A review of computational modeling in wastewater treatment processes

A Review of Computational Modeling in Wastewater Treatment Processes M. SaloméDuarte,* ,# Gilberto Martins, # Pedro Oliveira, # Bruno Fernandes, Eugénio C. Ferreira, M. Madalena Alves, Frederico Lopes, M. Alcina Pereira, and Paulo Novais Cite This: ACS EST Water 2024, 4, 784−804 Read Online ACCESS Metrics & More Article Recommendations ABSTRACT: Wastewater treatment companies are facing several challenges related to the optimization of energy efficiency, meeting more restricted water quality standards, and resource recovery potential. Over the past decades, computational models have gained recognition as effective tools for addressing some of these challenges, contributing to the economic and operational efficiencies of wastewater treatment plants (WWTPs). To predict the performance of WWTPs, numerous deterministic, stochastic, and time series-based models have been developed. Mechanistic models, incorporating physical and empirical knowledge, are dominant as predictive models. However, these models represent a simplification of reality, resulting in model structure uncertainty and a constant need for calibration. With the increasing amount of available data, data-driven models are becoming more attractive. The implementation of predictive models can revolutionize the way companies manage WWTPs by permitting the development of digital twins for process simulation in (near) real-time. In data-driven models, the structure is not explicitly specified but is instead determined by searching for relationships in the available data. Thus, the main objective of the present review is to discuss the implementation of machine learning models for the prediction of WWTP effluent characteristics and wastewater inflows as well as anomaly detection studies and energy consumption optimization in WWTPs. Furthermore, an overview considering the merging of both mechanistic and machine learning models resulting in hybrid models is presented as a promising approach. A critical assessment of the main gaps and future directions on the implementation of mathematical modeling in wastewater treatment processes is also presented, focusing on topics such as the explainability of data-driven models and the use of Transfer Learning processes. 1. INTRODUCTION Population growth and the change in the lifestyles and in the consumption patterns of humanity make it expectable that demand for water, energy, and other goods and services that require water will also increase, making this natural resource of primary importance with potential scarcity in some regions. 1 Trying to overcome this issue, Sustainable Development Goal 6 of Agenda 2030 (of the United Nations) aims to ensure availability and sustainable management of water and sanitation for all, by 2030. Specifically, target 6.3 intends: “By 2030, improve water quality by reducing pollution, eliminating dumping and minimizing release of hazardous chemicals and materials, halving the proportion of untreated wastewater and substantially increasing recycling and safe reuse globally”. 2 Therefore, concerns about the quality and quantity of clean water have been increasing. The improvement of the management of this natural resource has become one of the main research subjects nowadays. A large part of the population live in urban centers where municipal authorities provide services and infrastructures to guarantee access to clean water to the population, through the urban water cycle, a challenge that includes disposal and treatment of effluents and water supply. 3,4 To guarantee the water quality level, it is necessary to monitor its treatment in several wastewater treatment plants (WWTPs). Monitoring leads to the detection of failures in WWTPs, resulting in an improvement both in terms of quality and in reducing maintenance risks. 5−7 Managing WWTPs is an exhaustive and complex process, as it depends on uncontrollable factors Special Issue: Applications of Artificial Intelligence, Machine Learning, and Data Analytics in Water Environments Received: March 14, 2023 Revised: August 11, 2023 Accepted: August 11, 2023 Published: August 24, 2023 Reviewpubs.acs.org/estwater © 2023 The Authors. Published by American Chemical Society 784 https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 This article is licensed under CC-BY 4.0 Table 1. Summary of Parameters and Conditions of Studies Focused on Predicting WWTPs Effluent Operational Parameters AI algorithm input variable output variable objective model performance cross validation overfitting control correlation among input variables ref ANN month, volumetric flow rate of inflow, pH, temp., CODd, TSS, TNinf, TN of pretreated food waste leachate TNeff Machine learning models to predict 1-day interval TNeff 0.55 (R2)/ 0.56 (NSE) no no no 74 SVM 1.00 (R2)/ 1.00 (NSE) ANFIS pHinf, CODinf, TSinf, NH4+ free, NH4+-N and TKNinf TKNeff SVM and ANFIS for predicting the TKN removal from a domestic WWTP GBELL MF -TKNeff: 0.128 mg/L (RMSE) yes yes yes; effluent TKN at the time (t) is strongly correlated with the TKNinf; NH4+-N and the NH4+free 8 Trapezoidal MF -TKNeff: 0.532 mg/L (RMSE) SVM TKNeff: 0.155 mg/L (RMSE) FFNN pHinf, conductivity (Condinf), BODinf, CODinf and TNinf BODeff, CODeff, and TNeff (i) AI based models and conventional multilinear models for prediction of the WWTP performance considering different combinations of input parameters BODeff: 0.0065 (RMSE) yes; employed the holdout (leavegroup-out) yes no 73 CODeff: 0.0014(RMSE) TNeff: 0.0004 (RMSE) ANFIS BODeff: 0.0053 (RMSE) (ii) 3 ensemble techniques using the outputs of single models in order to improve the overall efficiency of the prediction performance CODeff: 0.0012 (RMSE) TNeff: 0.0005 (RMSE) SVM BODeff: 0.0080 (RMSE) CODeff: 0.0047 (RMSE) TNeff: 0.0013 (RMSE) MLR BODeff: 0.0077 (RMSE) CODeff: 0.0014 (RMSE) TNeff: 0.0006 (RMSE) CNN-LSTM Temp(inf), pHinf, NH3inf, inflow, CODinf Model 1: sewage inflow and the COD concentration (COD mass flow can be calculated from the prediction) COD mass flow prediction model based on a deep learning algorithm Model 1:48.0592 (RMSE) no yes, epoch adjustment no 78 Model 2:17.50 (RMSE) CNN Model 2: predicts the COD mass flow directly 23.86 (RMSE) LSTM 29.88 (RMSE) FFNN pHinf, TSSinf, BODinf, CODinf at the current time (t) and BODeff and CODeff at the previous time (t-1) BODeff and CODeff at time tAI models used for predicting BODeff and CODeff BOD: 0.0341 (RMSE) no yes yes 77 COD: 0.0299 (RMSE) ANFIS BOD: 0.0296 (RMSE) COD: 0.0272 (RMSE)) SVR BOD: 0.0346 (RMSE) COD: 0.0322 (RMSE) ACS ES&T Water pubs.acs.org/estwater Review https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 785 Table 1. continued AI algorithm input variable output variable objective model performance cross validation overfitting control correlation among input variables ref ARIMA BOD: 0.0345 (RMSE) COD: 0.0338 (RMSE) ANN BOD5inf, DOinf, CODinf, Temp(inf), TSSinf, turbidity (inf), and Ecinf BODeff, CODeff, and TSSeff ANN and M5 model tree for assessing the performance of WWTP and estimating the quality of effluent BODeff: 3.50 mg/L (RMSE) no no yes 75 CODeff: 3.43 mg/L (RMSE) TSSeff: 2.62 mg/L (RMSE) M5 model tree BODeff: 4.75 mg/L (RMSE) CODeff: 4.74 mg/L (RMSE) TSSeff: 4.60 mg/L (RMSE) SARIMAX TPeff TPeff This study aims to explore the application of ML models on big data for prediction of wastewater quality from different full-scale WWTP WWTP A: 0.01008 (MAE) no yes yes 79 WWTP B: 0.00530 (MAE) WWTP C: 0.01104 (MAE) GTB WWTP A: 0.01294 (MAE) WWTP B: 0.00724 (MAE) WWTP C: 0.01355 (MAE) RF WWTP A: 0.01276 (MAE) WWTP B: 0.00694 (MAE) WWTP C: 0.01321 (MAE) SVM WWTP A: 0.01290 (MAE) WWTP B: 0.00694 (MAE) WWTP C: 0.01233 (MAE) LSTM WWTP A: 0.01176 (MAE) WWTP B: 0.00645 (MAE) WWTP C: 0.01337 (MAE) ANFIS WWTP A: 0.01417 (MAE) WWTP B: 0.00780 (MAE) WWTP C: 0.01488 (MAE) FFNN inflow, outflow, CODinf, NH3inf, TNinf, TPinf, pHinf, CODeff, NH3eff, TNeff and TPeff CODeff and TNeff The goal of this study is to predict in real time, the water quality of WWTP, by using an improved FFNN coupled with an optimization algorithm CODeff: 6.3% (MAPE) yes yes no 76 TNeff: 3.6% (MAPE) IFFNN CODeff: 5.9% (MAPE) TNeff: 2.8% (MAPE) Ga-IFFNN CODeff: 3.7% (MAPE) TNeff: 0.6% (MAPE) ACS ES&T Water pubs.acs.org/estwater Review https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 786 Table 2. Summary of Parameters and Conditions of Studies Focused on Predicting WWTPs Inflow AI algorithm input variable output variable objective model performance cross validation overfitting control correlation among input variables ref MLP-ANN influent data, rainfall data, and radar reflectivity data Influent flow Neural network approach is used to predict influent flow in the WWTP t: 1.09 (MAE) no no no 87 t+15:1.48 (MAE) t+30:1.89 (MAE) t+60:2.75 (MAE) t+90:3.61 (MAE) t+120:4.46 (MAE) t+150:5.26 (MAE) t+180:6.02 (MAE) SVM rainfall values, the water levels of the Wisłok river, and WWTP sewage entrances inflow Different approaches of data mining to model the inflow of sewage into the WWTP Q(t-1): 2.963 (MAE) no no yes 93 P(t-1): 4.127 (MAE) h(t-1): 3.467 (MAE) Q(t-1), h(t-1): 2.854 (MAE) P(t-1), P(t-2): 4.011 (MAE) h(t-1), h(t-2): 3.551 (MAE) Q(t-1), Q(t-2): 2.815 (MAE) P(t-1), h(t-1): 2.966 (MAE) Q(t-1), P(t-1) 2.912 (MAE) Q(t-1), Q(t-2), h(t-1): 2.789 (MAE) Q(t-1), Q(t-2), h(t-1), P(t-1): 2.647 (MAE) Q(t-1), Q(t-2), h(t-1), P(t-1), P(t-2): 2.641 (MAE) RF Q(t-1): 2.859 (MAE) P(t-1): 4.127 (MAE) h(t-1): 3.553 (MAE) Q(t-1), Q(t-2): 2.767 (MAE) P(t-1), P(t-2): 4.056 (MAE) h(t-1), h(t-2): 3.507 (MAE) Q(t-1), h(t-1): 2.847 (MAE) P(t-1), h(t-1): 3.008 (MAE) Q(t-1), P(t-1): 2.721 (MAE) Q(t-1), Q(t-2), h(t-1): 2.786 (MAE) Q(t-1), Q(t-2), h(t-1), P(t-1): 2.651 (MAE) Q(t-1), Q(t-2), h(t-1), P(t-1), P(t-2): 2.617 (MAE) KNN Q(t-1): 2.936 (MAE) P(t-1): 4.509 (MAE) ACS ES&T Water pubs.acs.org/estwater Review https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 787 Table 2. continued AI algorithm input variable output variable objective model performance cross validation overfitting control correlation among input variables ref h(t-1): 3.686 (MAE) Q(t-1), Q(t-2): 2.965 (MAE) P(t-1), P(t-2): 4.434 (MAE) h(t-1), h(t-2): 3.696 (MAE) Q(t-1), h(t-1): 2.961 (MAE) P(t-1), h(t-1): 3.157 (MAE) Q(t-1), P(t-1): 2.87 (MAE) Q(t-1), Q(t-2), h(t-1): 2.995 (MAE) Q(t-1), Q(t-2), h(t-1), P(t-1): 2.718 (MAE) Q(t-1), Q(t-2), h(t-1), P(t-1), P(t-2): 2.682 (MAE) Kernel regression (K) Q(t-1): 2.935 (MAE) P(t-1): 4.189 (MAE) h(t-1): 3.691 (MAE) Q(t-1), Q(t-2): 2.842 (MAE) P(t-1), P(t-2): 4.122 (MAE) h(t-1), h(t-2): 3.939 (MAE) Q(t-1), h(t-1): 3.152 (MAE) P(t-1), h(t-1): 3.408 (MAE) Q(t-1), P(t-1): 3.193 (MAE) Q(t-1), Q(t-2), h(t-1): 3.100 (MAE) Q(t-1), Q(t-2), h(t-1), P(t-1): 3.002 (MAE) Q(t-1), Q(t-2), h(t-1), P(t-1), P(t-2): 2.925 (MAE) RF historical weather data: max temp., min temp., mean temp., heating degree days, cooling degree days, total rain, total snow, total precipitation, and accumulated precipitation daily wastewater inflow RF for wastewater inflow prediction confidential WWTP: 35.937 (RMSE) yes no Weather parameters were selected for each WWTP according to a correlation analysis and to the literature 92 Humber WWTP: 7.547 (RMSE) MLP confidential WWTP: 95.699 (RMSE) Humber WWTP: 19.269 (RMSE) SVM inflow rate, COD, BOD5, NH4+, and TKN current weather condition Soft-sensor for predicting the current weather signal accuracy mean of 33validation data sets yes In the strong filter row, the authors obtained excellent accuracy rates. This was caused by an overfit to the training data, as explored in validation phase COD and BOD5 are extremely correlated; NH4+and TKN are very correlated; therefore inflow, COD and NH4+were the selected variables 86 no filter: 0.45 smooth filter: 0.68 strong filter: 0.33 Gaussian Naive Bayes no filter: 0.41 smooth filter: 0.56 strong filter:0.39 DT no filter: 0.45 ACS ES&T Water pubs.acs.org/estwater Review https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 788 Table 2. continued AI algorithm input variable output variable objective model performance cross validation overfitting control correlation among input variables ref smooth filter: 0.75 strong filter: 0.33 KNN (1) no filter: 0.46 smooth filter: 0.85 strong filter: 0.35 KNN (3) no filter: 0.46 smooth filter: 0.82 strong filter: 0.35 RF no filter: 0.47 smooth filter: 0.84 strong filter: 0.33 MLP-ANN TN, NH4+, BOD, COD, mixed liquor suspended solids (MLSS), Mixed liquor volatile suspended solid (MLVSS), pH, DO TNinf Feature selection methods for enhancing the prediction performance of TN in the WWTPs scenario I: 77 ×10−3(RMSE) no no TN had a good correlation with NH4-N, COD, and BOD, and a weak correlation with pH and DO 88 scenario II: 79 ×10−3(RMSE) scenario III: 74 ×10−3(RMSE) scenario IV: 73 ×10−3(RMSE) RF scenario I: 96 ×10−3(RMSE) scenario II: 60 ×10−3(RMSE) scenario III: 55 ×10−3(RMSE) scenario IV: 55 ×10−3(RMSE) GBM scenario I: 78 ×10−3(RMSE) scenario II: 72 ×10−3(RMSE) scenario III: 68 ×10−3(RMSE) scenario IV: 68 ×10−3(RMSE) ACS ES&T Water pubs.acs.org/estwater Review https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 789 such as weather conditions or illicit discharges and water leaks. These factors cause variations in the flow and characteristics of the influent, requiring a more resilient and robust treatment. WWTPs aim to control all processes that ensure the quality of the water treatment, by minimizing simultaneously the environmental impacts and the operating costs. Over the last few decades, computational models have gained recognition as effective tools for addressing some of these challenges by contributing to the economic and operational efficiencies of WWTPs. In order to predict the performance of WWTPs, numerous deterministic, stochastic, and time series-based models have been developed. 8,9 These models can be used to predict the effluent parameters over the process and take preventive actions to avoid compromising its treatment quality. 10 Predictive models are conceived to help decision-makers understand the data and make predictions about it to reduce environmental risks. Some examples of predictive models are artificial neural networks (ANNs), support vector machines (SVMs), and recurrent neural networks (RNNs), among others. Besides the implementation of predictive models, modulation and detection of abnormal situations may also play important roles in WWTPs management. Anomaly detection for the cyber-physical system (CPS) is related to the identification of unfamiliar patterns of behaviors, i.e., the detection of potential intrusions as a deviation from normality (anomaly detection) that are not exhibited under normal operation. 6,11 These anomalies could result from the physical environment and human error, but also from standard bugs or incorrect or suboptimal configurations in the software. 11 The detection of anomalies plays a defensive role, at the same time that facilitates development, maintenance, and repairs of CPSs. 11 Deep neural networks (DNNs) and SVMs are some examples of models that can be used for anomaly detection. Mechanistic models, incorporating physical and empirical knowledge, are dominant as predictive models. Nevertheless, this type of model represent a simplification of reality, which results in an uncertainty of the models’ structure. 12 With a constant increase in the amount of available data, data-driven methods are becoming more and more attractive. In this kind of model, the structure is not explicitly specified, but it is instead determined by searching for relationships in the available data. 12 Over the last several years, some reviews on the application of AI models to water/wastewater treatment have become available, providing a systematic overview of the application of AI mainly in technology, both physical/chemical 13 and biological 14,15 treatments, and management. 16 For example, Safeer et al. 13 reviewed the recent advancements and applications of AI in water purification and wastewater treatment processes. Regarding water purification, this review emphasizes specific processes such as coagulation/flocculation, disinfection, membrane filtration, and desalination. Regarding the AI models for wastewater treatment, it focuses on membrane processes, and heavy metals and dyes. 13 The paper by Sundui et al. 14 explores the advancements and perspectives on utilizing ML algorithms to improve biological wastewater treatment processes, specifically in algae−bacteria consortia systems. The work of Singh et al. 15 focuses on the application of AI and ML techniques for monitoring and designing biological wastewater treatment systems. In the case of Fu et al. 16 their paper is a critical review of the role of deep learning in the field of urban water management. Since deep learning is a subset of ML, this is focused on only a part of the ML models. Nevertheless, it discusses broader aspects related to water management: water supply and distribution systems; urban flooding; cyber security; etc. Their review presents only a short section regarding wastewater treatment plants. Zhong et al. 17 explores the innovative ideas and tools that have emerged with the adoption of ML techniques to address various environmental challenges in the field of environmental science and engineering, presenting a broader view of the application of ML when compared with our review. These authors approach only a subsection regarding the modeling of biochemical wastewater treatment systems. Nevertheless, when comparing these reviews with the present work, we believe ours presents a wider perspective on wastewater treatment systems discussing both MM and ML to study factors such as effluent characteristics (Table 1), wastewater inflow rates (Table 2), anomaly detection (Table 3), and energy consumption optimization (Table 4) in WWTPs. This review also includes a section where the recent developments of hybrid models in wastewater treatment modeling are explored, since we believe that hybrid models that join the best of both (ML e MM) models are the best solution to improve model performance and model explainability. Finally, the main gaps and weaknesses, such as data size/periodicity, lack of transparency and explainability (blackbox approach), difficulty in predicting and responding to process disturbances, and the lower benchmark calculations until the moment are critically discussed. Table 3. Summary of Parameters of Studies Focused on Anomaly and Fault Detection in WWTPs AI algorithm target objective model performance (F-scores) ref DNN sensors Application of unsupervised machine learning to anomaly detection for a CPS 0.80281 11 SVM 0.79628 DBM influent conditions Application of unsupervised machine learning to anomaly detection for a CPS 0.98 (OCSVM) 96 RBM 0.99 (OCSVM) RNN 0.97 (OCSVM) RNNRBM 0.99 (OCSVM) Stand alone 0.98 (OCSVM) LSTM WWTP sensor data Method based on DNN (specifically, long short-term memory) compared with statistical and traditional machine learning methods 0.9267 97 PCASVM 0.8667 ACS ES&T Water pubs.acs.org/estwater Review https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 790 Table 4. Summary of Parameters and Conditions of Studies Focused on Predicting and Performing Energy Consumption Optimization in WWTPs AI algorithm input variable output variable objective model performance cross validation overfitting control correlation among input variables ref NN CODeff, TPeff, TNeff, BOD5eff, tCODinf, TPinf, TNinf, BOD5inf, Inflow, the price of energy and the removal performance of COD, TN, and TP energy cost ML was used to generate high-performing energy cost models for WWTP R2> 0.86 no no no 98 RF R2> 0.95 ANFIS DO, oxidation reduction potential (ORP), temp., NH4+, and NO3−in the oxidation tank, and the output TN airflow rate (Ua) and the internal recycle Qr Development of a model capable of estimating the process variables, providing the right amount of aeration to achieve an economical and efficient operation NO3−: 0.12 mg/L (MAE) no no no 100 NH4+: 0.04 mg/L (MAE) Ua: 22.43 N m3/h (MAE) PCACNNLSTM energy consumption, material consumption, and influent conditions BOD5eff, CODeff, Sseff, pHeff, TPeff, TNeff, NH3eff, E. coli, Mud vol energy and materials-saving management via deep learning for WWTPs BOD5eff: 1.2984 (RMSE) no no yes 101 CODeff: 3.5454 (RMSE) SSeff: 2.4698 (RMSE) pHeff: 0.8889 (RMSE) TPeff: 0.0829 (RMSE) TNeff: 2.9816 (RMSE) NH3eff: 0.6784 (RMSE) E.coli: 2.1633 (RMSE) Mud vol: 1.4207 (RMSE) DNN Temp. influent, recirculated sludge flow, influent flow energy consumption ANN for creating an optimal model of energy consumption in a WWTP 90−92% (R2) no no no 102 RF design treatment capacity, annual average load rate, and removal ratios (BODiinf/ BODeff, CODinf/CODeff, NH3ifnf/ NH3eff) energy consumption Energy consumption model of WWTPs through machine learning using data from 2472 WWTPs in China, employing the RF approach 0.106 kWh/m3(RMSE) no no yes 103 LSTM influent flow, COD, and TN removed energy consumption Developing, tuning, and evaluation of a set of candidate DL models with the goal of forecasting the energy consumption of a WWTP, using a recursive multistep approach Model 1 (Multi-Variate-Scenario 3): 729.73 (RMSE) yes yes yes; the influent flow had the highest correlation coefficient with the target parameter 99 Model 2 (Uni-Variate-Scenario 1): 913.90 (RMSE) GRU Model 1 (Uni-Variate-Scenario 1): 715.42 (RMSE) Model 2 (Uni-Variate-Scenario 1): 869.85 (RMSE) CNN Model 1 (Multi-Variate-Scenario 3): 690.00 (RMSE) Model 2 (Uni-Variate-Scenario 1): 869.78 (RMSE) ACS ES&T Water pubs.acs.org/estwater Review https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 791 2. MECHANISTIC WASTEWATER MODELS: A PIECE OF HISTORY Water quality modeling has evolved since the early years of the 20th century. The pioneering work of Streeter and Phelps (1925) 18 launched the basis for the evolution and development of mathematical models applied to water quality problems. Later, with the emergence of computational capabilities, it allowed the development of more complex models. Mechanistic models or deterministic models implement a set of differential equations reflecting the mass balance equations and other conserved quantities, for all involved compounds. 19 Back in 1987, Henze et al. developed the Activated Sludge Model No. 1 (ASM1), the first WWTP model well accepted by research community and industry. 20 ASM1 describes the removal of nitrogen and organic carbon compounds, with the simultaneous consumption of electron acceptors (nitrate and oxygen), in municipal activated sludge WWTPs. In ASM1, the biological reactions are defined according to the Monod kinetics, and the majority of the basic concepts were inspired from the activated sludge model developed by Dold et al. 21 This integrated model combined the chemical oxygen demand (COD) conservation with stoichiometry and kinetics, by expressing transformation rates in the form of derivatives. 21 Further developments led to the expansion of the ASM model to include biological phosphorus removal and chemical phosphorus removal via precipitation processes, ASM2 22 and ASM2d 23 models, as well as the ASM3, 24 which were intended to amend the ASM1 model flaws and facilitate the calibration. ASM2 22 and ASM2d 23 models include the description of biological P processes and chemical P removal via precipitation, with simultaneous nitrification-denitrification processes. Later, a new version of the ASM model, ASM3, 24 was developed, intending to amend the ASM1 model flaws that have emerged during its usage. ASM3 has almost the same objectives as ASM1, and supposedly is easier to calibrate. This new ASM version distinguishes the importance of storage polymers in the conversion of heterotrophic activated sludge, which is mainly achieved by converting the circular growth− decay−growth model, frequently known as death−regeneration concept, into a growth-endogenous respiration model. 19 To integrate all of these tools and guarantee their evaluation and comparison, several benchmark tools have been developed by Working Groups of COST Action 682 and 624, and later by the IWA Task Group of Benchmarking of Control Strategies. This benchmark platform defines the WWTP arrangement, the simulation model, influent data sets, test procedures, and evaluation criteria. 25 The Benchmark Simulation Model no. 1 - BSM1 25 was the first layout to be developed and is comprised by a five-compartment activated sludge reactor divided in two anoxic tanks and three aerobic tanks. It combines nitrification with predenitrification, which is usually used for nitrogen removal in municipal WWTP. BSM2 26 was developed to also integrate the sludge treatment. Finally, a Risk Module was proposed, 27 considering the microbiology-related settling problems (filamentous bulking sludge, filamentous foaming, or deflocculation), which cause several operational problems in WWTPs. 28 Additionally, BSM−UWS (urban wastewater system), established as an integrated model library aiming to simulate on a single platform the dynamics of flow rate and pollutant loads in all the subsystems of an urban wastewater system, and the BSM2G for predicting greenhouse gas emissions were also developed. 29 Benchmark calculations using ASM models offer several advantages. 30 ASM models are based on a scientific understanding of the biological and chemical processes, thus providing insights into the underlying mechanisms and dynamics of the treatment process. 19 They are flexible and can be customized to represent specific treatment configurations, operational conditions, and influent characteristics. ASM models can simulate the behavior of wastewater treatment processes and predict their performance under different scenarios, as well as quantify key performance indicators (KPIs), such as effluent quality, sludge production, nutrient removal efficiency, and energy consumption. 31 Optimization and troubleshooting efforts, and the estimation of resource requirements are other advantages of benchmark calculations using ASM models. 30 Nevertheless, ASM models also have some drawbacks such as model complexity, accuracy, data requirements, high uncertainty due to many simplifications and assumptions, lack of adaptability, insufficient model validation, and computational requirements. 32−34 One of the main drawbacks of mechanistic models is the need for model calibration. Model calibration is the adjustment of model parameters starting from a default parameter set, which is updated considering the fitting of experimental data with simulation results. This is a time-consuming step and hinders the broader application of these models. 32 In the calibration, it can be used nondynamic data (i.e.: composite 24 h samples) or dynamic data (dynamic profiles of influent and effluent composition). 22 The calibration can be carried out following a heuristic approach, considering the process understanding and the model structure or through a purely mathematical optimization process. 19 The first approach is more sensitive but requires a considerable level of expert knowledge of the process. Usually, the calibration process based on engineering (heuristic) approaches could be combined with the mathematical approach, by applying a sensitivity analysis to model parameters. 35 In addition, despite ASM models being widely accepted, some novel treatment processes, such as anaerobic ammonium oxidation processes 36 and membrane treatment, 37 are still lacking for standard modeling frameworks. 34 Also, digital twins or virtual replicas of water and wastewater treatment infrastructures have been developed. Some examples include simulation platforms such as EPANET for drinking water distribution network, collection systems (info works, SWMM) water-related domain (DHI) and water resources recovery facilities (Biowin, Aquasim, GPS-X, Sumo, Simba, WEST). 38 However, the limited prediction capabilities of mechanistic models hinder its application. In summary, the long history of ASM models application has demonstrated their effectiveness for the design, optimization, and operation of WWTP, as well as in the comprehension of involved processes. 39 In an attempt to adapt the models to changes in WWTPs, i.e., process upgrades and introduction of new treatments, or even more strict effluent discharge limits, new models and/or extensions to existing models have been developed. 33 However, these changes result in an increase in the model complexity, making them too parametrized and difficult to calibrate. 40 Therefore, their popularity has decreased over the last years, as can be observed in Figure 1, where the number of publications related to the Activate Sludge Model (one of the most used mechanistic models) is in decline. Although the mechanistic models represented by the ASM model have been widely used, in recent years, studying the wastewater treatment processes with the data-driven ACS ES&T Water pubs.acs.org/estwater Review https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 792 not generalize to a given training data set. In the case of time series, it is also necessary to use specific cross-validation, such as time series split, so that the test data set has more recent periodicity than those used in the train. Analyzing the reviewed studies, not all consider this vital aspect when conceiving time series forecasting models (Tables 1,2, and 4). Regarding anomaly detection models, there is still a lack of studies in WWTPs. The WWTPs must follow limits imposed on the emission of various substances present in the wastewater, thus leading to tight control of these values. However, there may be times when this control may fail due to multiple factors, such as a failure in one of the wastewater treatment processes at these facilities. In this sense, anomaly detection models can be advantageous, alerting people who work in WWTPs to some anomalous value in some processes carried out in the facilities. We can identify a practical example of this utility in energy consumption. If any of the processes use more energy than usual, it may indicate a failure in equipment used in the process in question, causing it to consume more energy. In this case, using an anomaly detection model can help to identify this problem more quickly, leading to faster action by the WWTPs’ interlocutors. Nevertheless, to study the best anomaly detection model for different data sets, it is necessary to label them by people specialized in the area to classify a value as an anomaly or not an anomaly. One of the aspects pointed out to data-driven models is their lack of transparency and an explanation of what happens in their process. Many companies today still have difficulty using AI models due to the lack of confidence and security in understanding the whole process. Therefore, it is essential to give interpretability to the black box that surrounds data-driven models. One of the future directions is the application of Explainable AI (XAI) to demonstrate the entire process performed within these algorithms, such as feature importance. 124 In addition, using a Transfer Learning process is another point to consider as a direction. This process aims to use a pretrained model on a given problem, applying it to another but within the same context. 125 For example, at the level of a WWTPs management entity, the use of a pretrained model for forecasting energy consumption in a given WWTP can be reused in a different WWTP, for a similar forecast. This way, a single trained model can be used in different WWTPs, within the same context. However, to use this process in the context of WWTPs, attention to the infrastructure will be necessary. In the case of a pretrained model for predicting energy consumption in a specific WWTP, if we use it to carry out the same prediction in a larger WWTP where overall energy consumption is higher, this type of approach will not have many effects because the model was trained in a range of smaller values due to less energy consumption. Nevertheless, to date the majority of the literature studies are based on specific study cases, and there is a lack of benchmark calculations. 126 Still, a few studies have looked into this issue. For example, Torregrossa et al. 98 benchmarked the classic cost approaches with the performance of neural network and random forest to estimate the cost function in WWTP. In addition, the BSM1 platform was used to simulate a reinforcement learning-based particle swarm optimization method to optimize the control setting in the sewage process in WWTPs. 127 The results of this approach demonstrated that the developed model could provide feasible treatment solutions while reducing the operating costs. Another benchmark calculation example was provided by Heo et al. 94 In this study, the authors developed a hybrid machine-learning algorithm to find optimal set points of multiple controllers under varying influent conditions. They applied the BSM2 to model the WWTP and test the multiobjective supervisory control strategy. 94 Data-driven models are built under a set of hyperparameters without any physical and biological meaning, lacking the processes’ interpretability achieved by the mechanistic models. In addition, large data sets are needed to represent the entire WWTPs’ operation, this being the only source of knowledge to the model. This fact makes model predictions difficult when the WWTPs are under environmental or process disturbances. 12 Thus, as a future direction, we envisage the combination of both model approaches (mechanistic and data-driven), as the pros of one tend to be the cons of the other, allowing the junction of expert knowledge with data. The construction of hybrid models applicable to WWTPs could rely on an AI layer overtaking the mechanistic framework, combining data-driven models into a single loop by employing cycle-consistent adversarial networks. Thus, the mechanistic framework will facilitate the interpretation of model results, while the datadriven model can provide the individual parameter calibration and model refinement. Furthermore, with the conception of hybrid models, it will be possible to cover some essential aspects in parallel and series approaches. Considering the hybrid models with a parallel approach, one of the objectives would be to explain the result coming from the ML models. Nowadays, ML models still have a gap in the interpretability of their results without any explanation, known as Blackbox. Using the series hybrid model design approach, the output generated by the ML models would feed the MM, with their input. Through this mechanism, the MM would explain the obtained results by the ML models, which could improve the decision-making process in WWTPs. On the other hand, using the parallel approach for designing these models, the focus would be on minimizing the prediction error of both models. ML models perform well with a greater amount of available data. However, if we consider small data sets or disturbances in the systems, the ML models cannot perform satisfactorily, since it had small data set to be trained and may not have knowledge to predict disturbances. 34 This would happen due to the lack of more data to learn these variations. In these cases, the MM may respond better to the variations presented in the data. By using the models in parallel, we will have a more accurate and calibrated forecast, always considering the model that obtains the best performance at the instant of time that we want to forecast. Nowadays, some studies already use hybrid models in the field of ML, such as CNN with GRUs or CNN with LSTM, to predict energy consumption 128,129 and Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) joining algorithms XGBoost and RF, for water quality prediction. 130 Compared with DL algorithms, this model gave better results. The principle of the hybrid model approach is to take the strengths of different models and their knowledge representations, 131,132 as the CNN-LSTM hybrid model utilizes the ability of the CNN to extract features and LSTM to handle time series and sequence data. This combination intends to minimize the minimization of RMSE. ACS ES&T Water pubs.acs.org/estwater Review https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 799 6. CONCLUSIONS Computational modeling has shown to be a promising tool to assist in the management of WWTPs. Recent years came with a shift from the traditional mechanistic models where the process design has a special role to data-driven models, where modeling is based on machine learning approaches, without providing any knowledge about the function of the system. Nevertheless, data-driven models present better prediction capabilities than mechanistic ones, and overall, they present smaller errors. Data collection and curation were identified as the main limitations to be overcome for a wider implementation of AI models. Prediction of influent flow and effluent characterization are the most studied applications. Nonetheless, there is room for significant developments in models for anomaly detection and energy consumption optimization, for example. Despite the availability of mechanistic models for the different elements of water and wastewater systems, a robust integration with data-driven models is still missing to achieve an optimal balance between their prediction capabilities and the required computational power. Thus, future research should focus on the implementation of combined mechanistic and data-driven models. This approach will contribute to the economic and operational efficiency of WWTPs increasing their environmental sustainability. ■AUTHOR INFORMATION Corresponding Author M. Salomé Duarte −CEB −Centre of Biological Engineering, University of Minho, 4710-057 Braga, Portugal; LABBELS −Associate Laboratory, 4710-057 Guimaraes, Portugal; orcid.org/0000-0003-4645-908X; Email: [email protected] Authors Gilberto Martins −CEB −Centre of Biological Engineering, University of Minho, 4710-057 Braga, Portugal; LABBELS −Associate Laboratory, 4710-057 Guimaraes, Portugal; orcid.org/0000-0001-7187-0538 Pedro Oliveira −ALGORITMI Centre, Department of Informatics, University of Minho, 4710-057 Braga, Portugal Bruno Fernandes −ALGORITMI Centre, Department of Informatics, University of Minho, 4710-057 Braga, Portugal Eugénio C. Ferreira −CEB −Centre of Biological Engineering, University of Minho, 4710-057 Braga, Portugal; LABBELS −Associate Laboratory, 4710-057 Guimaraes, Portugal M. Madalena Alves −CEB −Centre of Biological Engineering, University of Minho, 4710-057 Braga, Portugal; LABBELS −Associate Laboratory, 4710-057 Guimaraes, Portugal Frederico Lopes −Aguas do Norte, 4810-284 Guimaraes, Portugal M. Alcina Pereira −CEB −Centre of Biological Engineering, University of Minho, 4710-057 Braga, Portugal; LABBELS −Associate Laboratory, 4710-057 Guimaraes, Portugal; orcid.org/0000-0002-7110-1779 Paulo Novais −ALGORITMI Centre, Department of Informatics, University of Minho, 4710-057 Braga, Portugal Complete contact information is available at: https://pubs.acs.org/10.1021/acsestwater.3c00117 Author Contributions # M.S.D., G.M., and P.O. contributed equally to this paper. Notes The authors declare no competing financial interest. ■ACKNOWLEDGMENTS This work was supported by the Portuguese Foundation for Science and Technology (FCT) under the scope of the PAMWater Project (DSAIPA/Al/0099/2019), the AIM4Water Project (2022.06822.PTDC), and the strategic funding of UIDB/04469/2020 and UIDB/00319/2020 units. The work of P.O. was supported by the doctoral Grant PRT/BD/ 154311/2022 financed by the Portuguese Foundation for Science and Technology (FCT), and with funds from European Union, under MIT Portugal Program. ■REFERENCES (1) Howard, G.; Bartram, J.; Williams, A.; Overbo, A.; Fuente, D.; Geere, J.-A. Domestic Water Quantity, Service Level and Health, 2nd ed.; WHO, 2020. (2) Summary Progress Update 2021: SDG 6 �Water and Sanitation for All; United Nations: Geneva, Switzerland, 2021. (3) Thompson, T.; Sobsey, M.; Bartram, J. Providing Clean Water, Keeping Water Clean: An Integrated Approach. Int. J. Environ. Health Res. 2003,13 (SUPPL.1), S89. (4) Gleick, P. H. The Human Right to Water. In Water Policy; Edward Elgar Publishing, 1998; Vol. 1, pp 487−503. (5) Leigh, C.; Alsibai, O.; Hyndman, R. J.; Kandanaarachchi, S.; King, O. C.; McGree, J. M.; Neelamraju, C.; Strauss, J.; Talagala, P. D.; Turner, R. D. R.; Mengersen, K.; Peterson, E. E. A Framework for Automated Anomaly Detection in High Frequency Water-Quality Data from in Situ Sensors. Sci. Total Environ. 2019,664, 885−898. (6) Raciti, M.; Cucurull, J.; Nadjm-Tehrani, S. Anomaly Detection in Water Management Systems. Lect. Notes Comput. Sci. 2012,7130, 98−119. (7) Sun, C.; Puig, V.; Cembrano, G. Real-Time Control of Urban Water Cycle under Cyber-Physical Systems Framework. Water 2020, 12 (2), 406. (8) Manu, D. S.; Thalla, A. K. Artificial Intelligence Models for Predicting the Performance of Biological Wastewater Treatment Plant in the Removal of Kjeldahl Nitrogen from Wastewater. Appl. Water Sci. 2017,7(7), 3783−3791. (9) Zaghloul, M. S.; Achari, G. A Review of Mechanistic and DataDriven Models of Aerobic Granular Sludge. J. Environ. Chem. Eng. 2022,10 (3), No. 107500. (10) Malviya, A.; Jaspal, D. Artificial Intelligence as an Upcoming Technology in Wastewater Treatment: A Comprehensive Review. Environ. Technol. Rev. 2021,10 (1), 177−187. (11) Inoue, J.; Yamagata, Y.; Chen, Y.; Poskitt, C. M.; Sun, J. Anomaly Detection for a Water Treatment System Using Unsupervised Machine Learning. IEEE Int. Conf. Data Min. Work. ICDMW 2017,2017, 1058−1065. (12) Quaghebeur, W.; Torfs, E.; De Baets, B.; Nopens, I. Hybrid Differential Equations: Integrating Mechanistic and Data-Driven Techniques for Modelling of Water Systems. Water Res. 2022,213, No. 118166. (13) Safeer, S.; Pandey, R. P.; Rehman, B.; Safdar, T.; Ahmad, I.; Hasan, S. W.; Ullah, A. A Review of Artificial Intelligence in Water Purification and Wastewater Treatment: Recent Advancements. J. Water Process Eng. 2022,49 (April), No. 102974. (14) Sundui, B.; Ramirez Calderon, O. A.; Abdeldayem, O. M.; Lázaro-Gil, J.; Rene, E. R.; Sambuu, U. Applications of Machine Learning Algorithms for Biological Wastewater Treatment: Updates and Perspectives. Clean Technol. Environ. Policy 2021,23 (1), 127− 143. (15) Singh, N. K.; Yadav, M.; Singh, V.; Padhiyar, H.; Kumar, V.; Bhatia, S. K.; Show, P. L. Artificial Intelligence and Machine LearningACS ES&T Water pubs.acs.org/estwater Review https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 800 Based Monitoring and Design of Biological Wastewater Treatment Systems. Bioresour. Technol. 2023,369, No. 128486. (16) Fu, G.; Jin, Y.; Sun, S.; Yuan, Z.; Butler, D. The Role of Deep Learning in Urban Water Management: A Critical Review. Water Res. 2022,223, No. 118973. (17) Zhong, S.; Zhang, K.; Bagheri, M.; Burken, J. G.; Gu, A.; Li, B.; Ma, X.; Marrone, B. L.; Ren, Z. J.; Schrier, J.; Shi, W.; Tan, H.; Wang, T.; Wang, X.; Wong, B. M.; Xiao, X.; Yu, X.; Zhu, J. J.; Zhang, H. Machine Learning: New Ideas and Tools in Environmental Science and Engineering. Environ. Sci. Technol. 2021,55 (19), 12741−12754. (18) Streeter, H. W.; Phelps, E. B. A Study of the Pollution and Natural Purification of the Ohio River. Public Heal. Bull. no. 146. U.S. Public Heal. Serv. Washington, DC, USA 1925.DOI: 10.1002/ ange.19290420217. (19) Gernaey, K. V; van Loosdrecht, M. C. M.; Henze, M.; Lind, M.; Jørgensen, S. B Activated Sludge Wastewater Treatment Plant Modelling and Simulation: State of the Art. Environ. Model. Softw. 2004,19 (9), 763−783. (20) Henze, M.; Grady, L., Jr.; Gujer, W.; Marais, G.; Matsuo, T. Activated Sludge Model No 1. Water Sci. Technol. 1987,29. (21) Dold, P. L.; Ekama, G. A.; Marais, G. vR. A General Model for the Activated Sludge Process. In Water Pollution Research and Development; Jenkins, S. H., Ed.; Elsevier, 1981; pp 47−77. DOI: 10.1016/B978-1-4832-8438-5.50010-8. (22) Henze, M.; Gujer, W.; Mino, T.; Matsuo, T.; Wentzel, M. C. M.; Marais, G. V. R. Activated Sludge Model No. 2; London, UK, 1995. (23) Henze, M.; Gujer, W.; Mino, T.; Matsuo, T.; Wentzel, M. C.; Marais, G. v. R.; Van Loosdrecht, M. C. M. Activated Sludge Model No. 2d, ASM2d. Water Sci. Technol. 1999,39 (1), 165−182. (24) Gujer, W.; Henze, M.; Mino, T.; van Loosdrecht, M. Activated Sludge Model No. 3. Water Sci. Technol. 1999,39, 183−193. (25) Alex, J.; Benedetti, L.; Copp, J. B.; Gernaey, K. V; Jeppsson, U.; Nopens, I.; Pons, M.-N.; Rieger, L. P.; Rosén, C.; Steyer, J.-P.; Vanrolleghem, P. A.; Winkler, S. Benchmark Simulation Model No. 1 (BSM1), EA Lund University, 2008. (26) Alex, J.; Benedetti, L.; Copp, J.; Gernaey, K. V.; Jeppsson, U.; Nopens, I.; Pons, M.-N.; Steyer, J.-P.; Vanrolleghem, P. Benchmark Simulation Model No. 2 (BSM2). IWA Model. Integr. Assesment SG 2018. (27) Comas, J.; Rodríguez-Roda, I.; Gernaey, K. V.; Rosen, C.; Jeppsson, U.; Poch, M. Risk Assessment Modelling of MicrobiologyRelated Solids Separation Problems in Activated Sludge Systems. Environ. Model. Softw. 2008,23 (10−11), 1250−1261. (28) Flores-Alsina, X.; Comas, J.; Rodriguez-Roda, I.; Gernaey, K. V.; Rosen, C. Including the Effects of Filamentous Bulking Sludge during the Simulation of Wastewater Treatment Plants Using a Risk Assessment Model. Water Res. 2009,43 (18), 4527−4538. (29) Jeppsson, U. The Benchmark Simulation Modelling Platform − Areas of Recent Development and Extension. Lect. Notes Civ. Eng. 2017,4, 81−91. (30) Jeppsson, U.; Rosen, C.; Alex, J.; Copp, J.; Gernaey, K. V.; Pons, M. N.; Vanrolleghem, P. A. Towards a Benchmark Simulation Model for Plant-Wide Control Strategy Performance Evaluation of WWTPs. Water Sci. Technol. a J. Int. Assoc. Water Pollut. Res. 2006,53 (1), 287−295. (31) Mannina, G.; Reboucas, T. F.; Cosenza, A.; Sanchez-Marre, M.; Gibert, K. Decision Support Systems (DSS) for Wastewater Treatment Plants −A Review of the State of the Art. Bioresour. Technol. 2019,290 (July), No. 121814. (32) Hauduc, H.; Rieger, L.; Ohtsuki, T.; Shaw, A.; Takács, I.; Winkler, S.; Héduit, A.; Vanrolleghem, P. A.; Gillot, S. Activated Sludge Modelling: Development and Potential Use of a Practical Applications Database. Water Sci. Technol. 2011,63 (10), 2164−2182. (33) Corominas, L.; Rieger, L.; Takács, I.; Ekama, G.; Hauduc, H.; Vanrolleghem, P. A.; Oehmen, A.; Gernaey, K. V.; Van Loosdrecht, M. C. M.; Comeau, Y. New Framework for Standardized Notation in Wastewater Treatment Modelling. Water Sci. Technol. 2010,61 (4), 841−857. (34) Schneider, M. Y.; Quaghebeur, W.; Borzooei, S.; Froemelt, A.; Li, F.; Saagi, R.; Wade, M. J.; Zhu, J.-J.; Torfs, E. Hybrid Modelling of Water Resource Recovery Facilities: Status and Opportunities. Water Sci. Technol. 2022,85 (9), 2503−2524. (35) Petersen, B.; Gernaey, K.; Henze, M.; Vanrolleghem, P. A. Evaluation of an ASM1Model Calibration Procedure on a MunicipalIndustrial Wastewater Treatment Plant. J. Hydroinformatics 2002,4 (1), 15−38. (36) Baeten, J. E.; Batstone, D. J.; Schraa, O. J.; van Loosdrecht, M. C. M.; Volcke, E. I. P. Modelling Anaerobic, Aerobic and Partial Nitritation-Anammox Granular Sludge Reactors - A Review. Water Res. 2019,149, 322−341. (37) Mannina, G.; Di Bella, G.; Viviani, G. An Integrated Model for Biological and Physical Process Simulation in Membrane Bioreactors (MBRs). J. Membr. Sci. 2011,376 (1), 56−69. (38) Matheri, A. N.; Mohamed, B.; Ntuli, F.; Nabadda, E.; Ngila, J. C. Sustainable Circularity and Intelligent Data-Driven Operations and Control of the Wastewater Treatment Plant. Phys. Chem. Earth, Parts A/B/C 2022,126, No. 103152. (39) Gujer, W. Activated Sludge Modelling: Past, Present and Future. Water Sci. Technol. 2006,53 (3), 111−119. (40) Sin, G.; Al, R. Activated Sludge Models at the Crossroad of Artificial Intelligence�A Perspective on Advancing Process Modeling. npj Clean Water 2021,4(1), 1−7. (41) Zhao, L.; Dai, T.; Qiao, Z.; Sun, P.; Hao, J.; Yang, Y. Application of Artificial Intelligence to Wastewater Treatment: A Bibliometric Analysis and Systematic Review of Technology, Economy, Management, and Wastewater Reuse. Process Saf. Environ. Prot. 2020,133 (92), 169−182. (42) Mohri, M.; Rostamizadeh, A.; Talwalkar, A. Foundations of Machine Learning; MIT Press, 2018. (43) Fernandes, B.; Neves, J.; Analide, C. SafeCity: A Platform for Safer and Smarter Cities. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Demazeau, Y., Holvoet, T., Corchado, J. M., Costantini, S., Eds.; Springer International Publishing: Cham, 2020; Vol. 12092 LNAI, pp 412−416. DOI: 10.1007/978-3-030-497781_37. (44) Jordan, M. I.; Mitchell, T. M. Machine Learning: Trends, Perspectives, and Prospects. Science (80-.). 2015,349 (6245), 255− 260. (45) Sarker, I. H. Machine Learning: Algorithms, Real-World Applications and Research Directions. SN Comput. Sci. 2021,2(3), 1−21. (46) Cunningham, P.; Cord, M.; Delany, S. J. Supervised Learning. In Machine Learning Techniques for Multimedia: Case Studies on Organization and Retrieval; Cord, M., Cunningham, P., Eds.; Springer Berlin Heidelberg: Berlin, Heidelberg, 2008; pp 21−49. DOI: 10.1007/978-3-540-75171-7_2. (47) Pimenta, A.; Carneiro, D.; Neves, J.; Novais, P. A Neural Network to Classify Fatigue from Human−Computer Interaction. Neurocomputing 2016,172, 413−426. (48) Carneiro, D.; Pimenta, A.; Neves, J.; Novais, P. A Multi-Modal Architecture for Non-Intrusive Analysis of Performance in the Workplace. Neurocomputing 2017,231, 41−46. (49) Carneiro, D.; Pimenta, A.; Neves, J.; Novais, P. Non-Intrusive Quantification of Performance and Its Relationship to Mood. Soft Comput. 2017,21 (17), 4917−4923. (50) Chakraborty, T.; Kumar, A.; Chattopadhyay, S. Journal of Computational and Applied A Novel Distribution-Free Hybrid Regression Model for Manufacturing Process Efficiency Improvement. J. Comput. Appl. Math. 2019,362, 130−142. (51) Zhang, Y.; Gao, X.; Smith, K.; Inial, G.; Liu, S.; Conil, L. B.; Pan, B. Integrating Water Quality and Operation into Prediction of Water Production in Drinking Water Treatment Plants by Genetic Algorithm Enhanced Arti Fi Cial Neural Network. Water Res. 2019, 164, No. 114888. (52) Ghahramani, Z. Unsupervised Learning. Lect. Notes Comput. Sci. 2004,3176, 72−112. ACS ES&T Water pubs.acs.org/estwater Review https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 801 (53) Celebi, M. E.; Aydin, K. Unsupervised Learning Algorithms. Unsupervised Learn. Algorithms 2016, 1−558. (54) Kaelbling, L. P.; Littman, M. L.; Moore, A. W. Reinforcement Learning: A Survey. J. Artif. Intell. Res. 1996,4(12), 237−285. (55) Qiang, W.; Zhongli, Z. Reinforcement Learning Model, Algorithms and Its Application. In 2011 International Conference on Mechatronic Science, Electric Engineering and Computer (MEC); IEEE, 2011; pp 1143−1146. DOI: 10.1109/MEC.2011.6025669. (56) LeCun, Y.; Bengio, Y.; Hinton, G. Deep Learning. Nature 2015, 521 (7553), 436−444. (57) Jove, E.; Casteleiro-Roca, J. L.; Quintián, H.; Zayas-Gato, F.; Novais, P.; Méndez-Pérez, J. A.; Calvo-Rolle, J. L. A Comparative Study to Detect Flowmeter Deviations Using One-Class Classifiers. In Advances in Intelligent Systems and Computing; Herrero, A., Cambra, C., Urda, D., Sedano, J., Quintián, H., Corchado, E., Eds.; Springer International Publishing: Cham, 2021; pp 66−75. DOI: 10.1007/ 978-3-030-57805-3_7. (58) Lauzon, F. Q. An Introduction to Deep Learning. 2012 11th Int. Conf. Inf. Sci. Signal Process. Their Appl. 2012, 1438−1439. (59) Fernandes, B.; Vicente, H.; Ribeiro, J.; Analide, C.; Neves, J. Evolutionary Computation on Road Safety. In Lecture Notes in Computer Science; de Cos Juez, F. J., Villar, J. R., de la Cal, E. A., Herrero, A., Quintián, H., Sáez, J. A., Corchado, E., Eds.; Springer International Publishing: Cham, 2018; Vol. 10870 LNAI, pp 647− 657. DOI: 10.1007/978-3-319-92639-1_54. (60) Baldi, P. Autoencoders, Unsupervised Learning, and Deep Architectures. ICML Unsupervised Transfer Learn. 2012, 37−50. (61) Hajela, P.; Berke, L. Neural Networks in Structural Analysis and Design: An Overview. Comput. Syst. Eng. 1992,3(1−4), 525−538. (62) Palmeira, J.; Ramos, J.; Silva, R.; Fernandes, B.; Analide, C. A Machine Learning Approach to Forecast the Safest Period for Outdoor Sports. In Intelligent Data Engineering and Automated Learning -- IDEAL 2020; Analide, C., Novais, P., Camacho, D., Yin, H., Eds.; Springer International Publishing: Cham, 2020; pp 185− 196. (63) Albawi, S.; Mohammed, T. A.; Al-Zawi, S. Understanding of a Convolutional Neural Network. In 2017 International Conference on Engineering and Technology (ICET); 2017; pp 1−6. DOI: 10.1109/ ICEngTechnol.2017.8308186. (64) Barino, F. O.; Silva, V. N. H.; Lopez-Barbero, A. P.; De Mello Honorio, L.; dos Santos, A. B. Correlated Time-Series in Multi-DayAhead Streamflow Forecasting Using Convolutional Networks. IEEE Access 2020,8, 215748−215757. (65) Garg, S.; Jindal, H. Evaluation of Time Series Forecasting Models for Estimation of PM2.5 Levels in Air. 2021 6th Int. Conf. Converg. Technol. I2CT 2021 2021, 1−8. (66) Medsker, L. R.; Jain, L. C. Recurrent Neural Networks. Des. Appl. 2001,5, 64−67. (67) Fernandes, B.; Silva, F.; Alaiz-Moretón, H.; Novais, P.; Analide, C.; Neves, J. Traffic Flow Forecasting on Data-Scarce Environments Using ARIMA and LSTM Networks. In New Knowledge in Information Systems and Technologies; Rocha, A., Adeli, H., Reis, L. P., Costanzo, S., Eds.; Springer International Publishing: Cham, 2019; pp 273−282. (68) Zadeh, L. A. The Role of Fuzzy Logic in the Management of Uncertainty in Expert Systems. Fuzzy Sets Syst. 1983,11 (1−3), 199− 227. (69) Vafaie, H.; Jong, K. De. Genetic Algorithms as a Tool for Feature Selection in Machine Learning. In Proceedings Fourth International Conference on Tools with Artificial Intelligence TAI ’92; 1992; pp 200−203. DOI: 10.1109/TAI.1992.246402. (70) Al Aani, S.; Bonny, T.; Hasan, S. W.; Hilal, N. Can Machine Language and Artificial Intelligence Revolutionize Process Automation for Water Treatment and Desalination? Desalination 2019; pp 45884−96. . (71) Chau, K. A Review on Integration of Artificial Intelligence into Water Quality Modelling. Marine Pollution Bulletin 2006,52, 726− 733. (72) Abraham, A. Neuro Fuzzy Systems: State-of-the-Art Modeling Techniques. In Connectionist Models of Neurons, Learning Processes, and Artificial Intelligence; Mira, J., Prieto, A., Eds.; Springer Berlin Heidelberg: Berlin, Heidelberg, 2001; pp 269−276. (73) Nourani, V.; Elkiran, G.; Abba, S. I. Wastewater Treatment Plant Performance Analysis Using Artificial Intelligence - An Ensemble Approach. Water Sci. Technol. 2018,78 (10), 2064−2076. (74) Guo, H.; Jeong, K.; Lim, J.; Jo, J.; Kim, Y. M.; Park, J.-p.; Kim, J. H.; Cho, K. H. Prediction of Effluent Concentration in a Wastewater Treatment Plant Using Machine Learning Models. J. Environ. Sci. (China) 2015,32, 90−101. (75) Asami, H.; Golabi, M.; Albaji, M. Simulation of the Biochemical and Chemical Oxygen Demand and Total Suspended Solids in Wastewater Treatment Plants: Data-Mining Approach. J. Clean. Prod. 2021,296, No. 126533. (76) Xie, Y.; Chen, Y.; Lian, Q.; Yin, H.; Peng, J.; Sheng, M.; Wang, Y. Enhancing Real-Time Prediction of Effluent Water Quality of Wastewater Treatment Plant Based on Improved Feedforward Neural Network Coupled with Optimization Algorithm. Water (Switzerland) 2022,14 (7), 1053. (77) Nourani, V.; Asghari, P.; Sharghi, E. Artificial Intelligence Based Ensemble Modeling of Wastewater Treatment Plant Using Jittered Data. J. Clean. Prod. 2021,291, No. 125772. (78) Wang, Z.; Man, Y.; Hu, Y.; Li, J.; Hong, M.; Cui, P. A Deep Learning Based Dynamic COD Prediction Model for Urban Sewage. Environ. Sci. Water Res. Technol. 2019,5(12), 2210−2218. (79) Ly, Q. V.; Truong, V. H.; Ji, B.; Nguyen, X. C.; Cho, K. H.; Ngo, H. H.; Zhang, Z. Exploring Potential Machine Learning Application Based on Big Data for Prediction of Wastewater Quality from Different Full-Scale Wastewater Treatment Plants. Sci. Total Environ. 2022,832 (March), No. 154930. (80) Zaghloul, M. S.; Achari, G. Application of Machine Learning Techniques to Model a Full-Scale Wastewater Treatment Plant with Biological Nutrient Removal. J. Environ. Chem. Eng. 2022,10 (3), No. 107430. (81) Picos-Benítez, A. R.; Martínez-Vargas, B. L.; Duron-Torres, S. M.; Brillas, E.; Peralta-Hernández, J. M. The Use of Artificial Intelligence Models in the Prediction of Optimum Operational Conditions for the Treatment of Dye Wastewaters with Similar Structural Characteristics. Process Saf. Environ. Prot. 2020,143, 36− 44. (82) Jana, D. K.; Bhunia, P.; Das Adhikary, S.; Bej, B. Optimization of Effluents Using Artificial Neural Network and Support Vector Regression in Detergent Industrial Wastewater Treatment. Clean. Chem. Eng. 2022,3(April), No. 100039. (83) Meng, G.; Fang, L.; Yin, Y.; Zhang, Z.; Li, T.; Chen, P.; Liu, Y.; Zhang, L. Intelligent Control of the Electrochemical Nitrate Removal Basing on Artificial Neural Network (ANN). J. Water Process Eng. 2022,49, No. 103122. (84) Raza, A.; Bardhan, S.; Xu, L.; Yamijala, S. S. R. K. C.; Lian, C.; Kwon, H.; Wong, B. M. A Machine Learning Approach for Predicting Defluorination of PerAnd Polyfluoroalkyl Substances (PFAS) for Their Efficient Treatment and Removal. Environ. Sci. Technol. Lett. 2019,6, 624. (85) Di Fraia, S.; Massarotti, N.; Vanoli, L. A Novel Energy Assessment of Urban Wastewater Treatment Plants. Energy Convers. Manag. 2018,163, 304−313. (86) Hernández-del-Olmo, F.; Gaudioso, E.; Duro, N.; Dormido, R. Machine Learning Weather Soft-Sensor for Advanced Control of Wastewater Treatment Plants. Sensors (Switzerland) 2019,19 (14), 3139. (87) Wei, X.; Kusiak, A. Short-Term Prediction of Influent Flow in Wastewater Treatment Plant. Stoch. Environ. Res. Risk Assess. 2015,29 (1), 241−249. (88) Bagherzadeh, F.; Mehrani, M. J.; Basirifard, M.; Roostaei, J. Comparative Study on Total Nitrogen Prediction in Wastewater Treatment Plant and Effect of Various Feature Selection Methods on Machine Learning Algorithms Performance. J. Water Process Eng. 2021,41 (March), No. 102033. (89) Zhang, D.; Martinez, N.; Lindholm, G.; Ratnaweera, H. Manage Sewer In-Line Storage Control Using Hydraulic Model and ACS ES&T Water pubs.acs.org/estwater Review https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 802 Recurrent Neural Network. Water Resour. Manag. 2018,32 (6), 2079−2098. (90) Elman, J. Finding Structure in Time. Cogn. Sci. 1990,14 (2), 179−211. (91) Siegelmann, H. T.; Horne, B. G.; Giles, C. L. Computational Capabilities of Recurrent NARX Neural Networks. IEEE Trans. Syst. Man, Cybern. Part B 1997,27 (2), 208−215. (92) Zhou, P.; Li, Z.; Snowling, S.; Baetz, B. W.; Na, D.; Boyd, G. A Random Forest Model for Inflow Prediction at Wastewater Treatment Plants. Stoch. Environ. Res. Risk Assess. 2019,33 (10), 1781−1792. (93) Szelag, B.; Bartkiewicz, L.; Studzinski, J.; Barbusinski, K. Evaluation of the Impact of Explanatory Variables on the Accuracy of Prediction of Daily Inflow to the Sewage Treatment Plant by Selected Models Nonlinear. Arch. Environ. Prot. 2017,43 (3), 74−81. (94) Heo, S. K.; Nam, K. J.; Tariq, S.; Lim, J. Y.; Park, J.; Yoo, C. K. A Hybrid Machine Learning−Based Multi-Objective Supervisory Control Strategy of a Full-Scale Wastewater Treatment for CostEffective and Sustainable Operation under Varying Influent Conditions. J. Clean. Prod. 2021,291, No. 125853. (95) Harrou, F.; Dairi, A.; Sun, Y.; Senouci, M. Wastewater Treatment Plant Monitoring via a Deep Learning Approach. IEEE Int. Conf. Ind. Technol. 2018,2018 (1), 1544−1548. (96) Dairi, A.; Cheng, T.; Harrou, F.; Sun, Y.; Leiknes, T. O. Deep Learning Approach for Sustainable WWTP Operation: A Case Study on Data-Driven Influent Conditions Monitoring. Sustain. Cities Soc. 2019,50, No. 101670. (97) Mamandipoor, B.; Majd, M.; Sheikhalishahi, S.; Modena, C.; Osmani, V. Monitoring and Detecting Faults in Wastewater Treatment Plants Using Deep Learning. Environ. Monit. Assess. 2020,192 (2). DOI: 10.1007/s10661-020-8064-1. (98) Torregrossa, D.; Leopold, U.; Hernández-Sancho, F.; Hansen, J. Machine Learning for Energy Cost Modelling in Wastewater Treatment Plants. J. Environ. Manage. 2018,223 (July), 1061−1067. (99) Oliveira, P.; Fernandes, B.; Analide, C.; Novais, P. Forecasting Energy Consumption of Wastewater Treatment Plants with a Transfer Learning Approach for Sustainable Cities. Electron. 2021,10 (10). 1149. (100) Bernardelli, A.; Marsili-Libelli, S.; Manzini, A.; Stancari, S.; Tardini, G.; Montanari, D.; Anceschi, G.; Gelli, P.; Venier, S. RealTime Model Predictive Control of a Wastewater Treatment Plant Based on Machine Learning. Water Sci. Technol. 2020,81 (11), 2391−2400. (101) Wang, J.; Wan, K.; Gao, X.; Cheng, X.; Shen, Y.; Wen, Z.; Tariq, U.; Piran, J. Energy and Materials-Saving Management via Deep Learning for Wastewater Treatment Plants. IEEE Access 2020,8, 191694−191705. (102) Oulebsir, R.; Lefkir, A.; Safri, A.; Bermad, A. Optimization of the Energy Consumption in Activated Sludge Process Using Deep Learning Selective Modeling. Biomass and Bioenergy 2020,132, No. 105420. (103) Zhang, S.; Wang, H.; Keller, A. A. Novel Machine LearningBased Energy Consumption Model of Wastewater Treatment Plants. ACS ES&T Water 2021,1(12), 2531−2540. (104) Fan, Y.; Guo, Z.; Wang, J.; Zhang, B.; Shen, Y.; Gao, X. Online Learning-Empowered Smart Management for A2O Process in Sewage Treatment Processes. Environ. Res. 2022,210, No. 113015. (105) Hvala, N.; Kocijan, J. Design of a Hybrid Mechanistic/ Gaussian Process Model to Predict Full-Scale Wastewater Treatment Plant Effluent. Comput. Chem. Eng. 2020,140, 106934. (106) Coté, M.; Grandjean, B. P. A.; Lessard, P.; Thibault, J. Dynamic Modelling of the Activated Sludge Process: Improving Prediction Using Neural Networks. Water Res. 1995,29 (4), 995− 1004. (107) Anderson, J. S.; McAvoy, T. J.; Hao, O. J. Use of Hybrid Models in Wastewater Systems. Ind. Eng. Chem. Res. 2000,39 (6), 1694−1704. (108) Lee, D. S.; Jeon, C. O.; Park, J. M.; Chang, K. S. Hybrid Neural Network Modeling of a Full-Scale Industrial Wastewater Treatment Process. Biotechnol. Bioeng. 2002,78 (6), 670−682. (109) Lee, D. S.; Vanrolleghem, P. A.; Park, J. M. Parallel Hybrid Modeling Methods for a Full-Scale Cokes Wastewater Treatment Plant. J. Biotechnol. 2005,115 (3), 317−328. (110) Lessard, P.; Beck, M. B. Dynamic Modelling of the Activated Sludge Process: A Case Study. Water Res. 1993,27 (6), 963−978. (111) Zhao, H.; Isaacs, S.H.; Søeberg, H.; Kummel, M. A Novel Control Strategy for Improved Nitrogen Removal in an Alternating Activated Sludge Process�Part I. Process Analysis. Water Res. 1994, 28 (3), 521−534. (112) Li, K.; Duan, H.; Liu, L.; Qiu, R.; van den Akker, B.; Ni, B. J.; Chen, T.; Yin, H.; Yuan, Z.; Ye, L. An Integrated First Principal and Deep Learning Approach for Modeling Nitrous Oxide Emissions from Wastewater Treatment Plants. Environ. Sci. Technol. 2022,56 (4), 2816−2826. (113) Wu, X.; Zheng, Z.; Wang, L.; Li, X.; Yang, X.; He, J. Coupling Process-Based Modeling with Machine Learning for Long-Term Simulation of Wastewater Treatment Plant Operations. J. Environ. Manage. 2023,341 (May), No. 118116. (114) Cheng, X.; Guo, Z.; Shen, Y.; Yu, K.; Gao, X. Knowledge and Data-Driven Hybrid System for Modeling Fuzzy Wastewater Treatment Process. Neural Comput. Appl. 2023,35 (10), 7185−7206. (115) Wang, J. H.; Zhao, D. J.; Liao, W. S.; Mahmoud, M. S.; Guo, Z. W.; Li, H. M.; Gao, X.; Feng, D.; Shi, L. F.; Chen, Y. P.; Shen, Y. An Online Intelligent Management Method for Wastewater Treatment Supported by Coupling Data-Driven and Mechanism Models. J. Water Process Eng. 2023,53, No. 103653. (116) Yao, S.; Zhang, C.; Yuan, H. Emerging Investigator Series: Modeling of Wastewater Treatment Bioprocesses: Current Development and Future Opportunities. Environ. Sci. Water Res. Technol. 2022,8(2), 208−225. (117) Baker, R. E.; Pena, J. M.; Jayamohan, J.; Jérusalem, A. Mechanistic Models versus Machine Learning, a Fight Worth Fighting for the Biological Community? Biol. Lett. 2018,14 (5), 20170660. (118) Davenport, T. H. From Analytics to Artificial Intelligence. J. Bus. Anal. 2018,1(2), 73−80. (119) Harrou, F.; Cheng, T.; Sun, Y.; Leiknes, T.; Ghaffour, N. A Data-Driven Soft Sensor to Forecast Energy Consumption in Wastewater Treatment Plants: A Case Study. IEEE Sens. J. 2021,21 (4), 4908−4917. (120) Vasilaki, V.; Conca, V.; Frison, N.; Eusebi, A. L.; Fatone, F.; Katsou, E. A Knowledge Discovery Framework to Predict the N2O Emissions in the Wastewater Sector. Water Res. 2020,178 (2), No. 115799. (121) Demir, S.; Mincev, K.; Kok, K.; Paterakis, N. G. Data Augmentation for Time Series Regression: Applying Transformations, Autoencoders and Adversarial Networks to Electricity Price Forecasting. Appl. Energy 2021,304, No. 117695. (122) Fryer, D.; Strumke, I.; Nguyen, H. Shapley Values for Feature Selection: The Good, the Bad, and the Axioms. IEEE Access 2021,3, 3−10. (123) Lever, J.; Krzywinski, M.; Altman, N. Model Selection and Overfitting. Nat. Methods 2016,13 (9), 703−704. (124) Dosilovic, F. K.; Brcic, M.; Hlupic, N. Explainable Artificial Intelligence: A Survey. 2018 41st Int. Conv. Inf. Commun. Technol. Electron. Microelectron. MIPRO 2018 - Proc. 2018, 210−215. (125) Zhuang, F.; Qi, Z.; Duan, K.; Xi, D.; Zhu, Y.; Zhu, H.; Xiong, H.; He, Q. A Comprehensive Survey on Transfer Learning. Proc. IEEE 2021,109 (1), 43−76. (126) Corominas, L.; Garrido-Baserba, M.; Villez, K.; Olsson, G.; Cortés, U.; Poch, M. Transforming Data into Knowledge for Improved Wastewater Treatment Operation: A Critical Review of Techniques. Environ. Model. Softw. 2018,106, 89−103. (127) Lu, L.; Zheng, H.; Jie, J.; Zhang, M.; Dai, R. Reinforcement Learning-Based Particle Swarm Optimization for Sewage Treatment Control. Complex Intell. Syst. 2021,7(5), 2199−2210. ACS ES&T Water pubs.acs.org/estwater Review https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 803 (128) Shao, X.; Pu, C.; Zhang, Y.; Kim, C. S. Domain Fusion CNNLSTM for Short-Term Power Consumption Forecasting. IEEE Access 2020,8, 188352−188362. (129) Sajjad, M.; Khan, Z. A.; Ullah, A.; Hussain, T.; Ullah, W.; Lee, M. Y.; Baik, S. W. A Novel CNN-GRU-Based Hybrid Approach for Short-Term Residential Load Forecasting. IEEE Access 2020,8, 143759−143768. (130) Lu, H.; Ma, X. Hybrid Decision Tree-Based Machine Learning Models for Short-Term Water Quality Prediction. Chemosphere 2020, 249, No. 126169. (131) Syed, D.; Abu-Rub, H.; Ghrayeb, A.; Refaat, S. S. HouseholdLevel Energy Forecasting in Smart Buildings Using a Novel Hybrid Deep Learning Model. IEEE Access 2021,9, 33498−33511. (132) Dong, B.; Li, Z.; Rahman, S. M. M.; Vega, R. A Hybrid Model Approach for Forecasting Future Residential Electricity Consumption. Energy Build. 2016,117, 341−351. ACS ES&T Water pubs.acs.org/estwater Review https://doi.org/10.1021/acsestwater.3c00117 ACS EST Water 2024, 4, 784−804 804