Journal of Water Process Engineering 53 (2023) 103802 Contents lists available at ScienceDirect Journal of Water Process Engineering journal homepage: www.elsevier.com/locate/jwpe Nitrous oxide reduction in wastewater treatment plants by the regulation of the internal recirculation flow rate with a fuzzy controller I. Santín a, b, M. Meneses a, C. Pedret a, M. Barbu a, c, R. Vilanova a a Department of Telecommunications and Systems Engineering, School of Engineering, Autonomous University of Barcelona, 08193 Bellaterra, Barcelona, Spain b Salesian University School of Sarrià, Passeig Sant Joan Bosco 74, 08017 Barcelona, Spain c Department of Automatic Control and Electrical Engineering, “Dunarea de Jos” University of Galati, 800008 Galati, Romania A R T I C L E I N F O A B S T R A C T Keywords: The reduction of greenhouse gas emissions due to anthropogenic causes is one of the world’s main challenges to Wastewater biological treatment face climate change. Wastewater treatment plants are necessary to improve the quality of wastewater before it is Fuzzy control discharged into the receiving environment, but they have the disadvantage of generating nitrous oxide emissions Internal recirculation flow rate during the biological treatment, which is a potent greenhouse gas. Avoiding partial nitrification by increasing Benchmark simulation model no. 2 Control strategies dissolved oxygen is one of the ways to reduce these emissions. However, this article proposes to face a reduction of nitrous oxide emissions by reducing oxygen to minimum levels causing heterotrophic microorganisms to reduce nitrous oxide to dinitrogen. To achieve this objective, the present work proposes a regulation of the internal recirculation flow rate of the biological treatment by means of a fuzzy controller. This regulation is added to a usual control strategy in wastewater treatment plants, which achieves satisfactory results with respect to water quality and operational costs but that generates high nitrous oxide emissions. The Benchmark Simu­ lation Model no. 2 Gas is used as working scenario, which includes the two main nitrous oxide emission path­ ways: heterotrophic denitrification and ammonia oxidizing bacteria denitrification. The proposed internal recirculation manipulation is shown to achieve nitrous oxide reductions of 26.70 and 30.83 % in different time periods with a slight effluent quality improvement and an operational cost reduction. 1. Introduction necessary to reduce wastewater pollution before it is discharged into the receiving environment, improving its quality and thus maintaining the The excess of Greenhouse Gases (GHG) is one of the main global aquatic ecosystem life. Among the WWTP processes, the secondary problems, and their reduction is a primary objective in the fight against treatment is essential to reduce the concentrations of suspended solids, climate change. Nitrous oxide gas (N2 O) is a powerful GHG that con­ organic matter and nutrients. In the secondary treatment, biological tributes to global warming of the planet [1–4]. processes take place, where microorganisms achieve the aforemen­ On the other hand, wastewater treatment plants (WWTPs) are tioned reductions. However, GHG and especially N2 O are generated Abbreviations: ACC, ammonia cascade control; AE, aeration energy (kWh/day); AOB, ammonia oxidizing bacteria; ASM1, Activated Sludge Model no. 1; BOD5 , 5- day Biological Oxygen Demand (mg/l); BSM2, Benchmark Simulation Model no. 2; BSM2G, Benchmark Simulation Model no. 2 Gas; CO2 , carbon dioxide (kg/day); CODt , total chemical oxygen demand (mg/l); EC, external carbon (kg/day); EQI, Effluent Quality Index (kg of pollutants/day); GHG, greenhouse gases; HRT, hy­ draulic retention time (s); KL a, oxygen transfer coefficient (d− 1 ); KL ai , oxygen transfer coefficient in tank i (d− 1 ); N2 O, nitrous oxide gas; N2 O, i, nitrous oxide gas in tank i; N2 , nitrogen gas/dinitrogen; NOB, nitrite oxidizing bacteria; OCI, Operational Cost Index; PE, pumping energy (kWh/day); PI, Proportional-Integral; Q, flow rate (m3 /day); Qa , internal recirculation flow rate (m3 /day); Qin , influent flow rate (m3 /day); Qw , wastage flow rate (m3 /day); Qpo , flow rate from the primary clarifier (m3 /day); SNtot , total nitrogen concentration (mg/l); SNtot,e , total nitrogen concentration in the effluent (mg/l); SNH , ammonium and ammonia nitrogen concentration (mg/l); SNH,i , ammonium and ammonia nitrogen concentration in tank i (mg/l); SNH,in , ammonium and ammonia nitrogen concentration at the input of the primary clarifier (mg/l); SNH,e , ammonium and ammonia nitrogen concentration in the effluent (mg/l); SNO , dissolved nitric oxide concentration (mg/l); SN2 O , dissolved nitrous oxide concentration (mg/l); SN2 , dissolved dinitrogen concentration (mg/l); SNO2 , nitrite concentration (mg/l); SNO3 , nitrate concentration (mg/l); SNKj , Kjeldahl nitrogen (mg/l); SO , dissolved oxygen concentration (mg/l); SO,i , dissolved oxygen concentration in tank i (mg/l); SFA , free ammonia; SFNA , free nitrous acid; SS , readily biodegradable substrate; Tas , temperature (◦ C); TSS, Total Suspended Solids (mg/l); WWTP, wastewater treatment plants; XB,A1 , ammonia oxidizing bacteria concentration (mg/l); XB,A2 , nitrite oxidizing bacteria concentration (mg/l); XB,H , heterotrophic microorganisms concentration (mg/l). https://doi.org/10.1016/j.jwpe.2023.103802 Received 20 October 2022; Received in revised form 2 April 2023; Accepted 5 May 2023 Available online 2 June 2023 2214-7144/© 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by- nc-nd/4.0/). I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 during these processes. and ammonia nitrogen concentration (SNH ) in the fifth tank (SNH,5 ) The increasing interest in climate change due to GHG emissions has cascade controls that manipulate SO set-points to take into account both emphasized the need to establish approaches to better control and N2 O reduction and effluent quality. Boiocchi et al. [21] designs a fuzzy operate WWTPs at the plant-wide scale. This fact, jointly with the wide controller to manipulate the SO set-points taking into account SNH and acceptance of benchmark simulation models to compare the perfor­ nitrate concentration (SNO3 ) at the inlet and outlet of the aerobic reactor. mance of different control approaches as well as to evaluate them before Santín et al. [9] designs a fuzzy controller for a plant-wide control its practical implementation, motivated the inclusion of accounting for adding more manipulated variables, not only with the aim of reducing GHG emissions in plant-wide models. One of the most used plant-wide GHGs, but also to improve effluent quality and costs. However, the N2 O models that takes into account the GHG emissions is the BSM2G reduction is also carried out in this case by manipulating the SO set- Flores-Alsina et al. [5]. Several works in the literature have applied this points. In Solís et al. [20], two of the applied control strategies aim to model to study the effect on GHG emissions when implementing reduce N2 O by SNO2 and dissolved N2 O (SN2 O ) control, both by means different control/operational strategies Barbu et al. [6]; Flores-Alsina ofSO set-point manipulation. et al. [5,7]; Santín et al. [8,9]; Sweetapple et al. [10] being de facto The referred works regulate SO in the aerobic tanks of the biological reference model up to now for testing and comparing control strategies treatment, increasing it during the necessary periods to carry out com­ related to mitigation of N2 O emissions. There are however some con­ plete nitrification in order not to generate N2 O emissions through the siderations that need to be highlighted regarding the accuracy of the AOB denitrification pathway. However, in comparison with partial N2 O predictions read from this model. In fact, this model is based on the nitrification, by complete nitrification more SNO3 is generated and ASMN model suggested by Hiatt and Grady [11] as an extension to the consequently more total nitrogen (SNtot ), which is a nutrient that causes well-known ASM models Henze et al. [12]. It accounts for two nitrifying eutrophication. This is due to the fact that by partial nitrification, some populations; ammonia oxidizing bacteria (AOB) and nitrite oxidizing of the nitrogen dissolved in the water present in SNO , SN2 O and SN2 bacteria (NOB) and also incorporates four step denitrification. It is molecules is converted into gas. important to highlight that N2 O production is only considered during The present work also aims to reduce N2 O through the application of heterotrophic denitrification. Other important pathways like N2 O pro­ control techniques. However, this reduction is not carried out by an SO duction during nitrification are not considered. increase. The objective is to ensure that the biological treatment is Recent advances published recently in the literature reported capable of oxidizing SNH with the minimum possible SO added during the different extensions and improvements regarding the different N2 O periods when large N2 O emissions are most likely to occur. This results pathways and their consideration and incorporation into existing in partial nitrification, which could lead to N2 O emissions through the models (ASM variants) presenting them as improvements towards N2 O AOB denitrification pathway. However, if SO levels are low enough, emission considerations. See for example the reviews of Ni and Yuan heterotrophic microorganisms (XB,H ) consume the oxygen from N2 O, [13], Mannina et al. [14], Vasilaki et al. [15] and Vasilaki et al. [16]. reducing it to dinitrogen (N2 ). In addition, by the reduction of the in­ Even these works confirm the subject is reaching a stage of maturity, ternal recirculation flow rate (Qa ), the hydraulic retention time (HRT) there still do exists uncertainty and lack of agreement in a common can be increased. This fact allows the oxidation of the SNH peaks to be reference model. As pointed out in Vasilaki et al. [15], the complexity carried out for a longer period of time, reducing the N2 O emission peaks. and over-parameterization of such models, makes it difficult and a real It should be noted that the N2 O reduction to N2 depends on the amount challenge their practical application. Even those complexities, some of organic matter since there is a relationship between oxygen con­ recent results based on the ASM3d extensión can be found in Blomberg sumption and organic matter degradation by XB,H . To apply the pro­ et al. [17] where it was also tested on a full-scale plant. Same data was posed control in another plant, an analysis of the amount of organic taken in Maktabifard et al. [18] to also test the ASM2d extension pre­ matter present in the aerobic tanks should be carried out. The possibility viously used in Zaborowska et al. [19]. On another direction, in Solís of adding carbon in the aerated tanks can be assessed in the case of there et al. [20] an extension of the above mentioned BSM2G benchmark is not enough organic matter. scenario (BSM2-PSFe-GHG) including biological COD/N/P removal, Thus, the main novelty of this work compared to the above refer­ GHG emissions, and chemical and physico-chemical models to evaluate enced papers is the achievement of the N2 O reduction by reducing SO , resource recovery in Water resource Recovery Facilities is proposed. instead of increasing it. This is carried out using a control strategy Hence, even all the previously mentioned works show solid advances commonly applied in WWTPs as starting point. This control strategy is towards a better description of the N2 O pathways, therefore, the quan­ designed to obtain optimal results with respect effluent quality and tification of N2 O emissions, there is still the need to reach an agreement operational costs, but does not take into account N2 O emissions. Spe­ on a reference scenario to test and evaluate operation strategies in a cifically, the SNH cascade control implemented in Santín et al. [8] is comparable way. In this work, the original BSM2G framework is chosen used. This one obtains the best results in terms of water quality and as reference scenario just for an easy and fair comparison with previous costs, but emits higher N2 O concentrations compared to the other con­ works. The main goal is to show the advantages that can be achieved by trol strategies shown in the referred article. The objective is to maintain means of an appropriate operation of the internal recirculation control the SO regulation of this control strategy, but adding a regulation of Qa , handle. Therefore, adopting the same scenarios as in previous works. In which leads part of the water from the last biological treatment tank to case new models are taken, the figures of merit and the reference con­ the first one. ditions would be too different, and approaches unfairly compared. On A fuzzy controller is proposed to perform this Qa manipulation. This that basis, the conclusions stated by this work, must be taken with the control achieves a nitrification process improvement by reducing the uncertainty that may come from the actual arena of existing improved necessary SO levels to keep the established SNH set point. With the models. original control strategy, a partial nitrification takes place with high Although there are several works that apply control strategies in N2 O emissions. The objective of the Qa manipulation with the fuzzy WWTPs, the literature regarding the application of automatic control controller is to achieve a greater SO reduction and consequently a with the objective of reducing GHGs is scarcer. Some works such as greater N2 generation to reduce N2 O emissions, while maintaining Flores-Alsina et al. [5,7] and Barbu et al. [6] evaluate GHG emissions of optimal nitrogen levels and operational costs. The HRT increase by the the usual control strategies applied to improve effluent quality and/or Qa regulation also can help reduce the N2 O peaks. reduce operational costs. Other works such as Santín et al. [8]; Boiocchi The use of a fuzzy control approach is chosen because the fuzzy logic et al. [21]; Santín et al. [9] and Solís et al. [20] apply control strategies provides capabilities to introduce both an intuitive and formal repre­ with the aim of reducing GHGs and specifically N2 O. Santín et al. [8] sentation of the process as well as to include operator experience and applies a combination of nitrite concentration (SNO2 ) and ammonium 2 I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 Qin Qbypass Qpo po Qe Primary Secondary Activated sludge clarifier clarifier reactors Qa Qr Qw Thickener Anaerobic digester Storage tank Dewatering Sludge Removal Fig. 1. BSM2 plant with notation used for flow rates. process knowledge in the controller. This latest point is crucial in predicted N2 O values should be considered. gaining process operator confidence and for an easy way of posterior The paper is organized as follows. First, BSM2G working scenario is improvement of the controller by means of fine-tuning of actual fuzzy explained. Next, the proposed control strategies and the fuzzy control rules or even adding new ones to better reflect actual dynamics. Within designed are presented. After, results are shown, as well as the discus­ the wastewater treatment community, there have been several works sion about them. Finally, the most important conclusions are drawn. that show its application to different control problems. Different works can be found on the literature based on fuzzy logic. 2. Materials and methods From the initial works of Tong et al. [22] in the 80s, to different works that during the last decade have been approaching the dissolved oxygen The proposed control strategy is tested in BSM2G, which is an control and nutrient removal problems Meyer and Pöpel [23]; Traore extended version of Benchmark Simulation Model no. 2 (BSM2) [28]. et al. [24]; Yong et al. [25]; Baroni et al. [26]; Bertanza et al. [27], until BSM2G adds to BSM2 the assessment of the following GHGs: some latest results where the authors also show the potential of fuzzy methane (CH4 ), carbon dioxide (CO2 ) and N2 O. BSM2G was presented logic on plant wide approaches to deal with Greenhouse gas emissions in Flores-Alsina et al. [5] with the heterotrophic denitrification pathway Santín et al. [8,9]. This spectrum suggests fuzzy logic as a good candi­ for N2 O emissions. The same authors updated and provided a new date to introduce what in the paragraphs above authors have been version of BSM2G including the AOB denitrification pathway based on referencing as a complementary control action to improve greenhouse Guo and Vanrolleghem [29]. gas emissions of an existing operation strategy. Summarizing, the present paper proposes a N2 O emission reduction 2.1. Layout in WWTPs through applying the following novelties: The plant layout of the BSM2G (Fig. 1) is the same as that of the • Carrying out a partial nitrification by reducing SO in the tanks, BSM2, which includes a primary clarifier, a wastewater biological instead of a total nitrification by increasing SO . treatment, a secondary clarifier and a sludge treatment. • Manipulating Qa with a fuzzy controller, showing its effects on N2 O The biological treatment is carried out in five tanks connected in emissions. series, of which the first two are anoxic and the last three aerobic, with an internal recirculation from the latter to the first tank. Denitrification Starting from a commonly used control strategy, without modifying reactions take place in the anoxic tanks and nitrification reactions in the the SO control. aerobic tanks. Driven by the use of dynamic simulation models as a common Biological reactions inside the reactors are modeled by the Activated practice to assess and compare control strategies in WWTPs or to eval­ Sludge Model no. 1 (ASM1) [12], which is extended on the basis of Hiatt uate them before full-scale implementation, as mentioned above, this and Grady [11] and Mampaey et al. [30] to include, in addition to SNO3 , work uses Benchmark Simulation Model no. 2 Gas (BSM2G), which in­ the concentration of the SNO2 , nitric oxide (SNO ), SN2 O and dissolved N2 cludes the evaluation of GHG emissions in WWTPs. This version was (SN2 ). The equations of these concentrations and SNH are shown below: introduced by Flores-Alsina et al. [5] and later improved by the same authors, including two pathways for N2 O emissions: heterotrophic SNH =− 0.086⋅proc1 − 0.086⋅proc2.1 − 0.086⋅proc2.2 − 0.086⋅proc2.3 − 0.086⋅ denitrification and AOB denitrification. Even this model does not ⋅proc2.4 − 5.6416⋅proc3.1 − 0.086⋅proc3.2 +proc6 − 6.7794⋅proc9 − 6.7794⋅proc10 include all available knowledge at present date, it is the one that pro­ (1) vides the most suitable framework for comparing control strategies SNO3 = − 0.5756⋅proc2.1 + 16.6667⋅proc3.2 (2) under the same conditions. Therefore, previously highlighted un­ certainties (missing pathways in BSM2G) regarding the specific 3 I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 SNO2 = 0.5756⋅proc2.1 − 1.1522⋅proc2.2 + 5.5556⋅proc3.1 − 16.6667⋅proc3.2 − Table 1 − 6.6964⋅proc9 + 6.6964⋅proc10 Limits for the effluent pollutants. (3) Variable Value SNtot <18 mg/l SNO = 1.1522⋅proc2.2 − 1.1522⋅proc2.3 + 13.3869⋅proc9 − 13.3869⋅proc10 CODt <100 mg/l (4) SNH <4 mg/l TSS <30 mg/l SN2 O = 1.1522⋅proc2.3 − 1.1522⋅proc2.4 + 13.3869⋅proc10 (5) BOD5 <10 mg/l SN2 = 1.1522⋅proc2.4 (6) ( )( ) SFNA SFA where proc1 , proc2.1 , proc2.2 , proc2.3 , proc2.4 , proc3.1 , proc3.2 , proc6 , proc9 proc9 = μA1T ⋅ 0.0006 + SFNA ⋅ 0.0027 + SFA ⋅ and proc10 are biological processes, which are defined as follows: ( ) (20) ( )( ) SO SS SO ⋅ ( / ) ⋅XB,A1 proc1 = μHT ⋅ ⋅ ⋅XB,H (7) 11.4 − 35.0952⋅SO + SO2 0.0035 15 + SS 0.2 + SO ( )( ) SNO SFA where SS is the readily biodegradable substrate and μHT is: proc10 = μA1T ⋅ ⋅ ⋅ 1 + SNO 0.0027 + SFA ( ( ) )2 (21) μHT = 0.0625⋅(Tas + 20)⋅ 1 − e0.3⋅(Tas − 50) (8) ( ) SO ⋅ ( / ) ⋅XB,A1 ( )( )( ) 11.4 − 35.0952⋅SO + SO2 0.0035 SS SNO3 0.2 proc2.1 = μHT ⋅0.3⋅ ⋅ ⋅ ⋅XB,H (9) 20 + SS 1.5 + SNO3 0.2 + SO And the transfer to N2 O is as follows: ( )( )( )( ) proc2.2 = μHT ⋅0.3⋅ SS ⋅ SNO2 ⋅ 0.2 ⋅ 0.5 ⋅XB,H N2 O = 0.9137⋅pH⋅1.024Tas − 15 ⋅SNO ⋅vol (22) 20 + SS 0.3 + SNO2 0.2 + SO 0.5 + SNO (10) where vol is the volume of the tank. The sludge is deposited by gravity from the lower level of the sec­ ( )( )( ) SS SNO 0.2 ondary clarifier. Some of this biomass is directed to the sludge treatment proc2.3 = μHT ⋅0.6⋅ ⋅ ( 2 / ) ⋅ ⋅XB,H 20 + SS 0.04 + SNO + SNO 0.3 0.2 + SO (wastage) and the other part is recirculated to the biological treatment inlet. The secondary clarifier is modeled as a 10 layers non-reactive unit. (11) The sludge treatment consists of a thickener, an anaerobic digester ( SS )( SN2 O )( 0.2 )( 0.2 ) and a dewatering unit and includes the recirculation of water from the proc2.4 = μHT ⋅0.8⋅ 30 + SS ⋅ 0.02 + SN2 O ⋅ 0.2 + SO ⋅ 0.2 + SNO ⋅XB,H dewatering outlet and part of the thickener outlet to the first clarifier inlet. (12) The plant is designed for an average influent flow rate (Qin ) of ( )( )( ) 20,648.36 m3 /day with an average biodegradable Chemical Oxygen SFA SO 0.1 proc3.1 = μA1T ⋅ ( 2 / ) ⋅ ⋅ ⋅XB,A1 Demand (COD) of 592.53 mg/l. BSM2 defines a 609-day influent, 0.004 + SFA + SFA 0.1 0.6 + SO 0.1 + SFNA although only results from days 245 to 609 are evaluated. The capacity (13) of the different elements of the biological treatment is as follows: 900 m3 the first clarifier, 1500 m3 each anoxic tank, 3000 m3 each where XB,A1 is the concentration of AOB in mg/l and μA1T is: aerobic tank and 600 m3 the secondary clarifier. The resulting hydraulic ( ( ) )2 μA1T = 0.0255⋅(Tas + 15)⋅ 1 − e0.15⋅(Tas − 50) (14) retention time is 22 h. The influent wastewater composition follows the principles outlined in Gernaey et al. [31]. It includes diurnal flow rate while Tas is the temperature, SFA is free ammonia and SFNA is free nitrous and concentration variations, corresponding to diurnal pollutant flux acid defined as: profiles for the wastewater derived from households. Also, slowest ef­ fects are introduced such as a holiday effect (reducing the pollutant SNH ⋅10pH fluxes and the wastewater flow rate) and also a seasonal correction SFA = (15) e6344/(273.15+Tas ) + 10pH taking account of both yearly temperature variations (between 10 and 20 ◦ C). As the influent is the same as in BSM2, more information about it SNO2 SFNA = (16) can be found in Jeppsson et al. [32], Jeppsson et al. [33], Vrecko et al. 1 + e2300/(273.15+Tas ) ⋅10pH [34,37], Nopens et al. [35]. The influent characteristics also do consider ( )( )( ) the sewer model provided in Gernaey et al. [36]. SFNA SO 0.5 proc3.2 = μA2T ⋅ ( 2 / ) ⋅ ⋅ ⋅XB,A2 5⋅10− 6 + SFNA + SFA 0.1 1.2 + SO 0.5 + SFA 2.2. Evaluation criteria (17) The performance of control strategies is assessed by effluent quality, where XB,A2 is the concentration of Nitrite Oxidizing Bacteria (NOB) in operational costs and GHG emissions. mg/l and μA2T is: Effluent quality is determined by the percentage of time that ( ( ) )2 pollutant concentrations exceed the legally established limits and by the μA2T = 0.0235⋅(Tas + 25)⋅ 1 − e0.05⋅(Tas − 57) (18) Effluent Quality Index (EQI). proc6 = KaT ⋅SND ⋅XB,H (19) The legal effluent concentration limits of total nitrogen (SNtot ), total COD (CODt), SNH , Total Suspended Solids (TSS) and 5-day Biological Oxygen Demand (BOD5 ) are shown in Table 1. SNtot is the sum of SNO3 , SNO2 , SNO , SN2 O and Kjeldahl nitrogen (SNKj ), which includes organic nitrogen and SNH . This article only evaluates the violations of SNtot in the effluent (SNtot,e ) 4 I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 Fig. 2. Schemes of the original ACC with fixed Qa based on Santín et al. [8] (a) and ACC combining fixed and regulated Qa with the proposed fuzzy controller, depending on Tas (b)). and SNH in the effluent (SNH,e ), as the other pollutants are easily kept ment and CO2 due to electrical consumption have been evaluated, due to below the limits and there are only violations of them when there is a the specific objectives of the applied control strategy. As mentioned bypass due to a rain event that increases the flow rate above that allowed before, the heterotrophic denitrification and the AOB denitrification by the plant. pathways are taken into account for N2 O emissions, which are assessed EQI is measured in kg of pollutants per day and is calculated by on the basis of Hiatt and Grady [11] and Guo and Vanrolleghem [29], weighting the effluent concentrations of the different pollutants ac­ respectively. cording to the following equation: ⎛ 3. Control approach t=609days ∫ 1 ⎜ EQI = ⎝2⋅TSS(t) + COD(t) + 30⋅SNKj (t)+ During the nitrification process, SNH is oxidized by AOB, allowing the 1000⋅T (23) t=245days SNH reduction. The required SO in this process depends on the SNH ( ) ) +10⋅ SNO3 + SNO2 + SNO + SN2 O (t) + 2⋅BOD5 (t) ⋅Q(t)⋅dt values, so varying SO based on SNH is more efficient than keeping it fixed. This is the reason why the SO regulation in aerated tanks to control where T is the evaluation period and Q is the flow rate. SNH is a common practice in WWTPs, both in real plants and in research Costs are assessed by the Operational Cost Index (OCI), which is works such as Vrecko et al. [34,37]; Stare et al. [38]; Nopens et al. [35]; calculated by weighting the different costs according to the following Vrecko et al. [39]; Santín et al. [40]; Santín et al. [8]. These works apply equation: and evaluate different control techniques. Nevertheless, GHG emissions are not commonly evaluated in this control strategy. OCI = AE + PE + 3⋅SP + 3⋅EC + ME − 6⋅METprod + HEnet (24) The present work uses the Ammonia Cascade Control (ACC) applied in Santín et al. [8] as a starting point (Fig. 2a). The referenced research where AE is the aeration energy (kWh/day), PE is the pumping energy shows how the ACC obtains satisfactory results of EQI, OCI and SNtot,e (kWh/day), SP is the sludge production (kg/day), EC refers to the carbon limit violations, but GHG and specifically N2 O emissions result to be that could be added to improve denitrification (kg/day), ME is the very high. mixing energy (kWh/day), METprod is the produced methane (kWh/day) In this ACC applied in Santín et al. [8] Qa is kept fixed. In the present and HE is the heating energy (kWh/day). work a fuzzy controller is proposed to manipulate Qa with the aim of The GHG emissions are calculated according to the principles pro­ reducing N2 O emissions. The proposed Qa regulation is only applied posed by Hiatt and Grady [11] and Mampaey et al. [30]. The following with daily average Tas above 17 ◦ C because at low Tas the nitrification sources of GHG emissions are taken into account in BSM2G: the bio­ process is worse performed. Therefore, it is not possible to reduce SO to logical treatment, the sludge treatment, the difference between electric low concentrations by regulating Qa while keeping SNH below the consumption and electric generation, the EC production and the sludge established limits. This Tas has been chosen for this benchmark, through to be disposed. an approximate observation of the behavior of the plant. In the case of In the present work, only the N2 O emissions in the biological treat­ 5 I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 carrying out the present proposal in a real plant or another model, the Table 2 effect of Tas in the nitrification process should be taken into account and Ranges, types and parameters of the membership functions of the proposed fuzzy observe up to what Tas can be applied. When the average Tas is lower controller for Qa manipulation. than 17 ◦ C, Qa is kept fixed as in the original ACC. The control strategy Variables Range Membership Type Parameters composed of ACC and the fuzzy controller for Qa manipulation is shown functions in Fig. 2b. Inputs SNH,in [12 40] Low Z-shaped [12 24.2] Therefore, the possible scenarios for the application of Qa control are Medium Triangular [16 25 34] all those plants that use this usual ACC, with the objective of reducing High S-shaped [26 40] SNH,5 [1.35 Low Z-shaped [1.35 1.397] N2 O emissions without modifying the SO control, while maintaining 1.49] Medium-Low Triangular [1.35 1.397 satisfactory levels of EQI, limit violations of pollutants and OCI. It should 1.443] be noted that the proposed Qa manipulation can be applied in any other Medium Triangular [1.397 1.443 control strategy that regulates SO set-points based on SNH,5 , regardless of 1.49] the manipulation of other variables. In the case of operating with High S-shaped [1.443 1.49] Qin [1e + 04 Low Z-shaped [1e + 04 2.25e another SNH,5 set-point or with an alternative control technique to 5e + 04] + 04] manipulate the SO set-points or with a different WWTP, the proposed Qa Medium Triangular [1.7e + 04 manipulation would also be advantageous, although the fuzzy controller 2.75e + 04 4e + would have to be adapted by adjusting the ranges of some membership 04] High S-shaped [3.75e + 04 functions. 4.25e + 04] Output Qa [0 2e + Very-low Triangular [− 1000 0 1000] 3.1. Proposed fuzzy controller for internal recirculation flow rate 05] Low Triangular [0 1e + 04 2e + manipulation 04] Medium-Low Triangular [1e + 04 3e + 04 5e + 04] Fuzzy logic can be defined as a control approach based on human Medium Triangular [3e + 04 6e + 04 expertise. Fuzzy controller adapts the input and output variables into 9e + 04] suitable linguistic values by membership functions. Rules between input High Triangular [6e + 04 9e + 04 and output variables are established by words. Non-expert readers can 1.2e + 05] Very-high Triangular [9e + 04 2e + 05 found further information about fuzzy control in standard references 3.1e + 05] such as Klir and Yuan [41]. The FIS1 Editor from Matlab is used in this work for the implementation of the proposed fuzzy controller. The proposed Qa regulation requires a study of the plant and an 3.1.2. Membership functions exhaustive knowledge of the biological treatment operation. Conse­ The fuzzy control is defined by words instead of numbers and sen­ quently, a fuzzy controller based on the process knowledge is proposed tences instead of equations. However, process variables are measured in in this work. numbers instead of words. For this reason, the fuzzifier adapts the input The objective of the Qa manipulation is to reduce the SO required by variables into suitable linguistic values by membership functions. ACC, by increasing the oxygen consumption of SN2 O by XB,H that reduces Table 2 provides the information related to the membership func­ SN2 O to N2 . tions: range, fuzzy sets, type and parameters. Mamdani [42] is the method of inference used in this paper. The The ranges of SNH,in and Qin variations correspond to usual values of design of the proposed fuzzy controller is explained below, and its code the plant’s influent, and have been established on the basis of practical is shown in Annex I. knowledge (SNH,in : 12–40 mg/l and Qin : 1⋅104 - 5⋅105 m3 /day). Both inputs have three membership functions: “Low”, “Medium”, “High”. The 3.1.1. Inputs and output significant increase in the Qin with respect to SNH,in that occurs with The fuzzy controller has three inputs and one output. The inputs are rainfall events can be used to detect them. This fact can be observed with Qin , SNH at the WWTP inlet (SNH,in ), and SNH,5 . The output is Qa , which is the membership function ranges shown in Table 2 and the decision rules the variable to be regulated. The inputs and output of the proposed fuzzy explained in Section 3.1.3. Also, if the Qin value is “High” (higher than controller are shown in Fig. 2. 3.75⋅104 m3 /day), a rain event is considered regardless of the SNH,in Qin affects the HRT of the biological treatment. SNH,in provides in­ value, since the Qin values are above the usual ones during dry weather. formation on the SNH that will need to be oxidized in the biological The SNH,5 range (1.35–1.49 mg/l) has been established taking into treatment and thus on the SO required. While Qin has immediate effects account that ACC tries to control SNH,5 at the set-point of 1.5 mg/l by on SNH,5 , SNH,in has effects on SNH,5 with a time delay that depends on regulating SO . Therefore SNH,5 values below 1.35 mg/l are not usual. HRT, which in turn depends on Qin and Qa . During dry weather there is a Four membership functions have been established for SNH,5 : “Low”, relationship between the SNH,in and Qin values since both variables vary “Medium-Low”, “Medium” and “High”. Although the range of values is simultaneously during the day, and the Qin /SNH,in ratio allows to know if small, four membership functions are applied to reduce abrupt Qa var­ there is a period of dry weather. The membership functions and the iations. This is because the Qa regulation instantly affects the SNH,5 value, decision rules explained in Sections 3.1.1 and 3.1.2 explain specifically and the fuzzy controller regulates Qa based on SNH,5 . what ranges of values of Qin and SNH,5 are considered for dry weather. In Regarding the output variable (Qa ), the regulation range is estab­ rain events, there is a Qin increase and simultaneously a SNH dilution, lished between 0 and 2⋅105 by six membership functions: “Very-Low”, which results in an increase of the Qin /SNH,in ratio. However, there is also “Low”, “Medium-Low”, “Medium”, “High”, “Very-High”. However, the a HRT reduction. SNH,5 determines the SO required for the nitrification “Very-High” membership function is applied only in some cases such as process. Qa is regulated with the aim of reducing the SO required to keep rain events and unusual influent conditions. Therefore, the usual range SNH,5 at the set-point of 1.5 mg/l by the PI controllers of ACC. of operation is lower. The fixed Qa value applied in ACC equal to 61,944 m3/day is taken as a reference point. 3.1.3. Decision rules The proposed fuzzy controller includes 26 decision rules, which are 1 shown in Table 3. By these rules, a value is assigned to Qa for any FIS: Fuzzy Inference System 6 I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 Table 3 Decision rules of the proposed fuzzy controller for Qa manipulation. Rule SNH,in SNH,5 Qin Qa number 1 If is low And is low And is low Then is medium 2 If is low And is low And is Then is high medium 3 If is low And is low And is high Then is very- high 4 If is low And is medium- And is low Then is medium- low low 5 If is low And is medium- And is Then is medium low medium 6 If is low And is medium- And is high Then is high low 7 If is low And is medium And is low Then is low 8 If is low And is medium And is Then is medium- medium low 9 If is low And is medium And is high Then is medium 10 If is And is low And is low Then is high medium 11 If is And is low And is Then is high medium medium 12 If is And is low And is high Then is very- medium high 13 If is And is medium- And is low Then is medium medium low 14 If is And is medium- And is Then is medium medium low medium 15 If is And is medium- And is high Then is high medium low 16 If is And is medium And is low Then is medium- medium low 17 If is And is medium And is Then is medium- medium medium low 18 If is And is medium And is high Then is medium medium 19 If is high And is low – Then is very- high 20 If is high And is medium- And is low Then is high low 21 If is high And is medium- And is Then is high low medium 22 If is high And is medium- And is high Then is very- low high 23 If is high And is medium And is low Then is medium 24 If is high And is medium And is Then is medium medium 25 If is high And is medium And is high Then is high 26 – If is high – Then is very-low combination of input values. As mentioned above, the manipulation of Qa tries to reduce the SO necessary to maintain SNH,5 at the set-point of 1.5 mg/l, by increasing the oxygen consumption of SN2 O by XB,H that reduces SN2 O to SN2 (Eqs. (1), (3) and (12)). The decision rules are established taking into account the following premises to keep SNH,5 at 1.5 mg/l with lower SO by regulating Qa : • Effect of Qa on the inlet of the biological treatment: SNH is reduced by oxidation during the biological treatment. Therefore, the SNH,5 value is lower than the value of SNH at the input of the biological treatment. Hence, SNH at the secondary treatment inlet can be diluted by increasing Qa . Conversely, the dilution is lower if Qa is decreased. • Qa effect on the nitrification process: Increasing Qa results in an HRT decrease. This fact worsens the nitrification process and less SNH is oxidized. HRT increases if Qa decreases, enhancing the nitrification process and further reducing SNH . • Effect of a Qin increase due to a rain event: A rain event has a similar effect as a Qa increase, but without the possibility of control. Rain dilutes the pollutant concentration and specifically that of SNH , and Fig. 3. Graphic surfaces of the fuzzy control output related to the inputs. at the same time reduces HRT. Therefore, rain can be beneficial or 7 I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 Table 4 Simulation results of GHG emissions, effluent quality and operational costs of ACC with fixed Qa and regulating Qa with the proposed fuzzy controller for the two time periods evaluated. Evaluation Criteria From day 245 to day 338 From day 569 to day 609 ACC with fixed ACC with % of ACC with fixed ACC with % of Qa regulated Qa improvement Qa regulated Qa improvement GHG emissions N2 O emissions (kg CO2 eq/day) 12,669.41 8762.78 30.83 16,135.26 11,826.79 26.70 CO2 due to electric consumption 1955.98 1857.62 5.03 2071.06 1975.49 4.61 (kg CO2 /day) Total GHG (kg CO2 /day) 26,914.998 22,919.75 14.84 31,506.51 27,108.92 13.96 Operational OCI 8163.0014 7962.21 2.46 8326.22 8134.13 2.26 costs Pumping energy (kWh/day) 262 61.97 76,35 261.60 73.52 71.89 Areation energy (kWh/day) 3451.08 3189.094 7,59 3453.22 3434.42 0.54 Effluent EQI 3408.52 3379.35 0.86 4230.54 4188.24 1.01 Quality SNtot,e limits violations (% of time) – – – – – – SNH,e limits violations (% of time) – – – 0.18 0.18 – COD limits violations (% of time) – – – 0.34 0.34 – TSS limits violations (% of time) 0.045 0.045 – 1.17 1.17 – BOD5 limits violations (% of time) 0.045 0.045 – 1.07 1.07 – detrimental depending on whether or not it coincides with an SNH5 “High” membership functions ranges, because a possible rain event is peak, on its duration and on its intensity. considered. Although rain dilutes SNH , higher Qa values help to dilute it further. Nevertheless, Qa is reduced to increase HRT in the case that Fig. 3 shows the surface plots that relate the inputs and the output SNH,5 approaches the set-point of 1.5 mg/l, taking into account that based on decision rules. These surface plots are explained below: higher Qin values hinder the nitrification process by the HRT reduction. Fig. 3a shows the relationship of the output Qa with the inputs SNH,in If the SNH,in values are in the range of the membership function and SNH,5 setting the Qin value to 1⋅104 m3 /day. With this Qin , it is “Medium” (Fig. 3c) there is a Qa increase compared to Fig. 3b when Qin is considered that there is no rainfall event, regardless of the value of “Low”. This is because the SNH dilution must be greater if SNH,in in­ SNH,in . This Fig. 3a shows how Qa increases when SNH,in increases to creases. In this case (SNH,in is “Medium”), the rain event is only consid­ dilute SNH at the inlet of the biological treatment. However, Qa is always ered when Qin is “High”. If Qin is “Medium”, Qa keeps the same values as conditioned by SNH,5 , since if SNH increases in the tanks where nitrifi­ in Fig. 3b because SNH,in increases in the case of Fig. 3c, but a rain event cation takes place, Qa must be reduced to improve this process and thus is considered in Fig. 3b. In Fig. 3c, Qa is also reduced if SNH,5 increases. require less SO . Since ACC controls SNH,5 at the set-point of 1.5 mg/l, Qa If SNH,in increases to values within the membership function “High” is reduced to 0 when SNH,5 is equal to 1.49 mg/l or greater, trying to (Fig. 3d), Qa is increased to dilute SNH . As in the previous cases, the Qa maintain the set-point without requiring a SO increase. values are lower if SNH,5 increases, and are reduced to 0 if SNH,5 reaches Fig. 3b, c and d show the relationship between Qin and SNH,5 for the the set-point of ACC. If SNH,5 is “Low”, the Qa value is “Very-High” three membership functions of SNH,in : “Low”, “Medium” and “High”, regardless of the Qin value. If SNH,5 is “Medium-Low” or “Medium”, the respectively. Qa value is higher if Qin is “High”, since it corresponds to a rain event. Increasing Qa to dilute SNH is not usually necessary with SNH,in values belonging to the range of the membership function “Low” (Fig. 3b). However, Qa increases progressively if Qin reaches “Medium” and 4 10 10 ACC with fixed Qa N2O,3 (g N2O) 8 ACC with 6 regulated Qa 4 2 0 250 260 270 280 290 300 310 320 330 4 10 10 N O,4 (g N O) 8 2 6 4 2 2 0 4 250 260 270 280 290 300 310 320 330 10 10 N O,5 (g N O) 8 2 6 4 2 2 0 250 260 270 280 290 300 310 320 330 time (days) Fig. 4. Time evolution of N2 O, 3, N2 O, 4 and N2 O, 5 for ACC with fixed Qa and regulating Qa with the proposed fuzzy controller from day 245 to day 338. 8 I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 4 10 10 ACC with fixed Qa N2O,3 (g N2O) 8 ACC with 6 regulated Qa 4 2 0 570 575 580 585 590 595 600 605 4 10 10 N O,4 (g N O) 8 2 6 4 2 2 0 5704 575 580 585 590 595 600 605 10 10 N O,5 (g N O) 8 2 6 4 2 2 0 570 575 580 585 590 595 600 605 time (days) Fig. 5. Time evolution of N2 O, 3, N2 O, 4 and N2 O, 5 for ACC with fixed Qa and regulating Qa with the proposed fuzzy controller from day 569 to day 609. 40 (mg/l) 20 NH,in S 0 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 2.5 ACC with SNH,5 (mg/l) 2 fixed Qa 1.5 ACC with regulatedQ a 1 0.5 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 4 8 10 Q (m /d) 6 3 4 a 2 0 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 4 10 10 (g N O) N O,3 2 5 2 0 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 1.5 (mg/l) 1 0.5 O,3 S 0 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 time (days) Fig. 6. Time evolution of SNH,in , SNH,5 , Qa , SO,3 and N2 O, 3 for ACC with fixed Qa and regulating Qa with the proposed fuzzy controller during days 293 and 294. 4. Simulation results and discussion (fixed Qa ) and with ACC including Qa regulation with the proposed fuzzy controller during time periods when daily average Tas is equal to or Results of effluent quality, costs and GHG emissions have been ob­ greater than 17 ◦ C. These time periods are from day 245 to day 338 (93 tained by simulating BSM2G with Matlab. According to the simulation days) and from day 569 to day 609 (40 days). Out of these periods, the protocol, 609 days are simulated, of which from day 245 to day 609 can plant is operated just with ACC, therefore no comparison makes sense. be evaluated. On the other hand, Figs. 4, 5, 6, 7, 8, 9, 11 and 12 present the evo­ As mentioned in Section 3, the Qa regulation is only applied with lution over time of the main variables showing the differences between average daily Tas equal to or greater than 17 ◦ C since at high Tas the Qa regulated and fixed Qa . nitrification process is better performed, and it is possible to reduce SO to Next, these numerical results are discussed by analysing the figures, lower concentrations by regulating Qa . arguing the reasons for the improvements obtained by regulating Qa . In this way, at average daily Tas lower than 17 ◦ C the original ACC with fixed Qa is applied; while at average daily Tas equal to or greater 4.1. N2 O emissions than 17 ◦ C, Qa is regulated with the proposed fuzzy controller. Table 4 shows the numerical results obtained with the original ACC The addition of Qa regulation to the already applied ACC aims to 9 I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 40 (mg/l) 20 NH,in 0 S 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 2.5 (mg/l) ACC with 2 fixed Qa 1.5 ACC with NH,5 1 regulated Qa S 0.5 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 4 8 10 Q a (m 3/d) 6 4 2 0 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 5 4 10 (g N 2O) 2 N2O,4 0 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 1.5 SO,4 (mg/l) 1 0.5 0 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 time (days) Fig. 7. Time evolution of SNH,in , SNH,5 , Qa , SO,4 and N2 O, 4 for ACC with fixed Qa and regulating Qa with the proposed fuzzy controller during days 293 and 294. 40 (mg/l) 20 NH,in 0 S 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 2.5 (mg/l) ACC with 2 fixed Qa 1.5 ACC with NH,5 1 regulated Qa S 0.5 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 4 8 10 Q a (m 3/d) 6 4 2 0 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 104 4 (g N 2O) N2O,5 2 0 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 SO,5 (mg/l) 0.4 0.2 0 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 time (days) Fig. 8. Time evolution of SNH,in , SNH,5 , Qa , SO,5 and N2 O, 5 for ACC with fixed Qa and regulating Qa with the proposed fuzzy controller during days 293 and 294. reduce N2 O emissions without worsening effluent quality and opera­ are higher and other days when they are lower. For this reason, a day tional costs compared to ACC with fixed Qa . with a lower N2 O peak (day 293) and a day with higher N2 O peak (day Table 4 shows an N2 O reduction of 30.83 and 26.70 % achieved in 294) have been chosen in Figs. 6, 7 and 8 to discuss the time variation of the two evaluated periods by regulating Qa compared to fixed Qa . These the most important variables. SNH,in , SNH,5 and Qa are the same in all N2 O reductions can also be observed in Figs. 4 and 5, where the time three figures. Only the values of N2 O and SO are specific to each tank. variation of N2 O in tank 3 (N2 O,3), in tank 4 (N2 O,4) and in tank 5 (N2 O, The three figures show how Qa is progressively increased to dilute 5) is shown for the two time evaluated periods. A reduction of the daily SNH at the biological treatment inlet when SNH,5 is decreasing during the N2 O peaks by the proposed Qa regulation can be clearly observed, beginning of the day. The variation of SNH,5 is slight, but it should be especially in tanks 3 and 5. noted that the value ranges of the SNH,5 membership functions are short, This reduction is argued below by means of Figs. 6, 7 and 8, which as mentioned in Section 3.1.2. The further SNH,5 is from the ACC set- show the variations of SNH,in , SNH,5 , Qa , SO and N2 O for two days in the point, the lower SO is, reducing the risk of a N2 O increase. Conse­ three aerobic tanks. Only two days have been chosen for a better visu­ quently, Qa is increased to dilute SNH at the biological treatment inlet. alization. As shown in Figs. 4 and 5, there are days when the N2 O peaks The Qa increase is greater when SNH,in increases since a greater 10 I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 4 10 ACC with fixed Qa 9 ACC with regulated Q a 8 7 6 Q a (m3 /day) 5 4 3 2 1 0 250 260 270 280 290 300 310 320 330 time (days) 4 10 10 ACC with fixed Qa 9 ACC with regulated Qa 8 7 6 Q a (m3 /day) 5 4 3 2 1 0 570 575 580 585 590 595 600 605 time (days) Fig. 9. Time evolution of Qa for ACC with fixed Qa and regulating Qa with the proposed fuzzy controller from day 243 to day 338 (a) and from day 569 to day 609 (b). dilution is necessary, as long as SNH,5 does not reach 1.49 mg/l. Although Consequently, by the proposed Qa regulation, there is a reduction in the Qa increases, most of the time it is below the fixed value of 61,944 m3/ N2 O, 3 peak of approximately 15.4 kg on day 293 and almost 200 kg on day (Fig. 9). While a higher Qa increase would result in greater dilution, day 294. The main difference between both days is the SNH,in peak with it would take longer to subsequently decrease it when SNH,5 increases. an approximate variation of 5 mg/l that subsequently results in a SNH,5 The results of the Qa regulation are reflected in Figs. 6, 7 and 8, which increase and therefore higher needs of SO,3 . On both days the N2 O, 3 show the differences in SO and N2 O in the three aerobic tanks between emissions begin to increase when SO,3 reaches values of between 0.4 and regulating Qa and with fixed Qa . As explained in Section 3, the SO value 0.5 mg/l approximately and increase much more to extremely high regulated by the PI controller of ACC is the same for tanks 3 and 4 and values as SO,3 increases further. This is because there is not enough SO for half for tank 5. AOB to perform full nitrification and oxidize SNO2 to SNO3 (Eqs. (3), (2) As shown in Fig. 6, ACC with fixed Qa needs an SO,3 level of and (17)). If there is not enough SO , AOB also oxidize SNH with the ox­ approximately 0.5 mg/l on day 293 and 1 mg/l on day 294 to reduce ygen of SNO2 , reducing it to SNO and subsequently to SN2 O (Eqs. (1), (3), SNH,5 to the set-point of 1.5 mg/l. On the other hand, by regulating Qa (4), (5), (20) and (21)). When SO is less than 0.4 or 0.5 mg/l approxi­ with the proposed fuzzy controller, the maximum SO,3 needed is mately, SO is so reduced that XB,H consume more oxygen from SN2 O , approximately 0.37 mg/l on day 293 and 0.57 mg/l on day 294. reducing it to SN2 (Eqs. (5), (6) and (12)). Furthermore, due to the fact 11 I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 0.6 Tank 3 0.5 Tank 4 (mg/l) 0.4 0.3 NO 0.2 S 0.1 0 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 10 8 S (mg/l) 6 S 4 2 0 293 293.2 293.4 293.6 293.8 294 294.2 294.4 294.6 294.8 295 time (days) Fig. 10. Time evolution of SNO and SS for ACC with regulated Qa in tanks 3 and 4 during days 293 and 294. that the Qa reduction results in a HRT increase together with the SO proposed Qa regulation compared to ACC with fixed Qa . reduction, the AOB denitrification takes place more slowly, reducing the As mentioned above, the Qa regulation by the proposed fuzzy N2 O peaks. controller aims to reduce SO to very low values to achieve a greater SN2 O On day 294 there is a slight reduction in N2 O, 3 in a short period of reduction to SN2 by XB,H . This SO reduction also results in a reduction in time during the SO,3 increase, probably due to a slight SNH reduction. The CO2 due to electric consumption. N2 O generation by the AOB denitrification pathway is related to the SNH In addition, this SO reduction is achieved with a Qa regulation, whose since the higher the SNH , the greater its oxidation by AOB through the values are most of the time below the fixed value of 61,944 m3/day oxygen of SNO2 and SNO (Eqs. (1), (3), (4), (20) and (21)). (Fig. 9) established in ACC, and which corresponds to the default value Fig. 7 shows the variations of the variables in tank 4. As mentioned of BSM2 and BSM2G. This Qa reduction also results in a reduction in CO2 above, the SO set-point in tank 4 is the same as in tank 3, and therefore due to electric consumption. When SNH,5 reaches values above 1.49 mg/ the SO values are similar. However, the N2 O emissions are higher in tank l, Qa is reduced to zero, which leads to a large reduction in electrical 4 than in tank 3, especially on day 294, where the N2 O reduction energy during these time periods. compared to ACC with fixed Qa is smaller. As it is can be observed in Fig. 10, SNO increases and SS decreases as the wastewater advances 4.3. Operational costs through the tanks of the aerobic reactor. In addition, there is a greater SNO increase when SNH increases. Thus, the greater SNO , the greater As mentioned in Section 2.2, operational costs are evaluated using generation of SN2 O and consequently of N2 O (Eqs. (5) and (21)), while OCI. Table 4 shows an OCI reduction of 2.46 % from day 245 to day 338 the reduction of SS hinders the reduction of SN2 O to SN2 by XB,H (Eqs. (6) and of 2.26 % from day 569 to 609 by the proposed Qa regulation and (12)). This results in N2 O, 4 peaks of approximate 100 kg with fixed compared to fixed Qa . Qa and 20.6 kg with variable Qa on day 293 and 360 kg with fixed Qa and OCI includes the expenses of AE and PE (24). AE is mainly based on 244 with variable Qa on day 294. the value of KL a, which is directly related to SO . PE is based on the flow Fig. 8 shows the evolution over time of the same variables in tank 5 rates of the pumps used in WWTP and therefore partly on Qa . In this way, during the same days (293 and 294). ACC regulates the SO,5 set-point to the OCI reduction is related to the SO and Qa reduction achieved by the half the value of SO,3 and SO,4 set-point, with the objective of recircu­ proposed Qa regulation explained above, as the CO2 reduction due to lating less SO is recirculated to the anoxic tanks, which can worsen the electricity consumption. Specifically, ACC with regulated Qa achieves a denitrification process. The SO,5 reduction compared to SO,3 and SO,4 also reduction in average pumping electrical energy of 200 kWh/day in the results in a N2 O, 5 reduction. Specifically, SO,5 peaks are approximately first period of time evaluated and 188 kWh/day in the second period of 0.19 mg/l on day 293 and 0.3 mg/l on day 294 by regulating Qa , time. The average electrical energy for aeration is reduced by 262 kWh/ resulting in N2 O, 5 peaks of approximately 2.28 kg on day 293 and day and 19 kWh/day in the two time periods evaluated, respectively. 8.645 kg on day 294. The N2 O, 5 reduction compared to ACC with fixed Qa is much greater when SNH,in is higher (day 294) since ACC with fixed Qa increases SO,5 up to approximately 0.47 mg/l, which results in a 4.4. Effluent quality maximum peak of around 113 kg. This represents a reduction of over 104 kg on day 294 by regulating Qa in comparison with fixed Qa . As mentioned in Section 2.2, the effluent quality is evaluated by the EQI (1) and by pollutant limit violations. Table 4 shows an EQI reduction of 0.86 % from day 245 to day 338 4.2. CO2 due to electric consumption and of 1.01 % from day 569 to 609 by regulating Qa compared to fixed Qa in ACC. Table 4 shows that a CO2 reduction due to electric consumption of It is worth mentioning that ACC with fixed Qa is the control strategy 5.03 % is achieved in the time period between day 245 and day 338 and with the best EQI and lowest SNtot,e violations tested in Santín et al. [8] of 4.61 % between day 569 and day 609 by means of ACC with the (called CS2 in the referenced paper) being the improvement of the EQI 12 I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 12 ACC with fixed Q a 10 ACC with regulated Qa (mg/l) 8 tot,e 6 N S 4 2 250 260 270 280 290 300 310 320 330 2.6 2.4 2.2 (mg/l) 2 NH,e 1.8 S 1.6 1.4 1.2 250 260 270 280 290 300 310 320 330 time (days) 20 ACC with fixed Qa ACC with 15 regulated Qa (mg/l) Limit 10 tot,e N S 5 0 570 575 580 585 590 595 600 605 4.5 4 3.5 (mg/l) 3 2.5 NH,e 2 1.5 S 1 0.5 0 570 575 580 585 590 595 600 605 time (days) Fig. 11. Time evolution of SNtot,e and SNH,e for ACC with fixed Qa and regulating Qa with the proposed fuzzy controller from day 243 to day 338 (a) and from day 569 to day 609 (b). due to a SNtot,e reduction. ACC with regulated Qa keeps low SNtot,e values by events (Figs. 11 and 12). As shown in the figures, the SNH,5 levels remain partial nitrification (Fig. 11). Some of the nitrogen dissolved in the water most of the time below the established limit (4 mg/l), specially by the Qa present in SNO , SN2 O and SN2 molecules is converted into gas, but it does regulation where SNH,5 does not stray too far from the set-point estab­ not happen with total nitrification, which generates SNO3 . Therefore, lished in ACC (1.5 mg/l). However, SNH,e overcomes the limit of 4 mg/l with full nitrification higher SNtot,e is generated. because the flow rate in the biological treatment cannot exceed 60,000 On the other hand, the Qa regulation combines the SNH dilution and m3 /day and thus the wastewater is directed bypassed without being the nitrification process improvement, as explained above. Conse­ treated. The proposed Qa regulation takes into account rain events, quently, SNH,e peaks are reduced compared to fixed Qa (Fig. 11). This increasing Qa . It is considered that there is a rain event when there is an SNH,e reduction result in a slight EQI reduction. increase of Qin with respect to SNH,in , as long as SNH,5 does not exceed Regarding the established limit violations of nutrients, organic 1.49 mg/l. However, this regulation depends on SNH,5 and not on SNH,e . matter and TSS, Table 4 shows that there is no variations between ACC Therefore, it does not take into account the bypass when the biological with fixed Qa and with variable Qa . treatment cannot assume more flow. With respect to SNtot,e , there are no limit violations during the eval­ Organic matter (COD and BOD5 ) and TSS keep the same limit vio­ uated periods. Due to the SNtot,e reduction by partial nitrification, it is lations by regulating Qa as with fixed Qa . kept below the established limit (18 mg/l) even though it increases in It should be noted that the proposed Qa regulation aims to reduce rain events (Fig. 11). N2 O emissions of ACC with fixed Qa , without worsening the operational SNH,e violations occur on days 574 and 598 coinciding with rain cost and effluent quality indicators. Finally, not only they are not worsened, but a slight improvement is achieved. 13 I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 4 8 10 Q (m /day) 6 3 4 2 in 0 570 575 580 585 590 595 600 605 40 SNH (mg/l) 20 in 0 570 575 580 585 590 595 600 605 4 ACC with SNH,5 (mg/l) fixed Qa 3 ACC with 2 regulated Qa 1 Limit 570 575 580 585 590 595 600 605 SNH,e (mg/l) 4 2 0 570 575 580 585 590 595 600 605 4 10 Q (m /day) 15 10 3 5 a 0 570 575 580 585 590 595 600 605 time (days) Fig. 12. Time evolution of Qin , SNH,in , SNH,5 , Qa and SNH,e for ACC with fixed Qa and regulating Qa with the proposed fuzzy controller from day 569 to day 609. 5. Conclusions will be aimed at combining the proposed Qa manipulation with other control strategies applied in WWTP. The fact of using a fuzzy logic This work has presented the design of a fuzzy controller for the Qa controller to determine the complementary Qa actuation, makes this manipulation in the biological treatment of WWTPs with the aim of extension to other baseline control strategies (other than the ACC) reducing N2 O emissions. This is conceived as a complementary control more suitable. As the control logic is based on process dynamics action, to be applied jointly with another existing control strategy with reasoning, the controller input signals will be the same and there the aim of reducing GHG emissions. The Qa regulation has been added to may be the need to adapt the fuzzy logic controller membership the ACC control strategy, which with fixed Qa achieves satisfactory re­ functions. sults in terms of effluent quality and operational costs but with the drawback of large N2 O emissions. The main conclusions of the Qa Declaration of competing interest regulation in ACC by means of the proposed fuzzy controller are: Montse Meneses reports financial support was provided by Spanish • The Qa regulation in the periods of time in which Tas is greater than Ministry of Science and Innovation. 17 ◦ C reduces N2 O emissions by 26.70 and 30.83 % compared to keeping Qa fixed. Data availability • The Qa manipulation with the proposed fuzzy controller is mainly based on combining the SNH dilution at the entrance of the biological The controller code is added as an appendix. treatment with improving the nitrification process. This combination is carried out according to the priorities of each time frame, which are determined by the input variables of the controller (SNH,in , Qin , Acknowledgment SNH,5 ). • The Qa manipulation allows to oxidize SNH for a longer period of This work has received support from the Catalan Government under time. This fact results in a reduction of SN2 O peaks. In addition, SO is Project 2021 SGR 00197 and also by the Spanish Government under so reduced that XB,H consume more oxygen from SN2 O , reducing it to MICINN projects TED2021-129134B-I00 and PID2019-105434RBC33 SN2 . co-funded with the European Union ERDF funds. Marian Barbu was • The reduction of N2 O emissions is greater in tanks 3 and 5 than in financed by a Maria Zambrano fellowship from Spanish Ministry of tank 4, although in tank 4 the oxygen set-point is the same as in tank Science, Innovation and Universities, financed by European Union 3. The higher N2 O emissions are mainly due to higher SNO but also to ‘NextGenerationEU’ (Universitat Autonoma de Barcelona). lower SS , especially when Qin is higher. • The SO reduction by the proposed Qa regulation results in slight re­ Appendix A. [System] ductions in GHG emissions due to the electric consumption and in operational costs compared to fixed Qa . Name = ‘Proposed fuzzy controller’ • The Qa regulation keeps partial nitrification and improves the nitri­ Type = ‘mamdani’ fication process. These facts result in a maintenance of SNtot,e values Version = 2.0 and a slightly reduction of SNH,e peaks, which results in a slight NumInputs = 3 improvement in effluent quality compared to fixed Qa . NumOutputs = 1 • The proposed fuzzy controller is designed to be applied jointly with NumRules = 26 any other control strategy that manipulates the SO set points based AndMethod = ‘min’ on SNH,5 , regardless of the control of other variables. A future work OrMethod = ‘max’ ImpMethod = ‘min’ 14 I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 AggMethod = ‘max’ Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change, Cambridge university press, 2001. DefuzzMethod = ‘centroi d’ [2] B. Netz, O.R. Davidson, P.R. Bosch, R. Dave, L.A. Meyer, et al., Climate change [Input1] 2007: mitigation. contribution of working group iii to the fourth assessment report Name = ‘SNHin’ of the intergovernmental panel on climate change. summary for policymakers, in: Range = [12 40] Climate change 2007: Mitigation. Contribution of Working Group III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Summary NumMFs = 3 for Policymakers, 2007. MF1 = ‘Low’: ‘zmf’, [1 2 2 4. 2] [3] M.J. Kampschreur, H. Temmink, R. Kleerebezem, M.S. Jetten, M.C. van Loosdrecht, MF2 = ‘Medium’: ‘trimf”, [1 6 25 34] Nitrous oxide emission during wastewater treatment, Water Res. 43 (2009) 4093–4103. MF3 = ‘High’: ‘smf’, [2 6 40] [4] D. Richardson, H. Felgate, N. Watmough, A. Thomson, E. Baggs, Mitigating release [Input2] of the potent greenhouse gas n2o from the nitrogen cycle–could enzymic regulation Name = ‘SNH5’ hold the key? Trends Biotechnol. 27 (2009) 388–397. [5] X. Flores-Alsina, L. Corominas, L. Snip, P.A. Vanrolleghem, Including greenhouse Range = [1. 3 5 1. 4 9] gas emissions during benchmarking of wastewater treatment plant control NumMFs = 4 strategies, Water Res. 45 (16) (2011) 4700–4710. MF1 = ‘Low’: ‘zmf’, [1. 3 5 1. 3 9 7] [6] M. Barbu, R. Vilanova, M. Meneses, I. Santin, On the evaluation of the global impact of control strategies applied to wastewater treatment plants, J. Clean. Prod. MF2 = ‘Medium− Low’: ‘trimf’, [1. 3 5 1. 397 1. 4 4 3] 149 (2017) 396–405. MF3 = ‘Medium’: ‘trimf’, [1. 3 9 7 1. 443 1. 4 9] [7] X. Flores-Alsina, M. Arnell, Y. Amerlinck, L. Corominas, K.V. Gernaey, L. Guo, MF4 = ‘High’: ‘smf’, [1. 4 4 2 1. 4 9] E. Lindblom, I. Nopens, J. Porro, A. Shaw, L. Snip, P.A. Vanrolleghem, U. Jeppsson, Balancing effluent quality, economic cost and greenhouse gas emissions during the [Input3] evaluation of (plant-wide) control/operational strategies in WWTPs, Sci. Total Name = ‘Qin’ Environ. 466-467 (2014) 616–624. Range = [10000 50000] [8] I. Santín, M. Barbu, C. Pedret, R. Vilanova, Control strategies for nitrous oxide NumMFs = 3 emissions reduction on wastewater treatment plants operation, Water Res. 125 (2017) 466–477. MF1 = ‘Low’: ‘zmf’, [1 0 0 0 022500] [9] I. Santín, M. Barbu, C. Pedret, R. Vilanova, Fuzzy logic for plant-wide control of MF2 = ‘Medium’: ‘trimf’, [1 7 0 0 0 27500 40000] biological wastewater treatment process including greenhouse gas emissions, ISA MF3 = ‘High’: ‘smf’, [3 7 5 0 0 42500] Trans. 77 (2018) 146–166. [10] C. Sweetapple, G. Fu, D. Butler, Does carbon reduction increase sustainability? A [Output1] study in wastewater treatment, Water Res. 87 (2015) 522–530. Name = ‘Qa’ [11] W. Hiatt, J.C. Grady, An updated process model for carbon oxidation, nitrification, Range = [0200000] and denitrification, Water Environ. Res. 80 (11) (2008) 2145–2156. [12] M. Henze, C. Grady, W. Gujer, G. Marais, T. Matsuo, Activated Sludge Model 1, NumMFs = 6 Scientific and Technical Report No.1, IAWQ, London, UK, 1987. MF1 = ‘Low’: ‘trimf’, [0 10000 20000815] [13] B.-J. Ni, Z. Yuan, Recent advances in mathematical modeling of nitrous oxides MF2 = ‘Medium–low’: ‘trimf’, [1 0 0 0 0 30000 50000] emissions from wastewater treatment processes, Water Res. 87 (2015) 336–346. [14] G. Mannina, G. Ekama, D. Caniani, A. Cosenza, G. Esposito, R. Gori, M. Garrido- MF3 = ‘Medium’: ‘trimf’, [3 0 0 0 0 60000 90000] Baserba, D. Rosso, G. Olsson, Greenhouse gases from wastewater treatment - a MF4 = ‘Very–High’: ‘trimf’, [9 0 0 0 0 200000 310000] review of modelling tools, Sci. Total Environ. 551 (2016) 254–270. MF5 = ‘Very–low’: ‘trimf’, [− 1000 0 1000] [15] V. Vasilaki, T. Massara, P. Stanchev, F. Fatone, E. Katsou, A decade of nitrous oxide (n2o) monitoring in full-scale wastewater treatment processes: a critical review, MF6 = ‘High’: ‘trimf’, [6 0 0 0 0 90000 120000] Water Res. 161 (2019) 392–412. [Rules]. [16] V. Vasilaki, V. Conca, N. Frison, A. Eusebi, F. Fatone, E. Katsou, A knowledge 1 1 1, 3 (1): 1 discovery framework to predict the n2o emissions in the wastewater sector, Water 1 1 2, 6 (1): 1 Res. 178 (2020), 115799. [17] K. Blomberg, P. Kosse, A. Mikola, A. Kuokkanen, T. Fred, M. Heinonen, M. Mulas, 1 1 3, 4 (1): 1 M. Lübken, M. Wichern, R. Vahala, Development of an extended asm3 model for 1 2 1, 2 (1): 1 predicting the nitrous oxide emissions in a full-scale wastewater treatment plant, 1 2 2, 3 (1): 1 Environ. Sci. Technol. 52 (2018) 5803–5811. [18] M. Maktabifard, K. Blomberg, E. Zaborowska, A. Mikola, J. MÄ...kinia, Model- 1 2 3, 6 (1): 1 based identification of the dominant n2o emission pathway in a full-scale activated 1 3 1, 1 (1): 1 sludge system, J. Clean. Prod. 336 (2022), 130347. 1 3 2, 2 (1): 1 [19] E. Zaborowska, X. Lu, J. Makinia, Strategies for mitigating nitrous oxide production and decreasing the carbon footprint of a full-scale combined nitrogen 1 3 3, 3 (1): 1 and phosphorus removal activated sludge system, Water Res. 162 (2019) 53–63. 2 1 1, 6 (1): 1 [20] B. Solís, A. Guisasola, X. Flores-Alsina, U. Jeppsson, J.A. Baeza, A plant-wide model 2 1 2, 6 (1): 1 describing ghg emissions and nutrient recovery options for water resource recovery facilities, Water Res. 215 (2022), 118223. 2 1 3, 4 (1): 1 [21] R. Boiocchi, K.V. Gernaey, G. Sin, A novel fuzzy-logic control strategy minimizing 2 2 1, 3 (1): 1 N2O emissions, Water Res. 123 (2017) 479–494. 2 2 2, 3 (1): 1 [22] R. Tong, M. Beck, A. Latten, Fuzzy control of the activated sludge wastewater treatment process, Automatica 16 (1980) 695–701. 2 2 3, 6 (1): 1 [23] U. Meyer, H. Pöpel, Fuzzy-control for improved nitrogen removal and energy 2 3 1, 2 (1): 1 saving in wwt-plants with pre-denitrification, Water Sci. Technol. 47 (2003) 2 3 2, 2 (1): 1 69–76. 2 3 3, 3 (1): 1 [24] A. Traore, S. Grieu, S. Puig, L. Corominas, F. Thiéry, M. Polit, J. Colprim, Fuzzy control of dissolved oxygen in a sequencing batch reactor pilot plant, Chem. Eng. J. 3 1 0, 4 (1): 1 111 (2005) 13–19. 3 2 1, 6 (1): 1 [25] M. Yong, P. Yong-Zhen, W. Xiao-Lian, W. Shu-Ying, Intelligent control aeration and 3 2 2, 6 (1): 1 external carbon addition for improving nitrogen removal, Environ. Model. Softw. 21 (2006) 821–828. 3 2 3, 4 (1): 1 [26] P. Baroni, G. Bertanza, C. Collivignarelli, V. Zambarda, Process improvement and 3 3 1, 3 (1): 1 energy saving in a full scale wastewater treatment plant: air supply regulation by a 3 3 2, 3 (1): 1 fuzzy logic system, Environ. Technol. 27 (2006) 733–746. [27] G. Bertanza, L. Menoni, P. Baroni, Energy saving for air supply in a real wwtp: 3 3 3, 6 (1): 1 application of a fuzzy logic controller, Water Sci. Technol. 81 (2020) 1552–1557. 0 4 0, 5 (1): 1 [28] K. Gernaey, U. Jeppsson, P. Vanrolleghem, J. Copp, Benchmarking of Control Strategies for Wastewater Treatment Plants, Scientific and Technical Report No.23, IWA Publishing, London, UK, 2014. References [29] L. Guo, P.A. Vanrolleghem, Calibration and validation of an activated sludge model for greenhouse gases no. 1 (ASMG1): prediction of temperature-dependent N2O [1] J.T. Houghton, Y. Ding, D.J. Griggs, M. Noguer, P.J. van der Linden, X. Dai, emission dynamics, Bioprocess Biosyst. Eng. 37 (2014) 151. K. Maskell, C. Johnson, Climate Change 2001: The Scientific Basis: Contribution of 15 I. Santín et al. Journal of Water Process Engineering 53 (2023) 103802 [30] K.E. Mampaey, B. Beuckels, R.K.M.J. Kampschreur, M.C. van Loosdrecht, E. [36] K. Gernaey, C. Rosén, L. Benedetti, U. Jeppsson, Phenomenological modeling of I. Volcke, Modelling nitrous and nitric oxide emissions by autotrophic ammonia- wastewater treatment plant influent disturbance scenarios, in: 10th International oxidizing bacteria, Environ. Technol. 34 (12) (2013) 1555–1566. Conference on Urban Drainage. [31] K. Gernaey, C. Rosén, U. Jeppsson, Wwtp dynamic disturbance modelling–an [37] D. Vrecko, N. Hvala, A. Stare, O. Burica, M. Strazar, M. Levstek, P. Cerar, essential module for long-term benchmarking development, Water Sci. Technol. 53 S. Podbevsek, Improvement of ammonia removal in activated sludge process with (2006) 225–234. feedforward-feedback aeration controllers, Water Sci. Technol. 53 (4–5) (2006) [32] U. Jeppsson, C. Rosen, J. Alex, J. Copp, K. Gernaey, M.-N. Pons, P. Vanrolleghem, 125–132. Towards a benchmark simulation model for plant-wide control strategy [38] A. Stare, D. Vrecko, N. Hvala, S. Strmcnick, Comparison of control strategies for performance evaluation of WWTPs, Water Sci. Technol. 53 (2006) 287–295. nitrogen removal in an activated sludge process in terms of operating costs: a [33] U. Jeppsson, M.-N. Pons, I. Nopens, J. Alex, J. Copp, K. Gernaey, C. Rosen, J.- simulation study, Water Res. 41 (9) (2007) 2004–2014. P. Steyer, P. Vanrolleghem, Benchmark simulation model no 2: general protocol [39] D. Vrecko, N. Hvala, M. Strazar, The application of model predictive control of and exploratory case studies, Water Sci. Technol. 56 (8) (2007) 67–78. ammonia nitrogen in an activated sludge process, Water Sci. Technol. 64 (5) [34] D. Vrecko, K. Gernaey, C. Rosén, U. Jeppsson, Benchmark simulation model no 2 in (2011) 1115–1121. matlab-simulink: towards plant-wide wwtp control strategy evaluation, Water Sci. [40] I. Santín, C. Pedret, R. Vilanova, Applying variable dissolved oxygen set point in a Technol. 54 (2006) 65–72. two level hierarchical control structure to a wastewater treatment process, [35] I. Nopens, L. Benedetti, U. Jeppsson, M.-N. Pons, J. Alex, J.B. Copp, K.V. Gernaey, J. Process Control 28 (2015) 40–55. C. Rosen, J.-P. Steyer, P.A. Vanrolleghem, Benchmark simulation model no 2: [41] G. Klir, B. Yuan, Fuzzy Sets and Fuzzy Logic Volume 4, Prentice hall New Jersey, finalisation of plant layout and default control strategy, Water Sci. Technol. 62 (9) 1995. (2010) 1967–1974. [42] E. Mamdani, Application of fuzzy algorithms for control of simple dynamic plant, Proc. Inst. Electr. Eng. 121 (12) (1976) 1585–1588. 16