Available online at www.sciencedirect.

com

ScienceDirect
Energy Procedia 105 (2017) 94 – 101

The 8th International Conference on Applied Energy – ICAE2016

Optimal control of hydrokinetic-powered pumpback system

for a hydropower plant in dry season: A case study
Fhazhil Wamalwaa*, Sam Sichilalua,b, Xiaohua Xiaa
a
Centre of New Energy Systems, Department of Electrical, Electronic and Computer Engineering, University of Pretoria, Pretoria 0002,
South Africa
b
Mosi-o-tunya University of Science and Technology, Faculty of Engineering, Lusaka, Zambia

Abstract

Low hydropower generation in dry season due to low water levels in hydropower dams is a common problem, particularly
in drought-prone regions such as Southern Africa. In this paper, an optimal hydrokinetic-powered pumpback retrofit for
recycling a part of the dam’s downstream discharge to optimise the performance of the dam is proposed. The optimisation
problem is formulated as a multi-objective problem to minimise grid pumping energy, maximise the use of hydrokinetic
energy for pumping operation and maximise restoration of the dam volume through the pumpback operation. Simulation
results of the proposed model using a practical case study show the potential of the model to increase the energy yield of
the plant by 39 to 41.48% in the dry season.
© 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
© 2016 The Authors. Published by Elsevier Ltd.
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Selection and/or peer-review under responsibility of ICAE
Peer-review under responsibility of the scientific committee of the 8th International Conference on Applied Energy.
Keywords: Hydrokinetic; Hydropower ; Pumpback operation; Hydraulic head; Optimal control.

1. Introduction

Hydropower is the most matured renewable energy technology with a huge potential to substitute the
depleting fossil fuel sources. However, its main limitation is vulnerability to climate change factors [1], a
problem that has been experienced in recent years in vulnerable regions such as Southern Africa with
Tanzania, Zimbabwe and Zambia worst affected. The anticipated decrease in precipitation in Southern Africa
implies that the problem is bound to become more pronounced [1]. This problem underscores the need to re-
model the existing hydropower dams to optimise the economic value of the available water. One of the
strategies used to optimise hydropower generation in dry seasons is pumpback operation [2,3,4]. In these
references, pumpback operation is used to recycle a part of the downstream discharge back to the main dam
during the off-peak period to maintain a high water level in the dam for peak generation. However, the use
of grid power for pumping operation is uneconomical for high head applications due to high head and
pressure losses [5]. To minimise reliance on grid imported power for pumpback operation, this paper
proposes the use of hydrokinetic energy conversion (HEC) system to power the pumpback system. Optimal
control of the pumpback system has the potential to minimise the overall pumping energy demand over the
control horizon as demonstrated in [6].

* Corresponding author. Tel.: +27 617 565 168; fax: +27 12 362 5000.
E-mail address: Fazil.wamalwa@gmail.com.

1876-6102 © 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the scientific committee of the 8th International Conference on Applied Energy.
doi:10.1016/j.egypro.2017.03.285
 Fhazhil Wamalwa et al. / Energy Procedia 105 (2017) 94 – 101 95

2. Mathematics model formulation

Figure 1 shows the schematic layout of the proposed optimal control model. It comprises the conventional
hydropower system, the hydrokinetic energy conversion (HEC) system and the pumpback system. In Figure
1, P1(j), P2(j), P4(j), P5(j) and P6(j) are respectively, the power output of the hydro-turbine generator, the
power output of the HEC system, the pumping power demand supplied through a control switch u(j), the
excess hydrokinetic power exported to the grid and the grid power imported to offset pumping power deficit
in cases of low HK power generation. The fraction of HK power supplied to meet the pumping power demand
is denoted by P3(j) while ୭ ሺŒሻ and Š୳ ሺŒሻ are respectively, the net head and the depth of the dam. The
quantities, Qo(j), Qk(j) and Qi(j) are respectively, the turbine discharge for hydropower generation, the flow
rate of pump K and the in-stream discharge expressed in m3/s.

Q k (j) Grid
Qi P5 (j)

P1(j)

hu(j)
Main dam(U)

Q o (j)

P6 (j)
Ho(j) Penstock P2 (j)
Power
house P4 (j)

Gen u(j)
υ(m/s)
K

HK

Fig 1: A hydroelectric power scheme with a cascaded HK-powered pumpback system

2.1. Conventional hydropower model

The theoretical power output of a hydro-turbine generator, ܲଵ ሺ݆ሻሺ‫ܹܯ‬ǡ is a non-linear function of the
turbine discharge, Qo(j), and the head of the water fall, Ho(j), expressed as follows [2]:

P1 (j) ρgη e H o (j)Q o (j) u 106 , Qomin d Qo (j) d A c 2gH omax , (1)
96 Fhazhil Wamalwa et al. / Energy Procedia 105 (2017) 94 – 101

where ɏሺ‰Ȁଷ ሻ, ‰ሺȀ• ଶ ሻ and ߟୣ are respectively, the density of water, the gravitational acceleration and
the combined efficiency of the hydro-turbine and the generator. ‫ܣ‬௖ ሺଶ ሻ is the cross sectional area of the
penstock while ୭୫ୟ୶ is the maximum head of the system before spillage occurs. The turbine flow rate, Qo(j),
is controlled by a butterfly valve installed inside the penstock. At any given time ݆, the hydraulic head of the
system, ୭ ሺሻ, is constrained to lie within୭ ‫ א‬ሾ୭୫୧୬ ǡ ୭୫ୟ୶ ሿ where ୭୫୧୬ is the minimum head of the main
dam. For modelling simplicity, a cylindrical model with base area, ‫ܣ‬௨ (ଶ ) is assumed for the dam.
Therefore, the state dynamics of the dam can be expressed in discrete time domain as follows [7]:
ts
h u (j + 1) = h u (j) + [Qi + u(j)Q k - Qo (j)] , (2)
Au
where ݄௨ ሺ݆ሻ and ݄௨ ሺ݆ ൅ ͳሻ are respectively, the height of water in the dam at the end of discrete time period
Œ and at the end of the next time period, ݆ ൅ ͳ. ‫ݑ‬ሺ݆ሻ is a binary variable [0, 1] denoting the state of pump ‫ܭ‬
while ‫ݐ‬௦ is the system sampling time.

2. 2 The HEC system

A hydrokinetic turbine works in the same principle as that of a wind turbine except that it is powered by
the kinetic energy of the river currents. Thus the energy output of the HEC system is expressed as [8] :

P2 (j) 0.5Cp ηhkA t ρυ3 u 106 , (3)

where C p , ηhk and At (m 2 ) are respectively, the coefficient of performance, the efficiency of the HEC
system and the area swept by the HK turbine rotor while X (m/s) is the river current velocity.

2. 3 The pumpback model

The proposed pumpback system employs a constant speed pump controlled by an on/off switch, u(j).
Therefore, the electrical power, P4 (j) (MW), required to lift water through a differential head, ‫ܪ‬଴ (m) at a
discharge rate, Qk (m3/s) is expressed as follows [9]:

ρgH o Q k u(j)
P4 (j) u 10-6 , (4)
ηp
where ߟp is the combined efficiency of the pump and its drive motor. In the proposed model, the pumping
power demand will be met primarily by the HK power, ܲଷ ሺ݆ሻ. However, in case the pumping power demand
is more than the total HK power produced, then grip power import, ܲ଺ ሺ݆ሻ, will be brought in to offset the
deficit. Therefore, the power balance of the pumpback system can be expressed as follows:

P3 (j)  P6 (j) u(j)P4 . (5)

2.4. Objective function

The proposed optimal control model seeks to simultaneously minimise grid pumping energy,
N N

¦ t P ( j ) , maximise the use of hydrokinetic energy for pumping operation , ¦ t P ( j ) , and maximise
j 1
s 6
j 1
s 3

N
restoration of the volume of the dam through pumpback operation, ¦ t Q u( j) .
j 1
s k The control horizon is

24 hours with a sampling time ‫ݐ‬௦ ൌ ͲǤʹͷ݄ and the total number of samples, ܰ ൌ ͻ͸. The multi-objection
function is expressed as follows:
 Fhazhil Wamalwa et al. / Energy Procedia 105 (2017) 94 – 101 97

N N N
w1t s ¦ P6 (j)  w 2 t s ¦ P3 (j)  w 3 t s ¦ Q k u(j) , (6)
j 1 j 1 j 1
subject to the following constraints:
P1 (j)  P5 (j) Pld (j) , (7)
P3 (j)  P6 (j) u(j)P4 , (8)
P3 (j) + P6 (j) = n hkP2 (j) , (9)

h min
u d hu(j) d h max
u , (10)

P1min d P1 (j) d P1max , Qomin d Qo (j) d A c 2gH omax , (11)

P2min d P2 (j) d P2max , (12)

P6min d P6 (j) d P6max , (13)

P d P3 (j) d P
min
3
max
3 , (14)
u  [0,1] , (15)
୲୦
where Πis the Πsampling interval, nhk is a control variable denoting the number of HK generators in
3
parallel operation while ¦w
i 1
i 1 are the weighting factors that determine the relative importance of
each of the objective vectors of equation (6).

2.5 Case study

The case study is based on a 68 MW, 170 m head Pangani fall hydropower plant located on Pangani
River in Tanzania. Power generation of the plant in dry seasons has declined for the last 10 years due to low
dam water level occasioned by an increase in droughts in the region. On average, the in-stream flow rate of
Pangani river into Pangani falls dam is 12.5m3/s in dry season and 24.8m3/s in rainy season [10]. The dam
has a minimum and maximum hydropower operation capacity of ͲǤͻ ൈ ͳͲ଺ ଷ and ͳǤ͹ ൈ ͳͲ଺ ݉ଷ
respectively. In the case study, an initial water volume of ͳǤͳ ൈ ͳͲ଺ ଷ is assumed, which is a typical average
capacity of the dam in dry season. This volumetric capacity of the dam is modelled into a cylindrical model
with a base radius of 71.36 m and a depth of 68.75 m for ease of control. The CC035A river-in-stream HK
turbines manufactured by the Clean Current Renewable Energy Systems Inc. and the SJT vertical turbine
high lift pumps manufactured by Sulzar Ltd are used in the case study.

3. Results and discussions

The problem is a non-linear optimization problem solvable in MATLAB using OPTI toolbox over a 24-
h control horizon with sampling interval, – ୱ ൌ ͳͷ‹. Figure 2 shows the optimal switching operation, flow
rates and change in water level of the dam for the case of ‫ݓ‬ଵ ൌ  ‫ݓ‬ଶ ൌ  ‫ݓ‬ଷ ൌ ͲǤ͵͵. As shown in the first row
of the figure, the pump switch ‫ ݑ‬is off between 00:00 and 09:00. Afterwards, the optimal control (OC)
switches it on from 09:00 to raise the water level of the dam in anticipation of peak hydro-turbine discharge,
Qo, which occurs between 12:00 and 16:00 as shown in the far right of the first row of Figure 2. The OC
maintains the pump in operation mode for the remainder of the control horizon to keep a high water level in
the dam for optimal generation. A high water level in the dam is necessary to minimise the volume of water
discharged per each unit of hydropower generated. Also shown in the left first row figure is the pumping
power demand of pump K, which is 9.19 MW whenever the pump is in operation, otherwise it is zero. In the
far right of the first row of Figure 2, Qo denotes the penstock discharge for hydropower generation while Q -
3
inflow is the combined in-stream discharge, Qi, and the discharge of pump K, Q k in m /s. As shown, Qo varies
98 Fhazhil Wamalwa et al. / Energy Procedia 105 (2017) 94 – 101

in response to changes in P1 while Q-inflow oscillates between 12.5m3/s and 18.00 m3/s depending on the
switching instances of the pump switch, ‫[ ݑ‬Q-inflow=12.5 (18.00) when ‫ ݑ‬is off (on)].

Fig. 2. Optimal switching and flow rates of the model

In the second row of Figure 2 is shown the resultant change in the dam water level due to the turbine discharge
for hydropower generation, HQo(m), combined inflows, HQ-inflow (m), and the overall optimal level ୭୮୲ ሺሻ.
The varying dam water level shows the would-have-been dam water level scenarios; for the baseline HQo
(m), and the proposed optimal intervention, H opt (m). The difference shows the benefits of this model. The
optimal water level, Hopt (m) is a result of the combined inflows (Q k+Qi) less the turbine discharge, Qo. As
shown, without constant water inflows (Q k1+Qi), the dam water level would drop from the initial point of
176.00 m to 92.48 m at the end of the day. However, due to Q-inflow, the resultant change in the water level
of the dam, Hopt (m), drops from 176.00 to 110.40 m. The high water level in the dam, partly due to pumpback
operation, optimises hydropower generation of the dam by minimising the volume of water discharged for
each unit of power generated.
Figure 3 shows the power flows of the resultant hydropower plant retrofitted with a hydrokinetic-
powered pumpback system. P6 is zero throughout the control horizon while P 3 is zero between 00:00 and
09:00 because the pump is in off mode and as a result, all the 21.62 MW of the on-site generated HK power
is supplied to the grid. However, P 3 rises to 9.19 MW between 09:00 and 24:00 when the OC switches on
pump K to restore the water level of the dam. Correspondingly, P5 falls from 21.62 MW to 12.43 MW
between 09:00 and 24:00. On the other hand, hydro-turbine power output, ଵ ǡvaries in response to the grid
load demand, ୪ୢǡ  supplemented by on-site HK powerfed into the grid, P5. For instance, P1supplies 28.38
MW between 00:00 and 04:00 while in the same period, 21.62 MW is supplied by P5 to meet the committed
grid demand of 50.00 MW. An increase in Pld from 50.00 MW to 60.00 MW at 04:00 against a constant HK
power export, P5, of 21.62 MW results in a corresponding increase in P 1 from 28.38 MW to 38.38 MW.
Similarly, a decrease in P5 at 09:00 from 21.62 MW to 12.43 MW results in a further increase in P1 to 47.57
MW to meet the system load demand. Subsequently, the OC adjusts P1 and Qo in response to changes in Pld
and P5 in order to economise the available water in the dam over the control horizon. The inverse
complementary relationship of P1 and P5, especially as shown at 09:00 illustrates the potential of the proposed
OC to optimise the economic value of the available water in the dam. This is the major advantage of this
model over conventional existing hydropower systems.
 Fhazhil Wamalwa et al. / Energy Procedia 105 (2017) 94 – 101 99

Fig. 3. Flowrates and change in water level of the dam in dry season

Table 1 shows the daily energy flows of the proposed optimal control system on a typical day in the dry
season for different weighting factors. In the table, E3, 6 (MWh) denotes the pumping energy demand of the
model which is the summation of HK pumping energy, E3 (MWh), and grid pumping energy, E6 (MWh),
over the 24 h control horizon. The optimal energy output, E opt (MWh), is the result of summation of hydro-
turbine energy output, E1 (MWh), excess HK energy supplied to the grid load, E 5 (MWh), less the grid
pumping energy demand, E6 (MWh). As shown in the table, changes in weighting factors does not affect the
performance of the model except the case of allocating full optimisation priority to the minimisation of grid
pumping energy demand. In this case, the pumping power demand is 148.76MWh, which is the highest, with
39.82% increase in the energy yield of the resultant system. Otherwise, in all operation strategies, optimal
energy flows results in zero grid pumping energy demand, meaning the pumpback system is fully powered
by the HEC system, and an overall pumping energy demand of 137.87 MWh with 41.48% increase in the
overall energy yield of the system.

Table 1: Daily optimal energy production and demand during dry season

୧ ଵ ሺŠሻ ହ ሺŠሻ ଷ ሺŠሻ ଺ ሺŠሻ ଷǡ଺ ሺŠሻ ୭୮୲ ሺŠሻ Increase (%)
w1=w2=w3=0.33 918.89 381.11 137.87 0.00 137.87 1300.00 41.48
w2=w3=0;w1=1 929.77 370.23 148.76 0.00 148.76 1300.00 39.82
w1=w3=0;w2=1 918.89 318.81 137.87 0.00 137.87 1300.00 41.48
w2=w2=0;w3=1 918.89 381.11 137.87 0.00 137.87 1300.00 41.48

Figure 4 shows the results of the optimal switching, flow rates and changes in water level of the dam on a
typical day in rainy season. As shown in the far left of the first row, the pump switch, u, is off and as a result,
P4 is zero. In the rainy season, there is sufficient in-stream flow, Qi, (24.9m3/s) to maintain a high water level
in the dam and as a result, the OC keeps the pumpback system off. Shown in the second row of Figure 4 is
the change in water level of the dam in response to Q-inflow, HQ-inflow (m), and change in response to Qo, HQo
(m). Since the pumpback system is off, all the on-site generated HK energy is exported to the grid resulting
in low demand for P1. As shown in the far right of the first row of Figure 4, Q -inflow is far higher than Qo
between 00:00 and 04:00 and as a result, there is a slight increase in water level of the dam model as shown
in the second row of the figure. This is followed by a slight drop between 04:00 and 16:00 when Q o is slightly
higher than Q-inflow. However, the high Q-inflow as compared to Qo between 16:00 and 24:00 results in an
increase in the water level of the dam from 174.80 m to 191.30 which is way above the maximum limit of
177.50 m of the dam model; resulting in spillage. In the figure, a change in HQ-inflow from 176.00 m to 307.60
m implies that the dam will overflow if it was operated without turbine discharge for power generation, Qo,
on a typical day in the rainy season.
100 Fhazhil Wamalwa et al. / Energy Procedia 105 (2017) 94 – 101

Fig. 4. Flowrates and change in water level of the dam in rainy season

Figure 5 shows the optimal power flows of the proposed model on a typical day in the rainy season. As shown,
P3 and P6 are zero since the pumpback system is off. This results ina constant supply of 24.26 MW of HK power
to the grid throughout the control period.

Fig. 5.Optimal power flows of the system in the rainy season

Table 1 shows the energy balance of the proposed optimal control system on a typical day in the rainy season.
As shown, the pumping energy,E3,6 (MWh), is zero since the OC opts to operate the pumpback system only
in dry season. As a result, all the on-site generated HK energy of 518.98 MWh is supplied to the grid resulting
in 66.45% increase in the overall energy yield of the resultant system.

Table 2: Daily optimal energy production and demand during rainy season

୧ ଵ ሺŠሻ ହ ሺŠሻ ଷ ሺŠሻ ଺ ሺŠሻ ଷǡ଺ ሺŠሻ ୭୮୲ ሺŠሻ Increase (%)
w1=w2=w3=0.33 781.01 518.98 0.00 0.00 0.00 1300.00 66.45

4. Conclusion
 Fhazhil Wamalwa et al. / Energy Procedia 105 (2017) 94 – 101 101

Retrofitting a conventional hydropower plant with a pumpback system has the potential to boost the
energy output of the plant. This is achieved by keeping a high water level in the dam to minimise the amount
of water discharged per each unit of hydropower generated. In this paper, an on-site hydrokinetic energy
conversion system is proposed to power the pumpback system to minimise reliance on grid power for
pumping operation. The optimisation problem is formulated as a multi-objective problem to simultaneously
minimise the grid pumping energy, maximise the use of on-site generated hydrokinetic energy for pumping
operation and maximise restoration of the volume of the dam through pumpback operation. The application
of the proposed optimal control model to a case study increases the energy yield of the resultant system by
39 to 41.48% on a typical day in dry season and 66.45% on a typical day in the rainy season. The high
increase in the energy yield in the later case is occasioned by high exportation of the on-site generated
hydrokinetic energy to the grid since the pumpback system is off in the rainy season.

Acknowledgements
The authors thank the National Research Foundation (NRF) grant No.99766 South Africa, the University
of Zambia-Zambia, the MasterCard Foundation and the National Hub for Energy Efficiency and Demand
Side Management for financial and other support for this research.

References
[1] Van Valiet MTH, Wiberg D, Leduc S and Riahi K. Power-generation system vulnerability and adaptation to changes in climate
and water resources. Nature Climate Change 2016; 6:375-380.
[2] Zhao G and Davison M. Optimal control of hydroelectric facility incorporating pump storage. Renewable Energy 2009; 34:1064-
1077.
[3] Li W, Huang J, Li G and Wang Z. Research on optimizing operation of the single reservoir of hybrid pumped storage power
station. 2011 4th International Conference on Electric Utility Deregulation and Restructuring and Power Technologies.Weihai
-China: Ieee; 2011.
[4] Ribeiro AF. Optimal Control for a Cascade of Hydroelectric Power Stations: Case Study. CONTROLO’2014–Proceedings of
the 11th Portuguese Conference on Automatic Control. Lisbon: Springer; 2015.
[5] Nakayama Y, Introduction to fluid mechanics. Oxford: Butterworth-Heinemann; 1999.
[6] Zhuan X and Xia X. Optimal operation scheduling of a pumping station with multiple pumps. Applied Energy 2013;104: 250-
257.
[7] Tayebiyan A, Ali TAM, Ghazali AH and Malek MA. Optimization of Exclusive Release Policies for Hydropower Reservoir
Operation by Using Genetic Algorithm. Water Resource Management 2016; 30:1203-1216.
[8] Ashok S. Optimised model for community-based hybrid energy system. Renewable Energy 2007;32:1155-1164.
[9] Tang Y, Zheng G and Zhang S. Optimal control approaches of pumping stations to achieve energy efficiency and load shifting.
International Journal of Electrical Power and Energy Systems 2014;55:572-580.
[10] Luteganya K and Kizzy S. Hydroelectric power modelling study. Pangani river basin flow assessment, 2009.

Biography

Fhazhil Wamalwa, a Kenyan male, received the BSc. degree in electrical and
electronic engineering from Jomo Kenyatta University of Agriculture and
Technology, Kenya, in 2012 and a BEng (Hons) in electrical engineering from the
University of Pretoria, South Africa in 2016, where he is currently reading towards
a MEng. Electrical.