur, 2:18–19.

Bloomberg (2016). France burns coal like it’s 1984 as prices jump on
atomic woes. http://www.bloomberg.com/news/articles/2016-10-18/
france-burns-coal-like-its-1984-as-prices-jump-on-atomic-woes.

Bosker, D. (2013). Krooshekreiniger geplaatst bij stuw stipdonk. http://www.mfbosker.


com/nl/nieuws/2013/08/oplevering-stuw-stipdonk.

Bosman (2016). Bosman watermanagement. http://www.bosman-water.nl/.

95
96 BIBLIOGRAPHY

Brakkenhoff, D. (2013). Cie4431 - hydrological modeling - development and evaluation


of lumped conceptual models for the hesperange catchment. Technical report, Delft
University of Technology.

Buck, Hers, Afman, Croezen, Rooijers, van der Veen, van der Wijk, and Slot (2014).
Toekomst warmtekrachtkoppeling en warmtevoorziening industrie en glastuinbouw.
Technical report, CE Delft and DNV GL.

canonnoordoostpolder.nl (2016). Provincies en waterschap. http://www.


canonnoordoostpolder.nl/vii-bestuurd/29-regionaal-bestuur.

co2emissiefactoren.nl (2016). Lijst emissiefactoren elektriciteit. http:


//co2emissiefactoren.nl/lijst-emissiefactoren/#elektriciteit.

Courant, R., Friedrichs, K., and Lewy, H. (1928). ber die partiellen differenzengleichungen
der mathematischen physik. Mathematische Annalen, 100many(1):32–74.

Cultuurtechnische Vereniging (1988). Cultuur technisch vademecum. Cultuurtechnische


Vereniging, Rijkskantorengebouw Westraven, Utrecht, The Netherlands.

DNV GL (2015). Routekaart energieopslag 2030. Technical report, DNV GL and TU


Delft and Berenschot.

Douben, K., van Heeringen, K., IJpelaar, R., Van Loenen, A., and van de Wouw, M.
(2014). Noord-Brabantse waterbeheerders anticiperen samen op hoogwater. H2O.

Enercon (2016). Enercon’s efficient e-92 2.3 mw wind turbine. http://www.enercon.de/


en/products/ep-2/e-92/.

Entsoe (2016). entsoe transparancy platform. https://transparency.entsoe.eu/.

Esch, v. B. and Rommens, C. (2010). Visvriendelijke pompen. H2O, 16:29.

EU Regulation, C. (2007). No 1100/2007 of 18 september 2007 establishing measures for


the recovery of the stock of european eel. Official Journal of the European Union L,
248(17):22–9.

Fenicia, F., Savenije, H., Matgen, P., and Pfister, L. (2006). Is the groundwater reservoir
linear? learning from data in hydrological modelling. Hydrology and Earth System
Sciences Discussions, 10(1):139–150.

Fenicia, F., Savenije, H. H., Matgen, P., and Pfister, L. (2008). Understanding catchment
behavior through stepwise model concept improvement. Water Resources Research,
44(1).

Flowserve (2016). Flowserve lnn range. http://www.consolidatedpumps.com/


products/catalogue/Flowserve/LNN-Range.

Forrester, J. W. (1997). Industrial dynamics. Journal of the Operational Research Society,


48(10):1037–1041.

Frontier Economics (2015). Scenarios for the dutch electricity supply system. Technical
report, Frontier Economics.
BIBLIOGRAPHY 97

Greeven, S. (2015). Societys response to climate change. Master’s thesis, TU Delft.

Grogg, K. (2005). Harvesting the wind: the physics of wind turbines. Physics and
Astronomy Comps Papers.

GWW, S. D. (2012). Aanpak duurzaam gww. http://www.aanpakduurzaamgww.nl/.

Haasnoot, M., Kwakkel, J. H., Walker, W. E., and ter Maat, J. (2013). Dynamic adaptive
policy pathways: A method for crafting robust decisions for a deeply uncertain world.
Global Environmental Change, 23(2):485–498.

Hamarat, C., Kwakkel, J. H., Pruyt, E., and Loonen, E. T. (2014). An exploratory ap-
proach for adaptive policymaking by using multi-objective robust optimization. Simu-
lation Modelling Practice and Theory, 46:25–39.

Heijkoop, D., Malda, W., and Beukema, P. (2011). Het watersysteem als elektriciteits-
buffer. H2O, 44(5):4–5.

Hekkenberg, M. and Verdonk, M. (2014). Nationale energieverkenning 2014. Technical


report, Energieonderzoek Centrum Nederland.

Herman, J. D., Reed, P. M., Zeff, H. B., and Characklis, G. W. (2015). How should
robustness be defined for water systems planning under change? Journal of Water
Resources Planning and Management, 141(10):04015012.

HHSK (2014). Gemaal verdoold. https://www.schielandendekrimpenerwaard.nl/


over-ons/waterpeil/gemalen/gemaal-verdoold.

Hoogreemraadschap Delfland (2015). Investeringsplan en uitvoeringskrediet voor vervan-


gen van BOS.

Kaia (2014). Solar-dike. http://www.kaia.nl/solar-dike/.

Kamp, H. (2015). Maatregelen negatieve elektriciteitsprijzen in sde+. Letter to: De


Voorzitter van de Tweede Kamer der Staten-Generaal From: Minister van Economische
Zaken H.G.J. Kamp.

Kasprzyk, J., Nataraj, S., and Lempert, R. (2013). Many objective robust decision making
for complex environmental systems undergoing change. Environmental Modelling and
Software, 42:55–71.

Kasprzyk, J. R., Reed, P. M., and Hadka, D. M. (2016). Battling arrow’s paradox to
discover robust water management alternatives. J. Water Resour. Plann. Manage.,
142(2):04015053.

Kempener, R. and Vivero (2015). Renewables and electricity storage. A technology


roadmap for REmap, 2030.

Klein Tank, A., Beersma, J., Bessembinder, J., van den Hurk, B., and Lenderink, G.
(2014). Knmi ’14 klimaatscenario’s voor nederland. KNMI, De Bilt, the Netherlands.

Klimaatplein (2015). Hoe ziet jouw klimaatakkoord eruit? http://www.klimaatplein.


com/hoe-ziet-jouw-klimaatakkoord-eruit.
98 BIBLIOGRAPHY

KNMI (2016). Klimatologie - uurgegevens van het weer in nederland - download. http:
//projects.knmi.nl/klimatologie/uurgegevens/selectie.cgi.

Kuipers, H., van Esch, B., Meijburg, M., Grispen, J., Pothof, I., and van Heeringen, K.
(2016). Energie- en kostenreductie in boezem- en polderbemaling. Project proposal.

Kunst, J., Spaargaren, B., Vriese, F., Kroes, M., Rutjes, C., van der Pouw Kraan, E., and
Jonker, R. (2010). Gemalen of vermalen worden. STOWA Onderzoek naar visvrien-
delijkheid van gemalen.

Kwakkel, J. H. (2016). Exploratory modelling and analysis (ema) workbench. http:


//emaworkbench.readthedocs.io/.

Kwakkel, J. H., Eker, S., and Pruyt, E. (2016). How Robust is a Robust Policy? Com-
paring Alternative Robustness Metrics for Robust Decision-Making, pages 221–237.
Springer International Publishing.

Lempert, R., Popper, S., and Bankes, S. (2003). Shaping the next one hundred years:
New methods for quantitative, long-term policy analysis. Technical report, The RAND
Corporation.

Lievense, L. (1880). Plan lievense. http://www.overwind.nl/artikelen/


plan-lievense-2/.

MeteoGroup (2016). Neerslag bewaking. http://neerslag.nl/.

Ministers van de Benelux (1996). Beschikking m vrije migratie van vissoorten. Law.

Moore, F. C. and Diaz, D. B. (2015). Temperature impacts on economic growth warrant


stringent mitigation policy. Nature Clim. Change, 5(2):127–131.

NGS (2016). De nederlandse gemalenstichting. http://www.gemalen.nl/.

Nieuwkamer, R. and Klinge, M. (2016). Variantenstudie vispasseerbaar maken gemaal


krimpenerwaard. Technical report, Witteveen+Bo and Hoogheemraadschap van
Schieland en de Krimpenerwaard.

Noordzeeloket (2016). Windenergie op zee. https://www.noordzeeloket.nl/


functies-en-gebruik/windenergie/.

NVOE (2011). Energie uit oppervlaktewater. Technical report, Nederlandse Vereniging


voor Ondergrondse Energieopslag.

PBL (2016). Correction wording flood risks for the netherlands in


ipcc report. http://www.pbl.nl/en/dossiers/Climatechange/content/
correction-wording-flood-risks.

Programmabureau Warmte Koude Zuid-Holland (2016). Warmterotonde. http://


warmopweg.nl/programma/warmterotonde/.

Pruyt, E. (2013). Small System dynamics models for big issues: Triple jump towards
real-world complexitylrm. TU Delft Library.
BIBLIOGRAPHY 99

Pumpfundamentals (2016). Caviation prediction. http://www.pumpfundamentals.com/


help2.html.

Quartz (2016). Germany had so much renewable energy on sunday that it had to pay
people to use electricity. http://qz.com/680661/.

Rijkswaterstaat (2016). Pilot ”duurzaam waterbeheer”. Technical report, Nelen & Schu-
urmans and Delta and Rijkswaterstaat.

Rooijers, F., Schepers, B., van Gerwen, R., van der Veen, W., Verheij, F., de Buck, A.,
et al. (2014). Scenario-ontwikkeling energievoorziening 2030. Delft, The Netherlands:
Publicatienummer, 14:C93.

Rosenhead, J., Elton, M., and Gupta, S. K. (1972). Robustness and optimality as criteria
for strategic decisions. Journal of the Operational Research Society, 23(4):413–431.

RVO (2015). Infoblad trias energetica en energieneutraal bouwen. Infosheet.

Schoots, K. and Hammingh, P. (2015). Nationale energieverkenning 2015. Technical


report, Energieonderzoek Centrum Nederland.

Schoots, K., Hekkenberg, M., and Hammingh, P. (2016). Nationale energieverkenning


2016. Technical report, Energieonderzoek Centrum Nederland.

SD Society (2016). Core system dynamics modeling software. http://tools.


systemdynamics.org/core-sd-software/.

Shelanski, H. (2013). Refining estimates of the social cost of carbon. https://www.


whitehouse.gov/blog/2013/11/01/refining-estimates-social-cost-carbon.

Smith, P. (2014). Enercon installs north america’s first e-92 tur-


bines. http://www.windpowermonthly.com/article/1317092/
enercon-installs-north-americas-first-e-92-turbines.

Spaans Babcock (2016). Vijzelpompen schroef van archimedes. http://www.


spaansbabcock.com/nl.

Stewart, J. (2010). Calculus. Cengage Learning.

Stichting DE-on (2016). Duurzame energie- en ontwikkelingsmaatschappij flevoland.


http://www.deon-flevoland.org/.

Tauw (2015). Onderzoek energiereducende maatregelen poldergemalen. Technical report,


Tauw in commission of Zuiderzeeland.

The British Geographer (2016). Hydrographs and recur-


rence intervals. http://thebritishgeographer.weebly.com/
hydrographs-recurrence-intervals-and-drainage-basin-responses.html.

UvW, STOWA, and Rijk (2016). Gd-195 green deal energy uvw - rijk. http://www.
greendeals.nl/gd195-energie-unie-van-waterschappen-rijk/.

van Beijen, W. (2016). Basisspecificatie rijkswateren. Technical report, Rijkswaterstaat.


100 BIBLIOGRAPHY

van Druten, E. (2016). Introduction video pumping sation for the future. http://
emielvandruten.nl/.

van Haaren, J. (2015). The future of energy supply and demand of a dutch city in 2030.
Master’s thesis, University of Twente.

Ventana Systems (2016). Vensim dss. http://vensim.com/.

Versteeg, R. and Roelevink, A. (2011). Verbetering watersysteemmodellering, neer-


slagspreiding. Technical report, HKV.

Visser, M., Stoppelenburg, F., Versteeg, R., and Bakker, M. (2012). Watersysteemtoets
2012. Technical report, Zuiderzeeland and HVK.

Walker, W., Haasnoot, M., and Kwakkel, J. H. (2013). Adapt or perish: A review of
planning approaches for adaptation under deep uncertainty. Sustainability, 5(2):955–
979.

waterschadeschatter.nl (2014). Waterschadeschatter by stowa. http://www.


waterschadeschatter.nl.

Wauben, M. (2011). Motorselectie pompen gemaal verdoold. Technical report, Wit-


teveen+Bos.

Windstats.nl (2016). Statistieken over windenergie in nederland. http://www.


windstats.nl/.

Wolstenholme, E. F. (1990). System enquiry: a system dynamics approach. John Wiley


& Sons, Inc.

Wolters, e., Hakvoort, H., Bosch, S., Versteeg, R., Bakker, M., Heijkers, J., Talsma,
M., and Peerdeman, K. (2015). Meteobase: online neerslag-en referentiegewasver-
dampingsdatabase voor het nederlandse waterbeheer. http://meteobase.nl.

Xylem (2016). Cusomized pump stations. http://www.xylem.com/pumping/no/


expertise/flygt-custom-pump-stations.

Zeegers, I. (2015). Watersystemen als elektriciteitsbuffer. Technical report, Hoogheem-


raadschap Hollands Noorderkwartier.

Zuiderzeeland (2010). Gemalenplan 2011 – 2020. Technical report, Waterschap Zuiderzee-


land, Afdeling Zuiveringen en Gemalen.

Zuiderzeeland (2015). Verkenning energie en gemalen. Technical report, Water Board


Zuiderzeeland.

Zuiderzeeland (2016). Waterschappen gaan energie besparen met slim malen. http:
//www.zuiderzeeland.nl/actueel/persberichten/@23753/waterschappen-gaan/.
Appendix A

Expert interviews

Since many of the subjects that were invested this research were new and innovative,
little literature was yet available. Still these developments and innovations needed to be
modeled. Thus I gathered much information by interviewing many experts from different
companies and institutions. An overview of all interviews is presented in Table A.1 in
chronological order.

Interview with Company Subject Date


Bart Dekens Witteveen+Bos DSS 20-02-2016
Jules van Haaren Witteveen+Bos Energy market 25-02-2016
Barry Scholten IF Technology ATES Smart Polder 25-02-2016
Paulien Hoogvorst Witteveen+Bos Energy supply 01-03-2016
Davı́d Brakkenhoff Witteveen+Bos Hydrology 08-03-2016
J. Kwakkel & J. Timmermans TU Delft Simulation approach 31-03-2016
Ivo Pothof Deltares DSS 01-04-2016
Graduation commission TU Delft Kick-off 04-04-2016
A. Koffeman & C. Bakker Zuiderzeeland Case study Vissering 04-04-2016
F. Roest & M. Wauben Witteveen+Bos Pumping stations 05-04-2016
E. Moerman & C. Rood HHNK Case study De Poel 11-04-2016
Ronald van Nooyen TU Delft DSS 13-04-2016
Kaz Vermeer Peeeks Imbalance market 13-04-2016
Marijke Visser Zuiderzeeland Hydrology NOP 17-05-2016
Thomas van Egmond Rijkswaterstaat Tender Vissering 20-05-2016
Rudolf de Vetten Agro Energy APX energy market 24-05-2016

Table A.1: Overview of interviews

101
102 Expert interviews

A.1 Interview with Bart Dekens

Interview with Bart Dekens


Company Witteveen+Bos
Subject Decision support systems
Date 20-02-2016

Bart works at Witteveen+Bos in the water management group. He is involved the de-
velopment of the new decision support system DSS 2.0 for the Delfland water board. I
interviewed him to learn more about decision support systems DSS.

A.1.1 How are the pumping stations in Delfland currently controlled?

Delfland already works with a DSS, DSS 2.0 is just an update for this system. Delfland
has 56 polder pumping stations that are pumping from the polders to the drainage canal
and 6 larger drainage pumping stations that pump water from the drainage canal to
water bodies that directly convey to the sea. Only the latter 6 are affected by DSS, while
the polder pumping stations automatically switch on when the maximal water inside the
polder is reached and it keeps pumping until the minimal water level is reached. The
current DSS gives an advice to the operators how many m3 to pump from the drainage
canal. DSS can also autonomously execute the advice it produced. The current DSS is
considered to be too slow, as it only updates this advice every 15 minutes. In order to
effectively react to heavy rainfall it is important for the system to give an updated advice
more frequently.

A.1.2 Which inputs and models does DSS use?

The inputs for DSS are water level measurements in the drainage canal, some rain gauges
and a rain forecast. The current DSS has a simplified water model made up of nodes and
lines. DSS 2.0 has a more complex water model included in RTC-tools and SOBEK. The
advice procedure works as follows:

1. Every 5 minutes an updated rain forecast from HydroNET comes in.

2. Water level measurement are updated and a prediction of the water load is done.
The rain gauges are uses to verify if the predicted rain matches the rain that has
actually fallen.

3. RTC-tools perform real-time control by simulating different pumping strategies in


the water model. It optimizes for an objective function, thereby determining the
most optimal pumping strategy to achieve the desired water levels.

4. The determined pumping strategy is verified in SOBEK, which includes an even


more detailed water model, which eliminates some undesired boundary effects.

5. The updated pumping strategy is set into action.


A.1 Interview with Bart Dekens 103

A.1.3 What are the benefits of DSS?

DSS enables the water board to use water system more efficiently by anticipating better on
the weather forecast. DSS makes it possible to respond quicker and better to heavy rainfall
event, thereby reducing the change of flooding. DSS 2.0 will operate more autonomous
than predecessor, but the operators that work at the water boards will not be replaced.
DSS does not enable significant cost savings, although the energy use will be less. Pomp
curves are namely included in DSS, so if possible an energy efficient working point will
be chosen. Also outside water level, which is influenced by tides is included, so this can
be seen as flexible pumping based on hydraulic head forecast.

A.1.4 Which state of the art innovations are implemented in DSS 2.0?

Flexible pumping based on fluctuating energy prices is not included. Delfland left this out
of the scope of this project. (Even though Delfland showed in a feasibility study Heijkoop
et al. (2011) that it is possible). Why Delfland chose so is a guess, maybe their opinion
was that the project was complex enough already. Also the target water level is -0.43
m NAP, and the operators are quite keen on keeping the drainage canal at that level.
So while there might theoretically be a minimal and maximal level, giving some storage
volume for flexible pumping, in practice this margin is not usable.
In DSS 2.0 wind set-up is now included in the water model, which actually plays a
significant role in Delfland. The main improvement is that the system has become quicker
and better in anticipating.

A.1.5 What does a DSS cost?

Witteveen+Bos is working on this project with 4 employers. The project is performed in


cooperation with HKV consultancy. HKV is integrating DSS with Delft-FEWS which is
the flood early warning system, and works with 2 employers on the project. The budget
for both parties is about e300,000. Delfland also had some internal employers working
on the implementation of DSS 2.0.

A.1.6 Do other water boards also use DSS?

Some water boards do not use DSS. The Noordoostpolder for instance, has a very robust
and simple water system with broad drainage canals and has only 3 over-dimensioned
pumping stations. Water board Zuiderzeeland Zuiderzeeland (2010), managing the No-
ordoostpolder, does not need DSS to determine a pumping strategy.
Water board Hollands Noorderkwartier is using a DSS with ControlNEXT software from
the company Nelen and Schuurmans, but this is actually an adapted version of RTC-tools.
Hollands Noorderkwartier is starting this year with flexible pumping Zeegers (2015) based
on fluctuating energy prices.
Bart was also involved in the development of DSS-Brabant Douben et al. (2014) . In this
DSS the water boars of Branbantse Delfland, Aa en Maas and De Dommel are working
104 Expert interviews

together to better anticipate on high water. It makes sense that these water boards work
together, because their water systems are interconnected.
There is a nationwide SOBEK model that combines all the water models of the different
water boards. There could be a trend of more DSS models from different water boards
being merged of linked together.

A.1.7 What are the bottlenecks in implementing DSS 2.0?

The most difficult part of this project is gathering the right data and combining different
data flows. When collection and storage of different data-flows was set-up in the past,
there was no attention on doing this in a way that makes it easy to combine these data
flows in the future. Also the arrangement of some smaller polder pumping stations was
not properly documented, while this information was necessary for properly making the
water model.

A.1.8 Which innovations do you expect to breakthrough in pumping


stations in the coming years?

Bart expects more innovation to drainage system anticipate even better. Also changes on
the energy market may lead to new innovation for pumping stations.

A.1.9 Do you have any tips on my thesis project.

Bart thinks it will be an interesting project, but also a challenging one. If the design
model has to predict if flexible pumping makes sense, actually a simplified DSS has to
programmed to determine that.
A.2 Interview with Jules van Haaren 105

A.2 Interview with Jules van Haaren

Interview with Jules van Haaren


Company Witteveen+Bos
Subject Energy market
Date 25-02-2016

Jules works at Witteveen+Bos at the energy development group. For my research I


considered three power supply options: fixed energy price contract, fluctuating energy
price contract and self-generated energy. I asked Jules about his vision on these three
options and possible ways to model them.

A.2.1 How is the energy price in a fixed or fluctuating energy contract


build up in the Netherlands?

The energy price is build up out of three mayor elements: the delivery, distribution and
taxes. The delivery cost is based up out of the bare energy price, but also includes a
standing monthly charge and for gas it includes a regional charge for transport to your
region.
This bare energy price is determined by the Amsterdam Power Exchange (APX), which
changes every hour based on supply and demand. Fixed energy contracts are bases on the
average APX price. This average price has gradually decreased from about 80 e/MWh
in 2011 to about 50 e/MWh in 2015. But between October 2015 and January 2016 it
has suddenly dropped to about 30 e/MWh.
The distribution costs are a fixed monthly sum for the connection to the end user, based
on the capacity of the connection. The energy supplier (energy supplier) transfers this
sum to the distribution network operator (DNO). Note that if an end user has only short
term energy peaks it could be an interesting business case to invest in local energy storage
in combination with a lower capacity connection.
Energy taxes are payed per kWh and are divides in different divisions bases on yearly
consumption. The more one uses the less one pays and this can make a large difference.
A normal household pays a rate of e0.10 per kWh, but for there are also rates of e0.05
and e0.01. For the largest users above 10 million kWh are rates of e0.001 and e0.0005.
Note that this is a factor 200 difference with normal households. The Dutch government
does this to stimulate large companies to establish themselves in the Netherlands. For
the water boards this is beneficial because their energy use is very high.

A.2.2 Which parties generate future energy prices scenarios?

First thing to say is that the energy market is very difficult to predict. A possible source
could be the National Energy Vision NEV 2015 Schoots and Hammingh (2015) report
written by the Energy research Center Netherlands ECN, the Central Bureau for Statistics
CBS and the government. This report gives predictions with a certain bandwidth about
the energy market. Jules used the NEV 2014 Hekkenberg and Verdonk (2014) report
106 Expert interviews

which is the for this thesis report. The NEV 2015 report contains already quite some
adjustments compared to the NEV 2014 report. The NEV 2015 report is the most actual
report, but the last months the oil and energy prices have dropped so much that the NEV
2015 can already be considered outdated. Jules thesis report Smart City 2030 van Haaren
(2015) could also be an interesting source for my research, since it includes the analysis
of urbanization- and energy trends.

A.2.3 Do you have any suggestions on how to model the energy market?

Modeling the entire energy market in detail yourself is very difficult. Some simplifications
need to be done. Maybe also some existing models could be integrated or replicated.
An interesting existing model is EnergyPLAN, it simulates the operation of national
energy systems on an hourly basis, including the electricity, heating, cooling, industry,
and transport sectors. Another possibility is the Energy Transition Model developed by
the company Quintel Intelligence which is an energy modeling and research firm that
assists governments, companies and institutions around the world in determining and
quantifying their long-term energy strategies. Jules emphasizes that Quintel could be an
interesting company to talk to about how to model the energy market for this research.
A.3 Interview with Barry Scholten 107

A.3 Interview with Barry Scholten

Interview with Barry Scholten


Company IF Technology
Subject ATES Smart Polder
Date 25-02-2016

Barry works at IF Technology, a company specialized in geothermal energy and aquifer


thermal energy storage ATES. The Smart Polder concept is an innovation for pumping
stations that integrates ATES in pumping stations. Since this research investigates pump-
ing station innovations. I asked Barry some questions about the Smart Polder concept.

A.3.1 Is it correct that the Smart Polder concept contains three specific
technical innovations?

That is correct, it contains ATES, flexible pumping based on fluctuating energy prices
and the implementation of information and communication technologies at pumping sta-
tions. The latter one can be compared with a decision support system DSS. From these
innovations ATES is the core business of IF.

A.3.2 What is the current state of the Smart Polder project?

IF has currently 3 pilot project running with the ATES technology. In Delft and Wa-
geningen heat is extracted from the city canals in summer, stored in the ground and in
winter it is extracted again and used for heating. At the Parksluizen pumping station
in Rotterdam the system works just the other way around. There cooling is supplied
to the Erasmus Medical Center in summer. IF also performed a feasibility scan for the
water boards of Delfland, Rivierenland and Zuiderzeeland and in currently performing a
national scan. In March a new green deal between the Union of Water boards (UvW) and
the government will be signed. This deal will announce a number of innovating project
at that will be developed.

A.3.3 How does the ATES heating system work and how does the busi-
ness case look like?

In summer the canals of cities heat up to above 20 ◦C. This is bad for water quality
and also contributes to the urban heat island effect causing the city to be up to 4 ◦C C
warmer as its surroundings. This heat can be extracted from the canal water using a heat
exchanger and stored in an ATES. This works with a cold well and a warm well. 12 ◦C
C ground water from the cold well is pumped up and fed through the heat exchanger
where in interacts with the 20 ◦C C canal water. In the heat exchanger the groundwater
is heated up to about 18 ◦C C and then stored in the hot well. In winter this 18 ◦C can
be pumped up again and used as a warm source for a heat pump which heats tap and
heating water to over 60 ◦C C and delivers it to households. Using this a relatively warm
source for the heat pump increases to coefficient of performance COP of the heat pump.
108 Expert interviews

The COP expresses the thermal energy output over the electrical energy input. For this
system the heat pump has a COP of about 5, the ATES system has a COP of 40 (applied
twice a year) and the surface water extraction a COP of 100. The COP for total system
can then be calculated by 1 / (1/100+2/40+1/5) ≈ 4 . IF says that that their system has
a 25% better efficiency than traditional heat pump systems, so that corresponds with a
COP of 3.4. Households traditionally produce their own heat with a boiler for which they
have to buy gas with a high tax rate. Water boards pay little energy tax and because
the heat is produced more efficiently, making this a very interesting businesses case. The
taxes on heat are also low, as disclosed in the heat law. This law also imposes maximum
prices for heat. The payback time when considering a renovation of a pumping station
and integrating an ATES heating system is usually 8 to 14 years.

A.3.4 How does the cooling system work and how does the business
case look like?

In the Netherlands there is not only a demand for heating in the winter, but also for cooling
in the summer. For instance hospitals and data-centers have large cooling demand that
can be supplied with a ATES system. Now cold is stored in winter. This works also
with a warm and a cold well, but this time 12 ◦C C ground water from the warm well is
pumped up and fed through the heat exchanger where in interacts with 4 ◦C C surface
water. In the heat exchanger the groundwater is cooled down to about 6 ◦C C and then
stored in the cold well. In summer this 6 ◦C C groundwater can be pumped up again and
directly used in a heat exchanger for cooling. A heat pump is thus not needed. The COP
for total system can then be calculated by 1 / (1/100+2/40) ≈ 17.

The COP of traditionally air conditioning is 2.5 so improving this to 17 makes a great
businesses case for the water boards, even if the consumer also pays low energy taxes.
The payback time when considering a renovation of a pumping station and integrating
an ATES cooling system is usually 6 to 12 years.

For both the heating and the cooling ATES system key figures for the investment costs can
be found online at www.wkotool.nl. When already considering a renovation of a pumping
station the additional cost for installing a heat exchanger can be considered small.

A.3.5 When would you advise to integrate an ATES system at a pump-


ing station?

This is really a supply and demand question. Transporting thermal energy of long dis-
tances is not profitable, since it requires expensive infrastructure. Thus a demand for
either heat or cold is required near the pumping station. On the other hand also a ade-
quate supply of warm/cold water needs to be available at the pumping station. Another
point to be stressed is that in case of the Wageningen and Delft project there is a supply
of warm water in summer that is considered too warm for water quality in the canal. In
this case the urge for a ATES system comes from the supply side, since the city canal
waters require cooling.
A.3 Interview with Barry Scholten 109

A.3.6 Tauw presented the Riothermie concept that recovers thermal


energy from sewers. Does this concept compete with the Smart
Polder concept?

Tauw executed a pilot project with Riothermie at a swimming pool in Urk, which is
nominated for Vernufteling engineering price Biesboer (2016). Barry tells me this project
makes sense because the swimming pool has a year round heat demand and in the sewer
near the pool there is still some rest warmth from the showers. The disadvantage of
this concept lays in the fact that integrating the heat exchanger inside the sewer is quite
complex and expensive. At a pumping station there is no need to bring the heat exchanger
to the heat source, the heat exchanger can simply be installed in the pumping complex
and the pumps bring the water up by design.

A.3.7 What is the potential of flexible pumping based on fluctuating


energy prices, and where does it depend on?

Flexible pumping is not the specialization of IF, so IF does not have in detail calculation of
the potential energy savings. Water board Delfland reports that the potential is between
15 % to 25 % Heijkoop et al. (2011) and water board Holland Noorderkwartier reports 8
% to 24 % (Zeegers, 2015). Hollands Noorderkwartier is enrolling flexible pumping this
year for most of their pumping stations. The potential of flexible pumping depends most
on the storage capacity of the water system. If simplified this storage capacity can be seen
as the product of water area and the allowed margin in water levels. The prediction time
of weather forecast with high certainty, which is about two days, also plays a big role.
The average APX price has two peeks every day, one in the morning at about 9:00 and
the other one in the evening at about 19:00. The price during night-time is usually lowest.
This means that if a rain event can be predicted two days in advance, there is enough
flexibility in the system, because there are at least two nights at which the predicted
discharge can already be pumped out of the system for a low price. Working with just a
day and night tariff thus also helps, and quite some water boards already have that.

A.3.8 Do you think the potential is influenced by a correlation between


weather conditions and energy price?

The question if there is a correlation between the weather conditions and energy price
during pumping is interesting. Barry acknowledges my hypothesis that the days around
a rain event are probably cloudy and there is less solar power. On the other hand rainfall
could also cohere with higher wind speeds, leading to more wind power. It is interesting
to investigate this hypothetical correlation in this research. Maybe Joep Gispen form the
company Nelen & Schuurmans that designed flexible pumping for Hollands Noorderk-
wartier could help me answering this question.
110 Expert interviews

A.3.9 Do you have any suggestions for a location possible case study?
I was considering pumping station Vissering in Urk.

Coincidentally IF is already involved in the pumping station of Vissering. Nowadays the


pumping station is delivering heating for companies in area. IF investigated that the
demand for cooling is actually higher, especially by fishing companies. The infrastructure
for delivering heating might be used for delivering cooling in the future. Barry acknowl-
edges that this pumping station might be an interesting case study. Teun Wendt and
Cees Bakker are busy with the topic energy within water board Zuiderzeeland. Royal
Haskoning DHV is currently developing an energy vision for Zuiderzeeland. Barry does
not know if the company for the restoration and electrification of the pumping station is
already selected.
A.4 Interview with Paulien Hoogvorst 111

A.4 Interview with Paulien Hoogvorst

Interview with Paulien Hoogvorst


Company Witteveen+Bos
Subject Energy supply
Date 01-03-2016

Paulien works at Witteveen+Bos in the group electrical and ICT infrastructure. Her
group often designs energy supplies for hydraulic structures for the purpose of operation
and control. Their goal is to provide hydraulic structures with more sustainable energy.

A.4.1 Does a self-generated energy supply lead to improved reliability?

It is correct that availability and reliability are very important in hydraulic structures.
Reliability is the probability that a system will not fail during operation. Availability
represent the probability that a system operates properly when needed given it has not
failed or undergoing maintenance. The group of Paulien usually develops energy supplies
for movable bridges and sluices. Here there is always an emergency generator or energy
storage available for redundancy. Looking at pumping stations there is already some
redundancy in the system because there are multiple pumping stations that can drain
the water system. Self-generated energy supply with batteries might increase reliability,
because the system is no longer sensitive to regional power cuts.

A.4.2 Which power supply configurations do you consider in design?

There are three options we consider when applying self-generated energy at a hydraulic
structure. First there is energy neutral, which means there is net connection, but yearly
averaged the use balances with the generation. In a hybrid system only a certain per-
centage of the use is self-generated. Finally with the autonomous option the system does
not have a grid connection. In this case also energy storage will be needed. Last year a
student performed a study at Witteveen+Bos for which he made an analysis tool (Bierling
2015) to determine which of the above options is optimal for a certain hydraulic structure.
In this tool the demand pattern of the structure and supply pattern of an energy source
are compared.

A.4.3 Is it realistic to self-generate energy for a pumping station with


a capacity of about 1 MW?

This is a large energy demand but it is not impossible to self-generate it. Also a consider-
able storage capacity is needed then. I drafted five energy supply scenarios: fixed energy
price, day and night tariffs, hourly fluctuating price, hourly fluctuating price + storage
and self-generated + storage. Paulien thinks that all scenarios are worth comparing and
she thinks the last three are the most interesting.
112 Expert interviews

A.5 Interview with Davı́d Brakkenhoff

Interview with Davı́d Brakkenhoff


Company Witteveen+Bos
Subject Hydrology
Date 08-03-2016

Davı́d works at Witteveen+Bos works at Witteveen+Bos in the water management group


and in the group geo-hydrology I presented Davı́d with Figure A.1, the hydrological
model that I build based on The British Geographer (2016), consisting of reservoir stocks
connected by hydrological processes. I asked his opinion on this model and how it could
be improved.

Figure A.1: Hydrological model at 08-03-2016

A.5.1 Is my proposed model correct and complete?

The model is a good start. Note that form the unsaturated soil stock a transpiration therm
should be included, while in the there is only evaporation form the interception stock.
Also note that quite some polders in the Netherlands have polder pumping stations and
drainage canal pumping stations. In that case an extra fast reacting reservoir stock should
be added. The model could be improved by comparing it to Figure A.2 FLEX hydrological
model from Fenicia et al. (2006), but note that this model is not developed for polders.
In fact there does not exist a standardized hydrological model for polders. For the TU
Delft course CIE4431 - Hydrological Modeling that Davı́d took he built and calibrated
a hydrological model for the Hesperange Catchment Brakkenhoff (2013), building on
the theory of Fenicia et al. (2008). It appeared that including interception made no
significant changes to the calibration of the model. Introducing preferential recharge did
significantly improve model calibration. This preferential recharge represents a bypass of
A.5 Interview with Davı́d Brakkenhoff 113

the unsaturated soil via cracks and roots. Given the limestone geology it makes sense
that this therm plays a role.

Figure A.2: FLEX model. Source: Fenicia et al. (2006)

A.5.2 Which parts of the model are dominant?

The presented theory in Fenicia et al. (2008) is also adequate for building a hydrological
model in the Netherlands, although is should be kept in mind that different hydrological
processes might be dominant. Looking at the Dutch polders interception can most likely
not be neglected. Because there is nearly no surface slope more precipitation is inter-
cepted. Preferential recharge on the other hand can most likely be neglected because of
the soft subsoil in the Netherlands there is very little bypassing of the unsaturated soil.

A.5.3 What is an appropriate time step for model simulation?

That is a difficult question, because the different hydrological processes have their own
characteristic time scale. Surface runoff is on the time scale of minute to hours, while
groundwater flow could take days to weeks. Considering the fact that the model should
be simulated for the entire lifetime of a pumping station it make no sense to go much
smaller than the hours.

A.5.4 How can land use (change) be included?

The land use for a Dutch polder could indeed be split up into urban area, agricultural
land and forest (nature). Depending on the land use the parameters and relations in the
hydrological model are different. It is a possibility to include all parameters as a weighted
average for the different land uses. In that way future land use change can be taken into
account by changing the percentage of the different land uses.
114 Expert interviews

A.5.5 Model review on 22-04-2016

A while later Davı́d reviewed my hydrological model once more. At that moment it looked
like A.3. Davı́d said that normally runoff and percolation (deep infiltration) are usually
described by lag functions. Right now these flows were described by a characteristic
emptying time, which might also be acceptable, but I should ask the water boards if how
they define these flows. I asked Davı́d how to quantify evaporation and transpiration in
a good way. He said that I could compute it form radiation, temperature and wind data,
but the KNMI also gives daily potential evapotranspiration that could be used. Also note
that in summer water is let into the water system, but this volume is never measured.
This term can make it more difficult to validate the model with historical data.

Figure A.3: Hydrological model at 22-04-2016


A.6 Interview with Jan Kwakkel & Jos Timmermans 115

A.6 Interview with Jan Kwakkel & Jos Timmermans

Interview with Jan Kwakkel & Jos Timmermans


Company TU Delft
Subject Simulation approach
Date 01-04-2016

Jos and Jan are both working at the faculty of Technology, Policy and Management at
the TU Delft. Jan’s research interest is on model-based decision support. Jos focuses on
adaptive delta management, and he is also part of my graduation comity. I presented
them with my proposed simulation approach and asked feedback on it.

A.6.1 What is your opinion on the main question: ”Can multi-objective


robust simulation method be applied to the re-design of a pump-
ing station?”

Jan states that off course it can be applied, so a more relevant question is: ”What is the
value of applying MORSO to the re-design of a pumping station?”. Jos asks about the
word ”simulation” in MORSO, because often it is just presented as multi-objective robust
optimization (MORO). Jan replies that is research MORSO is technically speaking the
correct term, because a simulation model is run for a non-linear problem.

A.6.2 What is your opinion on the adapted version of the Adaptive


Robust Design process?

Jan agrees that the scheme of Figure A.4 properly illustrates the proposed approach, but
the word ”optimization” in the Insight Optimization loop is not correct in this context.
The loop represents a vulnerability analyses that explores the design space to create
insights. Insight or learning is a more appropriate term here.

Figure A.4: Suggested research process


116 Expert interviews

Jos asked about the choice for Vensim. I reply that Vensim works well for describing a
system in software that derives a set of differential equations. Jan adds it is justifiable to
describing a pumping station in a set of differential equations to learn about the system.
Vensim is simply a convenient tool for setting up these differential equations.

A.6.3 How can robustness be measured?

Jan says that for better understanding of robust simulations I should read Herman et al.
(2015) titled ”How Should Robustness Be Defined for Water Systems Planning under
Change?” There is a robustness metric defined for every performance indicator. After
the optimization a trade off between the different performance indicators can be made.
These simulation optimizations are run from Python and this Python script is coupled to
the Vensim model. These scripts already exists and can be used.

A.6.4 How do I perform sensitivity analysis?

On sensitivity analysis the work of Saltelli is relevant and basically there are factor is-
land identification, factor prioritization and factor mapping. Factor island identification
investigates which factors are important, factor prioritization also ranks them and factor
mapping helps finding subspaces in your uncertainty space that produce a certain result.
The goal of sensitivity analysis is to determine for the different performance indicators
which factors relevant.

A.6.5 One or two case studies, and how to present them?

Everybody agrees that, if possible, two case studies should be done. It is interesting to see
if different locations yield different results. There could be different ways to present the
outcomes of the research. A interactive tool could be made that allows decision makers
to interact with the model. Jan says I should not put too much time in designing a
nice looking tool. The question is how to show the Water Board the value of model.
Jos comes up with the alternative not to bother the Water Board too much with the
complex method, but instead present them with the resulting robust design and ask for
their opinion.

A.6.6 Further tips?

Take a look at the work of Patrick Reed, he is expert on model analysis in water systems.
Try to talk to Luciano Raso, he does a PhD at Water Resources Management at the TU
Delft. He does research on Optimal Control of Water Systems Under Forecast Uncertainty.
He can help you represent a Model Predictive Control (MPC). For climate and energy
market it is easiest to represent them with prespecified scenarios. If there is no direct
feedback between the pumping station, which probably is the case, and the energy market
there is no need to model the energy market endogenously. Do not forget that possible
exogenous interferences in the water system should be taken into account, and described
in different rule sets.
A.7 Interview with Ivo Pothof 117

A.7 Interview with Ivo Pothof

Interview with Ivo Pothof


Company Deltares
Subject Decision support systems
Date 01-04-2016

Ivo works at Deltares as a specialist in industrial fluid dynamics, and is also leader of
the sustainable energy research and development program. At Deltares, being a water
and subsoil research institute, this is linked to sustainable energy innovations in water
or subsoil. Two examples of ongoing research are smart pumping strategies at pumping
stations and anticipating control of smart thermal grids at the TU Delft.

A.7.1 Will we see more thermal grids coming up in the coming years?

There is a minimal scale for thermal grids to become economically feasible. Thermal
grids requires a large starting investment, and the problem right now is to find enough
financing to start large project. But a large collective thermal network has the advantage
that is also possible to connect other sustainable sources. South Holland is now seriously
considering the project Warmterotonde Programmabureau Warmte Koude Zuid-Holland
(2016), and in Amsterdam and Arhnem - Nijmegen also large collective thermal grids are
being planned. Ivo is sure at least some large thermal grids will be constructed, he is not
sure how many there will be build but he think 5 to 10 in the Netherlands. From the
first project lessons can be learned, that will determine to what scale this will finally be
enrolled in the Netherlands.

A.7.2 Which innovations involving pumping stations have the largest


potential?

A trend is that operational and pumping strategy is something that is more often already
take into account in the design phase, and that is also something my research does. Smart
pumping now being researched at Deltares, and this helps to safe energy at pumping
stations. The coupling with the flexible energy market helps to make better use of the
installed sustainable power generation. The flexible water storage is being made available
on the energy market. Those are the first innovations that will be in pumping stations.
Energy savings can be realized by replacing pump impellers of older pumping stations,
but that is not really innovation. Heat or cold supply via ATES at pumping stations
(Smart polder) has also potential, but a thermal demand has to be nearby to make this
concept interesting.
118 Expert interviews

A.7.3 Could it be interesting to use a pumping station as a turbine?

If the daily energy price fluctuating is large enough it could become interesting to start
pumping water back and forth, creating active energy storage. Ivo thinks this is not
feasible, because of a low efficiency for turbine mode and the investment cost for converting
the pumping station. I say that if a pumping station is considered that is being re-build
anyway the investment costs are not the issue. I also tell him the daily fluctuations in the
energy price are considerable. Ivo still thinks thinks it wont be feasible, but he suggest
contacting Nijenhuis ask them about efficiency for turbine efficiency. Then is would be
interesting to study when the business case this concept becomes interesting.

A.7.4 What are the different pump control strategies?

Pump control strategies are often developed in RTC-tools, which is based on C++ pro-
gramming languages, and on the input side different tools that can fed with XML files. In
newer applications the Modelica environment is used, which is a sort of generic multi do-
main solver, in which dynamic systems can be modeled Modelica transforms the problem
to differential equations and is thereby comparable to Vensim or Simulink in Matlab. To
find out if an anticipating controller has an advantage over a feedback controller, it might
give a good estimate to take historical data as if they were predictions. So than you take
a perfect prediction of the APX and rainfall, but it has to be examined to what extend
this assumption holds. For an anticipating controller it is important that it can not pump
below the minimum water level, even if that might be energetically optimal. Best way
of programming this anticipating is integrating to total rainfall, determining the number
of hours x to pump and that pumping the x cheapest hours of the day. Defining too
complicated formulations is not handy as that increases the computation time. Also in
defining the ”real world” pump controllers the challenge is making them much faster than
real time, so that multiple runs can be done to find an optimum. Finding an anticipating
controller that is simple and fast but that shows desired behavior, is really the challenge.
At Deltares they look also at non linear programming and mixed integer optimization,
which is a bit more fundamental. This means that not the best result form a lot of random
simulations is drawn, but really an optimum is sought within the defined boundaries. In
such an optimization minimal energy use, or minimal energy cost could be defined as a
target function. At this moment Deltares does not make a trade off between different
target functions. The most price efficient scenario, will probably not be the most energy
efficient scenario, but via indirect reductions this might still be the most CO2 efficient
scenario. The APX price is now demand driven, but might become supply driven. Ivo
says that looking at Germany might give a first glance at how sustainable supply might
influence the daily energy price. What Nelen & Schuurmans does at HHNK is calculating
the pumping demand for the next 24 hours and this energy demand are presented to
Eneco (Agro Energy) and they chose the pumping timing based on the APX prediction.

A.7.5 Which pump control strategies should I define and replicate in


my model?

I state I would like to define 3 or 4 different controls to compare, each with a predefined
investment cost. In this way it can be checked if on the long run the more complicated,
A.7 Interview with Ivo Pothof 119

and thereby more expensive, controls pay off. Ivo advices to start with standard feedback
controls, which are mostly proportionalintegral controls that always go to a target water
level. Next feedback control with different target water levels based on energy price. Note
that this approach could mean later response to heavy rainfall events. A feed-forward
control that uses a signal that is higher up in the system, like rainfall enables the bypass
the runoff delay in the hydrological model. In this way a quicker response to rainfall
events is possible, with a control that uses only real time data. That control only makes
sense if there is a significant delay in the runoff. Finally there is a anticipating control or
also called model predictive control (MPC), that was previously discussed.
Possible pump controls:

1. Standard feedback.

2. Variable target feedback.

3. Feed-forward.

4. Model predictive control (MPC).

Ivo is interested in the resulting controls that I will define. He wants to review them after
they are finished. We agree to meet again in about a month to discuss the results.
120 Expert interviews

A.8 Interview with A. Koffeman & C. Bakker

Interview with A. Koffeman & C. Bakker


Company Zuiderzeeland
Subject Case study Vissering
Date 04-04-2016

I was expecting to have a conversation with only Albert Koffeman, Cees Bakker, but a
total of seven persons form WB Zuiderzeeland were present, including Cees Bakker as
team-manager pumping stations Noordoostpolder (NOP) and Albertjan Corjanus team-
manager pumping stations Zuidelijk and Oostelijk Flevoland. The largest group present
were pumping station technicians Jacob Snoek (pumping station Colijn/Lovink), Simon
Gerssen (pumping station Vissering), Gerrit and Albert Koffeman (pumping station Col-
ijn), who are specialist on electrical motors and large pumps. Simon Gerssen was also
performing a feasibility study on APX based pumping. Also present were authors of the
report Zuiderzeeland (2015) Albert and Cees who were previously introduced and Willem
Zwaneveld Senior project-manager Electro.

A.8.1 Presentation Energy and Pumping stations

Abert started off with a presentation of the project Energy and Pumping stationsZuiderzeeland
(2015). Off all water boards Zuiderzeeland has the largest relative energy use in the wa-
ter system. Their pumping stations use 22,5 million kWh, with an energy bill off e2,7
million in 2014 and a CO2-emission of about 7000 ton/year. The project is split up into
the parts: More efficient pumping, less pumping, self generating energy, sustainable pur-
chase of energy. Main solutions that were considered are pumping on solar, pumping on
wind, thermal energy supply (Smart polder), buying energy locally and new pump con-
trols. Electrical motors Colijn and Lovink are currently at 92% efficiency. New induction
can be purchased up to 95% efficiency. Permanent magnet motors, which are applied in
pumping stations since a few years, even go up to 96% efficiency.

A.8.2 What are the goals for re-design of pumping station Vissering?

Zuiderzeelands wants to involve all stakeholders in this project, to include all possible
innovative ideas. Feasible innovations discovered in Zuiderzeeland (2015) are to be applied
in the re-design of Vissering in Urk in 2018. WB Zuiderzeelands want to make Vissering
the most sustainable pumping station of the Netherlands. This is translated in a target
of 30% energy use reduction. The more complex studies into potential improvements are
done, the more it appears that historical design choices were not bad at all. Achieving
these target will require an open mindset and an unconventional approach.

A.8.3 Could Vissering be a potential case study?

Cees thinks it would be interesting to apply my model to pumping station Vissering as


a case study. Especially because Vissering will re redesigned and the different design
A.8 Interview with A. Koffeman & C. Bakker 121

options have to be put into perspective. For a case study data about the water system
and pumping stations is needed, to load into the Vensim model. This data includes
rainfall, water levels and discharges. Willem says is will be possible to export a lot of
that from FEWS, to use in the model. Albert asked if the model will allow for studying
the functioning of different innovations and design choices, I tell him that that is correct.
The goal of the research is to create new insight for the water board in how to make
their pumping stations more robust and sustainable. Willem asks which water boards I
approached, I tell that I approached WB Zuiderzeeland because of the large energy use
at pumping stations and HHNK because they are already quite far with smart pumping.
Albert says that WB Zuiderzeeland went on a visit to HHNK twice to hear about the
smart pumping at HHNK. Simon asks if my model can also be used to determine the
potential energy bill reduction with APX based pumping. I reply that that is definitely
possible if the water system is properly modeled. Simon says he will be helped with a good
estimate of the potential reduction. Willem asks if I will look at different pump types in
my model, I tell him that I will, because it is easy to define a different pump, as a pump
is defined as nothing more than a capacity, pump (efficiency) curve and some measure
about fish safety. I ask if for the re-design of Vissering the pumps will be replaced. Cees
says all options are open, even a totally new pump way next to the existing ones. Budget
wise replacing pumps will be a bit more difficult as the total budget is e10 million and a
proposed payback time of 12 years. Albert adds that is innovations are applied that can
be subsidized these subsidies can be added to the budget. The project might get Stichting
DE-on (2016) financing. Willem is skeptical about pump replacement, because he thinks
the electrification will take up nearly all the budget. Cees says that the start is that all
options are open, even investing the entire budget in a single pump if that satisfies the
requirements and delivers the most sustainable pumping station.

A.8.4 Willem about the energy market

Public energy generation with solar and wind is increasing. Traditional power stations
remain for when there is no wind and sun. Energy supplier want to make profit, so they
will try to pass the cost off keeping open these power stations to the costumers by asking
a high price when there is no wind and sun. It is clear that there is now an unbalance on
the market. That that is a problem to energy suppliers is also clear. If energy suppliers
succeed in fixing the storage problem than the business case for smart pumping falls
apart. Albert replies that, whatever energy scenario we get, it is important for the water
boards to reduce energy use and CO2 emissions. Willem also adds that much of pumping
stations in Zuiderzeeland are already very efficient, near theoretical optimum. In that
case it becomes difficult to still realize optimizations with a certain payback time. Other
water boards have less optimum systems and there is more room for improvement.

A.8.5 What are the next steps for the case study?

I reply that is start with gathering data, to build the model. Next studies and simulations
can be done and conclusions can be drawn. In the end it might be interesting to present
the finding to decision maker at the water board. Willem confirms this and adds that
the purpose of the case study is to prove the working off the model/tool. He suggest
122 Expert interviews

making a tool that is not too complicated, and that is able to quickly show improvement
realized by different innovations. Albertjan says he is curious to know if hind casting with
historical data is good enough, or if it is necessary to look at actual weather predictions.
Simon adds that at HHNK they say that their predictions are 94% accurate. Willem
says that all sorts of innovations are already being investigated by different parties. Cees
concludes that that is indeed the case, but the tool of this research new in comparing the
different innovations and putting them into perspective. We conclude that Albert and
Cees will be the contacts for Zuiderzeeland. Albert will send me the presentation, the
report Zuiderzeeland (2015) and the so called ”pumping station bible” Cultuurtechnische
Vereniging (1988).

A.8.6 Tour through pumping station Vissering

Albert showed me around in the pumping station. He showed me one of the two that
natural gas engines as seen on figure A.5. With a 90 degree bevel gear the torque is
transmitted to the vertical pump shafts as shown in A.6. Vissering has three vertical
centrifugal pumps in a concrete volute housing. The two yellow colored shafts are driven
by the gas engines, while the green shaft in the middle is diesel driven. The diesel engine
dates form 1983, while the two gas engines date form 1999. These gas engines can also
produce electric energy via a dynamo, to supply to the grid. Waste heat produced by
these gas engines is delivered to neighboring industries via a thermal grid. Inside the
pumping station all sorts of pipelines and heat vessels were visible.

Figure A.5: Vissering gas engine Figure A.6: Vissering engine shafts
A.9 Interview with Friso Roest & Marcel Wauben 123

A.9 Interview with Friso Roest & Marcel Wauben

Interview with Friso Roest & Marcel Wauben


Company Witteveen+Bos
Subject Pumping stations
Date 05-04-2016

Marcel Wauben is one of my company supervisors of Witteveen+Bos, and is the head of


the group pumping stations, hydraulics and pipelines. Friso Roest is also in this group
and will design pumping station De Poel for HHNK. This conversation is in preparation
for the meeting at HHNK the following week.

A.9.1 Speaking to water boards

Marcel says I should also dare to speak to decision makers at the water boards, because
that is more about politics than engineering. Try for instance to talk to Hans Oosters,
he is the Dijkgraaf of Hoogheemraadschap van Schieland en de Krimpenerwaard, but has
recently become the chairman of the UvW. Since the UvW has recently signed the Green
Deal this research is relevant for them. Peter Hermans is also an interesting person to
talk to. Nowadays he works at Witteveen+Bos, but he has been board member of WB
Groot Salland in the past. WB Zuiderzeeland and HHNK say they have the intention to
become more sustainable, but talking to decision maker could clarify how concrete their
plans are.

A.9.2 How can we make pumping stations sustainable?

Marcel states that making pumping stations energy neutral is a good ambition. Solar,
wind and thermal energy supply at the pumping station are options to realize this ambi-
tion. Pumping station Vissering has an installed capacity of 3 MW. Solar panels on the
roof a pumping station alone can not provide such a capacity. A wind mill has a large
enough capacity, but than difficulties in spatial integration arise. The wind mill does not
have to be right next to the pumping station, but it would contribute to the sustainable
image of the WB. A correlation between pumping demand (rainfall) and supply availabil-
ity (wind and sunshine) should be investigated. Marcel remarks that rain fronts are well
predictable, but for local showers it is much more difficult to predict where they fall. Not
only water quantity, but also water quality is important. Friso says pre-pumping should
be seen as a trick that is not always beneficial for the water quality and quantity, if for
instance the weather prediction is wrong.

A.9.3 Could it be interesting to use a pumping station as a turbine?

In dry periods pumping stations have to let in water. This potential energy is currently
wasted. When a pump is installed that can also function as a turbine this energy can be
harvested. If the daily energy price fluctuating is large enough it could become interesting
to start pumping water back and forth, creating active energy storage. Marcel says this
124 Expert interviews

idea is in fact a smaller version of Plan Lievense (1880). Because of the small height
difference Marcel doubts if the efficiency will be high enough for the concept to be feasible.
The concept is also limited by water quality differences in and outside the drainage canal.
Friso suggest a thought experiment: Lets assume efficiency of pump and turbine mode
are both 70%, than the total storage efficiency is 49%. This means the energy selling
price should at least be double the purchasing price.

A.9.4 Model based versus traditional design

Friso says that if I want to explain the model to water boards it should not be too complex,
otherwise it will miss its target. Friso is curious what results the model will yield. He
hopes it can deliver new insights compared to traditional pumping station design. Friso
explains that traditional design bases the pump capacity on rules of thumb and historical
experience. For different capacity classes different pumps are available and based on
location and on the preference of the water board a pump is chosen. For the largest
pumping stations horizontal axial flow pumps and bulb pumps are available. For mid
range pumping station water boards for instance say: ”We don’t want a screw pump or
a concrete volute pump, we think a horizontal or vertical axial-flow pump fits best here”.
Next the best available matching technology and motor is chosen, and that usually results
in a variable-frequency drive. That is bluntly spoken how traditional pumping station
design goes.
A.10 Interview with Ernst Moerman & Coen Rood 125

A.10 Interview with Ernst Moerman & Coen Rood

Interview with Ernst Moerman & Coen Rood


Company HHNK
Subject Case study De Poel
Date 11-04-2016

Together with my colleague Friso Roest (Witteveen+Bos) I visited Erst Moerman and
Coen Rood at the Water Board Hoogheemraadschap Hollands Noorderkwartier (HHNK).
Ernst works at the engineering department of HHNK, and is the project leader for the
new design of pumping station De Poel. Coen Rood is specialist water management and
is team leader water level regulating structures. Friso will be doing the technical design
for De Poel.

A.10.1 What aspects are most important for HHNK in pumping station
design?

Ernst says water level management is the main function of a pumping station, thus
water quantity and water safety is most important for HHNK. After that the aspects
sustainability, fish safety, and life cycle costing (LCC) will be taken into account. Fish
safe or fish friendly pumps are larger because they have larger impeller openings for the
fish to pass. Friso says that since fish safe pumps are larger they are also a bit more
expensive and could have a lower efficiency (up to 10%). For ecologists fish safety is of
high priority. Fish safety also regulated by law. Note that if there is little potential fish
migration at a considered location it might not be worthwhile applying a fish safe pump.
Coen says that above a capacity of 1 m3 /s HHNK applies two pumps for redundancy.
Having multiple pumps also creates flexibility in pump strategies. Friso says applying a
small and a large pump can even increases this flexibility. But than again having more
and thereby smaller pumps has a negative impact on fish safety.

A.10.2 Can you tell more about your innovation smart pumping?

Smart pumping is now being enrolled for most of the pumping stations in the HHNK water
system. All drainage canal pumping stations of HHNK are controlled by ControlNEXT.
ControlNEXT gives an advice to the operators how many m3 of water to pump from the
drainage canal. Ernst and Coen admit that they are both not a specialists on the topic.
Chris Smit (HHNK) or Thomas Berends (Nelen & Schuurmans) can be contacted to learn
more about the topic.

A.10.3 Could De Poel be an interesting case study?

De Poel is possibly an interesting case study. A completely new pumping station will be
designed near to the original one. The design process currently ongoing, so recommenda-
tions from my case study could possibly still be implemented. Pumping station Kadoelen
is the second pumping station in water system of De Poel. Kadoelen has a screw pump
126 Expert interviews

and is currently more efficient that De Poel, thereby having a pumping preference in the
pumping strategy. The number of operational hours for De Poel used to be 1700 hours per
year. The case study can investigate if a change in the pumping strategy for De Poel and
Kadoelen might be interesting, thereby changing the number of operational hours for both
pumping stations. It is prespecified that the pumping capacity of De Poel will remain the
same in the new design. The case study can prove if this chosen capacity is optimal. Coen
says a rule of thumb for pumping capacity is 10 m3 /min per 100 ha, which is the same as
14,4 mm/day for rural areas. The water lever margin in the drainage canal of De Poel is
quite strict, since it is in a peat area that is prone to land subsidence. Coen says smart
pumping controls based on system modeling are interesting, but the ”fingerspitzengefhl”
of operators should not be underestimated. Coen thinks smart pumping controls are only
effective if operators can understand them, otherwise operator will overrule them without
knowing the consequences for the water system. This year pumping station Cees Mantel
will be completed near Schardam. This is a new pumping station in that water system,
so a pumping strategy still has to be determined. Coen states that a case study for this
water system could be interesting. I answer that it definitely is interesting, but I prefer
the case study of De Poel because this one still has to be designed. We all conclude that
the case study for De Poel will be done, and should be seen as a parallel project next to
the actual design of the pumping station.

A.10.4 What sustainable innovations is HHNK developing?

Ernst says last month the Green Deal Energy UvW et al. (2016) between the UvW,
STOWA and the government was signed. HHNK is sponsor of this deal, and is also
developing a number innovative projects. Smart pumping is a first example of these
innovations. HHNK produces biogas from sewage sludge. At this moment 45% of the
available sewage sludge is being transformed into biogas. In the coming years this capacity
will be expanded. HHNK is also starting with solar energy, and PV cells at pumping
station are a possible scenario. HHNK is also supporting application of the Sustainable
GWW Approach GWW (2012) in their projects. It might be interesting to look into
GWW (2012) for my research.
A.11 Interview with Ronald van Nooyen 127

A.11 Interview with Ronald van Nooyen

Interview with Ronald van Nooyen


Company TU Delft
Subject Decision support systems
Date 13-04-2016

Ronald educates the course of Operational Water Management, and is focused on control
theory. Thereby Ronald operating on the interface of mathematics and civil engineering.

A.11.1 Do you know the research Kuipers et al. (2016)?

Kuipers et al. (2016) titled ”Energy and cost reduction at drainage and polder pumping” is
a project proposal for a research to be conducted by Nelen & Schuurmans, TU/e, Deltares,
WB Zuiderzeeland and the eRisk Group. Ronald is not familiar with this research, but
wonders why TU/e is involved instead of TU Delft, as TU/e has no civil engineering
faculty. TU/e has quite some knowledge on electrical motors behavior though. It is
possible that this is an occasional coalitions.

A.11.2 What pumping station related research is TU Delft doing?

Gerrit Schoups is researching stochastic networks for simulating systems. Some PhDs
are looking into reservoir management in East Asia. Some PhDs are busy with model
predictive control (MPC) for different water systems. Ronald himself has lately been
occupied with sewer pumping stations, where preventing overflow is the first priority and
preventing problems at the water treatment plant.

A.11.3 Background on pump control

At the pumping station in IJmuiden MPC is applied. That is a very large pumping station
and in this control weather predictions, sluice operation, tides and energetically efficient
pumping are all included. Note that control is just strategy, but also in the hardware
of the pumping station it is possible to save energy. That goes as far as looking at the
energy use of the variable-frequency drive. Is is desirable to run the pump at the most
efficient turning speed, but this optimum turning speeds depends on the differential hight.
Ronald is convinced that for the very large pumping stations like IJmuiden MPC is useful
to save energy and money. For ordinary drainage canal pumping stations he doubts if
that is also the case, especially if the water level margin is small. Most water boards do
not want much variation in their water level. There are already unwanted water level
fluctuations because of wind upset, pumping control should not make these fluctuations
larger. In the past water level management was much more difficult as pumping was only
possible if there was wind to turn the wind mills.
128 Expert interviews

A.11.4 What would you advise in defining pump controls?

Ronald advices to make a distinction between normal operation and impending emer-
gencies. Because when there is large wind upset because of a storm or a heavy rainfall
event, energy optimization should not threaten water safety. I say the control should be
defined in such a way that there is no threat for water safety. Ronald says that that would
implicate that the energy optimizing control shuts itself down at some threshold. Ronald
thinks that a large water level exceedance event is much more expensive that what you
can safe with optimizing the control. The cost of exceedance off course depends on the
type of land use in the polder. Difficultly in MPC is that weather predictions for small
areas like a single polder are very bad. Especially for summer showers are hard to predict.
HHNK claims an accuracy op 94%, Ronald doubt if such high accuracy is possible, but
than the question is how HHNK defines accuracy. Ronald would prefer using historical
predictions instead of historical data assumed to be predictions. With historical predic-
tions one can also evaluate how accurately they represent the historical data. Current
controls are defined in a way that when a pump turns on in will pump fore a at least a
fixed number of hours at a fixer turning speed. At WB Rijnland a minimum of 4 hours of
pumping was defined, because of the start up cost and effort. Ronald thinks the standard
feedback control with only a target level are no longer used for drainage pumping stations.
I say that there is a standard feedback control at the NOP of WB Zuiderzeeland. The
NOP is quite different from other polders in the Netherlands. The NOP, with a size of
48 km2 was namely reclaimed from the IJsselmeer (former Zuiderzee). The NOP is one
big polder and has no smaller polders discharging into a drainage canal, as is the case in
most of the Netherlands. The tree pumping stations in the NOP can be seen as very large
polder pumping stations. This should be seen as a different case, compared to other water
systems including a drainage canal. The hydrological model is not only of importance for
the rainfall to runoff delay, but works also as a low-pass filter. That means that sudden
peaks in rainfall will be dampened a bit, and that is satisfying for the control.
A.12 Interview with Kaz Vermeer 129

A.12 Interview with Kaz Vermeer

Interview with Kaz Vermeer


Company Peeeks
Subject Imbalance market
Date 13-04-2016

Kaz is a master student Precision and Microsystems Engineering at the TU in Delft,


working part time at the start up company Peeeks. Peeeks started a year ago together
with Eneco. Kaz has been involved since the start of the company and he focuses on the
energy market.

A.12.1 What does Peeeks exactly do, and are they operating in the
APX or the energy imbalance market?

In principle Peeeks is focused on the imbalance market. As Peeeks is an energy supplier


without their own generation, they purchase energy on the APX market one day ahead.
So Peeeks operates on both the APX and imbalance market, but mainly focused on the
latter one because the price range is much higher there. Peeeks uses flexibility in energy
use of their customers on the imbalance market to save money for their customers.

A.12.2 Is my suggested pros and cons list for the APX and imbalance
market correct?

APX Market Energy Imbalance Market


+ Predicable 24 hours in advance + High price range (up to 200 e/MWh)
+ Fixed daily pattern − Unpredictable
− Small price margin (20 to 40 e/MWh) − Price peaks of short duration

Table A.2: Pros and cons of APX and imbalance market

Kaz says that this list is basically correct. At the imbalance market the risk is higher,
but potential gains are two.

A.12.3 Does Peeeks only support costumers that use energy, or also
costumers that generate or store energy?

At this moment Peeeks is only an energy supplier. Flexible energy use can be seen
as passive energy storage, where energy is stored within a companies business process.
Peeeks does not (yet) support active energy storage, where first energy is stored to deliver
it back to the network later. I tell Kaz about the possibility to generate energy at pumping
stations when letting in water. Kaz says that it technically possible for Peeeks to support
costumers that also produce energy.
130 Expert interviews

A.12.4 Is Peeeks involved in the smart pumping project of HHNK?

I tell Kaz about smart pumping at HHNK and that together with Eneco (Agro Energy)
they are using the APX market in their pumping strategy. Kaz did not know about
this project, but he replied via e-mail later: Peeeks was indeed involved in a pilot at
HHNK. They are now testing if it is also possible to run a pumping station coupled to
the imbalance market. Kaz agreed with my comment that for pumping stations that
can not switch on and off a lot the APX market is the safest option. Apparently there
are also pumping station that can switch relatively quick and seamlessly, and for those
pumping stations a coupling to the imbalance market is possible. Kaz recently also had a
conversation with WB Hollandse Delta. They also came to the conclusion that the ability
to quickly switch in the energy use is a difficult factor to overcome when water boards
want to make a coupling to the imbalance market.

A.12.5 Does Peeeks only program the supply control, or also the de-
mand control for the costumer?

This really depends on the costumer. The system requirements of the costumer are always
taken into account, but it is possible the costumer already has his own control to determine
his demand. Water boards are an example of parties that have their own demand control.
But for instant the cooling of an ice skate ring has only a sensor and the requirement of a
minimum and maximum temperature, and in that case Peeeks will program the complete
control, including the demand side.

A.12.6 How does this energy imbalance market work, and how does
Peeeks act on it?

The imbalance market is quite complex. One day ahead the distribution network operator
TenneT gathers a program of projections use and supply of all Program Responsible
parties (PV-party). This use and supply should be in balance. Next day there are always
PV-parties that do not match their projection, and that creates imbalance. TenneT
tries to balance this with their own active flexible capacity. PV-parties that deviated
from their program will pay a fine, which can either positive or negative depending on
if a party is increasing or decreasing the imbalance. Peeeks is also a PV-party that
deliberately deviates from their program, to decrease the imbalance, thereby receiving
”negative fines”. TenneT publishes the ”fine” or imbalance price of every past 15 minutes
with a 3 minute delay. Peeeks acts passively on the imbalance market by predicting
the imbalance price for the coming 15 minutes and acting on that by switching on/off
capacity at their customers. Predicting the next 15 minutes based on previous prices is
the specialty of Peeeks, and they have three employees working on machine learning to
improve their predictions. As an aggregator Peeks is a PV-party that ”speculates” on
the imbalance market for their pool of costumer, to realize a lower energy price for their
costumers. Peeeks is an energy supplier that offers two types of contracts: flex only and
fixed price. Flex only acts is coupled to the imbalance market and works on a profit
sharing basis, and in this case the costumer carries the risk. In the fixed price contract
Peeeks carries the risk. Peeeks is the first energy supplier that offers a fixed energy price
A.12 Interview with Kaz Vermeer 131

that is bellow the average APX price. This fixed price contract has the only requirement
that Peeeks is allowed to flexibly utilize the companies energy demand. The imbalance
market is also quite well explained on the TenneT website.

A.12.7 How do you think the energy markets will develop in the future?

The only thing that can be said with certainty is that a lot will change. The APX
market is now demand driven, but as more wind and solar enter the market it may
become supply driven. Nowadays the daily pattern has a morning and evening price
peak. In the future there might be more peaks on different moments of the day that
are driven by a surplus in supply, represented by negatives peaks (valleys) in the energy
price. Flexible energy use and energy storage might counteract these fluctuations, as
they have a peak shaving effect. So these fluctuations are depended on the percentage of
sustainable generation and the amount peak shaving that will be realized. The imbalance
market on the other hand depends on how accurate PV-parties can predict their supply
and demand program for the coming day. Predicting methods might improve, but the
increase in weather depend generation leads to more uncertainty. Kaz says that if the
energy system remains unaltered, while solar and wind generation increase, imbalance
prices will probably increase. Next to ECN NEV reports Schoots and Hammingh (2015)
also Stedin publishes energy scenarios that could be interesting to look at.
132 Expert interviews

A.13 Interview with Marijke Visser

Interview with Marijke Visser


Company Zuiderzeeland
Subject Hydrology NOP
Date 17-05-2016

Marijke works as a hydrologist at water board Zuiderzeeland. Marijke advised my in


parameterizing the hydrological model in Vensim to match the Noordoostpolder (NOP).

A.13.1 What input datasets should be use?

Marijke advised me to use the hourly datasets of meteobase.nl Wolters et al. (2015) as
input datasets for rainfall and evaporation. These dataset are generated by the KNMI
based on the KNMI ’14 climate scenarios Klein Tank et al. (2014). Marijke says the data
sets are based on weather station 260 De Bilt, which is outside of the NOP, but will be
applicable for the NOP. An area reduction factor should be applied to compensate for the
fact that rainfall data in one single point is used to represent the rainfall in the entire NOP.
Multiplying intense local showers over the entire polder overestimates the precipitation.
This area reduction factor is applied to intense rainfall event above a certain threshold.
The area reduction factors for the NOP can be found in Versteeg and Roelevink (2011).
Meteobase.nl also provides an universal area reduction factor calculator. Meteobase.nl
gives also Makking evapotranspiration for crops. For different types of crops a correction
factor should be applied to this evapotranspiration value.

A.13.2 Is my current hydrological model correct?

The system response looks good. Marijke thinks the magnitude of surface runoff in
the NOP is small. Instead much of the rainfall finds its way to the drainage canal via
horizontal soil drainage tubes, that discharge into ditches. These drainage tubes lay 1 m
below the surface and start discharging when the groundwater level is above this level.
Marijke advises to add these ditches and a drainage tube flow from the groundwater to
these ditches. The maximum discharge through the drainage tubes is 10 to 13 mm/day.
Water inlet is also not included. Water inlet is done in dry times to flush salty seep away
through the system. Water inlet is about 100,000,000 m3 a year, over the period half
March to half September, more exact data on this topic is not available. Furthermore the
wet area percentage is 1.5% and the interception maximum is 10 mm. Seep is between 1
and 1.5 mm/day.

A.13.3 Are there any specific (future) aspects of the NOP that should
be taken into account?

At Marknesse there are two hydraulic structures of importance, the sluice and the high
water pumping station. The sluice is opened every night to let water from the high
area down to the low area of the NOP where pumping stations Buma and Vissering are
A.13 Interview with Marijke Visser 133

located. In extreme rainfall events a high water pumping station called De Expansie,
pumps water from the intermediate area to the high area. Pumping station Smeenge
that discharges this water via the high area thereby unburdening the low area. The area
around Tollebeek and Nagele, near to Urk has problems with soil settlement. Vacuum soil
drainage is applied to locally guarantee low enough groundwater levels for agriculture.
Unfortunately vacuum soil drainage accelerates the soil settlement, which is currently 10
mm a year. It might be needed in the future to separate this area including Vissering from
the low area and create a very low area. Urbanization is not expected in the NOP, and
might it occur regulations limit the maximum discharge from urban area to 13 mm/day.
So urbanization is not topic likely to have impact in the NOP. Marijke says that an
accuracy of the hydrological model of about 10% would be fine for the purpose of this
thesis research.

minimal water level <pump1 capacity>


maximal water target and initial <pump2 capacity>
level drainage canal
wet area
<pump3 capacity>
percentage Drainage canal
water level
Beta Cr runoff time
evaporation
surface water
rain lookup Cumulative
Interception Ditches Drainage canal
surface runoff ditch runoff discharge
<Time> precipitation discharge
infiltration rho
unsaturated soil ditch runoff time <control1>
inlet
max
preferential
<month of year 0> <control2>
recharge
Unsaturated
inlet yearly total <control3>
evap lookup actual transpiration soil
seep ditches
potential percolation percolation max
<Time> Lp
transpiration
groundwater seep total
lookup Groundwater
drainage drainage depth
groundwater level
normal depth
seep groundwater drainage capacity
dry normal wet
rain24h lookup dry depth

24h precipitation dry normal wet and groundwater flow


<Time> groundwater flow
prediction vey wet time

delay time 24h precipitation 24h precipitation


prediction prediction delayed max

Figure A.7: Water system model


134 Expert interviews

A.14 Interview with Thomas van Egmond

Interview with Thomas van Egmond


Company Rijkswaterstaat
Subject Tender Vissering
Date 20-05-2016

Thomas van Egmond studied social geography. He has been working as contract manager
at Rijkswaterstaat for 9 years now. Thomas is responsible for the tender procedure of
big projects. At he moment he is temporary employed at WB Zuiderzeeland as contract
manager for the renovation of pumping station Vissering and the energy and pumping
stations project Zuiderzeeland (2015).

A.14.1 What is the goal of the renovation of Vissering?

Zuiderzeeland wants to build the most sustainable pumping station of the Netherlands.
It should not only be sustainable after opening, but should remain that for a long time.
Therefore a robust design or an adaptive plan is needed. I ask Thomas how sustainability
is measured, in terms of CO2 or energy use reduction? Thomas says this is not clearly
defined by Zuiderzeeland yet. A target of 30% energy use reduction was set, but it is not
clear if this target is feasible. Thomas also believes that in the light of climate change
CO2 reduction should be the dominant steering factor. Only after that should come the
factors energy use reduction and energy bill. I present Thomas with the four performance
indicator of my research and ask how he sees the order of importance. Thomas says
that off course water level management in the main function of a pumping station. For
Zuiderzeeland energy use and sustainability comes next. Nevertheless ecology and fish
safety should be taken into account and meet the standards. For a future-proof pumping
station a proper maintenance and operational plan is as important as the design.

A.14.2 Are there plans to generate energy at Vissering?

Thomas confirms that there are plans for PV cells to generate electricity at Vissering.
Energy generation is funded through the energy and pumping station project and is
separate form the renovation project. I tell Thomas that although it are two separate
project, they can not be handles fully separately. The renovation should take into account
the power supply. Supply and demand matching in important, especially as the net energy
metering regulation (salderen) is in all probability abolished by the government in 2020.

A.14.3 What is already specified for the renovation of Vissering?

The starting point for the renovation was the electrification of the two gas engines and
the diesel engine. The budget for the project is e10 million. For electrification of all three
pumps, the expected cost is e7 million. That means e3 million remains for sustainability
purposes. But is not determined that all three pumps have to be electrified now. A
new and more efficient pump line is also an option. A new pump line with the same
A.14 Interview with Thomas van Egmond 135

capacity as the other pumps of 13.3 m3 /s is expected to cost the full e10 million. There
a doubts if the concrete foundation which originates from 1940 is still of good quality. If
not it might be necessary to renovate the foundation or build a completely new pumping
station. The latter one is expected to cost between e30 and e40 million, and in both
cases extra budget would be required. For the electrical motors the choice is between
traditional induction motors, the more efficient permanent magnet motors or the unproven
but promising reluctance motors. To make a good consideration between these motors
the weighing between cost versus efficiency improvement should be put into perspective
to the potential efficiency improvident at other parts of the pumping station, like the
pumps. Because so many options are still open, first a study into the most cost effective
measures to make Vissering more sustainable is needed. Only after that requirements for
the renovation tender can be properly specified.
136 Expert interviews

A.15 Interview with Rudolf de Vetten

Interview with Rudolf de Vetten


Company Agro Energy
Subject APX energy market
Date 24-05-2016

Rudolf works at Agro Energy, a daughter company of Eneco, that is focused at flexible en-
ergy use at costumers. They are both product developers for flexible energy use software.
This also requires thorough knowledge of the energy market, which was also discussed in
this interview.

A.15.1 What is the relation between Agro Energy and Eneco?

Agro Energy was funded by Rabobank and LTO, and later taken over by Eneco. Com-
pared to other energy suppliers Eneco has a lot of installed solar and wind power and less
traditional power plant. To meet demand for their costumers they need to buy and sell a
lot on the APX market. Another way is to try to fix this supply and demand mismatch
together with costumers that are flexible in their demand. Agro Energy help Eneco in
matching their solar and wind supply with costumer demand, but still Agro Energy has to
purchase a lot at the APX market. Agro Energy a lot of greenhouse farmers as costumers,
but is also starting to focus on water boards.

A.15.2 How does the purchase of energy work?

The closer to the moment of use, the more expensive electricity becomes. At the APX
market one does the bidding for the next day at 12:00. It is possible to bid for a volume
of energy ”at any price” or bid at the certain price. For the latter one, it is necessary to
forecast APX prices for the next day to make proper bids. If too much or too little is
purchased, TenneT will charge one the imbalance price. With wind one does a prediction
for the next day, but the wind can suddenly drop and than imbalance is created. Agro
Energy tries to solve that imbalance with consumers that have an energy demand that is
flexible and can quickly be turned on and off.

A.15.3 What is the potential of flex for pumping stations?

Large pumping station consume a lot of power and are within the time of a weather
forecast quite flexible, which makes them interesting for Agro Energy. On the other hand
the ability to quickly be turned on and off is often not there, because for instance first
the suction line has to be filled with water via a smaller vacuum pump. For Agro Energy
it would be if a costumer is able to turn an installation on within a minute and if there
is no problem with turning if off within 5 minutes again. The Pilot at HHNK is based
on APX, so a day ahead strategy based on the APX prediction and the KNMI weather
prediction of 03:00 that leads to water level predictions from ControlNext that HHNK
delivers. Aspect that are also taken into account are installation characteristics like start
A.15 Interview with Rudolf de Vetten 137

up time, minimal running time, water level margin and pump capacity. During the next
day ControlNext checks every hour if the chosen strategy and actual precipitation will not
cause a water level margin exceedance. If needed ControlNext deviates form the chosen
strategy. The next step would be perform the optimization at every hour instead of just
once a day, for that trading at an intra-day market is needed. ETPA is a company that is
currently setting up such an intra-day market. The price profile for ETPA is expected to
combine the APX and the imbalance price. Rudolf thinks the optimal situation is when
a base strategy is base on APX day-ahead market and making corrections during the day
on the intra-day market.

A.15.4 How will the energy market develop?

The introduction of solar and wind power creates fluctuations in energy supply. Electricity
price is currently demand driven, but as more wind and solar enter the market it may
become supply driven. Nowadays the daily pattern has a morning and evening price
peak. In the future there might be more peaks on different moments of the day that
are driven by a surplus in supply, represented by negatives peaks (valleys) in the energy
price. Flexible energy use and energy storage will counteract these price fluctuations,
as they have a peak shaving effect. Rudolf says that if everybody would flex that there
would be no price fluctuations, but still he expects price fluctuation to increase in the
coming years as he thinks that more solar and wind capacity will be installed compared
to flex and storage capacity realized. The imbalance market on the other hand depends
on how accurate PV-parties can predict their supply and demand program for the coming
day. Rudolf says that nobody hold the through about future energy prices, so everyone
is thinking in terms of scenarios. With a range of scenarios it is possible to investigate
the risk and opportunities for a certain business process.

A.15.5 How can pumping stations reduce CO2 emissions?

The traditional way of reducing CO2 emissions is the purchase of green electricity via
Guaranty of Origin (GvO) certificates. Sustainable energy producers can sell these GvO
certificates per MWh. The time of generation and the time of use is decoupled. A problem
with GvO certificates is that for instance the Scandinavian countries with a lot of hydro
power sell most of their certificates to other countries, because these countries already
met their EU sustainable generation goals. For imported GvO certificates from abroad it
is accepted that one should take the emission factor of ”gray electricity”, because these
GvO certificates do not contribute to making the Dutch energy supply more sustainable
co2emissiefactoren.nl (2016). An alternative I propose to express the CO2 emission on
the basis of the current CO2 intensity of the energy mix. Then companies are able to
use their flexible capacity to promote the use of sustainable energy. Rudolf says that
Agro Energy had the same idea and he agrees that it stimulates the energy transition
by preventing congestion and creating a higher price for sustainable electricity. Rudolf
remarks that one should look out what claims you make about CO2 reduction, because
GvO certificates are at this moment binding and not the current energy mix. Rudolf says
that the energy mix through time can be downloaded from Entsoe (2016).
138 Expert interviews

A.16 Interview with Wouter Zomer

Interview with Wouter Zomer


Company BZ Innovatiemanagement
Subject Stichting Gemaal van de Toekomst
Date 31-05-2016

Wouter and his colleagues of BZ Innovatiemanagement are the initiators of Stichting


Gemaal van de Toekomst (SGvdT) , translated as foundation for the Pumping station of
the Future. Since my research topic is so close me and my supervisor Marcel made an
appoint with Wouter to talk about the Pumping station of/for the Future.

A.16.1 What is the goal of Stichting Gemaal van de Toekomst?

SGvdT was established to bring different parties like government, companies, colleges
and universities together to realize innovation in pumping stations. The foundations has
distinguished four goals: Technology, Nature and Environment, Research and Analysis
and the last one Fascinate and Bind, see Figure A.8. Wouter noticed that is in water
management big data is already used a lot, but in hydraulic engineering it is not. SGvdT
wants to motivate hydraulic engineers to start using the opportunities of big data. Wouter
invigorate this with the example of the IJkdijk (gauge dike), a test dike with measuring
instruments that was loaded until it failed. Different companies were asked to predict
when the dike would fail. Different hydraulic engineering companies has used their expe-
rience to make a good prediction, but in the end the best prediction came form a washing
machine builder named Siemens, that had just looked very good at the data. SGvdT will
first focus on the north of the Netherlands, because there is an opportune environment
for big data. The focus will also start with sewage pumping stations, like the one shown
in Figure A.9, because they are managed by the city councils instead of water boards.
Wouters says that city councils are more open for innovation than water boards, and also
the risks are lower in sewage pumping stations.

Figure A.8: SGvdT goals. Figure A.9: Gemaal van de Toekomst.


A.16 Interview with Wouter Zomer 139

A.16.2 How will SGvdT achieve these goals?

SGvdT is still in a starting phase, but intends to bring together parties that want develop
a sustainable pumping station of the future. These parties will sign a memorandum of
understanding, that expresses a convergence of will to develop this pumping station of the
future together. This will be achieved via research and pilot projects. Wouter says that
water boards have a lot of experiential knowledge, that should not be underestimated.
On the other hand water boards are often blinkered because of their experience. Wouter
emphasizes that sometimes it is needed to take a step back and take time to look at the
bigger picture. A student who is new in the branch is per definition not blinkered, and
thereby it far easier for a student to make this step back. That is why SGvdT want
students to perform research into the pumping station of the future. Wouter says that
my research could for example have been a SGvdT project.
Appendix B

DNV GL energy scenario graphs

B.1 Merit order plots

The merit order is the relation between price and (residual) load. As load increases
also power plants with higher marginal cost will be fired up. Renewables have near zero
marginal costs, and are thus at the bottom of the merit oder. The merit order here shown
only the residual load, as the varying supply of sustainable would disturb the graph.

Figure B.1: 2015 merit. Figure B.2: 2030 High CO2 price merit.
Source: APX (2016) Source: Buck et al. (2014)

Figure B.3: 2030 sustainable merit. Figure B.4: 2030 vision 3 merit.
Source: Buck et al. (2014) Source: DNV GL (2015)

141
142 DNV GL energy scenario graphs

B.2 Price versus CO2 intensity plots

The energy supply mix has a corresponding CO2 intensity. The merit order effect, com-
bined with the fact that renewables have near zero marginal costs, would suggest that
cheap energy is also clean energy. To investigate this the price versus CO2 intensity is
plotted below.

Figure B.5: 2015 Figure B.6: 2030 High CO2


Price vs CO2 intensity. Price vs CO2 intensity.
Source: Entsoe (2016) Source: Buck et al. (2014)

Figure B.7: 2030 sustainable Figure B.8: 2030 vision 3


Price vs CO2 intensity. Price vs CO2 intensity.
Source: Buck et al. (2014) Source: DNV GL (2015)
Appendix C

Python simulation code

Python was used therefore to simulate and analyze the model. Python executes simula-
tions by loading the ingoing input uncertainties and design choices to the PSST model,
then Vensim runs a simulation and Python extracts and stores the outgoing performance.
How the Python script does this, is described in the following pages.

143
144 Python simulation code

PSST_Python_simulation_code

September 27, 2016

In [ ]: '''
Created on Sep 1, 2016

authors: evandruten & jhkwakkel

subject: simulation file for Pumping Station Simulation & Testing model
'''

0.1 Loading required packages


In [ ]: from __future__ import (absolute_import, print_function, unicode_literals)

import openpyxl

import numpy as np

import os

import sys
sys.path.append(r'D:\OneDrive\Documenten\TU Delft\CIE5060-09 MSc Thesis\
Python\EMAworkbench-master_20-09-2016\src')
import ema_workbench

from ema_workbench.em_framework import (ScalarOutcome,


CategoricalParameter,
RealParameter,
Constant,
Category,
IntegerParameter,
perform_experiments,
TimeSeriesOutcome)
from ema_workbench.util import ema_logging
from ema_workbench.connectors.vensim import VensimModelStructureInterface
from ema_workbench.em_framework.samplers import sample_levers
from ema_workbench.util.utilities import save_results

1
145

0.2 Creating function for data import


In [ ]: class readin_excel(object):
def __init__(self, wd='./models', file_name='data_small.xlsx'):

fn = os.path.join(os.path.abspath(wd), file_name)
if not os.path.isfile(fn):
fn = file_name

self.wb = openpyxl.load_workbook(fn, read_only=True)

self.sheets = {'energy':self.wb.get_sheet_by_name('Energy'),
'climate':self.wb.get_sheet_by_name('Climate')}

self.s = '{}4'
self.e = '{}43827'
self.time = self._read_data('climate', 'A')

def _read_data(self, sheet, column):


data = tuple(self.sheets[sheet]
[self.s.format(column):self.e.format(column)])
data_values = []
for row in data:
for cell in row:
data_values.append(cell.value)
data_values = np.array(data_values)
return data_values

def __call__(self, sheet, column):


data_values = self._read_data(sheet, column)
return zip(self.time, data_values)
readin_excel = readin_excel()

0.3 Loading the data from Excel


In [ ]: climate = 'climate'
energy = 'energy'

hour_of_day = readin_excel(climate, 'C')


day_of_week = readin_excel(climate, 'D')
month_of_year = readin_excel(climate, 'E')
solar_radiation = readin_excel(climate, 'H')
U10_onshore = readin_excel(climate, 'I')
APX_2011_2015_price = (energy, 'B')
APX_2011_2015_CO2 = (energy, 'C')
BaU_2030_price = (energy, 'D')
BaU_2030_CO2 = (energy, 'E')
CO2tax_2030_price = (energy, 'F')

2
146 Python simulation code

CO2tax_2030_CO2 = (energy, 'G')


sustainable_2030_price = (energy, 'H')
sustainable_2030_CO2 = (energy, 'I')
vision3_2030_price = (energy, 'J')
vision3_2030_CO2 = (energy, 'K')
vision4_2030_price = (energy, 'L')
vision4_2030_CO2 = (energy, 'M')
rain_2011_2015 = (climate, 'F')
rain24h_2011_2015 = (climate, 'G')
evap_2011_2015 = (climate, 'K')
rain_2030_lower = (climate, 'L')
rain24h_2030_lower = (climate, 'M')
rain_2030_center = (climate, 'N')
rain24h_2030_center = (climate, 'O')
rain_2030_upper = (climate, 'P')
rain24h_2030_upper = (climate, 'Q')
evap_2030 = (climate, 'R')
rain_2050_GL = (climate, 'S')
rain24h_2050_GL = (climate, 'T')
rain_2050_GH = (climate, 'U')
rain24h_2050_GH = (climate, 'V')
rain_2050_WL = (climate, 'W')
rain24h_2050_WL = (climate, 'X')
rain_2050_WH = (climate, 'Y')
rain24h_2050_WH = (climate, 'Z')
evap_2050 = (climate, 'AA')

0.4 Setting up the simulation according to the XLRM framework


In [ ]: def end_state(values):
return values[-1]

class LazyCategory(Category):

@property
def value(self):
if self._value == None:
values = []
for arg in self.args:
values.append(readin_excel(*arg))
self._value = values
return self._value

def __init__(self, name, args, multivalue=False):


self.args = args
self._value = None
self.name = name
self.multivalue = multivalue

3
147

if __name__ == '__main__':

ema_logging.log_to_stderr(ema_logging.INFO)

model = VensimModelStructureInterface('PSST', wd=r'./models',


model_file = r'/PSST.vpm')

# (X) External factors and/or Uncertainties and/or States of the World


energy_cats = [
LazyCategory('APX_2011_2015',
[APX_2011_2015_price, APX_2011_2015_CO2]),
LazyCategory('BaU_2030',
[BaU_2030_price, BaU_2030_CO2]),
LazyCategory('CO2tax_2030',
[CO2tax_2030_price, CO2tax_2030_CO2]),
LazyCategory('sustainable_2030',
[sustainable_2030_price, sustainable_2030_CO2]),
LazyCategory('vision3_2030',
[vision3_2030_price, vision3_2030_CO2]),
LazyCategory('vision4_2030',
[vision4_2030_price, vision4_2030_CO2])]

climate_cats = [
LazyCategory('2011_2015',
[rain_2011_2015, rain24h_2011_2015, evap_2011_2015]),
LazyCategory('2030_lower',
[rain_2030_lower, rain24h_2030_lower, evap_2011_2015]),
LazyCategory('2030_center',
[rain_2030_center, rain24h_2030_center, evap_2030]),
LazyCategory('2030_upper',
[rain_2030_upper, rain24h_2030_upper, evap_2030]),
LazyCategory('2050_GL',
[rain_2050_GL, rain24h_2050_GL, evap_2050]),
LazyCategory('2050_GH',
[rain_2050_GH, rain24h_2050_GH, evap_2050]),
LazyCategory('2050_WL',
[rain_2050_WL, rain24h_2050_WL, evap_2050]),
LazyCategory('2050_WH',
[rain_2050_WH, rain24h_2050_WH, evap_2050])]

CO2_cats = [
Category('averaged CO2 intensity', 0),
Category('realtime CO2 intensity', 1)]

4
148 Python simulation code

model.uncertainties = [
CategoricalParameter('Energy scenario', energy_cats,
variable_name=['price lookup',
'CO2 lookup']),
CategoricalParameter('Climate scenario', climate_cats,
variable_name=['rain lookup',
'rain24h lookup',
'evap lookup']),
CategoricalParameter('CO2 switch', CO2_cats),
RealParameter('Beta', 0, 0.5),
RealParameter('rho', 0, 1),
RealParameter('Lp', 0, 1),
RealParameter('wet area percentage', 0.01, 0.05),
RealParameter('unsaturated soil max', 40, 800),
RealParameter('percolation max', 0.00416, 0.0313),
RealParameter('runoff time', 0.024, 96),
RealParameter('groundwater flow time', 480, 1920),
RealParameter('ditch runoff time', 1, 12),
RealParameter('drainage capacity', 0.417, 0.542),
RealParameter('drainage depth', -1200, -800),
RealParameter('seep total', 0.0188, 0.0227),
RealParameter('inlet yearly total', 0.0186, 0.0203),
RealParameter('delay time prediction', 0, 24),
RealParameter('capacity per pump', 12, 13.33),
RealParameter('pump efficiency original', 0.7, 0.85),
RealParameter('pump efficiency increase by inflow improvement', 0.01, 0.03),
RealParameter('pump inflow improvement investment per pump', 20000, 40000),
RealParameter('water level IJselmeer', -400, -100),
RealParameter('responsible for percentage of total area', 0.3, 0.7),
RealParameter('induction motor efficiency', 0.94, 0.96),
RealParameter('PM motor efficiency improvement', 0.01, 0.02),
RealParameter('induction motor investment per motor', 1.5e+006, 2.2e+006),
RealParameter('PM motor investment savings per motor', 0.05, 0.25),
RealParameter('smooting time smart variable', 48, 168),
RealParameter('cut fraction', 0.1, 0.5),
RealParameter('tax and network costs', 0.01381, 0.03471),
RealParameter('U100 over U10', 1, 1.5),
RealParameter('SDE solar', 0.090, 0.128),
RealParameter('SDE wind', 0.076, 0.082),
RealParameter('cost of solar', 900, 1200),
RealParameter('cost of wind', 1000, 1500),
RealParameter('social cost per ton CO2', 0.033, 0.2),
RealParameter('cost of flooding', 1, 10),
RealParameter('direct flooding cost ratio', 0, 1),
RealParameter('potential fish migration rate', 0.167, 2),
RealParameter('weight per fish', 0.5, 1),
RealParameter('cost of fish damage', 12, 15)]

5
149

# (L) Policy Levers and/or design Alternatives


motor_cats = [
Category('induction motor', 0),
Category('permanent magnet motor', 1)]

pump_cats = [
Category('original inflow', 0),
Category('improved inflow', 1)]

scenario_cats = [
Category('no scenarios', 0),
Category('weather scenarios', 1)]

smart_cats = [
Category('normal', 0),
Category('day night', 1),
Category('APX', 2),
Category('local', 3),
Category('APX and local', 4),
Category('CO2', 5)]

fish_cats = [
Category('original impellers', 0),
Category('change impellers', 1),
Category('new fish safe pumps', 2)]

model.levers = [
CategoricalParameter('number of pumps', [2, 3]),
CategoricalParameter('motor choice switch', motor_cats),
CategoricalParameter('pump inflow improvement switch', pump_cats),
CategoricalParameter('scenario switch', scenario_cats),
CategoricalParameter('smart switch', smart_cats),
CategoricalParameter('fish safe switch', fish_cats),
CategoricalParameter('percentage energy neutral', [0, 1]),
RealParameter('mix solar wind', 0, 1)]

# (R) Other lookups to make the Relationships in System complete


model.constants = [
Constant('solar radiation', solar_radiation),
Constant('U10 onshore', U10_onshore),
Constant('hour of day', hour_of_day),
Constant('day of week', day_of_week),
Constant('month of year', month_of_year)]

6
150 Python simulation code

# (M) Performance Metrics and/or Outcomes of Interest


model.outcomes = [
TimeSeriesOutcome('Energy bill'),
TimeSeriesOutcome('Cumulative max water level exceedance'),
ScalarOutcome('Investment costs', function=end_state),
ScalarOutcome('Operational costs', function=end_state),
ScalarOutcome('Operational costs incl indirect costs',
function=end_state),
ScalarOutcome('Energy bill endstate', function=end_state,
variable_name='Energy bill'),
ScalarOutcome('Energy purchased', function=end_state),
ScalarOutcome('Energy sold', function=end_state),
ScalarOutcome('Energy used', function=end_state),
ScalarOutcome('CO2 emitted', function=end_state),
ScalarOutcome('Social cost of CO2', function=end_state),
ScalarOutcome('Cumulative max water level exceedance endstate',
function=end_state,
variable_name='Cumulative max water level exceedance'),
ScalarOutcome('flooding cost', function=end_state),
ScalarOutcome('Cummulative fish damge',function=end_state),
ScalarOutcome('fish damage cost', function=end_state),
ScalarOutcome('Cumulative precipitation', function=end_state),
ScalarOutcome('Cumulative seep', function=end_state),
ScalarOutcome('Cumulative inlet', function=end_state),
ScalarOutcome('inflow', function=end_state),
ScalarOutcome('Cumulative transpiration', function=end_state),
ScalarOutcome('Cumulative discharge', function=end_state),
ScalarOutcome('outflow', function=end_state),
ScalarOutcome('local energy produced', function=end_state),
ScalarOutcome('local energy used', function=end_state)]

0.5 Perform and save experiments


In [ ]: nr_policies = 50
policies = sample_levers([model], nr_policies)[0]
policies = [policy for policy in policies]

nr_experiments = 1000
results = perform_experiments(model, nr_experiments,
policies=policies, parallel=True)

save_results(results, './data/{} scenarios - {} policies'.


format(nr_experiments, nr_policies))

You might also like