Environmental Modelling & Software 64 (2015) 164e176

Contents lists available at ScienceDirect

Environmental Modelling & Software

journal homepage: www.elsevier.com/locate/envsoft

A mathematical programming framework for early stage design of

wastewater treatment plants
Hande Bozkurt, Alberto Quaglia, Krist V. Gernaey, Gürkan Sin*
CAPEC-PROCESS, Department of Chemical and Biochemical Engineering, Technical University of Denmark, Building 229, DK-2800 Kgs. Lyngby, Denmark

a r t i c l e i n f o a b s t r a c t

Article history: The increasing number of alternative wastewater treatment technologies and stricter effluent re-
Received 4 June 2014 quirements make the optimal treatment process selection for wastewater treatment plant design a
Received in revised form complicated problem. This task, defined as wastewater treatment process synthesis, is currently based on
3 November 2014
expert decisions and previous experiences. This paper proposes a new approach based on mathematical
Accepted 24 November 2014
programming to manage the complexity of the problem. The approach generates/identifies novel and
Available online 12 December 2014
optimal wastewater treatment process selection, and the interconnection between unit operations to
create a process flow diagram. Towards this end, a superstructure approach is used to represent the
Keywords:
Design
treatment alternatives for reaction and separation. A generic process interval model is used to describe
Modeling each alternative in terms of inputeoutput mass balances including conversion and separation factors.
Superstructure optimization Next the design problem is formulated as a Mixed Integer (Non)linear Programming problem e MI(N)LP
Wastewater treatment e and solved. A case study is formulated and solved to highlight the application of the framework.
Process flow © 2014 Elsevier Ltd. All rights reserved.
Interconnection

1. Introduction The number of alternative processes to choose from has been

One of the most challenging steps in wastewater treatment
plant design is the selection of the treatment process defined as a
combination of unit operations and processes capable of meeting
effluent permit requirements (Tchobanoglous et al., 2003). We
suggest to call this particular task Wastewater treatment process
synthesis and define it as follows: Wastewater treatment process
synthesis is the step in the design of a wastewater treatment plant
where the design engineer selects unit processes (separation and/
or reaction including physical, chemical and biological processes)
from numerous alternatives and interconnects them to create the
process flow diagram. Hence the objective of process synthesis is to
find the best process flow diagram, among numerous alternatives,
for treating a given influent wastewater with its flow rate and
composition to meet predefined performance criteria including
effluent permit requirements as well as cost and technical
requirements.
* Corresponding author. CAPEC-PROCESS Research Center, Department of
Chemical and Biochemical Engineering, Technical University of Denmark, Building
229, DK-2800 Kgs. Lyngby, Denmark. Tel.: þ45 45 25 28 06.
E-mail addresses: handebozkurtt@gmail.com (H. Bozkurt), gsi@kt.dtu.dk
(G. Sin).
increasing steadily since the beginning of the 20th century, where
many wastewater treatment processes and technologies have been
developed to meet increasingly stringent performance demands
(Henze et al., 2008).
The current phase of development in municipal wastewater
treatment technologies was initiated with the stricter effluent limit
values imposed by both emission and immission based regulations.
In EU, for example, the Water Framework Directive (Directive 2000/
60/EC) regulates the ecosystems by setting quality objectives and
the urban effluent wastewater quality is controlled by the Urban
Wastewater Treatment Directive (91/271/EEC). A wide range of
parameters were included in several different regulations covering
both water and sludge disposal, i.e. organics, nutrients (total ni-
trogen content of the effluent wastewater should be as low as
10 mg/L while this value is 1 mg/L for total phosphorus as regulated
in 91/271/EEC), pathogens, heavy metals, emerging contaminants
etc. This resulted in development of new treatment technologies as
well as new process flow diagrams for WWTPs. The number of
available alternative technologies using physical, chemical and
biological means of treatment has thus increased considerably in
order to satisfy the high removal efficiencies required by the stricter
regulations.
The fact that the number of alternative wastewater treatment
technologies is growing steadily increases the importance of early-

http://dx.doi.org/10.1016/j.envsoft.2014.11.023
1364-8152/© 2014 Elsevier Ltd. All rights reserved.
 H. Bozkurt et al. / Environmental Modelling & Software 64 (2015) 164e176
stage decision making in WWTP design and retrofitting problems.
When the design procedure is divided into different stages as stated
in the development funnel approach, which is shown in Fig. 1
(Quaglia, 2013), the first stage of the funnel e the early stage de-
cision making stage e corresponds to the design stage where a
variety of concepts and ideas are generated and a high number of
alternatives are evaluated in a less detailed/simplified manner. The
alternatives that are proven to satisfy the criteria needed by the
designer can move further through the funnel to be investigated in
more detail. When considering the WWTP design case, early stage
decision making is mainly about: (a) Which treatment processes
and unit operations to select for a particular wastewater treatment
problem; and, (b) How to verify the rationale and ensure engi-
neering optimality of the decision. Often e if not always e such
decisions are multi-objective and multi-criteria based considering
economics, environmental, legal and social constraints.
Recently, the WWTP process selection and network design
problem has evolved from being a simple technical design problem
to a complex integrated decision making task, mainly because of
the many aspects that are being considered in the early decision
making stage (Hamouda et al., 2009). Currently, the early stage
decision making for WWTP design is mainly based on expert de-
cisions and previous experiences (Tchobanoglous et al., 2003). This
approach takes values like environmental issues, water reuse, by-
product recovery (if possible) and impacts of the selected treat-
ment technologies on the surrounding population into account and
identifies the alternatives based on experience, similar existing
solutions and brainstorming to come up with the most viable
WWTP network (Daigger, 2005). However, with the increased
complexity of the technologies for wastewater treatment and the
stricter limit values for effluents, making the most feasible decision
using this approach is expected to become harder and harder.
An alternative approach is to cast the decision problem using
mathematical programming which has been an active research area
in chemical process synthesis (Grossmann, 2005), but has also seen
various applications in the wastewater treatment field including
Rigopoulos and Linke (2002), Vidal et al. (2002), Alasino et al. (2007,
2010). While these studies provided valuable insights and showed
the promising potential of the optimization based approaches for
plant design, their scope was however rather limited and focused
on either optimizing a given treatment process or selecting the best
candidate process from a limited number of alternatives. To realize
the full potential of this approach, however, there are a number of
barriers that need to be addressed and solved including represen-
tation of the increasing number of processes and unit operations in
wastewater treatment processes as well as tackling the resulting
multi-disciplinary complexity of the optimization problem which
requires both competences and methods from optimization as well
as wastewater engineering disciplines. In particular, for formulating
a realistic wastewater treatment plant design problem, the
complexity of the mixed integer nonlinear programming problem
Fig. 1. Illustration of the development funnel approach, where the number of feasible process options is gradually reduced from left to right (Quaglia, 2013).
165
can grow exponentially, which needs an effective formulation and
analysis method, which is the focus of this study. Moreover, the
knowledge-based decision support systems developed for con-
ceptual WWTP design have been presented in the literature
covering many different aspects of early stage design including
technical, economical, environmental and social considerations;
however, these approaches do not cover the optimization step for
the process synthesis (Comas et al., 2003; Garrido-Baserba et al.,
2012).
In this study, we propose a superstructure based optimization
framework based on mathematical programming to manage the
complexity of the problem and generate/identify novel and optimal
process selection and interconnection to create a process flow di-
agram for design of WWTPs. The purpose of the framework is to
support the WWTP design specialists in the process of making early
stage design decisions by allowing them to compare several
different treatment technologies with respect to many different
criteria. The framework contains a superstructure method for
representing the design space, and a systematic method for
modeling and data collection which enables effective formulation
of a mixed integer nonlinear programming problem using a tem-
plate approach. The framework is explained in more detail in Sec-
tion 2, and is then applied to a case study for retrofitting the
Benchmark Simulation Model no. 2 (BSM2) plant as described in
Section 3. The framework is evaluated under a number of scenarios
and the results are critically analyzed and discussed in Section 3.5.
2. Framework for synthesis and design of WWTP networks
A superstructure based optimization framework was developed
and evaluated successfully for various network design problems
including soybean oil processing, biorefinery, oil refinery waste-
water treatment etc. (Quaglia et al., 2012). The framework, modi-
fied and adapted in the context of the WWTP design problem is
given in Fig. 2.
Each of the steps mentioned in Fig. 2 will now be explained in
more detail:
Step 1: In the first step the scope of the problem (wastewater
characterization, effluent limit definitions etc.) is defined together
with the objective function. The problem is defined as the treat-
ment of domestic wastewater where a set of pollutants such as
organic nitrogen need to be removed. The characterization of the
wastewater source and the sink limitations for different water,
solids and gas streams leaving the plant (for example effluent re-
strictions) are defined here. The objective function, which repre-
sents the total annualized cost (TAC), is formulated together with
different scenario definitions. In the below equation, TAC is defined
as the summation of operational (OPEX) and capital cost (CAPEX).
CAPEX
TAC ¼ OPEX þ (1)
t
166 H. Bozkurt et al. / Environmental Modelling & Software 64 (2015) 164e176
Fig. 2. The superstructure based optimization methodology adapted for design of optimal WWT systems (Quaglia et al., 2012).
Step 2: The second step comprises a superstructure definition
consisting of different water sources, different tasks for water and
sludge treatment together with sinks and process alternatives for
the defined tasks. A superstructure, for example shown in Fig. 3, is a
compact representation of different process alternatives (i.e.
treatment technologies in the wastewater treatment case). Process
steps, which are the columns, represent wastewater sources and
sinks for the effluent streams (effluent water, sludge, by-products
etc.) as well as different tasks to be carried out throughout the
network in order to connect sources and sinks. In each step, in the
rows, alternative treatment process intervals (e.g. separation (pri-
mary clarifier, secondary clarifier, membrane reactor, etc.) or reac-
tion (activated sludge for C, N or P removal together, nitritation,
anammox etc.) and their different configurations) responsible for a
specific task are placed. There are three different ways of con-
structing a superstructure. One of the methods is alternative
collection, in which all known WWTP network configurations are
arranged in a superstructure representation. The resulting super-
structure includes the known configurations and enables only to
screen among the known candidates, which does not allow the
selection of innovative technologies or configurations. The second
approach is the method of combinatorial synthesis. In this
approach, the superstructure is composed of all treatment tech-
nologies placed under the relevant task connected to the others in
every possible connection way (Fig. 3 is an example of a full
combinatorial superstructure). This approach results in a very large
Fig. 3. A representative superstructure for wastewater treatment networks.
search space and redundant configurations. A third approach is
called the insight-based approach, which is used in this study. The
latter approach takes into account expert knowledge to include the
well-known configurations together with the innovative technol-
ogies and configurations in the superstructure, as well as elimi-
nating the unfeasible and non-convenient alternatives and
connections (Quaglia et al., 2014a). At this point, the selection of
technologies to be placed in the superstructure and the connections
between the alternatives are defined by design experts with a prior
screening procedure. A representative superstructure example is
given in Fig. 3, where a treatment plant is defined as a sequence of
different treatment tasks in addition to wastewater sources and
sinks. Under each treatment task, a number of alternative tech-
nologies are listed. The superstructure is then formulated by
defining the feasible connection streams between treatment tasks.
In this example, a full-combinatorial connection between different
source/sink/process intervals is defined.
Step 3: In this step, a systematic data collection procedure is
used to design the treatment technologies placed in the super-
structure, which is illustrated in Fig. 4. In this context, we used the
commonly accepted design procedures given by the ATV design
standards (2000), Tchobanoglous et al. (2003), WEF manuals
(2010) and Henze et al. (2008) for the design of a treatment pro-
cess alternative for a given wastewater source. A particular treat-
ment process design is made by using system specific design
criteria like SRT and HRT together with temperature dependent
biokinetic constants, settling tank data and internal/external
recycle ratios. Here, the treatment technologies are designed at
their optimality by fixing the SRT and HRT values rather than
optimizing them and in a second step more rigorous models can be
used for optimization once the number of alternatives is reduced.
This two-tiered approach for optimization is chosen on purpose to
manage the complexity of the optimization problem which be-
comes otherwise intractable (Quaglia et al., 2014b). The output
from the design includes volumes, utility consumption (electricity,
chemicals, aeration), and sludge production data which are used to
calculate the capital and operational costs as well. The algorithms
for the design were implemented in a Matlab script to automate
this step and ensure consistency and reproducibility. The system
design parameters, e.g. SRT and HRT, are selected from a given
range that is obtained from engineering practice (Tchobanoglous
et al., 2003) so that they satisfy the effluent requirements with
 H. Bozkurt et al. / Environmental Modelling & Software 64 (2015) 164e176
Fig. 4. Illustration of the systematic data collection procedure.
minimum required capital and operational cost. The procedure is
iterative; i.e. the SRT and HRT are modified until converging to a
solution satisfying all the constraints.
Each process interval in the superstructure is structured using a
generic model which is illustrated in Fig. 5. The intervals are
composed of a number of phenomena namely: mixing of all the
flows entering the interval and the utilities added, reaction, sepa-
ration of flow for internal recycle and sludge wastage, waste sep-
aration, flow separation for external recycle and sending the flow to
the process intervals of the next step. The mathematical equations
defining the intervals are given in Table 1.
Using the generic model (Fig. 5, Table 1), the treatment alter-
natives are described by inputeoutput mass balances. In the mixing
equation, all different influent flows containing the same compo-
nent originating from the previous process intervals to the process
interval of interest (from k to kk) are mixed together by using the
first equation (1.1) and then the flow after mixing is obtained in
equation (1.2) by mixing the flow resulting from equation (1.1) with
the utility flow (where 0  ai,kk  1). The utility flow is calculated by
the third equation (1.3) (where mi,ii,kk is given as daily mass of utility
added/flow of corresponding component). The reaction in the
generic interval is defined so that the key reactant is converted to
the other components with a given conversion efficiency by using
Fig. 5. Generic process interval structure: mass inputeoutput model.
167
the utilities added, while maintaining the overall mass balance
within the system boundaries of the process interval. The reaction
equation (1.4) calculates the flow after reaction; the key reactant is
removed with specified conversion efficiency and the other com-
ponents are produced or removed according to the defined stoi-
chiometry. As shown in Fig. 6, the stoichiometry matrix (gi,kk,rr) is
defined by expressing the reactions occurring in a treatment
technology. After the design of the treatment technology, the flows
of the components after the reaction are known; therefore, the
resulting design data are converted into a matrix which the MI(N)
LP routine can use by applying equation (2). The next equation (1.5)
calculates the flow after waste separation, where any waste stream
which is separated from the main flow is defined by a waste split
factor. According to our design assumptions, the sludge is wasted
from the reaction tank; therefore, the corresponding flow (Fi;kk out3 ) is
separated from the main stream before any composition change
occurs. The amount of wastage is defined by the ratio SW. The flow
is then separated to different outflows with their respective com-
positions as it happens in the secondary sedimentation tank as
well, with a separate thickening factor applied for soluble and
out1 ) is
particulate components (Spliti,kk). The first outlet stream (Fi;kk
the effluent water stream and follows the water stream line,
whereas the concentrated stream is separated into two other
streams which are the sludge recycle flow (Fi;kkrec ) and the secondary
out2 ). The ratio between these two streams is deter-
outlet flow (Fi;kk
mined by the external recycle ratio. The outlet streams (FXi;k;kk ) are
then directed to the other intervals and the direction of the flow
leaving the process interval is determined by the binary variable
which contains superstructure information (Sk,kk). Its value is equal
to 1 if the connection is present and 0 if there is no connection
between two intervals. If a process interval is selected by the
optimization tool, then the binary variable assigned to it (ykk) is
equal to 1, otherwise it is 0. This is only possible by expressing
relations between binary and continuous variables, which is done
with the activation constraints. Logical cuts, represented by equa-
tion (1.12), are included to ensure the selection of only one process
interval from a process step, which thus prevents the selection of
undesired or infeasible solutions. The last equation (1.13) ensures
that the effluent limits are satisfied in the sink intervals for defined
168 H. Bozkurt et al. / Environmental Modelling & Software 64 (2015) 164e176

Table 1
Mathematical equations representing the generic process model in each interval of the superstructure.

Phenomena Equation Explanations

Mixing i,ii: Component index

X
in
Fi;kk ¼ Fi;k;kk (1.1) k,kk: Process interval index
in : Inflow to the process interval
Fi;kk
f
Fi;k;kk : Inflow of component i to process kk coming from k
mix : Flow after mixing
Fi;kk
Ri;kk : Utility flow
mix in
Fi;kk ¼ Fi;kk þ ai;kk *Ri;kk (1.2) ai;kk : Fraction of utility consumed
mi;ii;kk : Specific consumption of utility
Utility addition reac : Flow after reaction
Fi;kk
X 
Ri;kk ¼ mi;ii;kk *Fii;kk (1.3) gi;kk;rr : matrix representing reaction stoichiometry
ii qreact;kk;rr : Conversion efficiency of the key reactant react
w : Flow after waste separation
Fi;kk
Reaction
X   Wi;kk : Waste split factor
reac mix
Fi;kk ¼ Fi;kk þ gi;kk;rr *qreact;kk;rr *Freact;kk (1.4) out1 , F out2 , F out3 : Outlet streams from interval
Fi;kk i;kk i;kk
rr;react
Spliti;kk : Flow split factor
Waste separation SWkk : Sludge wastage flow rate ratio
  rec : External recycle flow rate
w
Fi;kk reac
¼ Fi;kk * 1  Wi;kk (1.5) Fi;kk
reckk : External recycle ratio
Flow separation X: 1,2,3 (representing three different outlet flow streams)
out1
Fi;kk w
¼ Fi;kk *Spliti;kk (1.6) Sk;kk : Binary variables containing superstructure information
ykk : Binary variable describing the process interval
xk : Variable bounded by xLO k
and xUP
k
Limi;kk : Effluent limit value of component i
out2 w out1 rec
Fi;kk ¼ Fi;kk  Fi;kk  Fi;kk (1.7)

out3 reac
Fi;kk ¼ Fi;kk *SWkk (1.8)

 
rec w out1
Fi;kk ¼ Fi;kk  Fi;kk *reckk (1.9)

outX
FXi;k;kk  Fi;kk *Sk;kk (1.10)

Activation
ykk *xLO UP
k  xk  ykk *xk (1.11)

Logical cuts
X
ykk  1 (1.12)
kk

Effluent limits
X
in
Limi;kk  Fi;kk (1.13)
f

components. Once the design data of the treatment technologies treatment network and the value of the objective function. With
are converted into model parameters, the generic process interval respect to the nature of the problem, the optimization problem can
model is run and validated by using the design data. result in a linear or non-linear formulation. In this step the MI(N)LP
problem is formulated and solved. The models represent the mass
! inputeoutput model for each process interval, process constraints,
mix  F reac
Fi;kk i;kk structural constraints, effluent limit constraints, cost models
gi;kk;rr ¼  (2) together with the objective function. The adapted MI(N)LP formu-
mix
Freact;kk  Freact;kk
reac
lation for the specific case of a WWTP design/retrofit study is
Step 4: The formulation of the MI(N)LP and its solution take described below. The detailed model is given elsewhere in Table 1.
place in the fourth step of the synthesis and design framework, and
results in the optimal network, fate of pollutants throughout the X CAPEXkk
Min OBJ ¼ OPEXkk þ (3)
t
kk

subject to;
Process model
 
h ai;kk ; mi;ii;kk ; gi;kk;rr ; qreact;kk;rr ; Wi;kk ; Spliti;kk ; SWkk ; reckk ¼ 0
Fig. 6. Illustration of the reaction occurring in the process interval and its
(4)
stoichiometry.
 H. Bozkurt et al. / Environmental Modelling & Software 64 (2015) 164e176 169

Process constraints 3.1. Problem formulation

 
g Sk;kk  0 (5) The problem is defined as treatment of domestic wastewater
comprising mainly COD, nitrogen and solids as pollutants. The
Structural constraints
wastewater composition is shown in Table 2. The objective is to
X design the WWTP network against the lowest operational (aeration
ykk  1 where y2f0; 1gn (6)
cost, sludge disposal cost, pumping and mixing cost as well as
kk
biogas price) and capital cost possible while satisfying the effluent
The optimal wastewater network design problem is then limitations for organic material and nitrogen.
formulated as an MI(N)LP in GAMS (GAMS Development
Corporation, 2011) and solved for different scenarios. The optimi-
zation model in GAMS consists of an objective function covering
3.2. Superstructure development
both operational (utility cost in terms of aeration, chemical addition
and electricity use, landfill cost together with mixing and pumping
The superstructure developed for this problem is shown in
cost, and income from biogas use) and capital costs (with regard to
Fig. 7. It consists of a domestic wastewater source interval (WW) in
the size of the units), logical constraints defining the process flow
the source column, and sinks for water, sludge and biogas in the
diagram of the solution and process constraints describing the
last column. Treatment technologies are located in between the
process models of each treatment technology in the superstructure.
source and the sinks, and include primary treatment, secondary
The formulated MI(N)LP problem was solved by using the DICOPT
treatment, tertiary treatment, disinfection and sludge treatment as
solver for the non-linear case and the CPLEX solver for the linear
the tasks. The primary treatment consists of a primary sedimen-
case in GAMS software.
tation tank (PC) and a by-pass interval (BP1). In the biological
Step 5: Uncertainty and sensitivity analysis are performed in
treatment task, there are three technologies (Modified Ludzack e
order to generate robust solutions. It is also important to show that
Ettinger e MLE, Oxidation ditch e OxD, and upflow anaerobic
the network is feasible over the whole uncertain domain defined
sludge blanket e UASB) and a by-pass interval (BP2). The tertiary
with respect to the uncertain input parameters. For instance, in the
treatment is mainly responsible for nitrogen rich wastewater
wastewater treatment case, the composition of the wastewater is
treatment and includes the following technologies: partial nitri-
highly uncertain over time and this uncertainty has to be taken into
fication (Sharon), partial nitrification and anaerobic ammonium
account during design studies. Therefore, in this step the design
oxidation (SharoneAnammox) and a by-pass interval (BP3).
problem is solved under uncertainty. For the selected parameters,
Disinfection technologies comprise different treatments: UV, O3,
in their defined uncertain domain, Monte Carlo sampling is per-
chlorine and by-pass (BP4). The sludge treatment line on the other
formed and the stochastic programming problem is reformulated
hand, consists of sludge stabilization options (AnD-Anaerobic
and solved for the different combinations of uncertain parameters
digester, AeD e Aerobic digester) receiving the sludge from a
resulting from the Monte Carlo sampling (Quaglia et al., 2013). The
Thickener.
results are then used for a reliability and robustness analysis of the
optimal solution for the treatment process and the resulting pro-
cess flow diagram.
3.3. Data collection

3. Case study: Benchmark wastewater treatment plant The development of the database for the represented super-
structure was done by following the steps of the systematic data
To highlight the presentation and application of the super- collection procedure given in Fig. 4. For illustration purposes, the
structure based optimization methodology, a simple yet represen- detailed calculations for a selected process interval (MLE) are given
tative wastewater treatment network problem was defined, with below. Moreover, the calculation steps for all other process in-
focus on a domestic WWTP layout design. The activated sludge tervals can be found in Appendix 1.
plant defined in the Benchmark Simulation Model no. 2 (Gernaey
et al., 2014) was selected, since this plant layout is generally Data Step 1: The characterization of the wastewater source is
known in the wastewater treatment modeling community. taken from the Benchmark Simulation Model no.2 (BSM2)
definition (Gernaey et al., 2014) and given in Table 2. The pol-
lutants are represented by Activated Sludge Model no.1 (ASM1)
Table 2
components (Henze et al., 2000). The effluent limits given in the
Influent wastewater characterization, average composition (Gernaey et al., 2014).
Urban Wastewater Treatment Directive (91/271/EEC) are taken
Description Component Value Unit as a reference for the sink intervals.
Soluble inert organic matter SI 27.23 g COD/m3 Data Step 2: The technology to be designed is selected as a Pre-
Readily biodegradable substrate SS 58.18 g COD/m3 denitrification activated sludge system with low SRT and the
Particulate inert organic matter XI 92.49 g COD/m3 configuration is specified as MLE technology.
Slowly biodegradable substrate XS 363.94 g COD/m3
Active heterotrophic biomass XB,H 50.68 g COD/m3
Data Step 3: The SRT and HRT of the system are selected as 14
Active autotrophic biomass XB,A 0 g COD/m3 days and 12 h respectively, whereas the anoxic to aerobic vol-
Particulate products arising from XP 0 g COD/m3 ume ratio is set to 0.6 (Tchobanoglous et al., 2003). All the
biomass decay temperature dependent biokinetic constants needed in the
Oxygen SO 0 g -COD/m3
design procedure are taken from the BSM2 (Gernaey et al., 2014)
Nitrate and nitrite nitrogen SNO 0 g N/m3
NHþ g N/m3 for a temperature of 15  C and given in Table 3. The external
4 þ NH3 nitrogen SNH 23.85
Soluble biodegradable organic nitrogen SND 5.65 g N/m3 recycle ratio is fixed to be 100% of the influent flow rate.
Particulate biodegradable organic nitrogen XND 16.13 g N/m3 Data Step 4: The design is done as follows (ATV design standards,
Alkalinity SALK 7 mol/m3 2000; Tchobanoglous et al., 2003; WEF, 2010; Henze et al.,
Flow rate Q 20,648 m3/d
2008):
170 H. Bozkurt et al. / Environmental Modelling & Software 64 (2015) 164e176

Fig. 7. Benchmark WWTP superstructure.

 The volume of the tanks is calculated by using the aerobic and  The amount of solids produced is found by using the following
anoxic HRT values and the influent flow rate Q. Accordingly Vae is equations, where S0 represents the influent total COD concentra-
calculated to be 5764 m3 and Van is 3459 m3. tion and CN represents the NH4eN to be oxidized
Vae ¼ HRTae *Q (7) (CN ¼ Sin in in out
NH þ SND þ XND  SNH  NS with NS being the nitrogen
content of biomass). (XBH ¼ 961 g COD/m3, XBA ¼ 182 g N/m3,
XP ¼ 833 g COD/m3, XI ¼ 1434 g COD/m3 and XT ¼ 3410 g/m3)
Van ¼ HRTan *Q (8)

The effluent COD (denoted as S) and ammonia (denoted as NH) SRT Yh ðS0  SÞ
XBH ¼ (11)
are calculated by using the following equations where S ¼ 3.29 g HRT 1 þ bh *SRT
COD/m3 and NH ¼ 0.64 g NH4eN/m3.

SRT Ya *CN
Ks *ð1 þ bh *SRTÞ XBA ¼ (12)
S¼     (9) HRT 1 þ ba *SRT
SRT* mh VVtot
an
 bh  1
h   i
XP ¼ fp *bh *XBH þ fp *ba *XBA *SRT (13)
KNH *ð1 þ ba *SRTÞ
NH ¼     (10)
SRT* ma VVtot  ba  1 SRT
XI ¼ XI0 *
ae
(14)
HRT

XT ¼ XBH þ XBA þ XP þ XI (15)

Table 3
Biokinetic parameters (Gernaey et al., 2014).  Calculations of the amount of oxygen consumed for cell decay
(ROb) and nitrification (ROn) are shown below. The variables MXBH
Parameter Unit Value
and MXBA refer to the total mass of microorganisms in the reaction
Ya g cell COD/g N 0.24 tank. As a result the oxygen consumption was calculated as 1372
Yh g cell COD/g COD 0.67 and 3883 kg/d for cell decay and nitrification, respectively.
fp dimensionless 0.2
KS g COD/m3 10        
bh 1/d 0.3 Vae Vae
ROb ¼ 1fp *bh *MXBH * þ 1fp *ba *MXBA *
mh 1/d 4 Vt Vt
KNH g NH3eN/m3 1
ba 1/d 0.05 (16)
ma 1/d 0.5

ROn ¼ 4:57*Q *CN (17)

 H. Bozkurt et al. / Environmental Modelling & Software 64 (2015) 164e176 171

Table 4
   
Process information for secondary treatment task process intervals.
1  fp *ba *MXBH * VVant
Properties Process interval DP3 ¼ (21)
2:86*Q
Corresponding technology Modified Oxidation Upflow
LudzackeEttinger ditch Anaerobic
(MLE) Sludge DP ¼ DP1 þ DP2 þ DP3 (22)
Blanket
(UASB)

Temperature ( C) 15 15 15 Seff
NO ¼ CN  DP (23)
SRT (days) 14 28 120
HRT (h) 12 24 14 Data Step 5: The separation step is designed by defining a
Reactor volume (m3) 9223 18,446 12,956
thickening factor for the sedimentation part of the separation.
Settler volume (m3) 3774 3411 e
Anoxic/Aerobic volume ratio 0.6 0.6 e These values differ for soluble components and particulate com-
MLSS (g/m )3
3410 3032 18,590 ponents. Accordingly, 48% of the soluble and 0.2% of particulates by
Wastage flowrate (% influent flow) 3.5 3.5 e mass leave with the water stream, while the rest is assumed to
Sludge recycle flow (% influent flow) 100 100 e settle in the sludge zone. The flow equal to the influent flow rate is
COD removal efficiency (%) 88.4 87.78 68.5
Total N removal efficiency (%) 77.2 78.48 e
directed to the influent of the same process interval as the sludge
recycle, and the rest is sent to the water effluent line. Moreover, the
volume of the sedimentation basin is calculated by assuming spe-
 The wastage flow rate, which is assumed to be wasted from the cific surface overflow rate (SOR) and solids loading rate (SLR) values
reaction tank, (Qw ¼ 658 m3/d) is calculated as follows. as well as a certain depth of the tank from a range given for circular
clarifiers (WEF, 2010). In this respect, SOR and SLR values for the
V system are selected as 1.6 m/h and 130 kg/m2*d, respectively
Qw ¼ (18)
SRT whereas the depth of the clarifier was 4 m. The resulting volume of
the sedimentation basin then became 3774 m3.
 In order to find the effluent nitrate concentration, the deni-
Secondary treatment process alternatives were designed and
trification potential (DP) is calculated by using the following
the features of the processes are summarized in Table 4. For the
equations where K2 refers to denitrification rate and was taken as
purpose of validating the systematic data collection procedure used
0.07 g NO3/(g cell COD*d) (Henze et al., 2008). Accordingly
in the case study, the design parameters and performance values of
DP1 ¼ 8.02, DP2 ¼ 12.78 and DP3 ¼ 15.13 g N/m3 which gives a total
the pre-denitrification activated sludge type of treatment units
denitrification potential (DP) of 35.93 g N/m3 and a nitrate effluent
obtained using the steady-state design model (Eqs. (7)e(23)) are
concentration of 10.14 g N/m3. DP1, DP2 and DP3 represent the
compared with the steady state results obtained from a simulation
denitrification potential for readily biodegradable COD, slowly
carried out using a rigorous model, i.e., Activated Sludge Model 1
biodegradable COD and biomass respiration, respectively.
(ASM1). The simulation with the rigorous ASM1 model was per-
  formed in Matlab/Simulink®. The results are presented in Table 5
Ssi * 1
Yh which indicates that the differences between the estimated
DP1 ¼ (19) removal efficiencies for COD, TN and TSS by the rigorous model
2:86
(ASM1) and the steady-state design model used in this study were
K2 *ðXS  SÞ*Yh *SRTan quite small, i.e. the average relative error is less than 1.5%, 5% and
DP2 ¼ (20) 1.5%, respectively. Therefore it is concluded that with the same
1 þ ðbh *SRTÞ
design values selected (temperature, SRT and HRT), the estimated

Table 5
Validation of systematic data collection procedure.

Parameter Rigorous model (ASM1) Steady-state design model Rigorous model (ASM1) Steady-state design model
(this study) (this study)

Corresponding technology Pre-denitrification Pre-denitrification Pre-denitrification Pre-denitrification

activated sludge activated sludge activated sludge activated sludge
Temperature ( C) 15 15 15 15
HRT (h) 12 12 24 24
SRT (days) 14 14 28 28
Anoxic/Aerobic volume ratio 0.6 0.6 0.6 0.6

COD (g COD/m3)
Influent 381.19 381.19 381.19 381.19
Effluent 47.50 44.16 46.56 42.59
Reduction (%) 87.54 88.42 87.78 88.83

Nitrogen (g N/m3)
Total N influent 54.14 54.14 54.14 54.14
SNO effluent 11.41 10.14 9.95 7.77
SNH effluent 0.17 0.64 0.07 0.39
Total-N effluent 12.80 12.32 11.65 9.51
Total-N reduction (%) 76.36 77.24 78.48 82.43

Suspended Solids (g COD/m3)

Influent 211.27 211.27 211.27 211.27
Effluent 12.95 10.62 11.97 9.44
Reduction (%) 93.87 94.97 94.33 95.53
172 H. Bozkurt et al. / Environmental Modelling & Software 64 (2015) 164e176

Table 6
Cost information for operational and capital cost items.

Cost item Unit Value/range Description/assumption

Electricity consumption of oxygen transfera kg O2/kwh 1.9e3.2 Coarse bubble diffusor

Typical chlorine demandb mg/L 10e25 For activated sludge type plant effluents
Sodium hypochlorite costc euro/kg 0.12 e
Typical ozone demandd mg/L 10e15 e
Energy requirement for ozonee kwh/kg O3 21e35.2 Sum of ozone generation (air feed), ozone contacting
and all other uses (on the average)
Electricity costf euro/kwh 0.0978 In Denmark for industry
Landfill costg euro/t 107 In Denmark, on the average
Biogas priceh eurocent/m3 methane 40.3 In Denmark (assumptions: 1 mol of methane is 24 L and 1
mol of methane accounts for 64 g of COD)
Capital cost e UASBi US$/m3 425 Based on the price level of 2006, for 100,000 PE
Capital cost e Aeration tanki US$/m3 175 Based on the price level of 2006, for 100,000 PE
Capital cost e Secondary settleri US$/m3 290 Based on the price level of 2006, for 100,000 PE
Capital cost e Primary settleri US$/m3 375 Based on the price level of 2006, for 100,000 PE
Capital cost e Sludge thickeneri US$/m3 400 Based on the price level of 2006, for 100,000 PE
Capital cost e Anaerobic digesteri US$/m3 350 Based on the price level of 2006, for 100,000 PE
a
Siemens (2009).
b
Tchobanoglous et al. (2003).
c
AWWA Michigan Section (2006).
d
Takahara et al. (2006).
e
Tchobanoglous et al. (2003).
f
URL1 (2013).
g
URL2 (2013).
h
Hahn et al. (2010).
i
Van Haandel and van der Lubbe (2012).

system performance results in terms of removal efficiencies were in
agreement with each other and therefore the systematic data
collection and design for treatment alternatives is considered to be
validated against the more rigorous model.
Data Step 6: The objective function represents the total annu-
alized cost (TAC) as shown in Eq. (1) and the optimization problem
is formulated to minimize the TAC. OPEX corresponds to the
operational cost and is composed of aeration, electricity, chemical
addition, pumping, mixing and landfill cost as well as biogas price
as an income. CAPEX, on the other hand represents the capital cost
and t is the lifetime of the treatment plant. All the cost data are
collected from information available in the open literature and
summarized in Table 6. Pumping (PE ¼ 0:004Qa þ 0:008Qr þ
0:05Qw Þ and mixing cost (ME ¼ 24*0:005Vi Þ in kWh/d are defined
as a function of flow (return sludge e Qr , internal recycle e Qa and
sludge wastage e Qw ) and tank volumes e Vi , respectively (Copp,
2002).
3.4. Generic process interval model and MILP formulation
One of the most challenging steps in optimization based ap-
proaches is the resulting mathematical complexity of formulating
and solving the optimization problem. To manage this complexity
and facilitate the effective formulation and analysis of the problem,
a separation principle was used that separates the database needed
for model parameters from MINLP formulation and solution in
GAMS. The procedure is explained as follows.
Once the data have been collected for all the process intervals,
they are stored as matrices in an MS Excel based structure. The data
in the matrices are sent to GAMS by using GDX (GAMS Data Ex-
change) utilities. GDXXRW is used in this respect, which is the
utility responsible for reading from and writing to an MS Excel
spreadsheet. Once the data are transferred to GAMS, the formulated
MILP problem, consisting of the generic equations defined in
Table 1, is solved. Note that when a new problem is defined, only
the database needs to be changed, while the generic MINLP model
can still be used. The data flow and problem formulation (partly as a
screenshot) can be seen in Fig. A.1. In Table 7, a summary of the
parameters used in the MILP problem formulation is presented. The
maintenance of the database will be ensured by the user by
selecting and using the updated model parameters in the model
generation step. The optimization methodology allows also
including expert knowledge about technology selection by means
of logical cuts in the superstructure definition. In this particular
case study, since it is known from expert knowledge that selection
of a high SRT activated sludge technology (in this case oxidation
ditch) together with the anaerobic digestion is not meaningful, this
combination is eliminated from the search space by simply
inserting a constraint in the problem formulation as
y(OxD) þ y(AnD)  1.
3.5. Results and discussion
3.5.1. Optimal process selection
The objective of the optimization problem is to select among the
treatment alternatives so that the resulting treatment process flow
diagram has the minimum TAC and at the same time satisfies the
effluent limits given in the Urban Wastewater Treatment Directive
(91/271/EEC). The formulated optimization problem is solved by
using GAMS (GAMS Development Corporation, 2011) using the
solver CPLEX. The details of the solution will be analyzed and dis-
cussed in this section for three scenarios corresponding to different
objective function formulations.
The optimal process selection and the process flow diagram
after solution of the model are given in Table 8 for three different
scenarios, and the process flow diagram for scenario 1 is illustrated
in Fig. A.2. The first scenario takes only the OPEX into account while
the other scenarios are based on TAC. In the third scenario a more
stringent total N limit is defined for the sink interval.
Under the conditions of the first scenario, tertiary treatment and
disinfection tasks are by-passed and the water stream is sent to the
water sink interval after being treated by the primary clarifier and
low SRT MLE system whereas the sludge is stabilized in the
anaerobic digester and sent to the sink interval. When the capital
cost is also added into the formulation of the objective function (i.e.
scenario 2), the network selection does not change; however an
Table 7
Summary table for the data collection.

Treatment Mixing ai,kk mi,ii,kk Key reactant(s) Reaction stoichiometry (gi,kk,rr) Reaction efficiency Sludge Waste Sludge recycle Flow separation (Spliti,kk)
process (react) (qreact,kk,rr) wastage separation (reckk)
(SWkk) (Wi,kk)

PC e e e e e e e e e 99.3% for solubles and

water50% for particulates
MLE SO to SS ratio 1 4.09 SS, SNH, XS With respect to SS 100% SS removal 3.5% of e 100% of 48% for solubles and water
SS ¼ 1, So ¼ 4.07, 96% SNH removal incoming incoming flow 0.2% for particulates
XI ¼ 0.77, XBH ¼ 0.6, XBA ¼ 0.19, 97% XS removal flow
XP ¼ 0.87
With respect to SNH

H. Bozkurt et al. / Environmental Modelling & Software 64 (2015) 164e176

SNH ¼ 1, SNO ¼ 0.25, XND ¼ 0.03
With respect to XS:
XS ¼ 1
OxD SO to SS ratio 1 4.29 SS, SNH, XS With respect to SS: 100% SS removal 3.5% of e 100% of 48% for solubles and water
SS ¼ 1, So ¼ 4.27, 98% SNH removal incoming incoming flow 0.2% for particulates
XI ¼ 0.77, 97% XS removal flow
XBH ¼ 0.15, XBA ¼ 0.14, XP ¼ 0.97
With respect to SNH:
SNH ¼ 1, SNO ¼ 0.20, XND ¼ 0.03
With respect to XS:
XS ¼ 1
UASB e e e SS SS ¼ 1, SNH ¼ 0.05, XI ¼ 0.002, 100% SS removal e 1 for CH4 e e
XS ¼ 2.91,
XBH ¼ 0.41, XP ¼ 0.36, XND ¼ 0.15,
CH4 ¼ 0.94
SHARON eSO to SNH ratio 1 3.14 SNH SNH ¼ 1, SNO ¼ 0.99, SO ¼ 6.18 50% SNH removal e e e e
eNaOH to SNH ratio 1 2.85
SHARON- SO to SNH ratio 1 1.78 SNH SNH ¼ 1, SNO ¼ 0.137, SO ¼ 2.16, 80% SNH removal e 1 for N2 e e
Anammox N2 ¼ 1.08
UV Electricity to water ratio 0 15.87 Pathogens Pathogens ¼ 1 100% pathogen e e e e
removal
Ozone Ozone to water ratio 0 12.5 Pathogens Pathogens ¼ 1 100% pathogen e e e e
removal
Chlorine eElectricity to water ratio 0 2.12e17.5 Pathogens Pathogens ¼ 1 100% pathogen e e e e
eChlorine to water ratio 1 removal
AnD e e e XBH, XBA, XS With respect to XBH: 100% XBH and XBA e 1 for CH4 e 50% for solubles and 0.7%
XBH ¼ 1, SNH ¼ 0.07, XP ¼ 0.02, removal for particulates
CH4 ¼ 0.81 80% XS removal
With respect to XBA:
XBA ¼ 1,
With respect to XS:
XS ¼ 1
AeD SO to XBH ratio 1 0.99 XBH, XBA With respect to XBH: 100% XBH and XBA e e e 50% for solubles and 0.7%
XBH ¼ 1, SO ¼ 1.35, SNO ¼ 0.07, removal for particulates
XP ¼ 0.27
With respect to XBA:
XBA ¼ 1

173
174 H. Bozkurt et al. / Environmental Modelling & Software 64 (2015) 164e176

Table 8 at higher temperatures or it should be integrated with effluent

Summary of results for different scenarios. polishing steps which are not included in the current superstruc-
Scenario Objective Total N Selected process flow Value of objective ture (i.e. facultative ponds, sand filtration, constructed wetlands,
function limit diagram function (unit trickling filters, physico-chemical treatment and activated sludge
(mg/L) cost) treatment) (Henze et al., 2008). Therefore this treatment alterna-
1 OPEX 15 WW-PC1-MLE-BP3-BP4- 219.051 tive has not been selected by the optimizer under either of the
Thickener-AnD-Discharge- scenarios as expected from the above-mentioned process engi-
Sludge-Biogas
neering expertise.
2 TAC 15 WW-PC-MLE-BP3-BP4- 912.080
Thickener-AnD-Discharge- The cost summary and the performance evaluation for the
Sludge-Biogas scenarios are presented in Table 9. MLE is favored both for its low
3 TAC 10 WW-PC-MLE-BP3-BP4- 912.080 operational cost (low utility requirement and high sludge produc-
Thickener-AnD-Discharge-
tion resulting in high biogas production) and capital cost.
Sludge-Biogas
The optimizer also gives the flow of components through the
selected process flow diagram. The stream table of the components
for the solution of Scenario 1 is shown in Table 10, where the flows
Table 9
Cost summary and performance evaluation for the different scenarios.
are given as the flows after reaction in each process interval. By
using this information the selected process flow diagram is evalu-
Unit Scenario 1 Scenario 2 Scenario 3 ated in terms of its performance, which is presented also in the
Objective function e OPEX & Total TAC & Total TAC & Total summary of the results in Table 9. It can be seen that both COD and
N limit of N limit N limit of total nitrogen concentrations are below the limits set by the reg-
15 mg/L of 15 mg/L 10 mg/L
ulations (limits for COD and total nitrogen are 100 g COD/m3 and
Aeration cost unit cost 229.187 229.187 229.187
Landfill cost unit cost 614.191 614.191 614.191 15 g N/m3, respectively). This analysis shows that the selected
Biogas price unit cost 695.058 695.058 695.058 process flow diagram is capable of decreasing the concentrations of
Electricity cost unit cost e e e the key pollutants below the effluent discharge limits while using
Pumping cost unit cost 43.687 43.687 43.687 the minimum amount of utilities.
Mixing cost unit cost 27.044 27.044 27.044
Capital cost unit cost e 693.029 693.029
Objective function value unit cost 219.051 912.080 912.080 3.5.2. Solution statistics
Effluent COD g COD/m3 39.16 39.16 39.16 The model and solution statistics of the problem are given in
Effluent Total N g N/m3 9.82 9.82 9.82
Table 11. Accordingly, the solution time required for the problem
containing 44,445 equations is 0.172 s.
expected increase in the objective function is observed due to
capital cost. 3.5.3. Uncertainty analysis
In the third scenario, when the effluent nitrogen limit value is To perform the uncertainty analysis in the context of a process
lowered down to 10 mg N/L, the optimizer once again selects the synthesis problem, the following three step procedure is used:
MLE technology coupled with primary clarifier to treat the waste- uncertainty characterization, uncertainty mapping and decision
water and anaerobic digester for sludge stabilization purposes. making under uncertainty.
Although the anaerobic treatment alternative of the UASB In the first step, the uncertain parameters are identified and a
coupled with the nitrogen rich wastewater treatment alternatives uniform probability distribution is assigned to characterize the
of the tertiary treatment task can satisfy the effluent total nitrogen
limit, the UASB alone cannot generate an effluent stream complying Table 11
with the COD effluent limit criteria when operated at such low Model and solution statistics.
temperatures. Experimental work also showed the decrease in the
Number of variables 42,319
efficiency of the UASB reactor with the decreasing temperature Number of binary variables 20
(Lew et al., 2004). Although anaerobic treatment of domestic Number of equations 44,445
wastewater has been successfully demonstrated in full scale all Objective function 1
Execution time (s) 0.172
over the world, the conclusion here, however, is that in order to
Solver CPLEX
comply with the regulations the system should either be operated

Table 10
Scenario 1 stream table.

Unit WW1 PC MLE BP3 BP4 Thickener AnD Discharge Sludge

Q m3/d 20,648 20,648 42,633 20,553 20,553 763 190 20,553 95

SI kg/d 562 562 1160.382 559.420 559.420 20.771 5.159 559.420 2.580
SS kg/d 1201 1201 e e e e 8.407 e 4.203
SO kg/d e e 2305.431 1111.448 1111.448 41.267 2.435 1111.448 1.217
SNO kg/d e e 341.204 164.494 164.494 6.108 0.360 164.494 0.180
SND kg/d 117 117 e e e e e e e
SNH kg/d 492 492 22.055 10.633 10.633 0.395 126.268 10.633 63.134
XI kg/d 1909 1.909 52,521.861 105.044 105.044 1811.664 2743.809 105.044 2724.602
XS kg/d 7514 7514 116.229 0.232 0.232 4.009 752.192 0.232 746.927
XBH kg/d 1046 1046 34,557.848 69.116 69.116 1192.022 e 69.116 e
XBA kg/d e e 6344.526 12.689 12.689 218.845 e 12.689 e
XP kg/d e e 29,234.094 58.468 58.468 1008.387 1031.705 58.468 1024.483
XND kg/d 333 333 5167.206 10.334 10.334 178.235 342.536 10.334 340.138
CH4 kg/d 4599.228
 H. Bozkurt et al. / Environmental Modelling & Software 64 (2015) 164e176 175

uncertainty range. The resulting distribution parameters such as

the minimum, the maximum and the mean values are recorded in
the database. The uncertain domain is then sampled using the Latin
Hypercube Sampling technique. In the Benchmark WWTP case
study, the uncertain parameters are chosen as the cost related pa-
rameters (oxygen transfer efficiency, electricity price and landfill
price), effluent total nitrogen limits and influent wastewater char-
acterization. The details of the uncertain domain definition and
sampling can be found elsewhere (Bozkurt et al., 2014) and the
table showing uncertain parameters and their domain is shown in
Appendix 2.
In the second step, the optimization problem was solved for 50
scenarios generated in the preceding sampling step. The analysis of
the optimization results indicated that two different WWTP net-
works were identified as optimal with different frequencies (see
details in Table 12). The cumulative distribution of the objective
function is illustrated in Fig. 8 where the x-axis shows the objective
function value, which represents operational and capital cost, and
the y-axis represents the probability that the value of the objective
function will be lower than the stated value on the x-axis. This
indicates that there is a significant uncertainty on the treatment Fig. 8. Cumulative distribution of the objective function.
cost (the objective function) which ranges from 693 to 1606 unit
cost. Compared with the deterministic solution case given in Wastewater Treatment Plant case study is 0 because the network
Table 9, it can be seen that 78% of the scenarios result in higher selection did not change under uncertain conditions (solution un-
objective function values and in 16% of the scenarios, a different der uncertainty).
network configuration is selected. Although the output from the
VSS ¼ Eq ðf ðxdet ; ydet ; qÞÞ  minðEq ðf ðx; y; qÞÞÞ (25)
uncertainty analysis very much depends on the defined domain of
input uncertainties (step 1), this comparative analysis already in- Finally, the UP indicates the cost of uncertainty by calculating
dicates the significance of considering uncertainty analysis for the difference between the objective function values of the solution
better informed decision making, at least compared to one-point under uncertainty and the deterministic solution. It was calculated
analysis (the case of the deterministic solution). as 229.246 for the case study under consideration.
In the last step of the uncertainty analysis, the optimization
problem is formulated as a stochastic programming problem and UP ¼ minðEq ðf ðx; y; qÞÞÞ  minðf ðx; yÞÞ (26)
solved using the sample average approximation (SAA) technique
The calculated EVPI value is very low as compared to the
(Birge and Louveaux, 1999). The selected WWTP network and the
objective function value of the deterministic solution which in-
cost breakdown for the solution under uncertainty are given in
dicates that the optimizer did not identify a better solution in the
Table 13.
design space. Hence this means that the current network selection
In order to summarize the results of the uncertainty analysis,
is mature to go further through the project development stages
several indicators are defined (Quaglia, 2013): Expected value of
(Fig. 1). The VSS value is found to be 0; since both stochastic and
perfect information (EVPI), value of stochastic solution (VSS) and
deterministic formulation ended up in the same network solution,
uncertainty price (UP).
the performance is the same and therefore the difference is 0. The
The EVPI represents the expected increase in the objective
UP on the other hand, is relatively high, 25% of the deterministic
function resulting from uncertainty. When EVPI is large (as
objective function value itself, which indicates that the un-
compared to the value of the objective function), the designer is
certainties inherent in the operation of the plant namely waste-
expected to work more on the design phase; a low EVPI, on the
water composition and load will likely increase the operation cost
other hand, indicates that the current design can further go through
of the project by that much within the project lifetime. Therefore,
the development funnel (Fig. 1). In the particular case study the
by considering all these indicators, and especially the UP indicator,
EVPI is calculated as 39.833 (4.36% of the deterministic value of
the user can conclude that the uncertainty in the model parameters
TAC).
affects the performance of the selected process network signifi-
cantly and the uncertainty should be reduced in order to achieve a
EVPI ¼ minðEq ðf ðx; y; qÞÞÞ  Eq ðminðf ðx; y; qÞÞÞ (24) more optimal design decision. If uncertainties affecting the system

In the VSS calculation, the difference between the performance

of the selected network (in the deterministic solution) under un- Table 13
Summary of SAA results.
certainty conditions and the solution of the problem under un-
certainty is calculated. The calculated VSS value for the Benchmark Network WW-PC-MLE-BP3-BP4-Thickener-AnD-
Discharge-Sludge-Biogas
Table 12 Aeration cost 461.773
Uncertainty mapping results. Landfill cost 615.289
Biogas price 699.496
Network Probability of Selected intervals
Electricity cost e
realization
Pumping cost 43.687
1 84% WW-BP1-MLE-BP3-BP4-Thickener-AnD- Mixing cost 27.044
Discharge-Sludge-Biogas Effluent penalty e
2 16% WW-BP1-UASB-Shar-An-BP4-BP5-Discharge- Capital cost 693.029
Sludge-Biogas Objective function value 1141.326
176 H. Bozkurt et al. / Environmental Modelling & Software 64 (2015) 164e176
cannot be decreased, e.g. by improving the available wastewater
characterization through a long term measurement campaign, then
one can consider designing a flexible network which is a solution
that is addressed elsewhere (Quaglia et al., 2013).
3.5.4. Future perspectives of the mathematical programming
framework
Current work focused on presenting the underlying founda-
tions, theory and practical implementation of the methodology
using an illustrative example. Future work will consider (a)
expanding and updating the superstructure database of technolo-
gies; (b) including resource recovery technologies for the WWTP;
(c) adding greenhouse gas emissions estimations in addition to the
energy consumption and production for sustainability analysis etc.;
and, (d) consideration of technical uncertainties such as key per-
formance parameters (yield, conversion, etc.) to make decision
making more robust (Sin et al., 2011; Quaglia et al., 2013).
Therefore as well, the expanded tool is expected to be an
enabling technology for design engineers for early stage idea gen-
eration and comparison of the treatment technologies at their
optimality, and for selection of the best among the alternatives with
respect to the desired criteria from the WWTP network (i.e. low
cost, low energy consumption, nutrient recovery, complying with
the effluent limitations, robustness etc.).
4. Conclusions
A superstructure based optimization methodology has been
developed to support optimal treatment process selection which is
a critical and challenging step in the early stage of wastewater
treatment plant design. A novel framework to help effective
formulation and management of the complexity of the optimiza-
tion problem is developed. The underlying theory and mathemat-
ical concepts, the required methods for its solution and analysis and
its practical implementation as a tool is presented, using the BSM1
as an illustrative case study. For the design space, the superstruc-
ture method is used representing several process alternatives,
which are mathematically described using a generic process model,
to enable the user to compare them in an optimization context. The
database consisting of all the necessary parameters for the generic
process models is developed for each alternative in the super-
structure and validated against a more rigorous model (ASM1). The
deterministic solution resulted in different objective function
values for different scenario definitions; however, when the data
uncertainty is taken into account considerable variations in the
network performance and objective function value were observed.
The tool is expected to support and facilitate generation and eval-
uation of ideas for identifying optimal solutions to design and
retrofitting of WWTPs. Future work will focus on further expansion
of the superstructure and its database (with MBR, SBR, sludge reject
water stream treatment alternatives, etc.), as well as on treatment
of uncertainties in the optimization based design of wastewater
treatment plants.
Appendix A. Supplementary data
Supplementary data related to this article can be found at http://
dx.doi.org/10.1016/j.envsoft.2014.11.023.
References
Alasino, N., Mussati, M.C., Scenna, N., 2007. Wastewater treatment plant synthesis
and design. Indust. Eng. Chem. Res. 46, 7497e7512.
Alasino, N., Mussati, M.C., Scenna, N., Aguirre, P., 2010. Wastewater treatment plant
synthesis and design: combined biological nitrogen and phosphorus removal.
Indust. Eng. Chem. Res. 49, 8601e8612.
ATV-DVWK Standard - A131E Dimensioning of single-stage activated sludge plants
(2000).
AWWA Michigan Section, 2006. Sodium Hypochlorite and Chlorine Gas: Design,
Handling and Storage Issues. Water Treatment Practices Committee, AWWA
Michican Section, Spring Regionals.
Birge, J.K., Louveaux, F., 1999. Introduction to Stochastic Programming. Springer,
New York, USA.
Bozkurt, H., Quaglia, A., Gernaey, K.V., Sin, G., 2014. Superstructure development
and optimization under uncertainty for design and retrofit of municipal
wastewater treatment plants. Comput. Aided Chem. Eng. 33, 37e42.
Comas, J., Alemany, J., Poch, M., Torrens, A., Salgot, M., Bou, J., 2003. Development of
a knowledge-based decision support system for identifying adequate waste-
water treatment for small communities. Water Sci. Technol. 48, 11e12,
393e400.
Copp, J.B., 2002. The COST Simulation Benchmark: Description and Simulator
Manual. Office for Official Publications for the European Communities,
Luxembourg.
Daigger G.T., 2005. Wastewater treatment plant of the future: decision analysis
approach for increased sustainability. Presented at the Water Security: Policies
& Investments Water Week 2005 Making Decisions on Water Quality. Wash-
ington DC.
GAMS Development Corporation, 2011. GAMS Development Corporation, Wash-
ington DC, USA.
Garrido-Baserba, M., Reif, R., Rodriguez-Roda, I., Poch, M., 2012. A knowledge
management methodology for the integrated assessment of WWTP configu-
rations during conceptual design. Water Sci. Technol. 66 (1), 165e172.
Gernaey, K.V., Jeppsson, U., Vanrolleghem, P.A., Copp, J.B. (Eds.), 2014. Bench-
marking of Control Strategies for Wastewater Treatment Plants. IWA Publishing,
London, UK. IWA Scientific and Technical Report No. 23.
Grossmann, I., 2005. Enterprise-wide optimization: a new frontier in process sys-
tems engineering. AICHE J. 51 (7), 1846e1857.
Hahn H., Rutz D., Ferber E., Kirchmayer, F., 2010. Examples for financing of biogas
projects in Germany, Austria, The Netherlands, Denmark and Italy. IEE Project
BiogasIN. Report no. D.3.2.,WP3.
Hamouda, M.A., Anderson, W.B., Huck, P.M., 2009. Decision support system in water
and wastewater treatment process selection and design: a review. Water Sci.
Technol. 60 (7), 1757e1770.
Henze, M., Gujer, W., Mino, T., Van Loosdrecht, M., 2000. IWA Scientific and Tech-
nical Report on Activated Sludge Models ASM1, ASM2, ASM2d and ASM3. IWA
Task group on Mathematical Modelling for Design and Operation of Biological
Wastewater Treatment.
Henze, M., van Loosdrecht, M.C.M., Ekama, G., Brdjanovic, D., 2008. Biological
Wastewater Treatment Principles, Modelling and Design. IWA Publishing,
Glasgow.
Lew, B., Tarre, S., Belavski, M., Green, M., 2004. UASB reactor for domestic waste-
water treatment at low temperatures: a comparison between a classical UASB
and hybrid UASB-filter reactor. Water Sci. Technol. 49 (11e12), 295e301.
Quaglia, A., Sarup, B., Sin, G., Gani, R., 2012. Integrated business and engineering
framework for synthesis and design of enterprise-wide processing networks.
Comput. Chem. Eng. 38, 213e223.
Quaglia, A., Sarup, B., Sin, G., Gani, R., 2013. A systematic framework for enterprise-
wide optimization: synthesis and design of processing networks under uncer-
tainty. Comput. Chem. Eng. 59 (5), 47e62.
Quaglia, A., 2013. An Integrated Business and Engineering Framework for Synthesis
and Design of Processing Networks. PhD Thesis. Technical University of
Denmark, p. 220.
Quaglia, A., Gargalo, C.L., Chairakwongsa, S., Sin, G., Gani, R., 2014a. Systematic
network synthesis and design: problem formulation, superstructure genera-
tion, data management and solution. Comput. Chem. Eng. 72, 68e86.
Quaglia, A., Pennati, A., Bogataj, M., Kravanja, Z., Sin, G., Gani, R., 2014b. Industrial
process water treatment and reuse: a framework for synthesis and design.
Industrial Eng. Chem. Res. 53, 5160e5171.
Rigopoulos, S., Linke, P., 2002. Systematic development of optimal activated sludge
process designs. Comput. Chem. Eng. 26, 585e597.
Siemens, 2009. Providing answers to the wastewater aeration market: air
compression solutions. Germany.
Sin, G., Gernaey, K.V., Neumann, M.B., van Loosdrecht, M.C.M., Gujer, W., 2011.
Global sensitivity analysis in wastewater treatment plant model applications:
prioritizing sources of uncertainty. Water Res. 45 (2), 639e651.
Takahara, H., Nakayama, S., Tsuno, H., 2006. Application of ozone to munic
ipal sewage treatment. International Conference Ozone and UV, Wasser
Berlin.
Tchobanoglous, G., Burton, F.L., Stensel, H.D., 2003. Wastewater Engineering:
Treatment and Reuse. McGraw-Hill Publishing, New York.
URL1, 2013. Europe's Energy Portal. www.energy.eu (visited on October 2013).
URL2, 2013. Confederation of European Waste-to-Energy plants. www.cewep.eu
(visited on October 2013).
Van Haandel, A.C., van der Lubbe, J.G.M., 2012. Handbook of Biological Wastewater
Treatment: Design and Optimisation of Activated Sludge Systems. IWA Pub-
lishing, London, UK.
Vidal, N., Ban ~ ares-Alc
antara, R., Rodrigues-Roda, I., Poch, M., 2002. Design of
wastewater treatment plants using a conceptual design methodology. Indust.
Eng. Chem. Res. 41, 4993e5005.
WEF and ASCE/EWRI, 2010. Design of Municipal Wastewater Treatment Plants. WEF
Press, Virginia.