Next Article in Journal
Challenges Associated with Membrane Separation of Polypeptides and Relevant Solution Strategies
Previous Article in Journal
Exploring the Influence of Extraction Methods, Solvents, and Temperature on Total Phenolic Recovery and Antioxidant Capacity in Olive Leaf Extracts: A Systematic Review with Quantitative Synthesis
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Review

Artificial Intelligence in Chemical Dosing for Wastewater Purification and Treatment: Current Trends and Future Perspectives

1
Sichuan Development Guorun Water Investment (SCD-GRWI) Co., Ltd., Chengdu 610047, China
2
MARA Key Laboratory of Development and Application of Rural Renewable Energy, Sichuan Institute of Rural Human Settlements, Biogas Institute of Ministry of Agriculture and Rural Affairs (BIOMA), Chengdu 610041, China
3
Research Center for Rural Energy and Ecology, Chinese Agricultural Academy of Sciences, Chengdu 610041, China
4
Zizhong Guorun Drainage Co., Ltd., Neijiang 641200, China
*
Authors to whom correspondence should be addressed.
These authors contributed equally to this work.
Separations 2025, 12(9), 237; https://doi.org/10.3390/separations12090237
Submission received: 15 July 2025 / Revised: 28 August 2025 / Accepted: 1 September 2025 / Published: 3 September 2025

Abstract

Recent concerns regarding artificial intelligent (AI) technologies have spurred studies into improving wastewater treatment efficiency and identifying low-carbon processes. Treating one cubic meter of wastewater necessarily consumes a certain amount of chemicals and energy. Approximately 20% of the total chemical consumption is attributed to phosphorus and nitrogen removal, with the exact proportion varying based on treatment quality and facility size. To promote sustainability in wastewater treatment plants (WWTPs), there has been a shift from traditional control systems to AI-based strategies. Research in this area has demonstrated notable improvements in wastewater treatment efficiency. This review provides an extensive overview of the literature published over the past decades, aiming to advance the ongoing discourse on enhancing both the efficiency and sustainability of chemical dosing systems in WWTPs. It focuses on AI-based approaches utilizing algorithms such as neural networks and fuzzy logic. The review encompasses AI-based wastewater treatment processes: parameter analysis/forecasting, model development, and process optimization. Moreover, it summarizes six promising areas of AI-based chemical dosing, including acid–base regents, coagulants/flocculants, disinfectants/disinfection by-products (DBPs) management, external carbon sources, phosphorus removal regents, and adsorbents. Finally, the study concludes that significant challenges remain in deploying AI models beyond simulated environments to real-world applications.

1. Introduction

Water is essential for all life activities on the Earth. However, water pollution is increasing progressively due to rapid industrial and technological development, unregulated urbanization, and rising demand for water resources driven by population growth [1,2,3]. In the past decade, despite China’s GDP nearly doubling, total water consumption has remained stable with “zero growth”. Similarly, total wastewater discharge has not increased alongside rapid economic development, as illustrated in Figure 1. This phenomenon can be attributed to ongoing efforts of Chinese government to conserve water, improve resource utilization, develop water/wastewater infrastructure, and ensure water security [4]. From 2015 to 2023, WW discharge increased by approximately 20.0 billion tons per year, reaching over 66.0 billion tons in 2023. The treated amount of WW has been increasing year by year, rising from 42.9 billion tons in 2015 to 65.2 billion tons in 2023, an increase of 62.5%. The average annual growth rate usually remains at approximately 1%. Moreover, as WWTPs discharge standards become increasingly stringent, the demand for effective nitrogen and phosphorus removal continues to rise. Although most WWTPs employ biological processes for nutrient removal, achieving compliance solely through these methods has become more challenging due to their susceptibility to various operational and environmental factors. Therefore, chemical dosing has become an essential strategy for controlling nutrient concentrations in WWTPs across China.
Currently, chemical dosing is an indispensable component of wastewater purification and treatment (WPT) to ensure effluent quality in WWTPs [3]. The main addition processes are recognized: pre-dosing ahead of treatment tank; simultaneous dosing in aeration tank effluent or sludge-returning pipelines; and post-dosing after secondary sedimentation tank [5,6,7]. These methods involve chemical, physical, and biological processes, including oxidation, sedimentation, filtration, electrochemical treatment, and coagulation/flocculation, to remove organic compounds, soluble pollutants, suspended solids, and nutrients from wastewaters [8]. The commonly used chemicals include: pH and alkalinity adjustment (e.g., NaOH, Ca(OH)2, HCl, H2SO4) [9,10], phosphorus precipitant (e.g., FeCl3, FeSO4, polyaluminum chloride (PAC), alum, lime) [5,7,11,12], flocculation/coagulation aids (e.g., anionic, cationic and non-ionic polyacrylamide (PAM), activated silica) [13,14], disinfectants (e.g., Cl2, NaOCl, NH2Cl, ClO2) [6,15], and external carbon sources (e.g., methanol, ethanol, sodium acetate, molasses) [16]. However, most methods still rely on the traditional “fixed dosing”, which is set according to experience in “kilograms per ton of water” and with an added safety factor, resulting in a 15% to 40% waste of the regents [17]. This further leads to excessive accumulation of salt ions, increases the amount of sludge and the cost of subsequent dewatering chemicals, and poses challenges to the stable operation of WWTPs.
In these circumstances, to ensure efficient treatment of WW economically and sustainably, researchers have introduced artificial intelligence (AI) technologies into WWTPs. With the help, the dosing process can be achieved precisely and efficiently, which is usually accomplished by combining online water quality sensor technology with intelligent algorithms to monitor and analyze pollutant concentration in real time. Then the systems automatically adjust the dosage of chemicals, thereby achieving improved treatment performances with less regent cost [5,7]. Furthermore, AI techniques are also capable of evaluating and predicting pollutant fluctuations and optimizing purification and treatment processes during WPT [3,17,18]. This can reduce manual intervention, enhance operational stability, and thus better protect the environment and conserve resources. The introduction of AI dosing control technology is of great significance for improving pollution control effect and operational efficiency of WWTPs [19].
AI methods are increasingly applied in water/wastewater treatment related fields, especially in recent years (Figure 2). The involved technologies refer to algorithms that perform tasks and make inferences by mimicking human intelligence. These methods specifically include three aspects such as model paradigms (like artificial neural networks (ANNs), decision trees (DT), etc.), optimization strategies (such as genetic algorithms (GA), particle swarm optimization (PSO), etc.) and dimensionality reduction (like principal component analysis (PCA), etc.). Among them, ANNs include deep neural networks (DNNs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), and fuzzy neural networks (FNNs) and adaptive network-based fuzzy inference system (ANFIS). DT includes K-nearest neighbor (KNN) and support vector machine (SVM). In terms of optimization strategies, it encompasses swarm intelligence optimization methods such as particle swarm optimization (PSO), genetic algorithm (GA), and genetic programming (GP), as well as hybrid application methods (embedding GA/PSO into parametric models like ANN, SVM, and FNN to complete feature selection and parameter tuning). In addition, in terms of dimensionality reduction processing, the PCA method is mainly used to perform pre-noise reduction processing on CNNs, RNNs or SVM, and can also be combined with KNN to improve computational efficiency. Hybrid AI models including ANN-GA, ANN-PSO, MLP-ANN, PSO-GA, and SVR-GA have also been applied in WPT processes [2,17]. Numerous studies have shown that AI techniques are effective in modeling and optimizing WPT processes for pollutants removal [20,21,22,23]. AI is a powerful tool for predicting nonlinear data relationships, enabling accurate, data-driven forecasts in WWTPs. Compared to traditional methods, AI models deliver more precise results. They also require fewer input parameters and reduce sampling costs. Furthermore, AI models have been used to assess how operating conditions and variables affect WWTP management [24].
This review provides an overview of the literature on the application of AI models, with a special focus on chemical dosing systems in WPT process. Rather than offering a comprehensive survey of all AI models or smart technologies used in WTP process, this review focuses on summarizing key findings from significant published studies and analyzing emerging trends in the future development of AI. The applications discussed are primarily related to modeling, prediction, and optimization of WPT processes, including water quality monitoring, laboratory-scale investigations, and chemical dosing procedures. Additionally, this review explores the techno-economic benefits of AI integration, cost estimation methodologies, and future prospects of AI implementation in WWTPs.

2. Research Trends on AI-Based Chemical Dosing in WWTPs

Bibliometric analysis offers a comprehensive overview and classification of AI and/or ML research in WWTP, helping to identify potential research directions. By applying this method, it enables a systematic evaluation of relevant studies, thereby effectively revealing research hotspots in AI/ML applications. A total of 2424 publications were categorized into four groups and are presented in Figure 2. The bibliometric analysis revealed that WPT+ML was the most prevalent topic, accounting for 90% (2184 articles) of all publications. In contrast, WPT+AI and WPT+AI+Dosing/dosages accounted for 4.1% (104 articles) and 2.8% (70 articles) of the records, respectively. Other categories, such as WPT+ML+Dosing/dosage, WPT+AI, were the least common, comprising less than 3.1% of the total records. This report focuses exclusively on major reviews and research articles, which constituted the vast majority (97.4%, 2361 articles) of published works. Given this distribution, our study concentrated specially on AI/ML applications in WWTPs. Table 1 summarizes commonly used AI algorithms in WWTPs. Traditional computer simulations involve researchers encoding physical laws and analytical methods for computers to follow. In contrast, AI takes inspiration from human infants, who acquire the ability to recognize objects without explicit instruction. AI systems identify patterns in data and automatically generate rules, rather than depending on predefined programming. Through such methods, computers can learn the characteristics of “objects”—such as numerical values, text, sound, even or images—by training models like ANNs, and accurately recognize them in new contexts. ANNs differ in structure and algorithm, including CNNs, DNNs, FNNs, and RNNs. Among single models, ANNs and PCA models are the most widely used approaches in the WPT+ Dosing/dosage field, whereas the ANN-GA model is more commonly employed in hybrid modeling. Compared to the period of 2015–2020, applications of ML-based dosing, ANN-based dosing and PCA-based dosing increased by 14.6-fold, 1.1-fold and 0.63-fold, respectively, during 2021–2025. Furthermore, the application of the hybrid ANN-GA model in dosing has seen significant research advancements, with three additional studies published in the past five years.

3. AI Models Used in WWTPs

Recent advances in AI have expanded its use in the WPT field. Researchers have applied several AI methods to analysis and predict influent/effluent quality and operation needs, supporting the development of AI-based WWTPs. These innovations help improve operational efficiency and decision-making in WWTPs. The most commonly utilized AI models for WWTPs in the literature are presented in Figure 3. These models, which are employed for prediction and optimization purposes, can be broadly categorized into three main types: deep learning (DL), machine learning (ML), and evolutionary algorithm (EA). The hierarchical relationships among these models are illustrated in Figure 4. As known, AI applications are essential through big data processing, which is defined by its large volume and high-speed input, often faster than the processing capacity [19,25]. Their applications involve technologies for storing, processing, and managing complex datasets that exceed the capabilities of traditional methods. In manufacturing, it can be linked to the internet of things (IoT)—a network of connected devices that collect, store, and share real-time data. In the WPT section, IoT devices gather data from households, distribution systems, and WWTPs [26]. Moreover, industrial automation is a major IoT application, improving process accuracy through online control and reducing human involvement. Smart sensors are now used in IoT-connected WPT process to achieve precision at online management. In the systems, smart pumps with IoT connectivity can reduce energy use in urban networks compared to traditional methods [27]. For ML, it enables computers to automatically learn from and analyze big data and make predictions about real-world events. ML algorithms such as DT and SVM are capable of processing large datasets generated by IoT devices, which could identify patterns, make predictions and turn raw data into useful insights. This helps optimize WPT systems, predict problems, and respond to changing environments. The core concept of ML is using algorithms that improve performance through learning from data [7]. Moreover, DL, a subfield of ML, utilizes hierarchical neural networks to simulate human brain functions. Indeed, most contemporary AI applications are built upon some form of DL [13].

3.1. Deep Learning (DL) Models

DL has emerged as a powerful tool in wastewater treatment due to its ability to handle complex, nonlinear relationships and large datasets. DL models are used for tasks like predicting water quality parameters, optimizing treatment processes, and controlling energy consumption. Common DL architectures include ANNs, CNNs, DNNs, FNNs, and RNNs, as shown in Figure 5. Moreover, DL models are being applied to classification tasks, such as analyzing activated sludge flocs in microscopy images, to gain insights into treatment performance. Despite the potential, DL faces challenges such as the need for large amounts of high-quality data and computational resources. Current research is focused on developing more sophisticated models and techniques to enhance the efficiency and sustainability of WPT processes. The following section briefly introduces these DL models.

3.1.1. Artificial Neural Networks (ANNs)

ANNs, inspired by biological neural networks, use interconnected nodes to process information by adjusting weights (Figure 5). These networks such as CNNs and DNNs typically consist of an input layer, an output layer, and one or more hidden layers, employing activation functions like sigmoid, tanh, and ReLU for complex nonlinear computations [20,24,28]. As more hidden layers are added, ANNs can build more complex models, increasing their expressive power. This allows ANNs to learn complex input-output relationships through loss function design and optimization. However, ANNs also face some challenges such as overfitting and the need for extensive data and computational resources. In detail, their performance would be limited by the quality and quantity of training data. The interpretability of ANNs would become difficult to understand the decision-making process. And the computational cost of training deep ANNs could be prohibitive, especially for real-time applications [29].

3.1.2. Deep Neural Networks (DNNs)

DNNs are a type of ANN with multiple hidden layers that enables the learning of hierarchical data representations (Figure 5c). These architectures allow DNNs to model complex nonlinear relationships and capture intricate patterns in large datasets. However, their deep structures necessitate substantial training data and computational resources, making the training process both intensive and time-consuming. Despite these challenges, DNN has shown significant promise in WPT applications. For example, DNNs have been used to optimize the control of WWTPs, balancing effluent quality, and operational cost [30]. Specifically, a study proposed a hybrid machine learning model that includes DNNs to predict coagulant dosing in drilling wastewater treatment, demonstrating high prediction accuracy and efficiency [31]. Additionally, another study utilized DNNs combined with other techniques to optimize the dosing of chemicals for phosphorus removal, achieving a 20–25% increase in flocculation efficiency compared to benchmark methods [32]. These applications highlight the potential of DNNs to enhance the performance and sustainability of WPT processes.

3.1.3. Convolutional Neural Networks (CNNs)

CNNs, a type of feedforward neural network, employ convolutional operations and features a deep hierarchical structure (Figure 5b). They have strong representation learning capabilities and have been extensively applied in various domains. CNNs extracts high-level features from input images using convolutional layers, reduces feature dimensions via pooling layers [33]. In WPT process, CNNs have been applied to optimize chemical dosing processes. For instance, a hybrid CNN-LSTM model has been used to predict effluent quality in WWTPs, which can help in optimizing chemical dosing to ensure the treated water meets the required standards [34]. Another study used a CNN-based approach to predict the optimal dosing of coagulants in WWTPs, achieving high prediction accuracy and efficiency [35]. Additionally, CNNs have been employed to monitor and control the dosing of chemicals for phosphorus removal, resulting in a significant improvement in treatment efficiency [36]. These results demonstrate the potential of CNNs to enhance the performance and sustainability of WPT processes, particularly in chemical dosing. However, despite its success, CNNs face challenges such as limited datasets and computational complexity.

3.1.4. Recurrent Neural Networks (RNNs)

RNNs are a class of neural networks with feedback connections that process sequence data by performing recursive computations across the temporal dimension of the input (Figure 5e). They offer advantages such as memory retention, parameter sharing, and tuning completeness, which make them effective at capturing nonlinear patterns in time series problems. The most widely used variant of RNNs is the long short-term memory (LSTM) network, which mitigates the gradient vanishing problem inherent in traditional RNNs through the incorporation of specialized gating mechanisms [8,33]. However, RNNs still face challenges such as vanishing and exploding gradients, which can hinder the learning of long-term dependencies. Moreover, it is computationally intensive and often requires careful hyperparameter tuning.
During the WPT process, RNNs—especially LSTMs—have been employed to optimize chemical dosing. For example, a recent study applied RNNs to predict optimal coagulant dosing in wastewater treatment plants, achieving high accuracy and operational efficiency [31]. Another study used RNNs to monitor and control nutrient removal dosing, outperforming conventional strategies [35]. These examples show the potential of RNNs to improve efficiency and sustainability in wastewater treatment, particularly in chemical dosing. Still, their computational complexity and training challenges must be addressed for effective real-time control applications.

3.1.5. Fuzzy Neural Networks (FNNs)

FNNs integrate fuzzy logic with ANNs to effectively handle uncertain or ambiguous problems. They use fuzzy logic to process input data, defining fuzzy relationships between inputs, outputs, and neurons through elements such as membership functions, fuzzy inference, and normalization. Then, FNNs are employed for training and generating outputs. They perform better than traditional neural networks in complex tasks and are widely used in pattern recognition, control systems, and predictive analysis. However, FNNs may not capture temporal dependencies as effectively as RNNs. They have also been used to optimize chemical dosing processes in WPT process. For instance, a recent study utilized FNNs to predict optimal coagulant dosing based on historical data, enhancing treatment efficiency and reducing costs [3]. Another study employed FNNs to monitor and control the dosing of chemicals for nutrient removal, achieving better performance compared to traditional mechanical models [37]. These applications highlight the ability of FNNs to improve the performance and sustainability of WPT processes.

3.2. Machine Learning (ML) Models

ML is a subfield of AI that focuses on developing algorithms and statistical models to enable computer systems to learn from data without explicit programming. The primary objective of ML is to construct predictive models that can make accurate predictions or decisions based on patterns learned from data [7,38,39]. Commonly used ML models include decision trees (DT), support vector machine (SVM), random forests (RF), K-nearest neighbor (KNN), principal component analysis (PCA), and particle swarm optimization (PSO). Figure 3 illustrates schematic diagrams of these ML models.

3.2.1. Decision Trees (DTs)

DTs are common ML algorithms used for both classification and regression tasks. They partition datasets into subsets based on feature values, forming a hierarchical tree structure with root, intermediate, and leaf nodes. Each node represents a feature, branches represent feature values, and leaf nodes indicate the final outcomes [40]. DTs are easy to understand, handle missing data and outliers well, and support feature selection. However, they are prone to overfitting, especially in high-dimensional datasets. To mitigate this, techniques like pruning are often employed. Compared to models like ANNs, DTs are simpler and more interpretable. Although ANNs can capture more complex patterns, they require larger datasets and greater computational resources. In WPT, DTs have been applied to optimize dosing processes. A previous study used hybrid DTs-based algorithm to predict optimal coagulant dosage using historical data, thereby improving treatment efficiency and reducing operational costs [41]. Figure 3 shows a typical DT architecture with its branches, root, intermediate, and leaf nodes.

3.2.2. K-Nearest Neighbor (KNN)

KNN is a simple and widely used ML algorithm for classification and regression tasks, predicting the label of a new data point based on the labels of its K nearest neighbors in the training dataset. It can model complex nonlinear relationships, adaptively adjust model complexity, and work with various data types through different distance metrics. However, it can be sensitive to noise and high-dimensional data and can be computationally intensive for large datasets. For wastewater treatment, KNN has been used to optimize chemical dosing systems by predicting optimal dosing amounts based on historical data, eventually improving treatment performance [41]. However, similar challenges such as high dimensionality and computational cost may also limit its direct application in real-world scenarios. Figure 3 shows an example of KNN before and after classifying a new data point (solid red circle).

3.2.3. Principal Component Analysis (PCA)

PCA using multivariate statistical ML algorithm reduces data dimensionality by transforming high-dimensional data into a lower-dimensional form while preserving most of the original information through orthogonal transformation [42]. It identifies principal components based on variable variances, which are mutually orthogonal. PCA is commonly applied in data clustering, image processing, natural language processing, and noise filtering. However, it is sensitive to outliers and assumes linear relationships between variables, which may not hold in real-world scenarios. Studies show that the application of PCA in environmental assessment can be problematic if not properly applied, as it may lead to misinterpretation of data due to its reliance on numerical analysis rather than direct value judgments. Despite these challenges, PCA remains a valuable tool for data reduction and pattern recognition in environmental and chemical engineering applications.

3.2.4. Particle Swarm Optimization (PSO)

PSO is also widely adopted for optimizing algorithms to solve complex problems. It simulates the swarm behavior observed in biological populations, such as flocks of birds or schools of fish, by iteratively adjusting each particle’s position and velocity to coverage toward optimal solutions. During each iteration, particles update their positions by combining their own best-known location (personal best) with the best-known position of the entire population (global best) [43,44]. Due to its reliance on swarm intelligence, PSO is capable of handling nonlinear and multimodal problems while effectively avoiding premature convergence to local optima. However, the algorithm is sensitive to initial parameter settings, which may necessitate multiple independent runs to ensure solution reliability. A schematic of PSO finding the global optimum is shown in Figure 3.

3.2.5. Support Vector Machine (SVM)

SVM uses kernel functions—such as polynomial and radial basis function kernels—to project input data into a higher-dimensional feature space, where it identifies the optimal hyperplane that separates different classes by maximizing the margin from the nearest support vectors [3]. SVM is particularly effective in handling high-dimensional and nonlinear datasets, and it avoids falling into local optima, a common issue in other optimization methods. However, its computational complexity can lead to slower performance on large-scale datasets, and it remains sensitive to noise and outliers. Moreover, the choice of kernel functions and hyperparameters significantly affects model performance and thus requires careful tuning. As illustrated in Figure 3, the SVM classification boundary (solid line) represents the optimal hyperplane, while the two parallel dashed lines delineate the margin boundaries that define the maximum separation between classes.

3.3. Evolutionary Algorithm (EA)

Besides ANNs and ML, evolutionary algorithms (EAs) are also utilized AI models, including genetic algorithm (GA), genetic programming (GP), and evolution strategies (ES) as shown in Figure 3. The following section presents a brief overview of GA, GP and ES.

3.3.1. Genetic Algorithm (GA)

GA is a search and optimization method inspired by the principles of biological evolution. By simulating natural genetic processes—such as selection, crossover, and mutation—it generates solutions that are better adapted to the problem environment. As illustrated Figure 3, GA is capable of exploring high-dimensional spaces without being confined to local optima, thereby providing robust global search capabilities and adaptability in addressing complex nonlinear problems [45]. Due to those characteristics, GA is extensively applied in fields such as ML, AI, control systems, optimization design.

3.3.2. Genetic Programming (GP)

GP, based on GA, automatically generates and selects computer programs inspired by biological evolution. It is commonly used for regression, classification, and optimization tasks. Unlike other ML algorithms, GP begins with a randomly generated initial population and evolves over generations using GA [46]. A schematic of GP is provided in Figure 3. Due to strong adaptability and generalization capabilities, GP is particularly suitable for solving highly nonlinear problems or those lacking explicit analytical solutions, such as image recognition and prediction.

3.3.3. Evolution Strategies (ES)

ES, also inspired by biological evolution, are primarily based on the principles of natural selection and genetic inheritance. ES have been successfully applied to address multi-objective control problems in the WPT process, enabling simultaneous improvements in energy efficiency and effluent quality. Furthermore, this algorithm has been integrated with advanced techniques such as adaptive differential evolution, facilitating more intelligent and dynamic optimization in the WPT operations [47].
Table 2 provides an overview of how AI/ML techniques are being applied across various aspects of WPT processes, including influent monitoring, energy consumption, nitrogen removal, operation control, sludge prediction, and effluent monitoring. Models such as ANNs, CNNs, MLP, SVM, and LSTMs are utilized to monitor the characteristics of incoming wastewater. They analyze a wide range of input parameters, including flow rate, temperature, pH, turbidity, and various indicators (e.g., organic loading rate (OLR), hydraulic retention time (HRT), total suspended solids (TSS), mixed liquor suspended solid (MLSS), COD, ammonium (AN), total phosphorus (TP), and total nitrogen (TN)), to predict and manage influent quality and process outcomes effectively. These techniques not only enhance operational efficiency but also promote environmental sustainability. As AI/ML technologies continue to evolve, their role in WPT is expected to expand, fostering innovation and improving overall process performance.
Table 1. An overview of common AI algorithm classes used in WWTPs.
Table 1. An overview of common AI algorithm classes used in WWTPs.
ModelsFull NameAdvantagesDisadvantagesWWTP ApplicationsRef.
ANNsArtificial Neural NetworksANNs can capture complex nonlinear relationships, continuously learn and adapt through training, and process data in parallel for efficient handling of large real-time datasets. ANNs are robust to noise and missing data.ANNs’ model architecture is complex, and the training and optimization process demands significant computational resources. ANNs lack interpretability.Predict effluent quality indicators such as TSS, BOD, COD, and energy consumption in WWTPs. Forecast treatment efficiency under varying conditions. Use process data monitoring and analysis to detect potential equipment or process issues in advance.[20,24]
CNNsConvolutional Neural NetworksTo learn features from images automatically and be ideal for image processing by focusing on local patterns. To reduce parameters and improves computational efficiency.Model and architecture of CNNs are inherently complex and multifaceted, requiring substantial computing power for their implementation. Moreover, they are computationally expensive and challenging to learn.Image recognition is used to identify equipment status and sludge morphology, and to monitor water quality changes during WPT process.[33,48]
DNNsDeep Neural NetworksHigh expressive power enables the model to learn complex patterns in data effectively. Automatic feature extraction reduces reliance on manual feature engineering.High model complexity makes it prone to overfitting. DNNs require high computational resources and strong hardware support.DNNs model complex relationships in wastewater treatment, predicts efficiency and energy consumption, and optimizes operational parameters.[49,50]
DTsDecision TreesEasy to understand, interpret and classify, with no requirement for prior processing.Low training efficiency, overfitting on complex datasets, and ineffective training outcomes due to sensitivity to noise data or outliers.Classifying process compliance and predicting contaminant removal efficiency, identifying equipment failures through decision-making rules.[40,51]
FNNsFuzzy Neural NetworksTo be capable of processing fuzzy and uncertain data, while effectively integrating expert. knowledge into the model through the application of fuzzy rules.FNNs are highly complex, with a complicated structure and training process. Optimizing both the neural network and fuzzy system parameters simultaneously makes training particularly challenging.FNNs are applicable to handling fuzzy and uncertain data and used for modeling the nonlinear relationships in the wastewater treatment process.[21,52]
PCAPrincipal Component AnalysisReduces dimensionality, is simple and straightforward to use.Loss of some crucial information and sensitivity to noise in the data. The features after dimensionality reduction are difficult to interpret.Used to reduce the dimension of sewage treatment data. Remove the noise components in the data.[22,42]
PSOParticle Swarm OptimizationStrong universality, high computational efficiency, as well as simplicity and ease of use.Sensitive to initial conditions and prone to discrete defects.Parameters for optimizing the sewage treatment model. Optimizing the fault diagnosis model.[43,44]
RFRandom ForestRelatively stable and resistant to noise and outliers. Handles continuous and categorical variables, even with missing or incomplete data. Simple to use and ideal for high-dimensional datasets.Decision tree density affects accuracy and robustness. Higher density increases model complexity, training time, and computational requirements. It is computationally expensive and requires deep trees to ensure correctness and robustness.Modeling DO in simple and hybrid systems and the removal efficiency in the adsorption process.[23,53,54]
RNNs/LSTMRecurrent Neural Networks/Long Short-Term MemoryRNNs/LSTM are capable of capturing context dependencies within sequences and effectively performing context modeling. With strong sequence processing capabilities, they are well-suited for handling inputs or outputs of variable lengths.When sequences are too long, gradients at early time steps may vanish or explode, causing training difficulties. Additionally, since each step depends on the previous one, parallelization is limited, leading to high computational costs.RNNs/LSTM are used to predict time series data in the wastewater treatment process, such as water quality changes. Potential equipment failures are identified through time series data.[55]
SVM/SVRSupport Vector Machine/Support Vector RegressionSVM/SVR demonstrates strong performance on small datasets and is capable of effectively handling both linear and nonlinear classification tasks. Moreover, it achieves favorable generalization performance even when trained with limited samples.High computational cost, long training time, and high memory requirements. Moreover, SVM/SVR is not suitable for multi-classification problems and needs to be extended to multi-class SVM.SVM/SVR is used to classify the water quality status in the sewage treatment process. Predict the treatment efficiency, energy consumption and other indicators.[3]
Table 2. AI application for treatment operation and predication in WWTPs.
Table 2. AI application for treatment operation and predication in WWTPs.
Process Application ObjectsAI/ML TechniquesInput/InfluentOutput/EffluentScaleRef.
WWTPInfluent monitoringLDA, MLP, SVMQ, Temp, pH, turbidity, SAC, conductivityCOD, ANFull-scale[19]
WWTPEnergy consumptionDNNsQ, QR, TempEnergy consumptionFull-scale [25]
Anammox + biocharNitrogen removalANNsHRT, NLR, AN, NO2-N TN removalLab-scale[28]
WWTPInfluent monitoringCNNs, LSTMQ, pH, COD, AN, TP, TNCOD, AN, TN, TPFull-scale[36]
WWTPOperation control and predicationANNs, ANFIS, AVG, WAVG, SVRQ, pH, Temp, OLR, HRT, TSS, MLSS, MLVSS, SVI, COD, BOD5, TKN, AN, TN, TPQ, Temp, pH, OLR, HRT, TSS, MLSS, MLVSS, SVI, COD, BOD5, AN, TN, TKN, TPLab-scale[56]
WWTPSludge predicationFCNNs, DT, KNN, KRR, LR, RE, SVR, XGBoostQ, Temp, pH, rainfall, OLR, HRT, TSS, MLSS, MLVSS, SVI, COD, BOD5, AN, TKN, TN, TP Sludge productionFull-scale[57]
WWTPSludge bulkingGPRQ, Temp, DO, COD, MLSS, SVISludge bulkingFull-scale[58]
WWTPEffluent monitoringLM-ANNQ, QR, ANANFull-scale[59]
WWTPOperation controlBP, PSO, Gaussian classificationpH, COD, TS, ANTN, AN, COD, Air flow rateFull-scale[60]
WWTPN2O emissionsADABoost, DNNs, DT, KNN, RF, XGBoostQ, Temp, DO, TSS, AN, NO2-N, NO3-NN2O emissionsFull-scale[61]
A/OCarbon and nitrogen removalRSMQ, Temp, pH, TOC, AN, TNTSS, TOC, AN, TNLab-scale[62]
Anammox + PNNitrogen removalBP-ANNQ, Temp, pH, ANCOD, ANLab-scale[63]

4. AI Techniques for Chemical Dosing in WWTPs

AI methods have been increasingly applied in the fields of water and wastewater treatment. This section outlines six commonly used external chemicals in WWTPs. Combining advanced chemical process with AI technologies can reduce the efforts required for data collection (Table 3), while the availability of more comprehensive datasets can enhance the accuracy of AI models [20]. Although this overview reflects current trends in the WPT field, it does not encompass all existing AI-related research. AI has been successfully employed to optimize adsorption, disinfection, and nutrients removal processes; moreover, ML models have been proven effective in predicting disinfection by-products (DBPs) concentrations, model adsorption and coagulation—sedimentation processes, and controlling chlorination operations [5]. The performance of these models is typically evaluated using standard statistical metrics, including mean absolute error (MAE), mean square error (MSE), correlation coefficient (R2), relative error (RE), and root mean square error (RMSE).

4.1. Acid–Base Regents

pH adjustment plays a critical role in WWTPs, particularly in maintaining treatment performance within a specified range. Many treatment processes need precise pH control and measurement to meet specific product standards, such as those involved in precipitation, biological nutrient removal, sludge treatment, and gas removal. Commonly used reagents for pH adjustment include hydrochloric acid (HCl) and sodium hydroxide (NaOH), with the choice depending on the specific treatment requirements. Due to its nonlinear behavior and the involvement of multiple parameters, pH prediction is a challenging task that requires effective modeling tools. Given the time-dependent and stochastic nature of these systems, new approaches are necessary to ensure accurate and stable pH predictions [9]. Recently, AI methods, particularly ML models, have been employed to predict pH based on historical data. Researchers have frequently incorporated pH as an input variable to enhance model performance across various applications (Table 2 and Table 3). For instance, an ANFIS achieved 96.2% accuracy in predicting the water quality index and 100% accuracy in classification [64]. Fuzzy systems like ANFIS have also demonstrated effectiveness in modeling complex systems, including those with time delays. Similarly, a hybrid model combining ANFIS and SVM achieve a high R2 value of 0.97 for predicting treatment efficiency [65]. SVMs are particularly effective in capturing complex relationships among multiple variables. In another study, ANNs integrated with GAs were used to optimize conditions that prevent color loss of bromophenol blue in wastewater [66]. However, due to the nonlinear and complex nature of pH dynamics, it is often used as an input rather than a target output in the models. AI techniques, therefore, are struggling with high-dimensional datasets.

4.2. Coagulants and Flocculants

Coagulation and sedimentation processes utilize chemicals such as coagulants and flocculants to enhance solid–liquid separation. Coagulants primarily target the removal of suspended solids (SS) and colloidal particles by inducing particle aggregation and facilitating settling through mechanisms including double layer compression, charge neutralization, and entrapment. Flocculants, on the other hand, promote the formation of larger, settleable flocs via adsorption bridging and surface adsorption, thereby improving separation efficiency. These chemicals are widely employed in WWTPs. Common coagulants, such as iron and aluminum salts, destabilize particles by compressing the double electric layer and neutralizing surface charges, which enhances particle aggregation and sedimentation. PAM, a commonly used flocculant, facilitates floc formation through adsorption bridging and surface interactions, thus improving settling and filtration performance. In practice, these two types of chemicals are often used together to maximize treatment efficiency. Recently, AI models have been introduced to overcome the limitations of traditional methods by accurately predicting organic pollutant levels (i.e., COD and BOD5) during coagulation. This advancement supports more informed decision-making, optimizes coagulation operations, and enhances overall system performance. For automatic dosing prediction, AI enables water treatment systems to calculate optimal dosages in real time based on current conditions and multiple variables [14]. This leads to increased automation, reduced manual intervention, improved dosing accuracy, and faster adaptation to variations in water quality.

4.3. Disinfectants and DBPs Management

Disinfection aims to inactivate or eliminate microorganisms and viruses, primarily through the use of chlorine-based disinfectants. While chlorination is highly effective, it also poses potential risks to human health. In addition to its potential for acute toxicity, chlorine can react with naturally occurring bromide and organic matter in WPT systems, leading to the formation of DBPs. These compounds are suspected to be carcinogenic and may cause adverse effects on reproductive systems, prompting increased regulatory scrutiny worldwide. However, the complete mechanism underlying DBP formation in water systems remains poorly understood, making their prediction and control a promising area for the application of AI technologies. AI offers significant potential for predicting, minimizing, and managing DBP formation through data-driven approaches [6]. Researchers have successfully predicted DBP levels in distribution systems and at consumer taps using key parameters—such as water temperature (Temp), pH, contact time, chlorine concentration, and total organic carbon (TOC)—as model inputs [67]. ANNs are the most widely used AI models for predicting chlorination outcomes and DBP formation [37,38], although other techniques like SVM and GAs have also been explored [39]. Overall, ANNs remain the most validated method for modeling chlorination and DBP formation.
For DBP management, ANNs have proven to be a powerful tool in WPT process. Their ability to model complex nonlinear relationships makes them highly effective for predicting DBPs, such as trihalomethanes (THMs) and haloacetic acids (HAAs). Khan et al. developed ANN models to predict THMs formation using nine parameters like dissolved organic carbon (DOC), UV254, chlorine dose, and pipe length, achieving high accuracy [68]. Compared to traditional linear models, ANNs better capture complex variable interactions, improving prediction accuracy. However, they may require more data and computing resources. In contrast, SVMs and RF can be more computationally efficient in some cases but may not handle complex relationships as well. Hong et al. used a hybrid method combining radial basis function ANNs with gray relational analysis to predict THM levels, showing that integrating ANNs with complementary techniques can enhance predictive performance [69]. In practical applications, ANNs have been used to optimize disinfection processes, effectively reducing DBP formation while ensuring proper disinfection. This approach improves water quality and supports the sustainability of WPT processes. However, challenges remain, especially regarding data quality and availability. Future research should focus on advancing on-site sensors and integrating ANNs with emerging technologies like IoTs and DL to build more intelligent systems [70]. This integration could further optimize chemical dosing, reduce DBP risks, and improve public health outcomes.

4.4. External Carbon Sources

Nitrogen discharge standards are becoming increasingly stringent to mitigate the risks of eutrophication [46]. To meet the requirement of lower effluent total nitrogen (TN) levels, WWTPs often supplement carbon sources during the denitrification process. Traditional dosing strategies typically rely on either analyzing carbon consumption mechanisms or exploring environmentally friendly and cost-effective alternatives. However, these methods often demand high technical expertise, limiting their rapid implementation in WWTPs. With the advancement of AI, various algorithms have been widely adopted, reducing dependence on expert knowledge and offering innovative solutions for carbon source dosing. Chen et al. applied a LSTM model to successfully predict carbon source dosage, achieving strong predictive performance [71]. However, due to the high computational complexity of LSTM, its application becomes time-consuming and resource-intensive when dealing with multiple parameters in the complex denitrification process. Zhou et al. proposed a high-precision BPNN model for the anoxic-aerobic (A2O) process, accurately predicting ammonia nitrogen (AN) removal efficiency with a fitting error of only 1.0% [72]. Sensitivity analysis showed that chemical oxygen demand (COD) had the most significant impact on the model, while AN and TP had less influence. However, low sensitivity to certain features may lead the model to neglect variations in those features, potentially resulting in prediction inaccuracies. Although ML models are effective in extracting features and addressing high-dimensional nonlinear problems, they suffer from poor interpretability, high data demands, substantial computational requirements, and length development cycles. In AI-based carbon source dosing systems, equipment is often incomplete, faulty, or poorly maintained, and operations are typically rough, resulting in limited availability of high-quality data. As a result, few studies have applied DL techniques to intelligent carbon source dosing.

4.5. Phosphorus Removal Regents (PRAs)

Phosphorus removal agents (PRAs) are chemical substances that promote phosphorus removal from wastewater through mechanisms such as precipitation, adsorption, or biological processes. These agents can be broadly classified into several categories, including iron salts, aluminum salts, calcium salts, composite agents, and specialized PRAs. The selection of an appropriate agent should consider various factors such as the characteristics of wastewater, the forms of phosphorus present, and the associated treatment costs. The main PRAs are as follows: (1) Iron-based PRAs include polymeric ferric sulfate (PFS), FeCl3, and FeSO4. These agents offer advantages such as rapid sedimentation, additional coagulation effects on organic and insoluble phosphorus, a wide applicable pH range (6.0–9.0), and a phosphorus removal rate exceeding 90%. However, they may cause effluent discoloration and exhibit strong corrosiveness to equipment. (2) Aluminum-based PRAs primarily consist of Al2(SO4)3 and PAC. They are cost-effective and provide good flocculation performance, making them suitable for concurrent phosphorus removal in primary sedimentation. However, residual aluminum ions can pose ecological risks, leading to their gradual replacement by iron salts. (3) Calcium-based PRAs include Ca(OH)2 and CaCl2. Their advantages are low cost, suitability for wastewater requiring pH adjustment, and a phosphorus removal rate above 90% under high pH conditions (10.5–11.0). Limitations include excessive sludge generation, susceptibility to pipe scaling, and restricted application scenarios. (4) Composite and specialized PRAs, such as polyaluminum ferric chloride (PAFC) and subphosphorus removal agents, integrate the properties of metal ions. These agents are adaptable to complex water quality and demonstrate high specificity for refractory phosphorus forms. ML models were employed to predict effluent phosphorus levels using nine years of data collected from a small WWTPs. Pearson correlation analysis was conducted to identify key input features among 42 variables and to reveal their underlying relationships [11]. Moreover, an LSTM model was utilized to forecast phosphorus removal in the form of phosphate through an activated sludge process, achieving an R2 value of approximately 0.7–0.8 [73]. A dynamic Benchmark Simulation Model (BSM2)-based framework was used to automate iron dosing control for phosphorus removal in a membrane biological reactor (MBR) system [12]. Various AI algorithms were applied by researchers to analyze nutrient removal and critical wastewater quality parameters, with results indicating that LSTM models incorporating a 2-day look-back period attained the highest prediction accuracy of 77% [74]. Although this study proposed a robust method for predicting effluent quality, it also emphasized the complexities involved in understanding and modeling nutrient removal dynamics within WWTPs.

4.6. Adsorbents

Adsorption process is widely recognized as an effective physicochemical method for removing contaminants during WPT process. Adsorption is an exothermic process in which target molecules, referred to as adsorbates, transfer from a liquid or gaseous phase to the surface of a solid material, commonly known as an adsorbent. Due to the complex interactions that influence the efficiency of adsorption process, determining key operational parameters and predicting removal performance can be challenging. Simplifying the complexity can help extend the service life of sorptive media and improve the overall effectiveness and regulatory compliance confidence of the WPT process. To optimize the process, researchers have developed AI-based models capable of accurately predicting adsorption behavior. These AI tools hold significant potential for supporting operational decision-making in adsorption-based treatment systems. Previous studies have used inputs such as pH, water temperature, adsorbent dose, contact time, and initial adsorbate concentration in ML modeling of adsorption processes in water streams contaminated with metals, industrial dyes, and organic compounds [75]. Some studies have focused on models that predict the adsorbate removal percentage, or adsorption efficiency [2]. In contrast, other research efforts have aimed at predicting adsorption capacity, dimensionless effluent concentrations, and the relative impact of input water quality parameters. These findings suggest that AI has the potential to enhance adsorption efficiency of real-world WPT systems.
Table 3. AI application for the control of chemical dosing during the WPT processes.
Table 3. AI application for the control of chemical dosing during the WPT processes.
Target ParameterExternal ChemicalsModelsInputOutputWPT ScaleRef.
PO43−PACFNNsCOD, BOD5, SS, AN, TN, TPCOD, BOD5, SS, AN, TN, TPFull-sacle[5]
Fecal coliformNaClOLSTMQ, COD, AN, NaClO dosing historyMPN, available chlorineLab-sacle[6]
TPAlumLightGBM, SGD, SVC, MLPQ, Temp, DO, MLSS, BOD5, SS, TP, TNSS, TP, dosageFull-sacle[7]
pHLimeKNN, XGBoostQ, pH, Temp, BOD, TSS, TKN, lime additionpHFull-sacle[9]
pHNaOHDNNs, LSTMpH, HRT, gas flowpHLab-sacle[10]
PhosphateAlumSVM, DT, RF, ANNs, LSTMQ, SS, Cl2, AN, BOD5, PPFull-sacle[11]
TPFeSO4, FeCl3BSM2Q, pH, DO, TSS, NO3-N, TPTOC, TP, BOD7, TSS, TDS, VAF, AN, NO3-NPilot-sacle[12]
TurbidityPACGAMTF, RF, LSTMQ, pH, Temp, alkalinity, conductivity, turbidity, HRTCoagulant dose, turbidityFull-sacle[13]
TurbidityCoagulantENN, RFQ, pH, Temp, turbidity, DO, organic carbonDO, organic carbon, turbidityFull-sacle[14]
TNCarbon sourceXGBoostQ, pH, COD, TP, TN, DO, AN, SSAN, COD, SS, pH, TPFull-sacle[16]
TNMLP, SVM, RFQ, COD, BOD, TKN, AN, TSSTNFull-sacle[38]
TMPNaClOLSTMTMP, AN, Permeability, fluxTMP, ANFull-sacle[39]
TKNANFIS, SVMpH, Temp, TS, COD, FA, AN, TKNTKNFull-sacle[65]
TNSodium acetateBPNNQ, Temp, DO, COD, TN, AN, total aeration, HRTTNFull-sacle[72]
PhosphateMetal saltsLSTMQ, pH, Temp, DO, SSPhosphate removal efficiencyFull-sacle[73]
PhosphorusNot addedXGBoost, LSTMTP, TSS, AN, MLSS, MLVSS, NO3-NTP_EffFull-sacle[74]
PhosphateNot addedANFIS, ANN, SVRpH, HRT, TP, electrode type, current intensityPhosphate removal efficiencyLab-sacle[76]
Cu removalNa2S, FeSO4GA, PSOCu_InfCu_EffLab-sacle[77]
Note: Flow rate (Q), Hydraulic retention time (HRT), Multilayer perceptron (MLP), Genetic algorithm (GA), Particle swarm optimization (PSO), Transmembrane pressure (TMP), Free ammonia (FA), Ammonia nitrogen (AN), Kjeldahl nitrogen (TKN), Most probable number (MPN).

5. Fault Detection, Diagnosis, and Prognosis in Chemicals Dosing Systems

AI methods provide significant opportunities for fault diagnosis, fault management, and decision-making in the operation of WWTPs operations (Figure 6). To maintain optimal performance, many WWTPs currently employ online sensors to collect data and detect issues. Some systems are even automated to adjust process parameters in real time, either with or without human supervision. Presently, Supervisory Control and Data Acquisition (SCADA) systems—comprising sensors and actuators, programmable logic controllers (PLCs), human–machine interfaces (HMIs), supervisory computers, and communication networks—enable remote operation and monitoring. These systems often eliminate the need for on-site personnel by allowing prompt adjustments to operational parameters [78]. SCADA systems have demonstrated effectiveness in improving the operational efficiency of WWTPs, reducing energy consumption, and assessing design improvements (Figure 6). Operators can utilize SCADA systems to enhance daily monitoring and control capabilities. Further advancements in these systems can be achieved through the integration of AI techniques. Consequently, fault detection can be automated by combining SCADA systems and AI algorithms. By identifying abnormal data patterns using LSTM, WWTPs can respond promptly to maintain system performance. However, many studies rely on ex-post analysis and lack real-time monitoring. They are often limited to simulations or pilot-scale plants, without proving effectiveness across different WWTPs [78]. In contrast, a recent study evaluated several AI models—SVM, MP, RF, LightGBM, and XGBoost—using unknown data from multiple plants. Gradient boosting methods enabled the detection of 96% of anomalies, with correct classification rates reaching 84% and 62%, respectively [79].
IoT communication technology can overcome the limitations of SCADA systems—such as high costs, limited scalability, and maintenance demands—by enabling data transfer from physical infrastructure to intelligent devices. This advancement facilitates the evolution of traditional SCADA systems into more sophisticated and efficient smart WWTP management frameworks [80]. Meanwhile, wireless sensor networks are widely adopted in wastewater monitoring due to their cost-effectiveness and efficiency. However, many of these networks lack clear decision-making and control mechanisms based on sensor data, even though microcontroller- and GSM modem-based systems have been developed for water quality monitoring. These systems often rely on third-party cloud platforms and lack alternative data routing when Wi-Fi is unavailable, thereby compromising the reliability of SCADA systems [81]. Communication among system entities is crucial in such frameworks. To enhance performance, these systems have evolved from isolated setups into interconnected networks. However, this evolution introduces a significant risk: increased vulnerability to cyberattacks through communication channels. As a result, physical infrastructures in WWTPs face heightened threats that can disrupt operations and endanger public health and the environment. AI techniques can detect and respond to ongoing cyberattacks in real time [82]. A new IoT architecture should be introduced to address challenges such as unstable underground internet connections, power supply in WWTPs, and the need for real-time control of all devices. Such IoT systems would autonomously monitor and sample wastewaters in WWTPs, incorporating specialized hardware, real-time anomaly detection algorithms, and methods for managing harmful discharges and chemical dosage.

6. Challenges and Future Perspectives

Although AI technologies offer significant advancements in WPT processes (Figure 6), they also present certain concerns, challenges, and potential drawbacks that warrant careful consideration. For instance, AI systems typically require substantial initial investments in technology, expertise, training, and infrastructure. While these costs can be quantified, a less apparent yet critical issue is the cost associated with false positives. In full-scale WWTPs, AI-generated false positives may lead to violations of treatment and discharge regulations, potentially causing environmental consequences that extend beyond immediate compliance issues. As AI becomes increasingly widespread, concerns arise regarding workforce adaptability and potential job displacement. Overreliance on AI systems may erode human expertise and diminish decision-making capabilities, ultimately reducing preparedness for unforeseen events. Furthermore, data quality significantly influences the effectiveness and accuracy of AI strategies. Poor data quality, caused by sensor malfunctions, equipment failures, or process variability, can reduce efficiency. Process variability may also lead to concept drift, rendering AI models inaccurate when applied to new data. To ensure long-term success, AI strategies must address sensor failures and incorporate regular model retraining, which may increase operational costs. Additionally, studies have been indicated that AI tools may lack transparency and introduce biases into decision-making processes. AI systems exhibit bias due to data collection practices and preprocessing techniques used during model development. Finally, implementing AI strategies requires a thorough review of data and network security policies. These principles should guide AI development and deployment standards of AI systems to ensure their reliability and ethical use.

7. Conclusions

This review summarizes commonly used AI models and their applications in WPT process, including common methods, laboratory research, and chemical dosing systems. AI models are increasingly adopted in WPT due to their robust learning and predictive capabilities. They have been effectively utilized for system modeling, parameter optimization, performance prediction, and contamination detection. However, challenges such as high data requirements, inadequate data management, limited interpretability, poor reproducibility, and a lack of physical meaning and standardization still hinder broader adoption. To address these issues, interdisciplinary collaboration among mathematicians, biologists, engineers, and computer scientists is essential to advance models and technology development. Furthermore, additional lab- and field-scale studies are necessary to understand system behaviors under diverse conditions and identify key underlying mechanisms. Hybrid AI models that integrate multiple techniques, particularly attention-based models, present promising solutions. The integration of data-driven and knowledge-driven approaches represents an emerging and valuable direction in WPT research.

Author Contributions

Conceptualization, J.J.; methodology, M.L.; formal analysis, M.L.; investigation, J.L., Q.T. and D.F.; resources, Q.T., D.F. and Y.D.; writing—original draft preparation, J.J.; writing—review and editing, J.J., M.L., B.C., X.W., L.Y., Y.W., X.X., L.W., J.L., Y.L. and N.Y.; supervision, Y.D., Y.L. and N.Y.; project administration, B.C. and X.W.; funding acquisition, Y.D., Y.L. and N.Y. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the Technology Innovation Project of SCD-GRWI (2024-KJCX-01), the National Key Research and Development Program of China (2024AYFD1600205), the Sichuan Provincial Natural Science Foundation of China (2024NSFSC0382), the Rongpiao Plan of Chengdu city, China, and the Agricultural Science and Technology Innovation Project of Chinese Academy of Agricultural Sciences (CAAS-ASTIP-2021-BIOMA-04).

Data Availability Statement

No new data were created or analyzed in this study.

Acknowledgments

During the preparation of this work, the authors used Kimi-k1.5 and Youdao-AIBox for study design, data collection, interpretation and language editing. The authors have reviewed and edited the output and take full responsibility for the content of this publication.

Conflicts of Interest

Author Yibo Du was employed by the company Sichuan Development Guorun Water Investment (SCD-GRWI) Co., Ltd. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

References

  1. Duarte, M.S.; Martins, G.; Oliveira, P.; Fernandes, B.; Ferreira, E.C.; Alves, M.M.; Lopes, F.; Pereira, M.A.; Novais, P. A review of computational modeling in wastewater treatment processes. ACS EST Water 2024, 4, 784–804. [Google Scholar] [CrossRef]
  2. 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. [Google Scholar] [CrossRef]
  3. Alprol, A.E.; Mansour, A.T.; Ibrahim, M.E.E.-D.; Ashour, M. Artificial intelligence technologies revolutionizing wastewater treatment: Current trends and future prospective. Water 2024, 16, 314. [Google Scholar] [CrossRef]
  4. Department of Urban Social and Economic Survey, National Bureau of Statistics. 9 National Urban Drainage and Wastewater Treatment in Past Years (1978–2023). In China Urban Statistical Yearbook 2023; China Statistics Press: Beijing, China, 2024. [Google Scholar]
  5. Lu, X.; Huang, S.; Liu, H.; Yang, F.; Zhang, T.; Wan, X. Research on intelligent chemical dosing system for phosphorus removal in wastewater treatment plants. Water 2024, 16, 1623. [Google Scholar] [CrossRef]
  6. Li, Q.; Cui, X.; Gao, X.; Chen, X.; Zhao, H. Intelligent dosing of sodium hypochlorite in municipal wastewater treatment plants: Experimental and modeling studies. J. Water Process Eng. 2024, 64, 105662. [Google Scholar] [CrossRef]
  7. Sun, J.; Xu, Y.; Yang, H.; Liu, J.; He, Z. Machine learning facilitated the conceptual design of an alum dosing system for phosphorus removal in a wastewater treatment plant. Chemosphere 2024, 351, 141154. [Google Scholar] [CrossRef]
  8. Sakkaravarthy, S.; Jano, N.A.; Vijayakumar, A. Overcoming challenges in traditional waste water treatment through AI-driven innovation. In The AI Cleanse: Transforming Wastewater Treatment Through Artificial Intelligence: Harnessing Data-Driven Solutions; Garg, M.C., Ed.; Springer Nature: Cham, Switzerland, 2024; pp. 53–81. [Google Scholar]
  9. Xu, Y.; Zeng, X.; Bernard, S.; He, Z. Data-driven prediction of neutralizer pH and valve position towards precise control of chemical dosage in a wastewater treatment plant. J. Clean. Prod. 2022, 348, 131360. [Google Scholar] [CrossRef]
  10. Panjapornpon, C.; Chinchalongporn, P.; Bardeeniz, S.; Jitapunkul, K.; Hussain, M.A.; Satjeenphong, T. Development of physics-guided neural network framework for acid-base treatment prediction using carbon dioxide-based tubular reactor. Eng. Appl. Artif. Intel. 2024, 138, 109500. [Google Scholar] [CrossRef]
  11. Xu, Y.; Wang, Z.; Nairat, S.; Zhou, J.; He, Z. Artificial intelligence-assisted prediction of effluent phosphorus in a full-scale wastewater treatment plant with missing phosphorus input and removal data. ACS EST Water 2024, 4, 880–889. [Google Scholar] [CrossRef]
  12. Kazadi Mbamba, C.; Lindblom, E.; Flores-Alsina, X.; Tait, S.; Anderson, S.; Saagi, R.; Batstone, D.J.; Gernaey, K.V.; Jeppsson, U. Plant-wide model-based analysis of iron dosage strategies for chemical phosphorus removal in wastewater treatment systems. Water Res. 2019, 155, 12–25. [Google Scholar] [CrossRef]
  13. Lin, S.; Kim, J.; Hua, C.; Park, M.-H.; Kang, S. Coagulant dosage determination using deep learning-based graph attention multivariate time series forecasting model. Water Res. 2023, 232, 119665. [Google Scholar] [CrossRef]
  14. Wang, D.; Chen, L.; Li, T.; Chang, X.; Ma, K.; You, W.; Tan, C. Successful prediction for coagulant dosage and effluent turbidity of a coagulation process in a drinking water treatment plant based on the Elman neural network and random forest models. Environ. Sci. Water Res. Technol. 2023, 9, 2263–2274. [Google Scholar] [CrossRef]
  15. Hossain, S.; Hewa, G.A.; Chow, C.W.K.; Cook, D. Development and comparison of water quality network model and data analytics model for monochloramine decay prediction. Water 2022, 14, 2021. [Google Scholar] [CrossRef]
  16. Yun, J.; Yu, Y.; Tao, C.; Zhai, M.; Zhang, H.; Chen, Y.; Li, H.; Zhang, B.; Ma, J. Machine learning-based optimization of enhanced nitrogen removal in a full-scale urban wastewater treatment plant with ecological combination ponds. Water Res. 2025, 285, 123976. [Google Scholar] [CrossRef]
  17. Capodaglio, A.G.; Callegari, A. Use, potential, needs, and limits of AI in wastewater treatment applications. Water 2025, 17, 170. [Google Scholar] [CrossRef]
  18. Nagpal, M.; Siddique, M.A.; Sharma, K.; Sharma, N.; Mittal, A. Optimizing wastewater treatment through artificial intelligence: Recent advances and future prospects. Water Sci. Technol. 2024, 90, 731–757. [Google Scholar] [CrossRef]
  19. Nguyen, X.C.; Nguyen, T.T.H.; Tran, Q.B.; Bui, X.T.; Ngo, H.H.; Nguyen, D.D. Artificial intelligence for wastewater treatment. In Current Developments in Biotechnology and Bioengineering; Bui, X.T., Nguyen, D.D., Nguyen, P.D., Ngo, H.H., Pandey, A., Eds.; Elsevier: Amsterdam, The Netherlands, 2022; Chapter 21; pp. 587–608. [Google Scholar]
  20. Mjalli, F.S.; Al-Asheh, S.; Alfadala, H.E. Use of artificial neural network black-box modeling for the prediction of wastewater treatment plants performance. J. Environ. Manag. 2007, 83, 329–338. [Google Scholar] [CrossRef]
  21. Honggui, H.; Ying, L.; Junfei, Q. A fuzzy neural network approach for online fault detection in waste water treatment process. Comput. Electr. Eng. 2014, 40, 2216–2226. [Google Scholar] [CrossRef]
  22. Liu, Y.; Pan, Y.; Sun, Z.; Huang, D. Statistical monitoring of wastewater treatment plants using variational bayesian PCA. Ind. Eng. Chem. Res. 2014, 53, 3272–3282. [Google Scholar] [CrossRef]
  23. Sun, J.; Guan, X.; Sun, X.; Cao, X.; Tan, Y.; Liao, J. Water quality prediction and carbon reduction mechanisms in wastewater treatment in Northwest cities using Random Forest Regression model. Sci. Rep. 2024, 14, 31525. [Google Scholar] [CrossRef]
  24. Jawad, J.; Hawari, A.H.; Javaid Zaidi, S. Artificial neural network modeling of wastewater treatment and desalination using membrane processes: A review. Chem. Eng. J. 2021, 419, 129540. [Google Scholar] [CrossRef]
  25. Oulebsir, R.; Lefkir, A.; Safri, A.; Bermad, A. Optimization of the energy consumption in activated sludge process using deep learning selective modeling. Biomass Bioenerg. 2020, 132, 105420. [Google Scholar] [CrossRef]
  26. Martínez, R.; Vela, N.; El Aatik, A.; Murray, E.; Roche, P.; Navarro, J.M. On the use of an IoT integrated system for water quality monitoring and management in wastewater treatment plants. Water 2020, 12, 1096. [Google Scholar] [CrossRef]
  27. Faisal, M.; Muttaqi, K.M.; Sutanto, D.; Al-Shetwi, A.Q.; Ker, P.J.; Hannan, M.A. Control technologies of wastewater treatment plants: The state-of-the-art, current challenges, and future directions. Renew. Sustain. Energ. Rev. 2023, 181, 113324. [Google Scholar] [CrossRef]
  28. Mojiri, A.; Ohashi, A.; Ozaki, N.; Aoi, Y.; Kindaichi, T. Integrated anammox-biochar in synthetic wastewater treatment: Performance and optimization by artificial neural network. J. Clean. Prod. 2020, 243, 118638. [Google Scholar] [CrossRef]
  29. Walling, E.; Vaneeckhaute, C. Developing successful environmental decision support systems: Challenges and best practices. J. Environ. Manag. 2020, 264, 110513. [Google Scholar] [CrossRef]
  30. Lu, H.; Meng, Z.H.; Zhang, B.; Song, S.; Zhan, S.Y.; Li, Y.; Wu, Q.L.; Wang, H.Z.; Guo, W.Q. Deep learning-based multiobjective optimization for balancing effluent quality, operational cost, and greenhouse gas emissions in wastewater treatment plant control. ACS EST Water 2024, 4, 2564–2577. [Google Scholar] [CrossRef]
  31. Kalhormohammadi, M.; Khoramipour, S. Predictive modeling of coagulant dosing in drilling wastewater treatment using artificial neural networks. Sci. Rep. 2025, 15, 30003. [Google Scholar] [CrossRef]
  32. Randive, P.; Bhagat, M.S.; Bhorkar, M.P.; Bhagat, R.M.; Vinchurkar, S.M.; Shelare, S.; Sharma, S.; Beemkumar, N.; Hemalatha, S.; Kumar, P.; et al. Adaptive optimization of natural coagulants using hybrid machine learning approach for sustainable water treatment. Sci. Rep. 2025, 15, 16096. [Google Scholar] [CrossRef]
  33. 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]
  34. Ullas, S.; Maheswari, B.U.; Ponnekant, S.; Kumar, T.M.M. A three stage attention enabled stacked deep CNN-BiLSTM (ASDCBNet) model for end-to-end monitoring of wastewater treatment plant. Appl. Water Sci. 2025, 15, 203. [Google Scholar] [CrossRef]
  35. Zamfir, F.-S.; Carbureanu, M.; Mihalache, S.F. Application of machine learning models in optimizing wastewater treatment processes: A review. Appl. Sci. 2025, 15, 8360. [Google Scholar] [CrossRef]
  36. Li, Y.; Kong, B.; Yu, W.; Zhu, X. An attention-based CNN-LSTM method for effluent wastewater quality prediction. Appl. Sci. 2023, 13, 7011. [Google Scholar] [CrossRef]
  37. Liu, J.; Long, Y.; Zhu, G.; Hursthouse, A.S. Application of artificial intelligence in the management of coagulation treatment engineering system. Processes 2024, 12, 1824. [Google Scholar] [CrossRef]
  38. Rios Fuck, J.V.; Cechinel, M.A.P.; Neves, J.; Campos de Andrade, R.; Tristão, R.; Spogis, N.; Riella, H.G.; Soares, C.; Padoin, N. Predicting effluent quality parameters for wastewater treatment plant: A machine learning-based methodology. Chemosphere 2024, 352, 141472. [Google Scholar] [CrossRef]
  39. Zhu, Y.; Wang, Y.; Zhu, E.; Ma, Z.; Wang, H.; Chen, C.; Guan, J.; Waite, T.D. Predicting membrane fouling of submerged membrane bioreactor wastewater treatment plants using machine learning. Environ. Sci. Technol. 2025, 59, 10010–10021. [Google Scholar] [CrossRef]
  40. Deepnarain, N.; Nasr, M.; Kumari, S.; Stenström, T.A.; Reddy, P.; Pillay, K.; Bux, F. Decision tree for identification and prediction of filamentous bulking at full-scale activated sludge wastewater treatment plant. Proces. Saf. Environ. 2019, 126, 25–34. [Google Scholar] [CrossRef]
  41. Achite, M.; Farzin, S.; Elshaboury, N.; Valikhan Anaraki, M.; Amamra, M.; Toubal, A.K. Modeling the optimal dosage of coagulants in water treatment plants using various machine learning models. Environ. Dev. Sustain. 2024, 26, 3395–3421. [Google Scholar] [CrossRef]
  42. Cui, F.; Kim, M.; Park, C.; Kim, D.; Mo, K.; Kim, M. Application of principal component analysis (PCA) to the assessment of parameter correlations in the partial-nitrification process using aerobic granular sludge. J. Environ. Manag. 2021, 288, 112408. [Google Scholar] [CrossRef]
  43. Izquierdo, J.; Montalvo, I.; Pérez, R.; Fuertes, V.S. Design optimization of wastewater collection networks by PSO. Comput. Math. Appl. 2008, 56, 777–784. [Google Scholar] [CrossRef]
  44. Fu, X.; Zheng, Q.; Jiang, G.; Roy, K.; Huang, L.; Liu, C.; Li, K.; Chen, H.; Song, X.; Chen, J.; et al. Water quality prediction of copper-molybdenum mining-beneficiation wastewater based on the PSO-SVR model. Front. Environ. Sci. Eng. 2023, 17, 98. [Google Scholar] [CrossRef]
  45. Achite, M.; Samadianfard, S.; Elshaboury, N.; Sharafi, M. Modeling and optimization of coagulant dosage in water treatment plants using hybridized random forest model with genetic algorithm optimization. Environ. Develop Sustain. 2023, 25, 11189–11207. [Google Scholar] [CrossRef]
  46. Elsayed, A.; Ghaith, M.; Yosri, A.; Li, Z.; El-Dakhakhni, W. Genetic programming expressions for effluent quality prediction: Towards AI-driven monitoring and management of wastewater treatment plants. J. Environ. Manag. 2024, 356, 120510. [Google Scholar] [CrossRef]
  47. Muschalla, D. Optimization of integrated urban wastewater systems using multi-objective evolution strategies. Urban. Water J. 2008, 5, 59–67. [Google Scholar] [CrossRef]
  48. 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]
  49. Jafar, R.; Awad, A.; Jafar, K.; Shahrour, I. Predicting effluent quality in full-scale wastewater treatment plants using shallow and deep artificial neural networks. Sustainability 2022, 14, 15598. [Google Scholar] [CrossRef]
  50. Sadoune, H.; Rihani, R.; Marra, F.S. DNN model development of biogas production from an anaerobic wastewater treatment plant using Bayesian hyperparameter optimization. Chem. Eng. J. 2023, 471, 144671. [Google Scholar] [CrossRef]
  51. Wang, D.; Thunéll, S.; Lindberg, U.; Jiang, L.; Trygg, J.; Tysklind, M. Towards better process management in wastewater treatment plants: Process analytics based on SHAP values for tree-based machine learning methods. J. Environ. Manag. 2022, 301, 113941. [Google Scholar] [CrossRef]
  52. Zhao, Z.; Wang, Z.; Yuan, J.; Ma, J.; He, Z.; Xu, Y.; Shen, X.; Zhu, L. Development of a novel feedforward neural network model based on controllable parameters for predicting effluent total nitrogen. Engineering 2021, 7, 195–202. [Google Scholar] [CrossRef]
  53. Ly, Q.V.; Truong, V.H.; Ji, B.; Nguyen, X.C.; Cho, K.H.; Ngo, H.H.; Zhang, Z. Exploring potential machine learning application based on big data for prediction of wastewater quality from different full-scale wastewater treatment plants. Sci. Total Environ. 2022, 832, 154930. [Google Scholar] [CrossRef]
  54. Wu, X.; Zheng, Z.; Wang, L.; Li, X.; Yang, X.; He, J. Coupling process-based modeling with machine learning for long-term simulation of wastewater treatment plant operations. J. Environ. Manag. 2023, 341, 118116. [Google Scholar] [CrossRef]
  55. Wang, H.C.; Wang, Y.Q.; Wang, X.; Yin, W.X.; Yu, T.C.; Xue, C.H.; Wang, A.J. Multimodal machine learning guides low carbon aeration strategies in urban wastewater treatment. Engineering 2024, 36, 51–62. [Google Scholar] [CrossRef]
  56. Liu, Y.; Guo, J.; Wang, Q.; Huang, D. Prediction of filamentous sludge bulking using a state-based gaussian processes regression model. Sci. Rep. 2016, 6, 31303. [Google Scholar] [CrossRef] [PubMed]
  57. Shao, S.; Fu, D.; Yang, T.; Mu, H.; Gao, Q.; Zhang, Y. Analysis of machine learning models for wastewater treatment plant sludge output prediction. Sustainability 2023, 15, 13380. [Google Scholar] [CrossRef]
  58. Zaghloul, M.S.; Achari, G. Application of machine learning techniques to model a full-scale wastewater treatment plant with biological nutrient removal. J. Environ. Chem. Eng. 2022, 10, 107430. [Google Scholar] [CrossRef]
  59. Newhart, K.B.; Marks, C.A.; Rauch-Williams, T.; Cath, T.Y.; Hering, A.S. Hybrid statistical-machine learning ammonia forecasting in continuous activated sludge treatment for improved process control. J. Water Process Eng. 2020, 37, 101389. [Google Scholar] [CrossRef]
  60. Mao, Z.; Li, X.; Zhang, X.; Li, D.; Lu, J.; Li, J.; Zheng, F. Optimization of effluent quality and energy consumption of aeration process in wastewater treatment plants using artificial intelligence. J. Water Process Eng. 2024, 63, 105384. [Google Scholar] [CrossRef]
  61. Khalil, M.; AlSayed, A.; Liu, Y.; Vanrolleghem, P.A. Machine learning for modeling N2O emissions from wastewater treatment plants: Aligning model performance, complexity, and interpretability. Water Res. 2023, 245, 120667. [Google Scholar] [CrossRef]
  62. Bustillo-Lecompte, C.F.; Mehrvar, M. Treatment of actual slaughterhouse wastewater by combined anaerobic–aerobic processes for biogas generation and removal of organics and nutrients: An optimization study towards a cleaner production in the meat processing industry. J. Clean. Prod. 2017, 141, 278–289. [Google Scholar] [CrossRef]
  63. Antwi, P.; Zhang, D.; Luo, W.; Xiao, L.w.; Meng, J.; Kabutey, F.T.; Ayivi, F.; Li, J. Performance, microbial community evolution and neural network modeling of single-stage nitrogen removal by partial-nitritation/anammox process. Bioresource Technol. 2019, 284, 359–372. [Google Scholar] [CrossRef]
  64. Hmoud Al-Adhaileh, M.; Waselallah Alsaade, F. Modelling and prediction of water quality by using artificial intelligence. Sustainability 2021, 13, 4259. [Google Scholar] [CrossRef]
  65. 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]
  66. 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]
  67. Okoji, A.I.; Okoji, C.N.; Awarun, O.S. Performance evaluation of artificial intelligence with particle swarm optimization (PSO) to predict treatment water plant DBPs (haloacetic acids). Chemosphere 2023, 344, 140238. [Google Scholar] [CrossRef] [PubMed]
  68. Khan, F.; Zuthi, M.F.R.; Rahman, M.S.; Akbor, M.A.; Hossain, M.D.; Bhuiyan, M.N.I. Predicting disinfection by-products (DBPs) in supply water within a real water distribution network using an artificial neural network. Ecotoxicol. Environ. Saf. 2025, 303, 118762. [Google Scholar] [CrossRef]
  69. Hong, H.; Zhang, Z.; Guo, A.; Shen, L.; Sun, H.; Liang, Y.; Wu, F.; Lin, H. Radial basis function artificial neural network (RBF ANN) as well as the hybrid method of RBF ANN and grey relational analysis able to well predict trihalomethanes levels in tap water. J. Hydrol. 2020, 591, 125574. [Google Scholar] [CrossRef]
  70. Akbar, M.A.; Selvaganapathy, P.R.; Kruse, P. Continuous monitoring of monochloramine in water, and its distinction from free chlorine and dichloramine using a functionalized graphene-based array of chemiresistors. ACS EST Water 2024, 4, 4041–4051. [Google Scholar] [CrossRef]
  71. Chen, Z.; Cheng, H.; Wang, X.; Chen, B.; Chen, Y.; Cai, R.; Zhang, G.; Song, C.; He, Q. Development and application of an intelligent nitrogen removal diagnosis and optimization framework for WWTPs: Low-carbon and stable operation. Water Res. 2024, 266, 122337. [Google Scholar] [CrossRef]
  72. Zhou, Z.; Wu, X.; Dong, X.; Zhang, Y.; Wang, B.; Huang, Z.; Luo, F.; Zhou, A. Carbon source dosage intelligent determination using a multi-feature sensitive back propagation neural network model. J. Environ. Manag. 2025, 376, 124341. [Google Scholar] [CrossRef]
  73. Hansen, L.D.; Stokholm-Bjerregaard, M.; Durdevic, P. Modeling phosphorous dynamics in a wastewater treatment process using Bayesian optimized LSTM. Comput. Chem. Eng. 2022, 160, 107738. [Google Scholar] [CrossRef]
  74. Inbar, O.; Avisar, D. Enhancing wastewater treatment through artificial intelligence: A comprehensive study on nutrient removal and effluent quality prediction. J. Water Process Eng. 2024, 61, 105212. [Google Scholar] [CrossRef]
  75. 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]
  76. Gholami Shirkoohi, M.; Tyagi, R.D.; Vanrolleghem, P.A.; Drogui, P. A comparison of artificial intelligence models for predicting phosphate removal efficiency from wastewater using the electrocoagulation process. Digit. Chem. Eng. 2022, 4, 100043. [Google Scholar] [CrossRef]
  77. Wang, K.J.; Wang, P.S.; Nguyen, H.P. A data-driven optimization model for coagulant dosage decision in industrial wastewater treatment. Comput. Chem. Eng. 2021, 152, 107383. [Google Scholar] [CrossRef]
  78. Mohanty, A.; Mohanty, S.K.; Mohapatra, A.G. Real-time monitoring and fault detection in AI-enhanced wastewater treatment systems. In The AI Cleanse: Transforming Wastewater Treatment Through Artificial Intelligence: Harnessing Data-Driven Solutions; Garg, M.C., Ed.; Springer Nature: Cham, Switzerland, 2024; pp. 165–199. [Google Scholar]
  79. Bellamoli, F.; Di Iorio, M.; Vian, M.; Melgani, F. Machine learning methods for anomaly classification in wastewater treatment plants. J. Environ. Manag. 2023, 344, 118594. [Google Scholar] [CrossRef]
  80. Humoreanu, B.; Nascu, I. Wastewater Treatment Plant SCADA Application. In Proceedings of the 2012 IEEE International Conference on Automation, Quality and Testing, Robotics (AQTR 2012), Cluj-Napoca, Romania, 24–27 May 2012; pp. 575–580. [Google Scholar]
  81. Salem, R.M.M.; Saraya, M.S.; Ali-Eldin, A.M.T. An industrial cloud-based IoT system for real-time monitoring and controlling of wastewater. IEEE Access 2022, 10, 6528–6540. [Google Scholar] [CrossRef]
  82. Yalçın, N.; Çakır, S.; Ünaldı, S. Attack detection using artificial intelligence methods for SCADA security. IEEE Internet Things J. 2024, 11, 39550–39559. [Google Scholar] [CrossRef]
Figure 1. China’s annual wastewater discharge and treatment trends from 2015 to 2023 (Data sourced from MOHURD, [4]).
Figure 1. China’s annual wastewater discharge and treatment trends from 2015 to 2023 (Data sourced from MOHURD, [4]).
Separations 12 00237 g001
Figure 2. The published papers on wastewater purification and treatment in conjunction with machine learning/artificial intelligence and dosing/dosage strategies, retrieved from all Web of Science databases for the period of 2015–2025, as of 25 May 2025.
Figure 2. The published papers on wastewater purification and treatment in conjunction with machine learning/artificial intelligence and dosing/dosage strategies, retrieved from all Web of Science databases for the period of 2015–2025, as of 25 May 2025.
Separations 12 00237 g002
Figure 3. Classification of artificial intelligence (AI) models used in domestic WWTPs.
Figure 3. Classification of artificial intelligence (AI) models used in domestic WWTPs.
Separations 12 00237 g003
Figure 4. The relationship among artificial intelligence (AI), machine learning (ML), and deep learning (DL).
Figure 4. The relationship among artificial intelligence (AI), machine learning (ML), and deep learning (DL).
Separations 12 00237 g004
Figure 5. The schematic diagrams illustrating the working mechanisms of ANNs (a), CNNs (b), DNNs (c), FNNs (d), and RNNs (e).
Figure 5. The schematic diagrams illustrating the working mechanisms of ANNs (a), CNNs (b), DNNs (c), FNNs (d), and RNNs (e).
Separations 12 00237 g005
Figure 6. Schematic of AI applications for water quality detection, chemical dosing optimization in WWTPs, as well as the prediction of various operational faults, such as treatment inefficiency, pipe corrosion, and sludge bulking.
Figure 6. Schematic of AI applications for water quality detection, chemical dosing optimization in WWTPs, as well as the prediction of various operational faults, such as treatment inefficiency, pipe corrosion, and sludge bulking.
Separations 12 00237 g006
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

Jin, J.; Liu, M.; Chen, B.; Wu, X.; Yao, L.; Wang, Y.; Xiong, X.; Wei, L.; Li, J.; Tan, Q.; et al. Artificial Intelligence in Chemical Dosing for Wastewater Purification and Treatment: Current Trends and Future Perspectives. Separations 2025, 12, 237. https://doi.org/10.3390/separations12090237

AMA Style

Jin J, Liu M, Chen B, Wu X, Yao L, Wang Y, Xiong X, Wei L, Li J, Tan Q, et al. Artificial Intelligence in Chemical Dosing for Wastewater Purification and Treatment: Current Trends and Future Perspectives. Separations. 2025; 12(9):237. https://doi.org/10.3390/separations12090237

Chicago/Turabian Style

Jin, Jie, Ming Liu, Boyu Chen, Xuanbei Wu, Ling Yao, Yan Wang, Xia Xiong, Luoyu Wei, Jiang Li, Qifeng Tan, and et al. 2025. "Artificial Intelligence in Chemical Dosing for Wastewater Purification and Treatment: Current Trends and Future Perspectives" Separations 12, no. 9: 237. https://doi.org/10.3390/separations12090237

APA Style

Jin, J., Liu, M., Chen, B., Wu, X., Yao, L., Wang, Y., Xiong, X., Wei, L., Li, J., Tan, Q., Fan, D., Du, Y., Lei, Y., & Yang, N. (2025). Artificial Intelligence in Chemical Dosing for Wastewater Purification and Treatment: Current Trends and Future Perspectives. Separations, 12(9), 237. https://doi.org/10.3390/separations12090237

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