lable at ScienceDirect

Environmental Modelling & Software 26 (2011) 492e497

Contents lists avai

Environmental Modelling & Software

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

Improving the performance of a WWTP control system by model-based setpointoptimisation

Javier Guerrero a,1, Albert Guisasola a,*, Ramon Vilanova b,2, Juan A. Baeza a,3

aDepartament d’Enginyeria Química, Escola d’Enginyeria, Universitat Autònoma de Barcelona, 08193 Bellaterra, Barcelona, SpainbDepartament de Telecomunicacions i Enginyeria de Sistemes, Escola d’Enginyeria, Universitat Autònoma de Barcelona, 08193 Bellaterra, Barcelona, Spain

a r t i c l e i n f o

Article history:Received 19 April 2010Received in revised form20 September 2010Accepted 6 October 2010Available online 18 November 2010

Keywords:A2/OControlOperational costsEffluent qualitySetpoint optimisationNutrient removal

* Corresponding author. Tel.: þ34 93 581 1879; faxE-mail addresses: franciscojavier.guerrero@uab

guisasola@uab.cat (A. Guisasola), ramon.vilanojuanantonio.baeza@uab.cat (J.A. Baeza).

1 Tel.: þ34 93 581 4798; fax: þ34 93 581 2013.2 Tel.: þ34 93 581 2197; fax: þ34 93 581 4031.3 Tel.: þ34 93 581 1587; fax: þ34 93 581 2013.

1364-8152/$ e see front matter � 2010 Elsevier Ltd.doi:10.1016/j.envsoft.2010.10.012

a b s t r a c t

The aim of this work was the improvement of a WWTP control system using a model-based setpointoptimisation. For this purpose, an anaerobic/anoxic/aerobic (A2/O) pilot WWTP was simulated using theIWA ASM2d model under different influent conditions. Several control strategies for an efficient bio-logical C/N/P removal were evaluated in this WWTP: i) open loop; ii) dissolved oxygen control in theaerated reactors; iii) maximum performance of nutrient removal; iv) optimised fixed setpoints for thecontrolled variables; v) daily optimised setpoints; vi) two different sets of optimised setpoints forweekdays and weekends and vii) hourly optimised setpoints. A single cost function based on theoperating costs by converting the effluent quality into monetary units was chosen for evaluating theplant performance (i.e. the control loops setpoints were optimised to obtain low effluent N and Pdischarges with the minimum costs). Setpoint optimisation was shown as a good tool to improve theperformance of the system. In this case study, control strategy (vi) was selected as the best choiceconsidering the trade-off cost-benefit. The optimised control system resulted in around a 45% decrease ofoperational costs with respect to the open loop scenario, a significant improvement of the effluentquality and a drastic decrease of the time above discharge limits.

� 2010 Elsevier Ltd. All rights reserved.

1. Introduction

Stringent legislation for wastewater treatment plants (WWTP)is currently a top driving force for the development of new treat-ment technologies and for the optimisation of the existing ones.Meeting stringent concentration requirements for C, N and Pdischargewithminimal costs has raised the need of amore efficientoperation. These plants can be redesigned to include new treat-ments or can be upgraded with new control structures. Althoughseveral solutions have been reported so far, a number of plants stilloperate without being updated.

Model-based optimisation of WWTP configuration has beenused for design purposes (Rivas et al., 2008; Ferrer et al., 2008),while the utilisation of automatic control systems has improved the

: þ34 93 581 2013..cat (J. Guerrero), albert.va@uab.cat (R. Vilanova),

All rights reserved.

performance of numerous WWTP (Benedetti et al., 2010; Cecil andKozlowska, 2010). However, little attention has been paid to thetuning of controllers (Ruano et al., 2010) or to the setpoint opti-misation for WWTP performance purposes (Stare et al., 2007).Additionally, the development of reliable models has provided toolsto allow the model-based optimisation of these control systems.For example, IWA ASM2d (Henze et al., 1999) is a complex kineticmodel able to describe biological C/N/P removal processes fromwastewater. Although this model has a large number of parameterswhich are difficult to indentify due to correlation problems(Machado et al., 2009a), it is able to provide an accurate descriptionof the process with its default parameter values.

With respect to control, single feedback controllers on essentialparameters have lead to better quality effluents in the last decades;however the efficiency of this strategy is limited by: i) the dynamicsof the influent or ii) the inherent complexity of the system sincecontrol actions applied in one unit can somehow affect posteriorsub processes (Alex et al., 2008). As proposed by Olsson et al.(2007), these problems could be overcome by integrating plant-wide control systems with a continuous retuning of the controlloops (e.g. via gain scheduling or using adaptive control) for theoptimisation of the overall plant operation.

J. Guerrero et al. / Environmental Modelling & Software 26 (2011) 492e497 493

On the other hand, most of the control strategies reported so farabout improvingWWTP operationwere based on C and N removal,while P removal wasn’t the focus yet (Baeza et al., 2002; Copp et al.,2002; Rivas et al., 2008; Benedetti et al., 2010). However, thecurrent knowledge gained on the enhanced biological phosphorusremoval (EBPR) process has raised the opportunity of developingnew control structures considering simultaneous C/N/P removal(Ingildsen et al., 2005). Designing new control strategies in sucha complex biological system is not a straightforward issue becauseof the high number of variables involved and the multivariablenature of the problem. In the case of model-based design, thecurrent biological models present a complex structure containinga large number of state variables that evolve transport and trans-formation processes (Sorour and Bahgat, 2006). In addition, thedesign of control structures for nutrient removal in WWTP shouldconsider the best pairing of controlled and manipulated variables(Machado et al., 2009b) because it will provide better systemcontrollability with fewer operating costs and the most effectivewastewater treatment.

Once designed, the efficiency of these new control strategiesneeds to be assessed including not only effluent quality but alsoplant economics. Balancing both issues is not a straightforwardtopic: the different parameters have to be properly weighted andeffluent fines and other costs as investment or man-power aretypically location dependent. Moreover, the problem can becomeevenmore complex if other criteria as environmental impact or riskof microbiology-related solids separation problems are evaluated(Flores et al., 2008). Therefore, to allow the comparison of controlstrategies, different authors proposed the evaluation of the plantperformance with a single cost function calculated with the maincosts involved in the plant operation and adding the effluentquality converted into monetary units (Vanrolleghem et al., 1996;Gillot et al., 1999).

In view of this background, the aim of this work was improvingthe performance of an A2/O WWTP with biological C/N/P removalconsidering criteria of effluent quality and operational costs (OC) bymeans of the improvement of its control system with a model-based setpoint optimisation. For this purpose, the WWTP wassimulated using IWA ASM2d and several control strategies wereevaluated using different influents.

2. Materials and methods

2.1. Plant description

The simulated plant (Fig. 1) is a continuous A2/O system for simultaneous C/N/Premoval consisting of four continuous stirred tank reactors (CSTRs) and one settler

Fig. 1. Scheme of the A2/O simulated pla

(modelled using the 10-layer model of Takács et al., 1991). The hydraulic modelmimicked the configuration of a real pilot plant (146L), where the best controlstrategies found in this work were to be further evaluated. The biological kineticmodel used to describe C/N/P removal was ASM2d (Henze et al., 1999). R1 is ananaerobic reactor favouring the uptake of organic matter by Polyphosphate Accu-mulating Organisms (PAO) and thus, further P removal. R2 is an anoxic reactorwhere the nitrate brought by the internal recycle (QRINT) is reduced by either thedenitrifying fraction of PAO (DPAO) or ordinary heterotrophic organisms (OHO). R3and R4 are two aerobic reactors where complete organic matter and P removal takesplace together with nitrification. The settler produces an effluent stream anda biomass enriched stream. Most of the latter is returned to R1 through the externalrecycle (QREXT) and the rest is purged (QW). The flow rate and the composition of theinfluent (QIN) varied in time according to the influents proposed by the IWA TaskGroup on Benchmarking (Gernaey and Jorgensen, 2004), being 0.25 m3 d�1 theaverage flow-rate value. Three different dynamic plant influents were simulated:Dry-2, Rain-2 and Storm-2. Each influent contained 14 days of data at 15-minintervals.

The simulated plant included four local control loops:

1. Dissolved Oxygen (DO) feedback PI-control in R3 and R4 using the oxygentransfer coefficient (kLa) as the manipulated variable.

2. Effluent ammonium was controlled by the DO setpoint in R3 and R4 (bothreactors had the same DO setpoint) using a cascade control structure. DOsetpoint limits were 0 and 4 mg DO$L�1.

3. Nitrate feedback PI-control in R2 by manipulating QRINT.4. Total suspended solids (TSS) feedback PI-control in R4 by acting in the QW. To

avoid the effect of a possible change in TSS concentration on the treatmentcapacity or the sludge age and in order to compare the removal efficiencyrelated only with the tested control strategies, TSS were considered as inven-tory variable (i.e. variables that must be controlled for a proper plantmanagement) and were controlled at a fixed setpoint of 4500 mgTSS L�1

(Steffens and Lant, 1999; Machado et al., 2009b).

2.2. Cost function development

Effluent quality and OC are the key parameters when evaluating the effective-ness of different wastewater treatment processes. The cost function proposed byVanrolleghem and Gillot (2002), which allows rewriting effluent quality in terms ofmonetary units, was adopted in the present work as the criterion for selecting thebest control proposal structure. The OC per m3 of influent of a WWTP can be esti-mated with equations (1)e(6). These equations are a modification of the method-ology described in Vanrolleghem and Gillot (2002). We propose to include theinfluent (QIN) in the cost calculations (equations 2e4, 6) in order to obtain the costsperm3 of wastewater treated. Thus, specific plant characteristics are avoided and thecomparison between different plants becomes easier.

OChV$m�3

i¼ gEðAE þ PEÞ þ gSPSP þ EF (1)

AE corresponds to energy invested in aeration, PE is the necessary pumpingenergy, SP the sludge production and EF the effluent fines, gE (0.1 V$kWh�1)represents the cost of 1 kWh and gSP (5$10�4 V$g�1) stands for the cost of thetreatment of 1 g of produced sludge (Stare et al., 2007). The aeration energy (AE)wascalculated as proposed in Jeppsson (2005) by using equation (2), where kLai is theglobal oxygen transfer coefficient [d�1] of each aerobic reactor.

nt for simultaneous C/N/P removal.

J. Guerrero et al. / Environmental Modelling & Software 26 (2011) 492e497494

AEhkWh m�3

i¼ 24

ðt2 � t1Þ$

Zt2t1

"1

QINðtÞ$X4i¼3

0:0007$ðkLaiðtÞÞ2$

Vi

Vref

!

þ 0:3267$kLai ðtÞ$

Vi

Vref

!!#dt (2)

The pumping energy (PE) was calculated with equation (3), where PF(0.04 kWh m�3) converts the pump flow into required energy (Copp et al., 2002).

PEhkWh m�3

i¼ PF

ðt2 � t1Þ$

Zt2t1

�1

QINðtÞðQRINTðtÞ þ QREXTðtÞ þ QWðtÞÞ

�dt (3)

Sludge production (SP) was calculated as equation (4):

SPhg m�3

i¼ 1

ðt2 � t1Þ$

Zt2t1

�1

QINðtÞ$TSSW ðtÞ$QWðtÞ

�dt (4)

The solids content in the purge (TSSW) was estimated via mass balance of thesettler (equation (5)), using the total suspended solids concentration in R4 (TSSR4),assuming negligible suspended solids concentration in the effluent and biomasshold up in the settler constant.

TSSW ðtÞhgTSS m�3

i¼�QINðtÞ þ QREXTðtÞQWðtÞ þ QREXTðtÞ

�$TSSR4ðtÞ (5)

Effluent fines (equation (6)) were calculated comparing the effluent ammonium,total nitrogen (TN) and phosphate with the discharge limits, being TN the sum ofnitrogen as ammonium and nitrate in the effluent (Vanrolleghem et al., 1996).

EFhVm�3

i¼ 1

ðt2 � t1Þ$

Zt2t1

�1

QINðtÞ$� X

j¼NH4 ;

TN;PO4

�QEFF ðtÞ$Daj$CEFF

j ðtÞ þ�QEFF ðtÞ$

hb0;j

þ�CEFFj ðtÞ � CL;j

�$�Dbj � Daj

i�$�Heaviside

�CEFFj ðtÞ � CL;j

�����dt

(6)

Where CjEFF and CL,j are the effluent concentration and discharge limit of the

pollutant “j”, respectively; Daj is the slope of the curve EF versus CjEFF when Cj

EFF islower than or equal to CL,j;Dbj is the slope of the same curvewhen Cj

EFF is higher thanCL,j and b0,j is the increment of fines when Cj

EFF was higher than CL,j. The Heavisidefunction is equal to onewhen Cj

EFF is greater than CL,j. Otherwise, its value is zero. Thevalues of all the parameters involved in the EF calculation are given in Table 1. Theparameters for ammonium and TN were obtained from Stare et al. (2007). Phos-phate-related parameters were assumed equal to ammonium parameters, except forthe effluent discharge limit.

2.3. Simulation and optimisation

The setpoint optimisation aimed to find the maximum performance thata specific control structure could achieve. Therefore, perfect knowledge of theinfluent characteristics was required (i.e. using the influents proposed by the IWATask Group on Benchmarking: Dry-2, Rain-2 and Storm-2).

The dynamic model was implemented in MATLAB� and integrated using ode15s,a variable order method recommended for stiff systems. Each control strategy wassimulated during 28 days under Dry-2 conditions. The setpoint values of the controlloops were optimised using a cost function calculated only with the last 14 dayssimulated. The starting point for each simulation was the steady state of thesimulation under open-loop conditions with Dry-2 influent during 100 days.

2.4. Evaluation of different optimisation methods

Optimisation of a complex system as the operation of a WWTP is a challengingtask, as the minimisation of functions depending on highly nonlinear dynamicsystemsmayeasily result in localminima. A previous test of different searchmethodswas required in order todeterminewhichmethod avoided localminima. For this aim,the Genetic Algorithm, the NeldereMead and Pattern Search (PS) methods weretested and the minimal costs were found with the PS method. Detailed informationabout this evaluation can be found in the Supplementary content.

Table 1Parameters used to evaluate the effluent fines.

Effluent Variable Daj (V$kg�1) Dbj (V$kg�1) b0,j (V$m�3) CL,j (mg L�1)

Ammonium 4.00 12.00 2.70 � 10�3 4.00Total Nitrogen 2.70 8.10 1.40 � 10�3 18.00Phosphate 4.00 12.00 2.70 � 10�3 1.50

2.5. Description of the simulated control strategies

The different control strategies and setpoint optimisation methodologies testedare next summarised:

� Open Loop (OL): The systemwas simulated under open-loop conditions (i.e. allthe control strategies were disabled, except for TSS control loop in R4). Theaeration was kept constant in R3 and R4 with a kLa of 600 d�1 and 400 d�1

respectively to avoid oxygen limitations. During this period, the flow rates ofQRINT and QREXT were 0.763 m3 d�1 (three times the average influent flow rate)and 0.250 m3 d�1 (the average influent flow rate), respectively. This configu-ration provides reasonable P removal according to Gernaey and Jorgensen,(2004).

� DO control (DOC): DO control was activated with a setpoint of 4 mg DO L�1 inR3 and R4. The flow rates were kept at the previous values.

� Maximum performance for nutrient removal (MPR): Control setpoints were fixedto obtain themaximum removal performance. Ammonia setpoint was 0mg L�1

and nitrate setpoint was optimised tominimise nitrate in the effluent (Table 2).DO setpoint limits were 0 and 4 mg DO L�1.

� Ammonium and nitrate fixed optimum setpoints (A&N-FOS): This strategy con-sisted of using fixed optimum ammonium and nitrate setpoints during thesimulated period.

� Ammonium and nitrate daily variable optimum setpoints (A&N-DVOS): The set-points were daily optimised according to the influent flow pattern of the plant.

� Ammonium and nitrate weekly variable optimum setpoints (A&N-WVOS): Themajor inlet variations take place between weekdays and the weekend. Thus,two different sets of setpoints were optimised, one for weekend and one forthe weekdays.

� Ammonium and nitrate hourly variable optimum setpoints (A&N-HVOS): Thecontrol setpoints were hourly optimised according to the influent flow patternof the plant.

3. Results and discussion

3.1. Comparison of the different control strategies and setpointoptimisations

Table 2 summarises the results for the different control strategiesdescribedabove. For the three influents tested, theoptimumNeNO3

�

and NeNH4þ setpoints to minimise OC were calculated. All the

proposed control strategies were more efficient (in terms of lowereffluent discharges and lower OC) than the open-loop operation.

Each of the control strategies was simulated for 28 days. More-over, some of these simulations were also run during 100 days inorder to confirm that 28 days was a good approximation witha balanced computation time, with results differing less than 1%. Inthe OL strategy, ammonium and nitrate concentrations were alwaysbelow the discharge limits (Table 2). However, the effluent averagephosphate concentration (9.35 mg PePO4

3- L�1 for Dry-2 influent)waswell above the discharge limit resulting in the highest OC amongall the control strategies proposed for the three studied influents.

Unnecessary aeration was avoided with the DOC strategy,leading to a decrease of aeration costs. Besides, less DO in thereactors involved lower nitrifying activity, lower nitrate production,lower nitrate recycle and better P removal. As a result, a reductionin the effluent fines was also observed for the three influents tested.

In the MPR strategy, ammonium setpoint was fixed to 0 and thenitrate setpoint in R2was optimised tominimise the effluent nitrateconcentration (Table 2). This strategy did not result in the minimalOC because of the high time above the discharge limits caused bythe high effluent phosphate concentration. Moreover, the ammo-nium concentration could not obviously be reduced to 0 mg L�1

even though the maximum aeration was reached (4 mg DO L�1).When A&N-FOS was tested (Fig. S1 in Supplementary content),

the major cost reduction appeared in the aeration term. Only thestrictly necessary oxygen for nitrification was supplied to achievethe optimised ammonium setpoint, which was increased withrespect to the MPR strategy (Table 2). The effluent fines decreaseddue to a significant reduction of the time that phosphate was abovethe discharge limits. The EBPR enhancement could be explained

Table 2Summary of the different control strategies for the Dry-2, Rain-2 and Storm-2 influents and the main results: operational costs, applied setpoints, time that effluent quality isabove the discharges limits and mean effluent concentrations. CI is the cost improvement with respect to the open-loop scenario expressed in percentage. wd Weekdays, weWeekend, ra Rain and st Storm. OL Open loop, DOC DO control, MPR Maximum performance for nutrient removal, A&N-FOS Ammonium and nitrate fixed optimum setpoints,A&N-DVOS Ammonium and nitrate daily variable optimum setpoints, A&N-WVOS Ammonium and nitrate weekly variable optimum setpoints and A&N-HVOS Ammonium andnitrate hourly variable optimum setpoints.

Operational Costs (V$m�3) Setpoints (mg L�1) Time above limits (d) Mean concentration (mg L�1)

AE PE SP EF OC CI (%) NeNH4þ NeNO3

� NeNH4þ TN PePO4

3-

Dry-2 OL 0.140 0.016 0.053 0.132 0.341 e e e 14.00 0.59 9.11 9.35DOC 0.099 0.016 0.054 0.122 0.291 14.83 e e 14.00 0.62 8.96 8.50MPR 0.096 0.025 0.048 0.138 0.308 9.70 0.00 1.40 14.00 0.66 7.52 10.11A&N-FOS 0.050 0.010 0.067 0.063 0.190 44.31 2.90 0.10 8.32 2.90 11.66 2.03A&N-WVOS 0.051 0.009 0.069 0.059 0.188 44.90 3.00wd 0.10wd 5.60 2.43 12.27 1.34

1.90we 0.06we

A&N-DVOS 0.050 0.008 0.069 0.061 0.188 44.90 Variable Variable 5.10 3.25 14.27 0.89A&N-HVOS 0.036 0.006 0.070 0.101 0.214 37.28 Variable Variable 12.44 7.02 15.49 0.33

Rain-2 OL 0.140 0.019 0.024 0.138 0.321 e e e 14.00 0.75 8.18 8.96DOC 0.096 0.019 0.025 0.129 0.270 15.93 e e 14.00 0.83 8.02 8.29MPR 0.098 0.027 0.010 0.140 0.275 14.28 0.00 1.25 14.00 0.71 6.95 9.37A&N-FOS 0.053 0.013 0.029 0.069 0.164 48.86 2.70 0.10 9.04 2.70 10.20 2.37A&N-WVOS 0.055 0.009 0.035 0.064 0.163 49.30 3.30wd 0.10wd 3.90 2.88 12.01 0.94

2.10we 0.06we

2.90ra 0.10ra

A&N-DVOS 0.056 0.008 0.036 0.063 0.163 49.24 Variable Variable 3.53 3.00 12.28 0.71A&N-HVOS 0.038 0.007 0.037 0.090 0.172 46.34 Variable Variable 9.05 5.74 14.01 0.50

Storm-2 OL 0.140 0.017 0.034 0.136 0.326 e e e 14.00 0.68 8.67 9.18DOC 0.098 0.018 0.035 0.126 0.277 15.10 e e 14.00 0.74 8.55 8.37MPR 0.098 0.027 0.031 0.141 0.296 9.25 0.00 1.40 14.00 0.74 7.29 9.78A&N-FOS 0.055 0.012 0.052 0.066 0.185 43.47 2.80 0.10 8.15 2.80 11.28 2.01A&N-WVOS 0.058 0.009 0.056 0.062 0.184 43.65 3.10wd 0.10wd 4.60 2.53 12.03 1.34

2.30we 0.06we

1.60st 1.30st

A&N-DVOS 0.057 0.009 0.056 0.063 0.184 43.50 Variable Variable 4.43 3.09 14.98 0.60A&N-HVOS 0.037 0.007 0.057 0.106 0.206 36.76 Variable Variable 11.37 5.95 13.73 0.70

J. Guerrero et al. / Environmental Modelling & Software 26 (2011) 492e497 495

due to the EBPR requirement of alternating conditions withdifferent electron acceptor conditions. Nitrate presence in theanaerobic reactor coming with the external recycle is one of themajor causes of EBPR failure since denitrifiers could outcompetePAO for the influent COD. The low nitrate recycled to R2 (to ensureoptimum nitrate setpoint) did not affect PAO activity (P release wasalso observed in R2, indicating anaerobic conditions instead ofanoxic). The previous control strategies only achieved good C and Nremoval, but were detrimental for PAO growth (steady state valuesof PAO population of 18 mg COD L�1 and 760 mg COD L�1 wereobtained for MPR and A&N-FOS, respectively). This fact reveals theimportance of a setpoint optimisation with a cost function whichweighs up all the nutrient concentrations in the effluent (includingP). In other words, the optimal setpoints chosen favoured phos-phorus removal, although there was not any specific control loopimplemented for P removal.

The simulated influent flow pattern presented significant timevariations mimicking real WWTP influents. Hence, the A&N-DVOSstrategy was believed to be a sensible alternative to reduce the OC(Fig. S2 in Supplementary content). The control setpoints wouldadapt the plant operation to the daily influent variations intensi-fying the C/N/P removal when necessary. Unexpectedly, theimplementation of this control strategy did not result in animportant OC reduction when compared to the A&N-FOS strategy(Table 2), despite a 40% decrease of time above discharge limits inthe three studied scenarios. The results of A&N-DVOS evidencedsubstantial differences between weekend and weekdays setpoints.For this reason, the utilization of two different sets of setpoints (onefor weekend and one for the weekdays) was tested in the strategyA&N-WVOS (Fig. 2). For Rain-2 and Storm-2 influents, a third set ofsetpoints was proposed; therefore differences in rain and stormperiods were also observed in daily setpoint optimisation. Never-theless, the results of A&N-WVOS were very similar to the ones

obtained with A&N-DVOS (Table 2); however this strategy alloweda reduction in the number of parameters to be optimised. Withthree different sets of setpoints (i.e. weekdays, weekends andstorm) the results were very close to the approach of A&N-DVOS.The implementation of A&N-WVOS strategy could be more feasiblein a real WWTP than the daily optimisation, although a lowincrease in time above discharges limits could be observed whencompared to A&N-DVOS.

When hourly optimised setpoints strategy was tested (A&N-HVOS), the OC increased with respect to A&N-FOS or A&N-WVOS.Although the system achieved the highest phosphate removal(Table 2), ammonium and nitrate removal was worsened and thetime above the discharged limits and the effluent fine costsincreased. The A&N-HVOS strategy aimed at finding the optimumset of setpoints for a specific hour without taking into account thestate of the system at the end of that hour, i.e. the initial conditionsfor the next optimisation. The influent pattern had load variationsalong the time, so the optimum setpoints applied in low loadperiods reduce the system treatment capacity. When peak loadappeared, the plant was not capable to remove efficiently thepollutants in a short term. This fact led the system to a new situationwhere the new optimised set of setpoints guaranteed theminimumOC in that hour, but did not reach a decrease in the total costs afterthe whole period of 14 days. When A&N-DVOS was applied, theaforementioned behaviour for A&N-HVOS was not observed due tothe daily variation distribution of the influent pattern.

Finally, a sensitivity analysis for the A&N-WVOS was performedin order to analyse how the values of the effluent fines would affectthe setpoint optimisation (Table S2 in Supplementary content). Forthis purpose, the effluent fines values (Table 1) were increased anddecreased � 50%. The costs related to the plant operation (aeration,pumping and the sludge production) presented a slight variation(less than �2.5%). However, the optimum setpoints were reduced

Fig. 2. A&N-WVOS control strategy behaviour for Dry-2 influent. (A) Ammonium R4; (B) Phosphate R4; (C) Total Nitrogen R4; (D) Nitrate R2; (E) TSS R4; (F) DO setpoint R4; (G)QRINT; (H) QW. Dashed lines belong to system measurements, dotted lines belong to the limit of pollutant (4 mg NeNH4

þ L�1, 15 mg TN L�1 and 1.5 mg PePO4�3 L�1) and solid lines to

optimised setpoints.

J. Guerrero et al. / Environmental Modelling & Software 26 (2011) 492e497496

when using higher effluent fines, which resulted in a decrease inthe effluent time above limits (�15.5%). Reducing the effluent fineshad a low effect in the time above discharge limits (1.3%). Theseresults suggest that the cost related to the plant operationmay havean excessive weight in contrast to the monetary effluent qualitypenalties. The weight selected for the effluent quality costs is a keyfactor to avoid not subordinating the costs of plant operation(aeration, pumping and the sludge production) to the quality of theeffluent. To overcome these problems, multi-criteria tools(Benedetti et al., 2010) could be a useful alternative since they allowoptimising a system with different criteria which are not condi-tional on each other.

4. Conclusions

The present study concludes that a model-based optimisation ofthe setpoints of WWTP control loops can improve the WWTPmanagement, providing low effluent discharges with minimal OC.This optimisation can be very useful in full-scale municipal WWTPwhere the implementation of nutrient sensors and control systemscan be payback in short periods.

In the case presented, a decrease of the OC up to 45% anda reduction up to 72% of the time that the effluent quality was above

the discharge limits were obtained when compared to the open-loop scenario. The implementation of a control strategy with themodel-based setpoint optimisation of ammonium and nitrateconcentration improves not only the removal of these compounds,but also enhances EBPR.

The implementation of different sets of setpoints for weekdays,weekends and storm or rain episodes (A&N-WVOS) was the mostefficient control strategy considering the OC and the time that theeffluent quality was above the discharge limits. Nevertheless, theutilization of a fixed set of setpoints during all the week (A&N-FOS)also provided reasonable performance.

Finally, it was observed that the hourly retuning of the controlsetpoints was not an efficient strategy because it increased the totalcosts after the whole period of 14 days. These results demonstratethat a more complex control strategy does not result always ina plant performance improvement compared to more simplestrategies.

Acknowledgment

This work was supported by the Spanish Ministerio de Ciencia yTecnología (CTQ2007-61756/PPQ and DPI2010-15230). The authorsfrom the Departament d’Enginyeria Química of the UAB are

J. Guerrero et al. / Environmental Modelling & Software 26 (2011) 492e497 497

members of the GENOCOV group (Grup de Recerca Consolidat de laGeneralitat de Catalunya, 2009 SGR 815).

Appendix. Supplementary data

Supplementary data associated with this article can be found inthe on-line version, at doi:10.1016/j.envsoft.2010.10.012.

References

Alex, J.L. Benedetti, L., Copp, J., Gernaey, K.V., Jeppsson, U., Nopens, I., Pons, M.N.,Rosen C., Steyer, J.P., Vanrolleghem, P., Winkler, S., 2008. Benchmark SimulationModel no.1 (BSM1). Tech. Report no. LUTEDX/(TEIE- 7229)/1e62/(2008).

Baeza, J.A., Gabriel, D., Lafuente, J., 2002. Improving the nitrogen removal efficiencyof an A2/O based WWTP by using on-line knowledge based expert system.Water Res. 36, 2109e2123.

Benedetti, L., De Baets, B., Nopens, I., Vanrolleghem, P.A., 2010. Multi-criteriaanalysis of wastewater treatment plant design and control scenarios underuncertainty. Environ. Model. Softw. 25, 616e621.

Cecil, D., Kozlowska, M., 2010. Software sensors are a real alternative to truesensors. Environ. Model. Softw. 25, 622e625.

Copp, J.B., Spanjers, H., Vanrolleghem, P.A., 2002. Respirometry in Control of theActivated Sludge Process: Benchmark Control Strategies. Scientific and Tech-nical Report no 11. IWA Publishing, United Kingdom.

Ferrer, J., Seco, A., Serralta, J., Ribes, J., Manga, J., Asensi, E., Morenilla, J.J., Llavador, F.,2008. DESASS: a software tool for designing, simulating and optimisingWWTPs. Environ. Model. Softw. 23, 19e26.

Flores, X., Rodríguez-Roda, I., Gürkan, S., Gernaey, K.V., 2008. Multi-criteria evalu-ation of wastewater treatment plant control strategies under uncertainty.Water Res. 42, 4485e4497.

Gernaey, K.V., Jorgensen, S.B., 2004. Benchmarking combined biological phosphorusand nitrogen removal wastewater treatment processes. Control Eng. Pract. 12(3), 357e373.

Gillot, S., De Clercq, B., Defour, D., Simoens, F., Gernaey, K., Vanrolleghem, P.A., 1999.Optimization of wastewater treatment plant design and operation usingsimulation and cost analysis. In: Proceedings of 72nd Annual WEF Conferenceand Exposition, New Orleans, USA, 9e13 October 1999.

Henze, M., Gujer, W., Mino, T., Matsuo, T., Wentzel, M.C., Marais, G.V.R., vanLoosdrecht, M.C.M., 1999. Activated sludge model no 2d, ASM2d. Water Sci.Technol. 39 (1), 165e182.

Ingildsen, P., Rosen, C., Gernaey, K.V., Nielsen, M.K., Guildal, T., Jacobsen, B.N., 2005.Modelling and control strategy testing of biological and chemical removal atAvedore WWTP. Water Sci. Technol. 53 (4e5), 105e113.

Jeppsson, U., 2005. Aeration and mixing energy for BSM1, BSM1_LT and BSM2.http://benchmarkwwtp.org Benchmark Internal Report September 2005,available for download from.

Machado, V.C., Tapia, G., Gabriel, D., Lafuente, J., Baeza, J.A., 2009a. Systematicidentifiability study based on the Fisher information matrix for reducing thenumber of parameters in calibration of an activated sludge model. Environ.Model. Softw. 24, 1274e1284.

Machado, V.C., Gabriel, D., Lafuente, J., Baeza, J.A., 2009b. Cost and effluent qualitycontrollers design based on the relative gain array for nutrient removal WWTP.Water Res. 43 (20), 5129e5141.

Olsson, G., Nielsen, M., Yuan, Z., Lynggaard-Jensen, A., Steyer, J.P., 2007. Instru-mentation, Control and Automation in Wastewater Systems. Scientific andTechnical Report no 15. IWA Publishing, London.

Rivas, A., Irizar, I., Ayesa, E., 2008. Model-based optimisation of wastewater treat-ment plants design. Environ. Model. Softw. 23, 435e450.

Ruano, M.V., Ribes, J., Sin, G., Seco, A., Ferrer, J., 2010. A systematic approach for fine-tuning of fuzzy controllers applied to WWTPs. Environ. Model. Softw. 25,670e676.

Sorour, M.T., Bahgat, L.N.F., 2006. Simulation analysis of ASM/Takács models in theBSM1 configuration. Environ. Technol. 27, 1163e1170.

Stare, A., Vrecko, D., Hvala, N., Strmcnik, S., 2007. Comparison of control strategiesfor nitrogen removal in an activated sludge process in terms of operating costs.Water Res. 41 (9), 2004e2014.

Steffens, M.A., Lant, P.A., 1999. Multivariable control of nutrient-removing activatedslugde systems. Water Res. 33 (12), 2864e2878.

Takács, I., Patry, G.G., Nolasco, D., 1991. A dynamic model of the clarificationthickening process. Water Res. 25 (10), 1263e1271.

Vanrolleghem, P.A., Gillot, S., 2002. Robustness and economic measures ascontrol benchmark performance criteria. Water Sci. Technol. 45 (4e5),117e126.

Vanrolleghem, P.A., Jeppsson, U., Carstensen, J., Carlsson, B., Olsson, G., 1996.Integration of wastewater treatment plant design and operation ea systematic approach using cost functions. Water Sci. Technol. 34 (3e4),159e171.