Next Article in Journal
Optimizing Power Market Clearing with Segmented Electricity Prices: A Bilevel Model
Previous Article in Journal
Seismic Characterization of a Landslide Complex: A Case History from Majes, Peru
Previous Article in Special Issue
Research on Rural Wastewater Treatment Technology in Northwest China Based on Anaerobic Biofilm Coupled with Anaerobic Baffle Plate Reactor (ABR) Technology
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:

A Review on Applications of Artificial Intelligence in Wastewater Treatment

Institute of Agri-Biological Environmental Engineering, College of Biosystems Engineering and Food Science, Zhejiang University, Hangzhou 310058, China
Key Laboratory of Equipment and Informatization in Environment Controlled Agriculture, Ministry of Agriculture and Rural Affairs, Key Laboratory of Intelligent Equipment and Robotics for Agriculture of Zhejiang Province, Hangzhou 310058, China
School of Life and Environmental Science, Wenzhou University, Wenzhou 325035, China
National & Local Joint Engineering Research Center for Ecological Treatment Technology of Urban Water Pollution, Wenzhou University, Wenzhou 325035, China
Key Laboratory of Coastal Environment and Resources of Zhejiang Province, School of Engineering, Westlake University, Hangzhou 310030, China
School of Mathematics and Physics, Wenzhou University, Wenzhou 325035, China
Environmental Engineering Program, University of Northern British Columbia, Prince George, BC V2N 4Z9, Canada
Authors to whom correspondence should be addressed.
Sustainability 2023, 15(18), 13557;
Submission received: 30 June 2023 / Revised: 29 July 2023 / Accepted: 10 August 2023 / Published: 11 September 2023


In recent years, artificial intelligence (AI), as a rapidly developing and powerful tool to solve practical problems, has attracted much attention and has been widely used in various areas. Owing to their strong learning and accurate prediction abilities, all sorts of AI models have also been applied in wastewater treatment (WWT) to optimize the process, predict the efficiency and evaluate the performance, so as to explore more cost-effective solutions to WWT. In this review, we summarize and analyze various AI models and their applications in WWT. Specifically, we briefly introduce the commonly used AI models and their purposes, advantages and disadvantages, and comprehensively review the inputs, outputs, objectives and major findings of particular AI applications in water quality monitoring, laboratory-scale research and process design. Although AI models have gained great success in WWT-related fields, there are some challenges and limitations that hinder the widespread applications of AI models in real WWT, such as low interpretability, poor model reproducibility and big data demand, as well as a lack of physical significance, mechanism explanation, academic transparency and fair comparison. To overcome these hurdles and successfully apply AI models in WWT, we make recommendations and discuss the future directions of AI applications.

1. Introduction

Water resources, one of the most significant elements in human life and production processes, are now under serious threat from harmful pollutants caused by human activities and natural processes [1]. A large amount of wastewater is produced every day, most of it contains toxic pollutants and is directly released into the environment without being treated or reused [2]. In general, untreated sewage is rich in various nutrients, organic matters, suspended solids (SSs), organic micropollutants and pathogenic and nonpathogenic microorganisms [3]. Depending on the source, wastewaters can be classified into six categories: municipal, domestic, industrial, medical, agricultural and nuclear. Among them, municipal and domestic wastewaters are the most abundant, and the research on WWT is concentrated on these two types of sewage [4]. Because wastewaters with different sources may differ significantly and have different physical and chemical properties, it is important to assess their characteristics before choosing the appropriate treatment process. In order to protect the limited water resources, environment and human health and meet the growing water demand, it is imperative to explore the treatment of wastewaters and their reuse as a resource. Wastewater treatment (WWT) removes contaminants from sewage involving a combination of physical, chemical and biological processes [5], and produces clean water that is safely released back into the environment. The key issues of WWT are how to reduce water pollution to a safe level efficiently while producing fewer negative impacts on the environment and decreasing energy consumption. Overall, WWT, as an indispensable process for water resources reuse and sustainable development, is essential for protecting public health and the environment and has been widely used in industrial and agricultural fields [6]. Effective and innovative technologies are urgently needed to improve efficiency, reduce cost and decrease the energy consumption of WWT [7,8].
Artificial intelligence (AI) refers to the ability of a computer program to realize autonomous learning, reasoning, judgment and decision making by simulating human intelligence. AI, one of the most impressive inventions during this century, is developing rapidly and has been widely applied in many areas such as natural language processing (NLP), computer vision (CV) and autopilot. Benefiting from its high efficiency, AI can be used for classification and regression analysis of massive amounts of data generated anytime and anywhere, thus energizing industries and promoting the development of all walks of life greatly. With the rapid development of computer technology, machine learning (ML), as an important branch of AI, uses data, algorithms, statistics and mathematical optimization to imitate the way that humans learn, gradually improving its accuracy and achieving artificial intelligence. Due to the advent of the big data era and increasingly strong supercomputing capabilities, ML is becoming more and more popular and has been successfully implemented in industry, agriculture, medicine, environmental protection, scientific research and other fields.
The WWT process mainly consists of water quality monitoring, laboratory-scale research and process design. AI models are becoming more and more popular in wastewater-related fields, especially in recent years (see Figure 1), and have been employed for the prediction and optimization of the WWT process [9,10]. In previous WWT-related research, AI models have shown very good prediction and optimization performances [11], and have been successfully applied to WWT process design [10,12], water quality monitoring [13,14], WWT process parameters optimization [15,16] and WWT process performance prediction [17,18]. These pieces of research have demonstrated that an AI model, as a powerful tool, has achieved great success in the applications of WWT-related fields. However, most of the review works of AI applications in WWT focus on some technique or process design aspects of WWT, such as adsorption processes, membrane bioreactors, membrane processes and WWTP. A comprehensive review of AI applications in the WWT process involving water quality monitoring, laboratory studies and real process design has rarely been seen until now. Additionally, there are few review articles available, which comprehensively introduce the commonly used AI models in WWT and summarize their advantages, disadvantages and proposals.
In this review, an overview of the literature on the applications of AI models and smart technologies, with a special focus on most ML in WWT, is presented. This review is not intended to cover all the applications of AI, ML and smart technologies in WWT, but rather to summarize the key findings of these important published works and analyze future development trends of AI in WWT. The WWT-related applications are mainly concentrated in the modeling, prediction and optimization of water and wastewater treatment processes, containing water quality monitoring, laboratory-scale research and process design. The remainder of this review is organized as follows: Section 2 systematically introduces the commonly used AI models in WWT and summarizes the strengths and weaknesses of each model. Section 3 outlines the applications of AI models in WWT, including water quality monitoring for data acquisition, laboratory-scale research and process design. Section 4 presents challenges and future perspectives of AI applications in WWT. Finally, Section 5 ends with a conclusion.

2. AI Models

The most commonly used AI models for WWT in the literature are shown in Figure 2. These models used for prediction and optimization can be classified into three main categories: Artificial Neural Network (ANN), Machine Learning (ML) and Search Algorithm (SA).

2.1. Artificial Neural Network (ANN)

ANN is a mathematical model that imitates the behavioral characteristics of a biological neural network (NN) to process information. In ANN models, unit nodes are used to simulate neurons, and information processing is achieved by adjusting the weights of interconnection among a large number of nodes (neurons) in the neural network. Usually, ANN consists of an input layer, an output layer, and some hidden layers between the input and output layers. In ANN, many variable weights between neurons and active functions, such as sigmoid, tanh and ReLU functions, are used to perform complex nonlinear computation [19]. As the number of hidden layers of ANN increases, ANN can build more complex nonlinear models and its expression ability enhances. Thus, ANN can be trained to learn complex nonlinear relationships between inputs and outputs by constructing and optimizing loss functions [20]. The most commonly used ANN models mainly include Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), Fuzzy Neural Networks (FNNs) and Deep Neural Networks (DNNs). The basic architectures of ANN, RNN, CNN, FNN and DNN models are shown in Figure 3, where Figure 3a displays an ANN architecture, including neurons represented by the circles; an input layer fed by input variables 1, 2, , n; two hidden layers in the middle; and an output layer with output variables 1, 2, , n. Next, we will present a brief introduction of these ANN models.

2.1.1. Recurrent Neural Network (RNN)

RNN is a class of NN with feedback connections that takes sequence data as inputs and makes recursion in the evolution direction of the sequence. RNN has the abilities of memory, parameter sharing and Turing completeness, so it has certain advantages in learning the nonlinear characteristics of time series problems. The most commonly used RNN is long short-term memory (LSTM), which solves the gradient disappearance problem in traditional RNN by adding additional gated units [21]. RNN achieves great success in the applications of water and WWT, water quality management and water-based agriculture. A simple RNN architecture is shown in Figure 3b, where the output of the hidden layer is stored in the memory “W”, which can be considered as another input in the RNN.

2.1.2. Convolutional Neural Network (CNN)

CNN is a class of Feedforward Neural Networks with convolutional computation and deep structure. It is one of the representative algorithms of deep learning (DL) with representation learning ability and has been widely used in computer vision, natural language processing and other fields. CNN extracts the complex features of input images through the convolutional layers, reduces the feature dimension through the pooling layers and, finally, realizes the task of classification or regression through the fully connected layers [22]. Figure 3c depicts a typical CNN architecture, which consists of input, output, convolutional, pooling and fully connected layers.

2.1.3. Fuzzy Neural Network (FNN)

FNN is a hybrid NN model that combines the advantages of fuzzy logic and ANN to handle problems with uncertainty or ambiguity. FNN uses fuzzy logic reasoning to process the input data and then applies ANN to train and output the results. FNN has a similar structure to traditional NN, but it uses fuzzy logic (membership function, fuzzy inference and normalization) to describe the fuzzy relationship between inputs and outputs, as well as the connection weights among neurons. A typical FNN architecture, including input, membership function, fuzzy inference, normalized and output layers, is shown in Figure 3d. FNN has some advantages in solving the problems that are difficult for traditional NN, and has been widely used in pattern recognition, control system, predictive analysis and so on.

2.1.4. Deep Neural Network (DNN)

DNN is a type of ANN with multiple hidden layers between the input and output layers. The deep architecture of DNN enables it to learn hierarchical representations of data, where higher-level features are learned by combining lower-level features in successive layers. Similar to other neural networks, DNN consists of more hidden layers and neurons and has been widely used for learning highly nonlinear mappings from inputs to outputs or capturing complex patterns in data. However, DNN needs a large amount of data to train because of the complex network architecture, which makes training difficult and computation expensive. Figure 3e displays a common DNN architecture with input, output and multiple hidden layers.

2.2. Machine Learning (ML)

ML is a subfield of AI that focuses on the development of algorithms and statistical models, and enables computer systems to automatically learn from data without being explicitly programmed [23]. The primary goal of ML is to build predictive models that can make accurate predictions or decisions based on what it has learned from data [24]. The most commonly used ML models include Principal Component Analysis (PCA), Decision Tree (DT), Support Vector Machine (SVM), Particle Swarm Optimization (PSO), Random Forest (RF), K-Nearest Neighbor (KNN), Self-Organizing Map (SOM) and Adaptive-Network-Based Fuzzy Inference System (ANFIS). Figure 4 and Figure 5 present a schematic diagram of ML models, including PCA, DT, SVM and PSO and RF, SOM, KNN and ANFIS, respectively.

2.2.1. Principal Component Analysis (PCA)

PCA is a simple multivariate statistical machine learning algorithm and a commonly used data dimensionality reduction technique that can convert high-dimensional data to low-dimensional data while retaining most of the original data information by an orthogonal transformation [25]. It extracts some of the largest principal components based on the variances of variables for a better understanding of the system, which are orthogonal to each other. PCA is widely used in areas such as data clustering, image processing, natural language processing, noise filtering and other fields. A schematic diagram of PCA is presented in Figure 4a, where “ PC 1 ” and “ PC 2 ” denote the first and second principal components of datasets, respectively.

2.2.2. Decision Tree (DT)

DT is a common ML algorithm that can be used for classification and regression problems. DT divides the datasets into different subsets, and each subset is further divided according to the value of a certain feature. This process can be regarded as the construction of a tree with root, intermediate and leaf nodes, where each node represents a feature. Each branch represents the value of the feature, and every leaf node denotes the real classification or regression result [26]. DT is easy to understand and interpret, and can handle missing data, outliers and nonlinear relationships with high accuracy. It can also be used for feature selection, but it is easy to overfit, especially for higher-dimensional datasets. Figure 4b exhibits a typical DT architecture with branches and root, intermediate and leaf nodes.

2.2.3. Support Vector Machine (SVM)

SVM is a commonly used ML algorithm for classification and regression problems. It maps data to a higher dimensional feature space using kernel functions, such as polynomial kernel and radial basis kernel functions. It then searches an optimal hyperplane that separates different categories of data in the feature space based on the greatest distance from the nearest data point to the hyperplane [27]. SVM can deal with high-dimensional data and nonlinear relationships, and avoid the problem of local optimal solutions. However, it is inefficient for big datasets, sensitive to noise and outliers and may have difficulty in selecting suitable kernel functions and hyperparameters. Figure 4c displays an example of SVM for classification, where the solid line is a hyperplane acting as a decision boundary and the two parallel dashed lines represent spacing boundaries.

2.2.4. Particle Swarm Optimization (PSO)

PSO is a commonly used optimization algorithm for solving complex optimization problems. It imitates the swarm behavior of biological populations, such as birds or fishes, by constantly adjusting the individual position and velocity to search the optimal solutions. In each iteration of the algorithm, each individual represented by a particle updates its position and velocity and moves to the best-known position based on the optimal position in the population and the defined rule [28]. PSO can deal with various complex nonlinear problems and avoid falling into local optimal solutions, but it is sensitive to the initial conditions of the problem, resulting in the requirement of running many times to obtain good results. A schematic diagram of PSO to obtain the global optimal solution is shown in Figure 4d.

2.2.5. Random Forest (RF)

RF is an integrated ML algorithm and is mainly used for classification and regression problems. Its main idea is to improve the accuracy and stability of the model by integrating multiple DTs on data samples. In RF, to improve the diversity of the model, each DT is a basic classifier that is trained on a random subset, usually randomly drawn from the original datasets. The final classification or optimization is obtained by voting or averaging the results of multiple DTs [29]. RF can deal with high-dimensional data and complex nonlinear problems, and avoid overfitting. However, it may be sensitive to certain noises and outliers and requires more computing time and resources to train, especially under the condition of high tree density. Figure 5a depicts a schematic diagram of RF for classification by integrating n DTs.

2.2.6. Self-Organizing Map (SOM)

SOM is a commonly used ML algorithm for clustering and dimensionality reduction and only consists of input and output layers. It projects high-dimensional input data into a low-dimensional space and associates similar input data through a competitive process while preserving the topology of the data. SOM, as an unsupervised ML method, can reduce the dimensionality of complex datasets in a low-dimensional mapping space, which makes it easier to visualize and classify data points [30]. It is an effective visualization method to help understand high-dimensional data and has been widely used in data visualization, clustering, classification and other fields owing to its simplicity and practicality. A schematic diagram of SOM is presented in Figure 5b.

2.2.7. K-Nearest Neighbor (KNN)

KNN is a simple and commonly used ML algorithm used for classification and regression. It predicts the label of a new data point according to the label of K neighbors closest to the new data point in the training datasets. KNN has the ability to deal with complex nonlinear relationships between inputs and outputs, adjust the complexity of the model adaptively and adapt to various data types using different distance measurements. However, it may be sensitive to high-dimensional datasets and noise, and computationally expensive for big datasets. Figure 5c displays an example of KNN before and after classification for the new data (solid red circle) when K = 3 and K = 5 .

2.2.8. Adaptive-Network-Based Fuzzy Inference System (ANFIS)

ANFIS is a new hybrid intelligent inference system based on fuzzy logic and neural network for regression, classification and prediction. It effectively combines the fuzzy inference capacity of fuzzy logic with the learning ability of neural networks to realize the adaptive learning of fuzzy inference systems [31]. ANFIS and FNN have some similarities, but the network structure, computational method and application scope are different. A typical ANFIS architecture with input, membership function, firing strength calculation, normalized, linear combination and output layers is shown in Figure 5d. ANFIS can be used for modeling nonlinear and multivariable complex systems and has been widely applied in the areas of fuzzy control, prediction and classification.

2.3. Search Algorithm (SA)

Except for ANN and ML, SAs are also commonly used AI models, including Genetic Algorithm (GA) and Genetic Programming (GP), as shown in Figure 6. A brief introduction to GA and GP is given in the following.

2.3.1. Genetic Algorithm (GA)

GA is a commonly used search and optimization method based on the principle of biological evolution. It imitates the genetic process in nature and obtains excellent individuals better adapted to the environment through genetic manipulation. The genetic manipulation mainly includes selection, crossover and mutation, as depicted in Figure 6a. GA can search the optimal solution in multidimensional spaces and is not limited by the local optimal solution. Thus, it has good global search ability and strong adaptability and can deal with complex nonlinear optimization problems [32]. GA is widely applied in ML, AI, control systems, optimization design and other fields.

2.3.2. Genetic Programming (GP)

GP is an evolutionary computing technique based on GA that automates the generation and selection of computer programming inspired by biological evolutionary processes to perform regression, classification and optimization. Unlike other ML algorithms, GP is performed automatically by randomly generating an initial population and then using GA to evolve [33]. A schematic diagram of GP is displayed in Figure 6b. GP has good adaptability and generalization ability, and is often used to solve problems that are highly nonlinear or have no explicit analytic form, such as image recognition and prediction.
In fact, each AI model has advantages, disadvantages and application scopes. Table 1 lists the commonly used AI models for WWT, their purposes, advantages and disadvantages. An appropriate AI model should be selected carefully depending on its advantages and the characteristics of the problem to be solved, so as to get the best results. Meanwhile, Table 2 summarizes the commonly used activation functions of AI models for WWT. The expression and output range of every activation function are presented, and the frequency of use in the literature is marked with the blue symbol “”. The more symbols there are, the higher frequency of its use. Additionally, we present a general flow chart in Figure 7 to illustrate how AI models are applied in WWT. To evaluate the performance of AI models, some commonly used indicators are required, for example, mean squared error (MSE), root mean squared error (RMSE), sum of squared error (SSE), mean absolute error (MAE) and coefficient of determination ( R 2 ). The definitions and details of these common indicators are omitted here; interested readers can refer to the literature [34].

2.4. Hybrid AI Models

Hybrid AI models can take full advantage of individual models to improve the prediction or optimization performance of AI models by integrating two or more of the above AI models. As shown in Table 1, every AI model has some drawbacks used for WWT. Hybrid AI models overcome the major disadvantages of a single AI model and, thus, show stronger learning and prediction abilities in dealing with more complex nonlinear problems. In the literature, GA, PSO, RNN and SVM are the commonly used AI models for combination with other AI models to obtain a more effective hybrid AI model with better performance, such as GA-SVR, GA-ANN, GA-FNN, PSO-RNN, PSO-SVM, PSO-ANN, ANN-GANN and SVM-SA [49]. Hybrid AI models have shown their great potential in solving new or difficult environmental problems related to sewage and are receiving increasing attention from researchers [50,51,52].

3. Applications of AI Models in WWT

AI models along with the Internet of Things (IoT) framework and conventional methods are helpful for the design of smart WWT systems and the reuse of sewage [53]. An AI model is a useful and powerful tool for the modeling, prediction and optimization of the WWT process and has been widely applied in various aspects of WWT, such as the removal of dyes, heavy metals, nutrients, organics, solids, microbial contamination, drugs and pesticides from water [49,54,55,56]. From the viewpoint of research scale, the applications of AI models are mainly in laboratory-scale research and process design. In practical applications, process design usually consists of process parameter optimization and process performance prediction. In optimization and prediction, a large amount of data is required to establish and train AI models, which can be achieved by monitoring water quality.

3.1. Water Quality Monitoring for Data Acquisition

In the applications of AI models in WWT, water quality monitoring is an important method to obtain water quality parameters or data. A lot of data are available by using sensors to continuously monitor influent and effluent water quality [34]. Although different numbers of datasets as inputs are employed for different studies, the percentage of data for training and testing AI models almost remains the same. Most studies use 60–80% data for training and the remaining data for testing. Some researchers have made an attempt to design various sensors to enable rapid and accurate real-time monitoring and WWT process automation by real-time sensing, data analysis and online controls [57,58,59,60]. For the monitoring of influent water, some water quality parameters, such as BOD, COD, pH, DO, flow rate, temperature and initial pollutant concentration, are easily obtained and used for the inputs of AI models, while for the monitoring of effluent water, some water quality parameters, such as effluent BOD, COD, pH, DO and pollutant concentration, are usually used to evaluate the effect of WWT or the performance of wastewater treatment plants (WWTPs).
An increasing number of measured data and AI models, as well as multivariate statistical methods, have made data-driven modeling and real-time prediction attractive. Post et al. [61] combined a CNN model with laser-induced Raman and fluorescence spectroscopy (LIRFS) to achieve real-time monitoring of the micropollutants of WWTP with a correlation coefficient of R 2 = 0.74 for all samples. The results show that this method can lead to high-precision measurement results, reach detection limits and detect micropollutants that cannot be monitored using the monitoring methods of WWTPs. The combination of sensitive fluorescence measurements with very specific Raman measurements supplemented with AI is a promising real-time monitoring tool for image recognition in WWT-related fields, such as microbiological water quality tests, micropollutants identifications and even device-specific adjustments in WWTP management. Based on PCA, ANN and multivariate statistical process control, Lee et al. [38] developed a real-time remote monitoring system for WWTP to monitor operating statuses and provided the key information needed for efficient operation from the experts. Mustafa et al. [62] reviewed the applications of IoT and AI models in water quality monitoring and prediction with high accuracy, which can provide safe water quality services for users and an important basis for government water quality management decision making. For water quality monitoring using soft sensors, a literature review was performed by Haimi et al. [63] to summarize the applications of data-derived soft sensors for the monitoring and online prediction of biological WWTP. Ching, So and Morck [64] also conducted a systematic review of advances in soft sensors for the online monitoring of WWTP. Schneider et al. [65] performed an experimental study to identify sensors to monitor on-site WWTP without sensor maintenance over one year and showed that robust soft sensors can be reasonably designed to meet real-time monitoring tasks while reducing the maintenance frequency dramatically.

3.2. Laboratory-Scale Research

Owing to low cost, short experiment period, security controllability, easy operation and good repeatability, laboratory-scale research has been an important tool for developing new process designs and WWT technologies. We conclude the common applications of AI models for laboratory-scale research in Table 3. In laboratory-scale research, the applications of AI models in WWT mainly focus on the design and optimization of membrane processes or bioreactors. Some researchers have made comprehensive reviews of the applications of AI models for WWT using membrane processes or membrane bioreactors (MBRs) in recent studies [66,67,68]. MBR is an efficient and useful wastewater treatment process combining biological (microbial) treatment with membrane filtration. It is a hybrid system essentially that integrates a conventional biological treatment system and critical physical liquid–solid separation functions achieved by membrane filtration equipment. Because MBR combines the technical superiorities of membrane-based physical separation and microorganism-based biodegradation, it has several advantages over the traditional activated sludge process, such as higher biomass concentration, less sludge production, shorter hydraulic retention time (HRT), eliminating the need for secondary clarifiers, smaller plant space requirement and improved effluent quality [69]. Thus, as one of the most important innovative technologies in WWT, MBR is an efficient tool for sustainable wastewater management and has been widely used for the treatment of various municipal and industrial wastewater. Recently, Rahman et al. [70] performed a review of the historical advancement in MBR technology toward sustainable wastewater management. Tomczak and Gryta [71] outlined energy-efficient anaerobic MBR (AnMBR) technology for WWT and demonstrated that AnMBRs have lower energy demand than typical WWTPs.
A variety of anaerobic [72] and aerobic [73] MBRs have been developed and applied in WWT, but a main barrier for the widespread application of membrane processes or MBRs is membrane fouling since it significantly decreases the performance and lifespan of membranes, resulting in increased maintenance and operating costs. Membrane fouling induces a decrease in the permeability of membranes, and is a complex and inevitable phenomenon, which is attributed to the accumulation and adsorption of the pollutants in wastewater on the membrane surface and inside the membrane pores. Overall, membrane fouling is the key impact factor limiting the performance and cost of membrane processes, and researchers have been trying to develop efficient and sustainable fouling control strategies. Tomczak, Grubecki and Gryta [74] proposed a method for membrane fouling control in an MBR using 1% NaOH solutions and demonstrated that the method is effective in restoring the initial membrane performance. In the membrane processes or MBR-related application studies, how to control membrane fouling and optimize MBR performance effectively and economically are two of the most central questions for the rapid commercialization of and large-scale applications in WWT [75,76,77].
For the applications of AI models to predict or control membrane fouling, the commonly studied membrane types are forward osmosis, reverse osmosis, nanofiltration, ultrafiltration and microfiltration [78]. Chen et al. [79] applied a radial basis function (RBF) ANN model to quantify interfacial energy with a randomly rough membrane surface in the membrane fouling process. They showed that the RBF-ANN model can well capture the complex relationships between interfacial energy and key influencing factors. Jawad et al. [80] adopted an ANN model to predict permeate flux for a lab-scale forward osmosis process with high accuracy, R 2 = 0.973 . The results indicate that the ANN model performs better than the MLP model, and a lower number of hidden layers and a higher number of neurons are helpful to improve the accuracy of the ANN model. Subsequently, they presented a hybrid ANN-RSM model to further simulate the forward osmosis process and predict membrane flux [81]. In the developed hybrid model, the ANN model predicting the membrane flux is used for the experimental design, while the RSM model is used for optimization. The prediction performance for the ANN and RSM models are R 2 = 0.98036 and 0.9408, respectively.
For the applications of AI models to optimize MBR performance, various operating parameters, such as temperature, pH, DO, salinity, HRT, pollution load, BOD, COD and pollutant concentration, are considered to determine the optimum processing condition [82]. Zaghloul et al. [83] proposed a five-stage ML model to simulate and predict the behaviors of aerobic granular sludge (AGS) reactors using 475 days of data collected from three lab-based reactors and adopted an ensemble of ANN, SVR and ANFIS models to improve the predictive performance. They found the model can forecast the behaviors of AGS reactors with average R 2 = 95.7%, RMSE = 0.032 and MAPE = 3.7%. Ren et al. [84] used a Backpropagation Neural Network (BPNN) model to simulate the removal of COD f ilt and COD by conducting a pilot-scale submerged MBR to treat high-strength Chinese traditional medicine wastewater and confirmed that the model can accurately predict the removal rates and help to obtain the optimum operational conditions. Cai et al. [85] conducted an aerobic–anaerobic micro-sludge MBR (O-AMSMBR) to study the effect of pH on pollutant removal performance of a reactor using Wavelet Neural Network (WNN) and BPNN models and showed that pH is a key factor affecting the COD and TN removal efficiencies of O-AMSMBR. To optimize the processing efficiency of MBRs, they further studied the effects of various ecological factors on effluent marine domestic sewage by implementing an air-lift multilevel circulation MBR and analyzed their impacts on the O-AMSMBR performance using the BPNN model [86]. The results show that the order of relative importance for the ecological factors is pH ≈ MLSS > HRT > COD, which indicates that pH is significant and should be considered in implementing AI models to evaluate the effectiveness of MBR systems.
Table 3. Applications of AI models for laboratory-scale research.
Table 3. Applications of AI models for laboratory-scale research.
AI Model UsedInput VariablesOutput VariablesRemarksReactor TypeRef.
ANN, ANFIS,Influent NH 4 N , PO 4 3 ,Effluent COD,New multi-stage ML model forAGS reactor[83]
SVMpH, OLR, HRT, etc. NH 4 N andbetter prediction of AGS reactor
PO 4 3 performance. An ensemble of ML
for more accurate predictions.
ANNCOD, MLSS, MLVSS,TransmembraneNew ANN model to accuratelyAnoxic–[87]
pH, DO, Alkalinity, TN,pressure (TMP)predict membrane fouling.aerobic
TP, NO 3 N , NH 4 N Identify an optimal parameter setMBR
to predict TMP using ANN.
ANN, ANFISInfluent COD, pH, oilBiogasNew ANN and ANFIS models toUASB[88]
and grease removal, etc.productionpredict biogas production from
spearmint essential oil WWT.
Obtain the best BP-ANN and
ANFIS topologies.
ANNHRT, temperature,MethaneANN model to forecast biogasAnaerobic[89]
composition andproductionproduction and identify thereactor
chemical dose optimum process conditions.
Chemical treatment enhances
anaerobic digestion efficiency.
ANNVolatile solid, pH,BiogasANN model to predict biogasAnaerobic[90]
organic load rate, HRT,productionproduction from food, fruits andreactor
temperature, reactor vegetable wastes. Assess different
volume ANN topologies and build database.
ANNMLSS concentration,TMP and CODANN model to simulate andMBR[91]
HRT and timeremovalpredict TMP and COD removal
percentagepercentage of MBR. Protein in
biofilm/cake EPS is the dominant
fouling factor.
PCA, fuzzyTMPPrincipalPCA and FC to assess membraneMBR[92]
clustering (FC) components offouling. PCA-FC model for
TMPmembrane fouling control.
ANNFlux, aeration ratio,TMPMathematical and ANN modelsIntermittently[93]
initial TMP, operating to predict membrane fouling.aerated MBR
time, etc. Mathematical model has a better
stability and ANN has a better
prediction performance.
RecurrentInfluent COD, NH 4 N ,MembraneIntelligent detecting systemMBR[94]
fuzzy NNpH, BOD, SS, TP, etc.permeabilityto evaluate MBR performance.
Suitable for online detecting
membrane fouling.
MLP, ANN,Time, TSS, influentTMP orGA-ANN model to evaluate membraneSubmerged[95]
GACOD, SRT, MLSSpermeabilityfouling. GA-ANN predicts TMP andMBR
permeability accurately.
ANNInfluent concentrationsEffluentNew ANN model to predict biofilmBiofilm[96]
of COD, NH 4 N andconcentrations ofsystem performance. The new modelsystem
TN, etc.COD, NH 4 N , TNperforms the best.
WNN, BPNNpH, sludge loading,EffluentWNN and BPNN models to study O AMSMBR [85]
salinity, COD or TNconcentrations ofthe effect of pH on pollutant
volume loading rateCOD or TNremoval. pH is the key factor
for biodegradation.
BPNNInfluent concentrationsEffluentNew BPNN model to simulate AnMBRAnMBR[97]
of COD, BOD, etc.concentrationsperformance. AnMBR can treat
of COD, etc.pharmaceutical wastewater efficiently.
ANN, GAConductivity, organicCOD removalNew ANN-GA model to predict andUASB[98]
loading rate,efficiencyoptimize COD removal efficiency.
temperature ANN-GA improves reactor performance.

3.3. Process Design

WWT process involves complex process design and operating conditions, and AI models have shown great advantages in minimizing or reducing the complexities of the WWT process. AI models can effectively and easily establish a complex relationship between the various input and output variables. The commonly used input variables are time, temperature, pH, initial concentration of pollutants and influent water quality parameters, and the output variables are mainly the removal efficiency and the adsorption efficiency of contaminants or effluent water quality parameters. AI models have been successfully applied in various aspects of WWT, such as the prediction of effluent water quality and WWT performance, as well as the optimization of energy consumption and operating parameters [96]. From the perspective of process design of WWT, the applications of AI models are concentrated in the optimization of process parameters and the prediction of process performance [99].

3.3.1. Process Parameters Optimization

The main purpose of process parameter optimization is to reduce costs and increase the efficiency of WWT. Table 4 summarizes the applications of AI models for the optimization of process parameters. Nayak et al. [100] used a hybrid ANN-GA model to predict the optimal process conditions for enhancing the biomass of the green microalga in an algal biorefinery. They found 4-12-1 topology is the optimal network architecture with a maximum correlation coefficient R = 0.9947 and minimum MSE, and these parameters improved the algal biomass productivity by about 57% and had a CO 2 sequestration rate of 578.1 ± 23.1 mg L 1 d 1 and a COD reduction of 95.9 ± 2.4%. Qi et al. [39] applied RSM, ANN-PSO and ANN-GA models to study the decontamination of methylene blue (MB) from simulated wastewater using mesoporous rGO/Fe/Co nanohybrids. The results show that the ANN-PSO model has the best performance among these models in the prediction of the optimum conditions for decontamination efficiency. The mesoporous nanohybrids could be used as a low-cost and fast decontaminant material to treat organic contaminants or other pollutants in wastewater. Martín de la Vega and Jaramillo-Morán [40] adopted SOM to identify four key parameters of a municipal WWTP running by monitoring Oxidation–Reduction Potential (ORP) and DO based on three thousand two hundred aeration–non-aeration cycles. This method can improve the removal efficiency of nutrients in WWTP.
For process parameter optimization, Picos-Benítez et al. [101] utilized an ANN-GA model to predict the treatment performance of sulfate wastewaters with Bromophenol blue dye using an electro-oxidation (EO) process and obtained the optimum operational conditions. They found that the AI model is a powerful tool in designing and controlling the WWT processes. For MB WWT, ANN and ANFIS models were employed by Aghilesh et al. [102] to obtain the optimum conditions for MB removal using low-cost agricultural waste (sugarcane bagasse and peanut hulls). They also performed Fourier Transform Infra-Red (FTIR) spectral analysis to confirm the biosorption and the distinguished prediction performance of these AI models for biosorption. In fact, FTIR is a useful spectroscopic technique and has been widely used for the detection and analysis of pollutants, such as microplastics in water [103], table salts [104] and nitrates from agricultural fertilizers in soil [105,106]. FTIR spectra can be used to analyze the chemical composition of pollutants and identify functional groups, providing important information about them present in compounds, complex substances and bio-sorbent surfaces, but spectra interpretation is time-consuming. To reduce the time to analyze functional groups so as to facilitate the interpretation of FTIR spectra, Enders et al. [107] developed the first generalizable model based on CNN to identify functional groups in gas-phase FTIR spectra. The results demonstrate that CNN models are effective at identifying spectral features and can be extended to other micropollutants or chemical identification application fields with a lot of spectral examples. Overall, image-based AI models coupled with spectroscopy techniques, such as FTIR, SEM and Raman spectroscopy, are useful for the identification and detection of contaminants, especially micropollutants in water-related fields.
For a WWT system, Li et al. [43] proposed a hybrid deep leaning CLSTMA model based on CNN, LSTM and AM to monitor and model the water quality of a paper industrial WWT system for cleaner production. Compared with CNN, LSTM and CLSTM models, the CLSTMA model has better performance in monitoring and modeling water quality. An intelligent WWT system based on AI models and sensors was presented by Miao et al. [41] to assist in managing sewage treatment. A Gated Recurrent Unit (GRU) model performs better than LSTM and SVR models, and the intelligent WWT system can be extended to small-scale sewage industries in sustainable cities. Rodríguez-Rangel et al. [36] also explored five AI models to simulate and predict the biomass production of carbohydrates in WWT systems, considering the interactions of nutrients, carbon, biomass growth and population. The results indicate that the CNN-1D model has better performance than other models and can approximate system dynamics. For the study of WWTP, Hwangbo et al. [42] used DNN and LSTM to predict the N 2 O emission rate and identify the key parameters affecting the characteristics of N 2 O high emission. They found that the LSTM model performs better than the DNN model, and a hybrid model combining mechanistic with DL models is helpful in quantitatively describing and understanding complex N 2 O emission dynamics from WWTPs. Zhu, Jiang and Feng [37] also proposed an upgraded feedforward NN with the least square SVM (FFNN-LSSVM) method to forecast the effluent BOD/NH3-N of a WWTP. The proposed model has high predictive accuracy, limited computation duration and a simple calculation mechanism, and performs better than existing techniques in wastewater quality prediction.

3.3.2. Process Performance Prediction

Process performance is an important aspect of the WWT process and has been the focus of researchers’ attention. The common process performance predictions include the removal efficiency of pollutants, WWTP performance, optimal process condition and effluent quality. A brief summary of the applications of AI models for the prediction of process performance is presented in Table 5. For the prediction of WWTP performance, Nourani et al. [108] analyzed Nicosia WWTP performance using single AI and ensemble models. The results show that the ANFIS model performs better than other single AI models, and the NN ensemble model has the best prediction performance among the ensemble models. Xie et al. [109] also combined improved Feedforward Neural Network (IFFNN) with GA to predict the real-time effluent water quality of a WWTP in Jiangsu Province, and found the IFFNN-GA model enhances prediction performance by 52.3% (COD) and 72.6% (TN) compared with the traditional FFNN model. Deep cascade-forward backpropagation (DCB) and DL time series forecasting (DLTSF) models were presented by El-Rawy et al. [110] to predict the effluent water quality of the El-Berka WWTP and evaluate its treatment performance.
For the removal of contaminants, Bisaria et al. [111] employed ANN and ANFIS models to simulate the adsorption process of chlorpyrifos (CPS) using Trapa bispinosa peel (UFBC) and validated the effectiveness of adsorption experimentally. The results illustrate that UFBC is a sustainable and effective adsorbent with low equilibrium time and high adsorption capacity for CPS removal. For the prediction of effluent quality, Yang et al. [112] proposed a dynamic PCA-NARX model to predict the effluent quality and made potential real-time adjustments for WWTP operations. The results show that the dynamic model has better performance than static ANN models in modeling effluent quality. Nnaji et al. [113] predicted COD and CTSS removal efficiencies from textile wastewater using complex salt–Luffa cylindrica seed extract (CS-LCSE) as a coagulant based on RSM, ANN and ANFIS models. The results demonstrate that the ANFIS model has the best predictive performance with a higher R 2 value (0.9997 and 0.9996 for CTSS and COD removals) and a lower MSE value (0.0002643 and 0.0038472 for CTSS and COD removals). For the prediction of optimal process conditions, Mahmoud et al. [114] employed an ANN model to predict the COD removal efficiency from domestic wastewater by preparing Fe/Cu NPs under different operating conditions. The adsorption isotherm, kinetic studies and RSM results indicate that Fe/Cu NPs are an effective adsorbent material for COD removal, and the ANN model is useful to explore the optimum removal condition.
More and more studies have demonstrated that the utilization of AI in WWTPs can significantly enhance WWT efficiency and decision making, decrease environmental impacts and improve their performance (see Table 5) because AI algorithms can optimize various processes and operational conditions, such as temperature, pH, COD, chemical dosage, flow control and energy consumption. This leads to more efficient resource utilization, reduced operational costs and energy consumption, and improved overall system performance. Additionally, a combination of AI technologies with the use of sensors can realize real-time monitoring and control, resulting in decreased response time, system failure risks and human resource costs. Overall, WWTPs that utilize AI technologies have shown superior performance compared to those without AI. The incorporation of AI enables enhanced efficiency, real-time monitoring, predictive maintenance, data-driven decision making and reduced environmental impact. Implementing AI in WWTPs can lead to more sustainable and effective management of our water resources.
Besides the applications of AI models in the prediction of process performance and the optimization of process parameters, AI models can be used in the pretreatment of wastewater to improve the pretreatment accuracy and the adaptive control accuracy of the system [115,116]. Some researchers have employed various AI models and their variants to comprehensively evaluate the water quality of rivers [117], lakes [118] and reservoirs [119].
Table 5. Applications of AI models for the prediction of process performance.
Table 5. Applications of AI models for the prediction of process performance.
AI Models UsedInput VariablesOutput VariablesRemarksCountryRef.
FFNN, ANFIS,Influent pH, BOD,Effluent BOD,Used 3 AI models to predict NicosiaCyprus[108]
SVMCOD, conductivity,COD and TNWWTP performance. NN ensemble
and TN model is more robust and reliable.
IFFNN, GAInfluent waterEffluent waterNew IFFNN-GA model to enhanceChina[109]
quality,quality predictionreal-time prediction of WWTP effluent
flow rate, time t + 1 quality.
FFNN, LSTMInfluent TSS,Effluent TSS,Different AI models to predictEgypt[110]
BOD, COD,BOD, COD,effluents and performance of WWTP.
ammonia andammonia andRecommend DCB and DLTSF models
sulfidesulfidefor evaluation and prediction.
ANN, ANFISContact time,AdsorptionANN and ANFIS models to predictIndia[111]
adsorbent dose,efficiencythe adsorption capacity of CPS
pH, etc. by UFBC. New material is a sustainable
and effective adsorbent.
PCA,pH, COD, BOD,Effluent COD andNew PCA-NARX model to predictChina[112]
NARX NN,TN, TP, SS, NH 4 + TNeffluent water. Dynamic model
ANNand chromaticity performs better than static model.
ANN, ANFISpH, dosage andCOD and CTSSANFIS model outperformsNigeria[113]
stirring timeremovalover ANN and RSM models.
ANNpH, NP dose,COD removalNew ANN model to predict COD-[114]
contact time, etc.efficiencyremoval efficiency. Fe/Cu NPs
are strong absorbents.
ELM, GA,Flow rate,COD at time tKELM-SSA model performs betterIran[120]
SSA, PSOtemperature, pH, than other AI modes in predicting
NH 4 , EC, COD at real-time water quality due to
time t 1 the combination of SSA.
ANN, PCAType 0041,Sludge volumeANN and multivariate statistics toSouth[121]
Gordonia spp., etc.index (SVI)predict sludge volume index andAfrica
assess filamentous bacteria.
WNNInfluent COD,Effluent COD,New WNN model can accurately evaluateChina[122]
NH 4 + N , salinity NH 4 + N the effect of salinity and predict
pollutant removal processes.
WNNTN and CODEffluent COD,WNN model can forecast COD and TNChina[123]
loading rates,TNremovals and help long-term stable
HRT, pH operation of WWTP.
ANN, GATime, OLR, RT,COD, TOC,ANN-GA model can predict hypersaline   Malaysia[124]
TDSMLSS, oil inoily WWT processes and evaluate MBR
FFNN, ANFIS,Influent BOD,Effluent BOD and       AI models to predict WWTP effluentIran[125]
SVRCOD, TSS andCOD at time tparameters. Using jittering and
pH, etc. ensemble models simultaneously
increases prediction accuracy.
SVM, ANFISInfluent pH, TS,Effluent KjeldahlSVM and ANFIS models to predictIndia[126]
COD, etc.Nitrogenremoval efficiency of Kjeldahl
concentrationNitrogen. SVM can evaluate
WWTP efficiency.
ANNpH, adsorbentColor removalANN and other models to simulateEgypt[127]
dose, contactefficiencyadsorption processes. GT-nZVI has
time, etc. a strong color removal ability for
textile wastewater.

4. Challenges and Future Perspectives

Although AI models have many advantages over traditional models and have achieved great success in WWT-related fields demonstrated by all of the aforementioned studies, their disadvantages and limitations hindering widespread applications in WWT should not be ignored. The challenges and future perspectives of AI applications in WWT are summarized as follows:
  • An AI model such as ANN can describe complex nonlinear relationships between multiple inputs and multiple outputs, but it is a black-box and data-driven model essentially. That is to say, the AI model just offers a mapping relationship between inputs and outputs, but it cannot provide any mechanism information about the problem to be studied. AI models have proven to be a powerful tool and show good prospects in engineering applications for WWT fields; however, they have a long way to go in scientific research. The main reason is that the underlying mechanisms behind many WWT-related issues in the current research are still not clear. Although the AI models used in the above studies show good performances in solving specific problems and usually exhibit problem dependence, whether they can be applied to other WWT-related problems and their application scopes in WWT need to be studied. More importantly, traditional mathematical models [128], such as known knowledge, principles and equations, are called white-box models, which elucidate the underlying mechanisms, but have difficulty describing the complex nonlinear relationships between inputs and outputs of AI models. Combining an AI model (data-driven model) with a traditional mathematical model (knowledge-driven model) can dramatically reduce data requirements and allow for easily obtaining meaningful results [129]. The integrated black-box with white-box model is a promising tool for the study of the underlying mechanisms of WWT systems.
  • AI models usually have low interpretation because their parameters, such as neurons, hidden layers, weights and biases, have no physical significance, resulting in low interpretation. Another drawback of an AI model is poor reproducibility because of the random weights and biases [67]. Additionally, training NN, especially DNN, is difficult; thus, it is hard to obtain the optimal network parameters. An improperly trained NN may converge to a local minimum. Generally, NN provides different solutions under the conditions of different network architectures and parameters. There is no standard way to determine the best network architecture so far, which is often problem-dependent. Trial and error seems to be the only way, but this easily leads to overfitting or underfitting. For a particular real problem, the appropriate selection of AI models, inputs, outputs, model architecture and datasets is vital for the results. How to construct AI models reasonably needs to be further studied. Moreover, more theoretical studies on AI techniques are needed to mitigate the difficulties of NN training, parameter optimization, poor reproducibility and low interpretation, so as to promote the development of AI applications in WWT.
  • AI models are heavily dependent on data, and big data are required in the training or learning process to guarantee prediction or optimization accuracy. In the WWT applications mentioned above, AI models are applied to different WWT-related fields and a lot of data with different formats and types have been collected, resulting in poor data management and difficulty in reuse by other researchers. Moreover, the data and source code used in the studies are rarely made public for various reasons, which leads to a lack of academic transparency. This is another reason why it is difficult to reproduce the results in the literature. Furthermore, researchers have used some statistical methods to evaluate the accuracy or precision of their AI models in almost every paper reviewed in this journal, such as R 2 , MAE, MSE, RMSE, MAPE and SSE. The absence of open source code and data makes it difficult to fairly compare the performance of different AI models [12]. Due to the lack of benchmarks, standardization and fair comparison, it is hard for researchers to judge which AI model performs better for a specific real problem. Therefore, in order to reduce experiment costs, achieve fair comparisons and promote the widespread application of AI models in WWT, raw data and source code are encouraged to be made public and shared, and benchmark and standardization should be established.

5. Conclusions

This review summarizes the commonly used AI models and their applications in WWT, ranging from water quality monitoring, laboratory-scale research to process design. AI models are becoming more and more popular in WWT-related fields because of their strong learning and accurate prediction abilities. They have been successfully applied to model WWT systems, optimize process parameters, predict process performance and identify and detect contamination. Although AI models have many advantages and have become very useful tools for the treatment of wastewater, their disadvantages and limitations should not be ignored. Big data demand; poor data management; low interpretability; poor model reproducibility; and a lack of physical significance, mechanism explanation, academic transparency, standardization and fair comparison are important obstacles to the AI applications in relevant areas of WWT.
In order to overcome these hurdles and successfully apply AI models to WWT, mathematicians, biologists, engineers and computer experts should cooperate and develop new models or innovative technologies to design optimal WWT systems. Additionally, more studies from lab to field scales are needed to understand the complex behaviors of WWT systems with varying effect factors and to explore the mechanisms of key problems involved in WWT. Furthermore, hybrid AI models that integrate the advantages of two or more AI models and the newly emerging attention-based AI models could be solutions to complex water-treatment-related problems. The fusion of data-driven and knowledge-driven AI models is a new and promising method that is receiving increasing attention in WWT-related fields.

Author Contributions

Conceptualization, Y.W., Y.C. and D.L.; methodology, Y.W., Y.C., H.L., Q.G., C.D. and D.L.; validation, Y.W., Y.C., H.L., Q.G. and D.L.; formal analysis, Y.W., Y.C., H.L., Q.G., C.D. and D.L.; investigation, Y.W., Y.C., M.Z. and D.L.; writing—original draft preparation, Y.W. and Y.C.; writing—review and editing, Y.W., Y.C., C.D., M.Z. and D.L.; visualization, Y.W. and Y.C.; supervision, Y.W., M.Z. and D.L.; project administration, Y.W., M.Z. and D.L.; funding acquisition, Y.W., M.Z. and D.L. All authors have read and agreed to the published version of the manuscript.


This research was funded by the Key Research and Development Program of Zhejiang Province (grant No.: 2022C02045), the National Natural Science Foundation of China (grant No.: 12202319) and the National Key Research and Development Program of China (grant No.: 2018YFE0103700).

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Not applicable.


The authors sincerely thank the reviewers for their constructive comments to improve the manuscript.

Conflicts of Interest

The authors declare no conflict of interest.


The following abbreviations are used in this manuscript:
AIArtificial Intelligence
NNNeural Network
DLDeep Learning
ANNArtificial Neural Network
MLMachine Learning
SASearch Algorithm
RNNRecurrent Neural Network
CNNConvoluted Neural Network
FNNFuzzy Neural Network
DNNDeep Neural Network
PCAPrincipal Component Analysis
DTDecision Tree
SVMSupport Vector Machine
PSOParticle Swarm Optimization
RFRandom Forest
KNNK-Nearest Neighbor
SOMSelf-Organizing Map
ANFISAdaptive-Network-based Fuzzy Inference System
GAGenetic Algorithm
GPGenetic Programming
BPNNBackpropagation Neural Network
WNNWavelet Neural Network
RSMResponse Surface Methodology
RBFRadial Basis Function
MLPMulti-Layer Perceptron
SVRSupport Vector Regression
GRUGated Recurrent Unit
LSTMLong Short-Term Memory
IoTInternet of Things
LSSVMLeast Square Support Vector Machine
GA-SVRGenetic Algorithm–Support Vector Regression
ELMExtreme Learning Machine
R 2 Coefficient of Determination
MSEMean Squared Error
SSESum of Squared Error
RMSERoot Mean Square Error
MAPEMean Absolute Percentage Error
MAEMean Absolute Error
WWTWastewater Treatment
WWTPWastewater Treatment Plant
MBRMembrane Bioreactor
UASBUp-flow Anaerobic Sludge Blanket
AGSAerobic Granular Sludge
TMPTransmembrane Pressure
BODBiological Oxygen Demand
CODChemical Oxygen Demand
DODissolved Oxygen
TNTotal Nitrogen
TPTotal Phosphorus
HRTHydraulic Retention Time
SSSuspended Solid
TSSTotal Suspended Solid
CTSSColor Total Suspended Solid
MLSSMixed Liquor Suspended Solid
MLVSSMixed Liquor Volatile Suspended Solid
TOCTotal Organic Carbon
TICTotal Inorganic Carbon
OLROrganic Loading Rate
TDSTotal Dissolved Solid
RTReaction Time
SRTSludge Retention Time
ECElectrical Conductivity
EPSExtracellular Polymer
MBMethylene Blue


  1. Ray, S.S.; Verma, R.K.; Singh, A.; Ganesapillai, M.; Kwon, Y.-N. A holistic review on how artificial intelligence has redefined water treatment and seawater desalination processes. Desalination 2023, 546, 116221. [Google Scholar] [CrossRef]
  2. Boretti, A.; Rosa, L. Reassessing the projections of the world water development report. NPJ Clean Water 2019, 2, 15. [Google Scholar] [CrossRef]
  3. Chahal, C.; Van Den Akker, B.; Young, F.; Franco, C.; Blackbeard, J.; Monis, P. Pathogen and particle associations in wastewater: Significance and implications for treatment and disinfection processes. Adv. Appl. Microbiol. 2016, 97, 63–119. [Google Scholar] [PubMed]
  4. Jasim, N.A. The design for wastewater treatment plant (WWTP) with GPS X modelling. Cogent Eng. 2020, 7, 1723782. [Google Scholar] [CrossRef]
  5. Zaibel, I.; Arnon, S.; Zilberg, D. Treated municipal wastewater as a water source for sustainable aquaculture: A review. Rev. Aquacult. 2022, 14, 362–377. [Google Scholar] [CrossRef]
  6. Ren, J.; Shen, W.; Man, Y.; DOng, L. Applications of Artificial Intelligence in Process Systems Engineering; Elsevier: Amsterdam, The Netherlands, 2021. [Google Scholar]
  7. Malviya, A.; Jaspal, D. Artificial intelligence as an upcoming technology in wastewater treatment: A comprehensive review. Environ. Technol. Rev. 2021, 10, 177–187. [Google Scholar] [CrossRef]
  8. Ma, Z.; Cheah, W.Y.; Ng, I.S.; Chang, J.S.; Zhao, M.; Show, P.L. Microalgae-Based Biotechnological Sequestration of Carbon Dioxide for Net Zero Emissions. Trends Biotechnol. 2022, 40, 1439–1453. [Google Scholar] [CrossRef]
  9. 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, 102974. [Google Scholar] [CrossRef]
  10. Shirkoohi, M.G.; Tyagi, R.D.; Vanrolleghem, P.A.; Drogui, P. Artificial intelligence techniques in electrochemical processes for water and wastewater treatment: A review. J. Environ. Health Sci. Eng. 2022, 20, 1089–1109. [Google Scholar] [CrossRef]
  11. Bhardwaj, A.; Kishore, S.; Pandey, D.K. Artificial Intelligence in Biological Sciences. Life 2022, 12, 1430. [Google Scholar] [CrossRef]
  12. Lowe, M.; Qin, R.; Mao, X. A Review on Machine Learning, Artificial Intelligence, and Smart Technology in Water Treatment and Monitoring. Water 2022, 14, 1384. [Google Scholar] [CrossRef]
  13. Gaudio, M.T.; Coppola, G.; Zangari, L.; Curcio, S.; Greco, S.; Chakraborty, S. Artificial Intelligence-Based Optimization of Industrial Membrane Processes. Earth Syst. Environ. 2021, 5, 385–398. [Google Scholar] [CrossRef]
  14. Wang, G.; Su, W.; Hu, B.; AL-Huqail, A.; Majdi, H.S.; Algethami, J.S.; Jiang, Y.; Ali, H.E. Assessment in carbon-based layered double hydroxides for water and wastewater: Application of artificial intelligence and recent progress. Chemosphere 2022, 308, 136303. [Google Scholar] [CrossRef] [PubMed]
  15. Bhagat, S.K.; Pilario, K.E.; Babalola, O.E.; Tiyasha, T.; Yaqub, M.; Onu, C.E.; Pyrgaki, K.; Falah, M.W.; Jawad, A.H.; Yaseen, D.A.; et al. Comprehensive review on machine learning methodologies for modeling dye removal processes in wastewater. J. Clean. Prod. 2023, 385, 135522. [Google Scholar] [CrossRef]
  16. Hasan, M.A. An emergent addition for the optimal systemization of wastewater utilization plants using artificial intelligence. Water Sci. Technol. 2021, 84, 2805–2817. [Google Scholar] [CrossRef]
  17. Yaseen, Z.M. An insight into machine learning models era in simulating soil, water bodies and adsorption heavy metals: Review, challenges and solutions. Chemosphere 2021, 277, 130126. [Google Scholar] [CrossRef]
  18. Liu, S.; Lo, C.K.Y.; Kan, C.w. Application of artificial intelligence techniques in textile wastewater decolorisation fields: A systematic and citation network analysis review. Color. Technol. 2022, 138, 117–136. [Google Scholar] [CrossRef]
  19. Basheer, I.A.; Hajmeer, M. Artificial neural networks: Fundamentals, computing, design, and application. J. Microbiol. Meth. 2000, 43, 3–31. [Google Scholar] [CrossRef]
  20. Ding, S.; Li, H.; Su, C.; Yu, J.; Jin, F. Evolutionary artificial neural networks: A review. Artif. Intell. Rev. 2013, 39, 251–260. [Google Scholar] [CrossRef]
  21. Sherstinsky, A. Fundamentals of recurrent neural network (RNN) and long short-term memory (LSTM) network. Physica D 2020, 404, 132306. [Google Scholar] [CrossRef]
  22. Alzubaidi, L.; Zhang, J.; Humaidi, A.J.; Al-Dujaili, A.; Duan, Y.; Al-Shamma, O.; Santamaría, J.; Fadhel, M.A.; Al-Amidie, M.; Farhan, L. Review of deep learning: Concepts, CNN architectures, challenges, applications, future directions. J. Big Data 2021, 8, 53. [Google Scholar] [CrossRef] [PubMed]
  23. Samuel, A.L. Some studies in machine learning using the game of checkers. IBM J. Res. Dev. 1959, 3, 210–229. [Google Scholar] [CrossRef]
  24. Dey, A. Machine Learning Algorithms: A Review. Int. J. Comput. Sci. Inf. Technol. 2016, 7, 1174–1179. [Google Scholar]
  25. 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, 148. [Google Scholar] [CrossRef] [PubMed]
  26. Mining, W.I.D. Data mining: Concepts and techniques. Morgan Kaufinann 2006, 10, 559–569. [Google Scholar]
  27. Yekkehkhany, B.; Safari, A.; Homayouni, S.; Hasanlou, M. A comparison study of different kernel functions for SVM-based classification of multi-temporal polarimetry SAR data. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2014, 40, 281. [Google Scholar] [CrossRef]
  28. Cura, T. A particle swarm optimization approach to clustering. Expert Syst. Appl. 2012, 39, 1582–1588. [Google Scholar] [CrossRef]
  29. Genuer, R.; Poggi, J.-M.; Tuleau-Malot, C. Variable selection using random forests. Pattern Recogn. Lett. 2010, 31, 2225–2236. [Google Scholar] [CrossRef]
  30. Qu, X.; Yang, L.; Guo, K.; Ma, L.; Sun, M.; Ke, M.; Li, M. A survey on the development of self-organizing maps for unsupervised intrusion detection. Mobile Netw. Appl. 2021, 26, 808–829. [Google Scholar] [CrossRef]
  31. Karaboga, D.; Kaya, E. Adaptive network based fuzzy inference system (ANFIS) training approaches: A comprehensive survey. Artif. Intell. Rev. 2019, 52, 2263–2293. [Google Scholar] [CrossRef]
  32. Holland, J.H. Genetic algorithms. Sci. Am. 1992, 267, 66–73. [Google Scholar] [CrossRef]
  33. Espejo, P.G.; Ventura, S.; Herrera, F. A survey on the application of genetic programming to classification. IEEE Trans. Syst. Man Cybern. C 2009, 40, 121–144. [Google Scholar] [CrossRef]
  34. Singh, N.K.; Yadav, M.; Singh, V.; Padhiyar, H.; Kumar, V.; Bhatia, S.K.; Show, P.-L. Artificial intelligence and machine learning-based monitoring and design of biological wastewater treatment systems. Bioresour. Technol. 2023, 369, 128486. [Google Scholar] [CrossRef]
  35. Guo, Z.; Du, B.; Wang, J.; Shen, Y.; Li, Q.; Feng, D.; Gao, X.; Wang, H. Data-driven prediction and control of wastewater treatment process through the combination of convolutional neural network and recurrent neural network. RSC Adv. 2020, 10, 13410–13419. [Google Scholar] [CrossRef] [PubMed]
  36. Rodríguez-Rángel, H.; Arias, D.M.; Morales-Rosales, L.A.; Gonzalez-Huitron, V.; Valenzuela Partida, M.; García, J. Machine learning methods modeling carbohydrate-enriched cyanobacteria biomass production in wastewater treatment systems. Energies 2022, 15, 2500. [Google Scholar] [CrossRef]
  37. Zhu, J.; Jiang, Z.; Feng, L. Improved neural network with least square support vector machine for wastewater treatment process. Chemosphere 2022, 308, 136116. [Google Scholar] [CrossRef] [PubMed]
  38. Lee, M.W.; Hong, S.H.; Choi, H.; Kim, J.-H.; Lee, D.S.; Park, J.M. Real-time remote monitoring of small-scaled biological wastewater treatment plants by a multivariate statistical process control and neural network-based software sensors. Process Biochem. 2008, 43, 1107–1113. [Google Scholar] [CrossRef]
  39. Qi, J.; Hou, Y.; Hu, J.; Ruan, W.; Xiang, Y.; Wei, X. Decontamination of methylene Blue from simulated wastewater by the mesoporous rGO/Fe/Co nanohybrids: Artificial intelligence modeling and optimization. Mater. Today Commun. 2020, 24, 100709. [Google Scholar] [CrossRef]
  40. Martín de la Vega, P.; Jaramillo-Morán, M. Obtaining Key Parameters and Working Conditions of Wastewater Biological Nutrient Removal by Means of Artificial Intelligence Tools. Water 2018, 10, 685. [Google Scholar] [CrossRef]
  41. Miao, S.; Zhou, C.; AlQahtani, S.A.; Alrashoud, M.; Ghoneim, A.; Lv, Z. Applying machine learning in intelligent sewage treatment: A case study of chemical plant in sustainable cities. Sustain. Cities Soc. 2021, 72, 103009. [Google Scholar] [CrossRef]
  42. Hwangbo, S.; Al, R.; Chen, X.; Sin, G. Integrated model for understanding N2O emissions from wastewater treatment plants: A deep learning approach. Environ. Sci. Technol. 2021, 55, 2143–2151. [Google Scholar] [CrossRef] [PubMed]
  43. Li, X.; Yi, X.; Liu, Z.; Liu, H.; Chen, T.; Niu, G.; Yan, B.; Chen, C.; Huang, M.; Ying, G. Application of novel hybrid deep leaning model for cleaner production in a paper industrial wastewater treatment system. J. Clean. Prod. 2021, 294, 126343. [Google Scholar] [CrossRef]
  44. Ma, J.; Ding, Y.; Cheng, J.C.; Jiang, F.; Xu, Z. Soft detection of 5-day BOD with sparse matrix in city harbor water using deep learning techniques. Water Res. 2020, 170, 115350. [Google Scholar] [CrossRef] [PubMed]
  45. Chatterjee, S.; Sarkar, S.; Dey, N.; Ashour, A.S.; Sen, S.; Hassanien, A.E. Application of cuckoo search in water quality prediction using artificial neural network. Int. J. Comput. Intell. Stud. 2017, 6, 229–244. [Google Scholar] [CrossRef]
  46. Li, J.; Sharma, K.; Li, W.; Yuan, Z. Swift hydraulic models for real-time control applications in sewer networks. Water Res. 2022, 213, 118141. [Google Scholar] [CrossRef]
  47. Aka, R.J.N.; Wu, S.; Mohotti, D.; Bashir, M.A.; Nasir, A. Evaluation of a liquid-phase plasma discharge process for ammonia oxidation in wastewater: Process optimization and kinetic modeling. Water Res. 2022, 224, 119107. [Google Scholar] [CrossRef]
  48. Cangialosi, F.; Bruno, E.; De Santis, G. Application of machine learning for fenceline monitoring of odor classes and concentrations at a wastewater treatment plant. Sensors 2021, 21, 4716. [Google Scholar] [CrossRef]
  49. Alam, G.; Ihsanullah, I.; Naushad, M.; Sillanpää, M. Applications of artificial intelligence in water treatment for optimization and automation of adsorption processes: Recent advances and prospects. Chem. Eng. J. 2022, 427, 130011. [Google Scholar] [CrossRef]
  50. Bagheri, M.; Akbari, A.; Mirbagheri, S.A. Advanced control of membrane fouling in filtration systems using artificial intelligence and machine learning techniques: A critical review. Process Saf. Environ. 2019, 123, 229–252. [Google Scholar] [CrossRef]
  51. Kosko, B. Neural Networks and Fuzzy Systems: A Dynamical Systems Approach to Machine Intelligence; Prentice-Hall Inc.: Upper Saddle River, NJ, USA, 1992. [Google Scholar]
  52. 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. 2020, 133, 169–182. [Google Scholar] [CrossRef]
  53. Krishnan, S.R.; Nallakaruppan, M.K.; Chengoden, R.; Koppu, S.; Iyapparaja, M.; Sadhasivam, J.; Sethuraman, S. Smart Water Resource Management Using Artificial Intelligence—A Review. Sustainability 2022, 14, 13384. [Google Scholar] [CrossRef]
  54. Altowayti, W.A.H.; Shahir, S.; Othman, N.; Eisa, T.A.E.; Yafooz, W.M.S.; Al-Dhaqm, A.; Soon, C.Y.; Yahya, I.B.; Che Rahim, N.A.N.b.; Abaker, M.; et al. The Role of Conventional Methods and Artificial Intelligence in the Wastewater Treatment: A Comprehensive Review. Processes 2022, 10, 1832. [Google Scholar] [CrossRef]
  55. Abidli, A.; Huang, Y.; Ben Rejeb, Z.; Zaoui, A.; Park, C.B. Sustainable and efficient technologies for removal and recovery of toxic and valuable metals from wastewater: Recent progress, challenges, and future perspectives. Chemosphere 2022, 292, 133102. [Google Scholar] [CrossRef] [PubMed]
  56. Fan, C.; Lv, C.; Wang, Z.; Wu, S.; Jin, Z.; Bei, K.; He, S.; Kong, H.; Zhao, J.; Zhao, M. Influence of regular addition of ore on treatment efficiency and aquatic organisms in living machine system for black water treatment. J. Clean. Prod. 2022, 341, 130928. [Google Scholar] [CrossRef]
  57. Zhang, W.; Tooker, N.B.; Mueller, A.V. Enabling wastewater treatment process automation: Leveraging innovations in real-time sensing, data analysis, and online controls. Environ. Sci. Water Res. Technol. 2020, 6, 2973–2992. [Google Scholar] [CrossRef]
  58. Carreres-Prieto, D.; Garcia, J.T.; Carrillo, J.M.; Vigueras-Rodriguez, A. Towards highly economical and accurate wastewater sensors by reduced parts of the LED-visible spectrum. Sci. Total Environ. 2023, 871, 162082. [Google Scholar] [CrossRef]
  59. Emaminejad, S.A.; Morgan, V.L.; Kumar, K.; Kavathekar, A.; Ragush, C.; Shuai, W.; Jia, Z.; Huffaker, R.; Wells, G.; Cusick, R.D. Statistical and microbial analysis of bio-electrochemical sensors used for carbon monitoring at water resource recovery facilities. Environ. Sci. Water Res. Technol. 2022, 8, 2052–2064. [Google Scholar] [CrossRef]
  60. Zhang, W.; Wei, S.P.; Winkler, M.K.H.; Mueller, A.V. Design of a Soft Sensor for Monitoring Phosphorous Uptake in an EBPR Process. Acs Est. Engg. 2022, 2, 1847–1856. [Google Scholar] [CrossRef]
  61. Post, C.; Heyden, N.; Reinartz, A.; Foerderer, A.; Bruelisauer, S.; Linnemann, V.; Hug, W.; Amann, F. Possibilities of Real Time Monitoring of Micropollutants in Wastewater Using Laser-Induced Raman & Fluorescence Spectroscopy (LIRFS) and Artificial Intelligence (AI). Sensors 2022, 22, 4668. [Google Scholar]
  62. Mustafa, H.M.; Mustapha, A.; Hayder, G.; Salisu, A. Applications of IoT and Artificial Intelligence in Water Quality Monitoring and Prediction: A Review. In Proceedings of the 2021 6th International Conference on Inventive Computation Technologies (ICICT), Coimbatore, India, 20–22 January 2021; pp. 968–975. [Google Scholar]
  63. Haimi, H.; Mulas, M.; Corona, F.; Vahala, R. Data-derived soft-sensors for biological wastewater treatment plants: An overview. Environ. Modell. Softw. 2013, 47, 88–107. [Google Scholar] [CrossRef]
  64. Ching, P.M.; So, R.H.; Morck, T. Advances in soft sensors for wastewater treatment plants: A systematic review. J. Water Process Eng. 2021, 44, 102367. [Google Scholar] [CrossRef]
  65. Schneider, M.Y.; Carbajal, J.P.; Furrer, V.; Sterkele, B.; Maurer, M.; Villez, K. Beyond signal quality: The value of unmaintained pH, dissolved oxygen, and oxidation-reduction potential sensors for remote performance monitoring of on-site sequencing batch reactors. Water Res. 2019, 161, 639–651. [Google Scholar] [CrossRef] [PubMed]
  66. Kamali, M.; Appels, L.; Yu, X.; Aminabhavi, T.M.; Dewil, R. Artificial intelligence as a sustainable tool in wastewater treatment using membrane bioreactors. Chem. Eng. J. 2021, 417, 128070. [Google Scholar] [CrossRef]
  67. Jawad, J.; Hawari, A.H.; Zaidi, S.J. Artificial neural network modeling of wastewater treatment and desalination using membrane processes: A review. Chem. Eng. J. 2021, 419, 129540. [Google Scholar] [CrossRef]
  68. Narayanan, C.; Narayan, V. Biological wastewater treatment and bioreactor design: A review. Sustain. Environ. Res. 2019, 29, 33. [Google Scholar] [CrossRef]
  69. Iorhemen, O.T.; Hamza, R.A.; Tay, J.H. Membrane bioreactor (MBR) technology for wastewater treatment and reclamation: Membrane fouling. Membranes 2016, 6, 33. [Google Scholar] [CrossRef]
  70. Rahman, T.U.; Roy, H.; Islam, M.R.; Tahmid, M.; Fariha, A.; Mazumder, A.; Tasnim, N.; Pervez, M.N.; Cai, Y.; Naddeo, V. The advancement in membrane bioreactor (MBR) technology toward sustainable industrial wastewater management. Membranes 2023, 13, 181. [Google Scholar] [CrossRef]
  71. Tomczak, W.; Gryta, M. Energy-Efficient AnMBRs Technology for Treatment of Wastewaters: A Review. Energies 2022, 15, 4981. [Google Scholar] [CrossRef]
  72. Kamali, M.; Gameiro, T.; Costa, M.E.V.; Capela, I. Anaerobic digestion of pulp and paper mill wastes–An overview of the developments and improvement opportunities. Chem. Eng. J. 2016, 298, 162–182. [Google Scholar] [CrossRef]
  73. Lee, W.J.; Ng, Z.C.; Hubadillah, S.K.; Goh, P.S.; Lau, W.J.; Othman, M.; Ismail, A.F.; Hilal, N. Fouling mitigation in forward osmosis and membrane distillation for desalination. Desalination 2020, 480, 114338. [Google Scholar] [CrossRef]
  74. Tomczak, W.; Grubecki, I.; Gryta, M. The Use of NaOH Solutions for Fouling Control in a Membrane Bioreactor: A Feasibility Study. Membranes 2021, 11, 887. [Google Scholar] [CrossRef] [PubMed]
  75. Aslam, M.; Charfi, A.; Lesage, G.; Heran, M.; Kim, J. Membrane bioreactors for wastewater treatment: A review of mechanical cleaning by scouring agents to control membrane fouling. Chem. Eng. J. 2017, 307, 897–913. [Google Scholar] [CrossRef]
  76. Guo, W.; Ngo, H.-H.; Li, J. A mini-review on membrane fouling. Bioresour. Technol. 2012, 122, 27–34. [Google Scholar] [CrossRef] [PubMed]
  77. Shahbeig, H.; Mehrnia, M.R.; Tashauoei, H.R.; Rezaei, M. Role of zeolite in reducing membrane fouling in a hybrid membrane bioreactor system applied for wastewater treatment. Desalin. Water Treat. 2017, 98, 52–58. [Google Scholar] [CrossRef]
  78. Niu, C.; Li, X.; Dai, R.; Wang, Z. Artificial intelligence-incorporated membrane fouling prediction for membrane-based processes in the past 20 years: A critical review. Water Res. 2022, 216, 118299. [Google Scholar] [CrossRef] [PubMed]
  79. Chen, Y.; Yu, G.; Long, Y.; Teng, J.; You, X.; Liao, B.-Q.; Lin, H. Application of radial basis function artificial neural network to quantify interfacial energies related to membrane fouling in a membrane bioreactor. Bioresour. Technol. 2019, 293, 122103. [Google Scholar] [CrossRef] [PubMed]
  80. Jawad, J.; Hawari, A.H.; Zaidi, S. Modeling of forward osmosis process using artificial neural networks (ANN) to predict the permeate flux. Desalination 2020, 484, 114427. [Google Scholar] [CrossRef]
  81. Jawad, J.; Hawari, A.H.; Zaidi, S.J. Modeling and sensitivity analysis of the forward osmosis process to predict membrane flux using a novel combination of neural network and response surface methodology techniques. Membranes 2021, 11, 70. [Google Scholar] [CrossRef]
  82. Jin, Z.; Lv, C.; Zhao, M.; Zhang, Y.; Huang, X.; Bei, K.; Kong, H.; Zheng, X. Black water collected from the septic tank treated with a living machine system: HRT effect and microbial community structure. Chemosphere 2018, 210, 745–752. [Google Scholar] [CrossRef]
  83. Zaghloul, M.S.; Iorhemen, O.T.; Hamza, R.A.; Tay, J.H.; Achari, G. Development of an ensemble of machine learning algorithms to model aerobic granular sludge reactors. Water Res. 2021, 189, 116657. [Google Scholar] [CrossRef]
  84. Ren, N.-q.; Yan, X.-f.; Chen, Z.-b.; Hu, D.-x.; Gong, M.-l.; Guo, W.-q. Feasibility and simulation model of a pilot scale membrane bioreactor for wastewater treatment and reuse from Chinese traditional medicine. J. Environ. Sci. 2007, 19, 129–134. [Google Scholar] [CrossRef] [PubMed]
  85. Cai, Y.; Ben, T.; Zaidi, A.A.; Shi, Y.; Zhang, K.; Lin, A.; Liu, C. Effect of pH on pollutants removal of ship sewage treatment in an innovative aerobic-anaerobic micro-sludge MBR system. Water Air Soil Poll. 2019, 230, 163. [Google Scholar] [CrossRef]
  86. Cai, Y.; Li, X.; Zaidi, A.A.; Shi, Y.; Zhang, K.; Sun, P.; Lu, Z. Processing efficiency, simulation and enzyme activities analysis of an air-lift multilevel circulation membrane bioreactor (AMCMBR) on marine domestic sewage treatment. Period. Polytech. Chem. Eng. 2019, 63, 448–458. [Google Scholar] [CrossRef]
  87. Schmitt, F.; Banu, R.; Yeom, I.-T.; Do, K.-U. Development of artificial neural networks to predict membrane fouling in an anoxic-aerobic membrane bioreactor treating domestic wastewater. Biochem. Eng. J. 2018, 133, 47–58. [Google Scholar] [CrossRef]
  88. Heydari, B.; Sharghi, E.A.; Rafiee, S.; Mohtasebi, S.S. Use of artificial neural network and adaptive neuro-fuzzy inference system for prediction of biogas production from spearmint essential oil wastewater treatment in up-flow anaerobic sludge blanket reactor. Fuel 2021, 306, 121734. [Google Scholar] [CrossRef]
  89. Almomani, F. Prediction of biogas production from chemically treated co-digested agricultural waste using artificial neural network. Fuel 2020, 280, 118573. [Google Scholar] [CrossRef]
  90. Neto, J.G.; Ozorio, L.V.; de Abreu, T.C.C.; Dos Santos, B.F.; Pradelle, F. Modeling of biogas production from food, fruits and vegetables wastes using artificial neural network (ANN). Fuel 2021, 285, 119081. [Google Scholar] [CrossRef]
  91. Hazrati, H.; Moghaddam, A.H.; Rostamizadeh, M. The influence of hydraulic retention time on cake layer specifications in the membrane bioreactor: Experimental and artificial neural network modeling. J. Environ. Chem. Eng. 2017, 5, 3005–3013. [Google Scholar] [CrossRef]
  92. Maere, T.; Villez, K.; Marsili-Libelli, S.; Naessens, W.; Nopens, I. Membrane bioreactor fouling behaviour assessment through principal component analysis and fuzzy clustering. Water Res. 2012, 46, 6132–6142. [Google Scholar] [CrossRef]
  93. Wang, Z.; Wu, X. Mathematical and artificial neural network models to predict the membrane fouling behavior of an intermittently-aerated membrane bioreactor under sub-critical flux. CLEAN–Soil Air Water 2015, 43, 1002–1009. [Google Scholar] [CrossRef]
  94. Han, H.; Zhang, S.; Qiao, J.; Wang, X. An intelligent detecting system for permeability prediction of MBR. Water Sci. Technol. 2018, 77, 467–478. [Google Scholar] [CrossRef] [PubMed]
  95. Mirbagheri, S.A.; Bagheri, M.; Bagheri, Z.; Kamarkhani, A.M. Evaluation and prediction of membrane fouling in a submerged membrane bioreactor with simultaneous upward and downward aeration using artificial neural network-genetic algorithm. Process Saf. Environ. 2015, 96, 111–124. [Google Scholar] [CrossRef]
  96. Shi, S.; Xu, G. Novel performance prediction model of a biofilm system treating domestic wastewater based on stacked denoising auto-encoders deep learning network. Chem. Eng. J. 2018, 347, 280–290. [Google Scholar] [CrossRef]
  97. Huang, B.; Wang, H.-C.; Cui, D.; Zhang, B.; Chen, Z.-B.; Wang, A.-J. Treatment of pharmaceutical wastewater containing β-lactams antibiotics by a pilot-scale anaerobic membrane bioreactor (AnMBR). Chem. Eng. J. 2018, 341, 238–247. [Google Scholar] [CrossRef]
  98. Picos-Benítez, A.R.; López-Hincapié, J.D.; Chávez-Ramírez, A.U.; Rodríguez-García, A. Artificial intelligence based model for optimization of COD removal efficiency of an up-flow anaerobic sludge blanket reactor in the saline wastewater treatment. Water Sci. Technol. 2017, 75, 1351–1361. [Google Scholar] [CrossRef]
  99. Bein, E.; Zucker, I.; Drewes, J.E.; Huebner, U. Ozone membrane contactors for water and wastewater treatment: A critical review on materials selection, mass transfer and process design. Chem. Eng. J. 2021, 413, 127393. [Google Scholar] [CrossRef]
  100. Nayak, M.; Dhanarajan, G.; Dineshkumar, R.; Sen, R. Artificial intelligence driven process optimization for cleaner production of biomass with co-valorization of wastewater and flue gas in an algal biorefinery. J. Clean. Prod. 2018, 201, 1092–1100. [Google Scholar] [CrossRef]
  101. 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. 2020, 143, 36–44. [Google Scholar] [CrossRef]
  102. Aghilesh, K.; Kumar, A.; Agarwal, S.; Garg, M.C.; Joshi, H. Use of artificial intelligence for optimizing biosorption of textile wastewater using agricultural waste. Environ. Technol. 2023, 44, 22–34. [Google Scholar] [CrossRef]
  103. Yonkos, L.T.; Friedel, E.A.; Perez-Reyes, A.C.; Ghosal, S.; Arthur, C.D. Microplastics in four estuarine rivers in the Chesapeake Bay, USA. Environ. Sci. Technol. 2014, 48, 14195–14202. [Google Scholar] [CrossRef]
  104. Lee, H.-J.; Song, N.-S.; Kim, J.-S.; Kim, S.-K. Variation and uncertainty of microplastics in commercial table salts: Critical review and validation. J. Hazard. Mater. 2021, 402, 123743. [Google Scholar] [CrossRef] [PubMed]
  105. Linker, R.; Shmulevich, I.; Kenny, A.; Shaviv, A. Soil identification and chemometrics for direct determination of nitrate in soils using FTIR-ATR mid-infrared spectroscopy. Chemosphere 2005, 61, 652–658. [Google Scholar] [CrossRef] [PubMed]
  106. Anıl, I.; Golcuk, K.; Karaca, F. ATR-FTIR spectroscopic study of functional groups in aerosols: The contribution of a Saharan dust transport to urban atmosphere in Istanbul, Turkey. Water Air Soil Pollut. 2014, 225, 1898. [Google Scholar] [CrossRef]
  107. Enders, A.A.; North, N.M.; Fensore, C.M.; Velez-Alvarez, J.; Allen, H.C. Functional group identification for FTIR spectra using image-based machine learning models. Anal. Chem. 2021, 93, 9711–9718. [Google Scholar] [CrossRef] [PubMed]
  108. Nourani, V.; Elkiran, G.; Abba, S.I. Wastewater treatment plant performance analysis using artificial intelligence – an ensemble approach. Water Sci. Technol. 2018, 78, 2064–2076. [Google Scholar] [CrossRef]
  109. 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 2022, 14, 1053. [Google Scholar] [CrossRef]
  110. El-Rawy, M.; Abd-Ellah, M.K.; Fathi, H.; Ahmed, A.K.A. Forecasting effluent and performance of wastewater treatment plant using different machine learning techniques. J. Water Process Eng. 2021, 44, 102380. [Google Scholar] [CrossRef]
  111. Bisaria, K.; Singh, R.; Gupta, M.; Mathur, A.; Dixit, A. Novel acoustic-activated alkali-functionalized Trapa bispinosa peel biochar for green immobilization of chlorpyrifos from wastewater: Artificial intelligence modelling and experimental validation. Biomass Convers. Bior. 2022, 1–20. [Google Scholar] [CrossRef]
  112. Yang, Y.; Kim, K.-R.; Kou, R.; Li, Y.; Fu, J.; Zhao, L.; Liu, H. Prediction of effluent quality in a wastewater treatment plant by dynamic neural network modeling. Process Saf. Environ. 2022, 158, 515–524. [Google Scholar] [CrossRef]
  113. Nnaji, P.C.; Anadebe, V.C.; Onukwuli, O.D.; Okoye, C.C.; Ude, C.J. Multifactor optimization for treatment of textile wastewater using complex salt–Luffa cylindrica seed extract (CS-LCSE) as coagulant: Response surface methodology (RSM) and artificial intelligence algorithm (ANN–ANFIS). Chem. Pap. 2022, 76, 2125–2144. [Google Scholar] [CrossRef]
  114. Mahmoud, M.S.; Mahmoud, A.S. Wastewater treatment using nano bimetallic iron/copper, adsorption isotherm, kinetic studies, and artificial intelligence neural networks. Emergent Mater. 2021, 4, 1455–1463. [Google Scholar] [CrossRef]
  115. Chan, H.; Nai-He, Y. A pretreatment method of wastewater based on artificial intelligence and fuzzy neural network system. J. Intell. Fuzzy Syst. 2020, 39, 1711–1720. [Google Scholar] [CrossRef]
  116. Şenol, H. Methane yield prediction of ultrasonic pretreated sewage sludge by means of an artificial neural network. Energy 2021, 215, 119173. [Google Scholar] [CrossRef]
  117. Song, C.; Yao, L.; Hua, C.; Ni, Q. Comprehensive water quality evaluation based on kernel extreme learning machine optimized with the sparrow search algorithm in Luoyang River Basin, China. Environ. Earth Sci. 2021, 80, 521. [Google Scholar] [CrossRef]
  118. Yan, J.; Xu, Z.; Yu, Y.; Xu, H.; Gao, K. Application of a hybrid optimized BP network model to estimate water quality parameters of Beihai Lake in Beijing. Appl. Sci. 2019, 9, 1863. [Google Scholar] [CrossRef]
  119. Allawi, M.F.; Jaafar, O.; Mohamad Hamzah, F.; Abdullah, S.M.S.; El-Shafie, A. Review on applications of artificial intelligence methods for dam and reservoir-hydro-environment models. Environ. Sci. Pollut. Res. 2018, 25, 13446–13469. [Google Scholar] [CrossRef]
  120. Alavi, J.; Ewees, A.A.; Ansari, S.; Shahid, S.; Yaseen, Z.M. A new insight for real-time wastewater quality prediction using hybridized kernel-based extreme learning machines with advanced optimization algorithms. Environ. Sci. Pollut. Res. 2022, 29, 20496–20516. [Google Scholar] [CrossRef]
  121. Deepnarain, N.; Nasr, M.; Kumari, S.; Stenström, T.A.; Reddy, P.; Pillay, K.; Bux, F. Artificial intelligence and multivariate statistics for comprehensive assessment of filamentous bacteria in wastewater treatment plants experiencing sludge bulking. Environ. Technol. Inno. 2020, 19, 100853. [Google Scholar] [CrossRef]
  122. Cai, Y.; Zaidi, A.A.; Shi, Y.; Zhang, K.; Li, X.; Xiao, S.; Lin, A. Influence of salinity on the biological treatment of domestic ship sewage using an air-lift multilevel circulation membrane reactor. Environ. Sci. Pollut. Res. 2019, 26, 37026–37036. [Google Scholar] [CrossRef]
  123. Cai, Y.; Li, X.; Zaidi, A.A.; Shi, Y.; Zhang, K.; Feng, R.; Lin, A.; Liu, C. Effect of hydraulic retention time on pollutants removal from real ship sewage treatment via a pilot-scale air-lift multilevel circulation membrane bioreactor. Chemosphere 2019, 236, 124338. [Google Scholar] [CrossRef]
  124. Pendashteh, A.R.; Fakhru’l-Razi, A.; Chaibakhsh, N.; Abdullah, L.C.; Madaeni, S.S.; Abidin, Z.Z. Modeling of membrane bioreactor treating hypersaline oily wastewater by artificial neural network. J. Hazard. Mater. 2011, 192, 568–575. [Google Scholar] [CrossRef] [PubMed]
  125. Nourani, V.; Asghari, P.; Sharghi, E. Artificial intelligence based ensemble modeling of wastewater treatment plant using jittered data. J. Clean. Prod. 2021, 291, 125772. [Google Scholar] [CrossRef]
  126. 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, 3783–3791. [Google Scholar] [CrossRef]
  127. Karam, A.; Zaher, K.; Mahmoud, A.S. Comparative Studies of Using Nano Zerovalent Iron, Activated Carbon, and Green Synthesized Nano Zerovalent Iron for Textile Wastewater Color Removal Using Artificial Intelligence, Regression Analysis, Adsorption Isotherm, and Kinetic Studies. Air Soil Water Res. 2020, 13, 1178622120908273. [Google Scholar] [CrossRef]
  128. Guo, Q.; Wang, Y.; Dai, C.; Wang, L.; Liu, H.; Li, J.; Tiwari, P.K.; Zhao, M. Dynamics of a stochastic nutrient–plankton model with regime switching. Ecol. Model. 2023, 477, 110249. [Google Scholar] [CrossRef]
  129. 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, 2816–2826. [Google Scholar] [CrossRef]
Figure 1. Number of publications related to artificial intelligence in wastewater. Number of publications was obtained by using Elsevier’s Scopus database with queries TITLE-ABS-KEY (terms). Inset: network visualization of research topics related to artificial intelligence in wastewater, generated by VOSviewer based on keywords.
Figure 1. Number of publications related to artificial intelligence in wastewater. Number of publications was obtained by using Elsevier’s Scopus database with queries TITLE-ABS-KEY (terms). Inset: network visualization of research topics related to artificial intelligence in wastewater, generated by VOSviewer based on keywords.
Sustainability 15 13557 g001
Figure 2. Classification of AI models for WWT.
Figure 2. Classification of AI models for WWT.
Sustainability 15 13557 g002
Figure 3. A basic architecture of ANN models. (a) ANN; (b) RNN; (c) CNN; (d) FNN; (e) DNN.
Figure 3. A basic architecture of ANN models. (a) ANN; (b) RNN; (c) CNN; (d) FNN; (e) DNN.
Sustainability 15 13557 g003
Figure 4. A schematic diagram of ML models. (a) PCA; (b) DT; (c) SVM; (d) PSO.
Figure 4. A schematic diagram of ML models. (a) PCA; (b) DT; (c) SVM; (d) PSO.
Sustainability 15 13557 g004
Figure 5. A schematic diagram of ML models. (a) RF; (b) SOM; (c) KNN; (d) ANFIS.
Figure 5. A schematic diagram of ML models. (a) RF; (b) SOM; (c) KNN; (d) ANFIS.
Sustainability 15 13557 g005
Figure 6. A schematic diagram of SA models. (a) GA; (b) GP.
Figure 6. A schematic diagram of SA models. (a) GA; (b) GP.
Sustainability 15 13557 g006
Figure 7. Flow chart for the applications of AI models in WWT.
Figure 7. Flow chart for the applications of AI models in WWT.
Sustainability 15 13557 g007
Table 1. Commonly used AI models for WWT, their purposes, advantages and disadvantages.
Table 1. Commonly used AI models for WWT, their purposes, advantages and disadvantages.
AI ModelsPurposesAdvantagesDisadvantagesRef.
RNNRegressionSuitable for time series modelingComputationally expensive[35]
ClassificationNo limit to the length of inputsTraining difficulty
CNNRegressionSuitable for image-related modelingComputationally expensive[36]
ClassificationExtracting important features of imagesTraining difficulty
FNNRegressionEasy to implement and interpretComputationally expensive[37]
ClassificationSuitable for complex nonlinear problemsComplex model architecture
DNNRegressionAccurate and fast predictionComputationally expensive[36]
ClassificationSuitable for complex nonlinear problemsTraining difficulty
Prediction Easy to overfit
PCAClusteringSimple and easy to implementMay lose some important information[38]
Reduces dimensionalitySensitive to noise data
DTRegressionEasy to understand, interpret and classify    Low training efficiency-
ClassificationNo need to preprocessNot suitable for imbalanced datasets
SVMRegressionCan handle high-dimensional problemsComputationally expensive[37]
ClassificationSuitable for complex separable datasetsNot suitable for larger datasets
PSORegressionSimple and easy to useSensitive to initial conditions[39]
ClassificationHigh computational efficiencyNot suitable for discrete problems
ClusteringStrong universality
RFRegressionSimple and easy to useNeed dense decision trees to guarantee[36]
ClassificationSuitable for high-dimensional datasetsaccuracy and robustness
PredictionStrong generalizationComputationally expensive
KNNRegressionSimple and easy to useComputationally expensive[36]
ClassificationSuitable for nonlinear classificationHigh memory consumption
SOMClusteringSuitable for high-dimensional datasetsHigh computational complexity[40]
Reduces dimensionNot suitable for missing datasets
ANFISRegressionCombine the advantages of ANN and FISComputationally expensive[37]
ClassificationUse determination and fuzzy dataHard to define appropriate
Prediction membership function
GARegressionSuitable for complex nonlinear problemsDifficult to train[39]
ClassificationSupport multi-objective optimizationPoor local search ability
OptimizationEfficient and flexibleNot suitable for high dimensions
GPRegressionSuitable for complex optimization problemsMany control variables-
ClassificationOptimize by automatic searchConverge slowly
Optimization Not suitable for high dimensions           
Table 2. Commonly used activation function of AI models for WWT. The blue symbol “*” represents the frequency of use in the literature.
Table 2. Commonly used activation function of AI models for WWT. The blue symbol “*” represents the frequency of use in the literature.
Activation FunctionExpressionOutput RangeRef.
Sigmoid *** f ( x ) = 1 1 + e x ( 0 , 1 ) [41]
Tanh *** f ( x ) = tanh ( x ) ( 1 , 1 ) [42]
ReLU ** f ( x ) = max ( 0 , x ) [ 0 , + ) [36]
Leaky ReLU * f ( x ) = max ( ξ x , x ) ( , + ) [43]
ELU ** f ( x ) = a ( e x 1 ) , x < 0 x , x 0 ( a , + ) [44]
Heaviside * f ( x ) = 0 , x < T 1 , x T [ 0 , 1 ] [45,46]
Ramp * f ( x ) = 0 , x < T 1 x T 1 T 2 T 1 , T 1 x T 2 1 , x T 2 [ 0 , 1 ] [47]
Linear * f ( x ) = x ( , + ) [48]
Table 4. Applications of AI models for the optimization of process parameters.
Table 4. Applications of AI models for the optimization of process parameters.
AI Models UsedInput VariablesOutput VariablesRemarksPollution TypeRef.
ANN, GALight intensity,BiomassNew ANN-GA model toGreen[100]
photoperiod,productivitypredict optimal processmicroalga
temperature and conditions of an algal
initial pH biorefinery. Productivity
improved by 57%.
ANN, GA, PSOInitial MBDecontaminationANN-PSO model to predictMB[39]
concentration,efficiencythe optimum process conditions.
temperature, pH rGO/Fe/Co nanohybrids can treat
and contact time organic contaminants effectively.
SOMCOD, BOD,Organic overload,SOM model to optimize workingBiological[40]
TSS, TN,working conditionsconditions. Obtain key parametersnutrient
TP, etc. and working conditions of
biological nutrient removal.
ANN, GAElectrolysis time,DiscolorationANN-GA model to optimizeDye[101]
flow, currentefficiencyprocess conditions. AI can
density, pH, dye design, control and
concentration operate EO process.
ANN, ANFISTemperature, pH,RemovalAI model to predict biosorptionMB[102]
bio-sorbent andefficiency of MBand obtain optimum conditions.
dye concentration    Agricultural waste for effective
biosorption of textile wastewater.
CNN, LSTM, AMInfluent COD,Effluent COD andNew hybrid CLSTMA model toPaper[43]
SS, flux, DO, pHSSmonitor water quality for cleanerwastewater
and temperature production with low cost.
SVR, LSTM, GRU    Inflow andOutflow CODIntelligent WWT system basedCity[41]
outflow COD on ML and sensors. Applied itsewage
temperature to a fine chemical plant.
DNN, LSTMDO, NO 3 N , N 2 O concentration Integrating mechanistic and N 2 O [42]
NH 4 + N , influent DL models is very useful for
and air flow rates, understanding N 2 O emission
temperature dynamics.
ANN, CNN, LSTMMixed liquor,CarbohydrateUsed 5 AI models to forecastCarbohydrate[36]
KNN, RFbiomass, greencontentbiomass production. CNN-1D           
algae, etc. model performs the best.
ANN, SVM, FNNInfluent waterEffluentNew FFNN-LSSVM model toNH3-N[37]
quality, flow rate,BOD/NH3-Nforecast water quality andnitrogen
etc. optimize process parameters.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Wang, Y.; Cheng, Y.; Liu, H.; Guo, Q.; Dai, C.; Zhao, M.; Liu, D. A Review on Applications of Artificial Intelligence in Wastewater Treatment. Sustainability 2023, 15, 13557.

AMA Style

Wang Y, Cheng Y, Liu H, Guo Q, Dai C, Zhao M, Liu D. A Review on Applications of Artificial Intelligence in Wastewater Treatment. Sustainability. 2023; 15(18):13557.

Chicago/Turabian Style

Wang, Yi, Yuhan Cheng, He Liu, Qing Guo, Chuanjun Dai, Min Zhao, and Dezhao Liu. 2023. "A Review on Applications of Artificial Intelligence in Wastewater Treatment" Sustainability 15, no. 18: 13557.

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop