Abstract:Enhancing particle removal from combined sewer overflows using ballasted flocculation is a critical strategy for drainage pollution on rainy days. However, traditional optimization methods, which rely on empirical knowledge and one-factor-at-a-time experiments, struggle to systematically unravel the complex interactions among multiple parameters, resulting in a lack of theoretical guidance for process design and inconsistent operational performance. This study established a machine learning-based method for data-driven optimization and mechanistic interpretation of the ballasted flocculation process, aiming to identify key parameters and improve clarification efficiency. It integrated a comprehensive dataset of 580 experimental records, encompassing 14 feature parameters across reagent properties, operational conditions, and medium characteristics. Multiple machine learning models, including Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM), were developed to predict clarification efficiency. Results demonstrate that the LightGBM model achieves the best predictive accuracy and generalization capability, with a root mean square error (E_RMS) of 7.31% and a coefficient of determination (R²) of 0.74. Interpretable shapley additive explanations analysis reveals that operational parameters contribute most significantly to clarification performance (contribution rate of 38.6%), substantially outperforming medium and reagent parameters. Settling time and coagulation velocity gradient are identified as the key drivers. Furthermore, by integrating water treatment chemistry and kinetic principles, the analysis elucidates the dynamic coupling effect between initial turbidity and coagulation velocity gradient and reveals a significant synergistic effect between settling time and specific gravity of media. Experimental verification using actual combined sewer overflows confirms that the optimized parameter combination achieves over 95% particulate removal, with strong agreement between predicted and actual values. This study establishes a research framework comprising "data-driven modeling, parameter interpretation and recommendation, and experimental validation", providing a theoretical basis and methodological support for the precise regulation and intelligent decision-making in ballasted flocculation.

Abstract:To address the limitations of traditional single models when dealing with complex water quality prediction and other real-world challenges, this study proposed the model simple averaging method and two Stacking integration strategies, namely the weighted integration based on the Least Square (LS) method and the Stacking integration with the Extra Trees Regressor (ETR) as the meta-learner. The full-process hourly monitoring data of a certain drinking water treatment plant were collected on-site, and the original data set was subjected to strict preprocessing. On this basis, the predictive performance of all individual models and ensemble models was comprehensively evaluated and compared on the test set. Results indicate that the three proposed ensemble models consistently outperform all single models across all four water quality indicators. Notably, these models achieve a substantial reduction in prediction error. Among these, the LS-weighted model performs the best in terms of average error, with an MAPE value of approximately 1.08%, while the ETR-weighted model has an average error that is slightly higher by 0.02%. Furthermore, this study investigates the weight allocation of base learners in the two Stacking models, revealing that the ETR-weighted model demonstrates greater adaptability. Specifically, this model demonstrates the ability to respond rapidly to complex water quality variations through nonlinearly dynamic weight adjustment, particularly during high-risk events such as sudden turbidity spikes. This study provides a new perspective for the intelligent management of drinking water treatment. The proposed Stacking ensemble model can provide reliable data support for the stable operation and efficiency optimization of water treatment plants. Its strong adaptability offers crucial risk warnings for process disturbances and water quality anomalies, facilitating timely adjustments in subsequent treatment processes.

Abstract:Currently, the operating conditions of the secondary water supply pump set are disconnected from the water demand of the pipeline network. The pump set has operated in the low-efficiency zone for a long time under the constant-pressure and variable-flow mode of the most unfavorable conditions, resulting in high system energy consumption. To change the operation mode of secondary water supply, this paper proposed a flow prediction method for secondary water supply systems based on the long short-term memory (LSTM) neural network model to achieve the transformation of pump set water supply from the traditional constant-pressure and variable-flow mode to the variable-pressure and variable-flow mode. Based on the flow monitoring data of 30 communities in China, this paper formed a characteristic sample set of secondary water supply flow, proposed to use LSTM for data mining to establish a model that can accurately predict the flow of secondary water supply, and used mean absolute error (E_MA), mean absolute percentage error (E_MAP), and root mean square error (E_RMS) to evaluate the predicted values against the true values. The evaluation results show that both E_MA and E_RMS are 4.32, and the lowest E_MAP is only 0.12%, which is far below the engineering reference of 4.94%, demonstrating high prediction accuracy and reliability. From the evaluation results, the LSTM flow prediction model established in this study can be applied to the flow prediction of secondary water supply systems, providing an innovative idea and method for the transformation of secondary water supply from constant-pressure and variable-flow mode to variable-pressure and variable-flow mode, offering a feasible implementation plan for promoting the energy-saving renovation of secondary water supply and having great engineering application value for promoting the digital transformation of the water supply industry.

Abstract:Emerging contaminants have structural diversity, environmental persistence, and biological toxicity, which pose potential risks to ecosystems and human health. Life Cycle Assessment (LCA) is an important method for evaluating their environmental impacts, but it has problems such as a lack of basic data and difficulty in accurate quantification. To address these issues, this paper proposed an Artificial Neural Network (ANN) model based on Principal Component Analysis (PCA) to optimize input features. By performing PCA dimensionality reduction on a large number of molecular descriptors, main features were extracted to reduce data redundancy and dimensional burden for achieving the prediction of environmental impacts of emerging contaminants. Four models with different feature retention ratios were constructed, namely NOPCA (all descriptors), PCA95 (retaining principal components with 95% cumulative variance contribution rate), PCA85 (retaining principal components with 85% cumulative variance contribution rate), and PCA75 (retaining principal components with 75% cumulative variance contribution rate), and the differences in prediction performance were evaluated and compared. The results show that moderate feature extraction can improve the generalization ability of the model; in the prediction tasks of global warming potential (GWP) and human toxicity (HTP), the model constructed using PCA85-preprocessed descriptors performs the best, with test-set R² of 0.64 and 0.73, respectively; in the prediction tasks of fossil energy depletion (FDP) and terrestrial acidification (TAP), the model achieves optimal performance under PCA95, with test-set R² of 0.77 and 0.76, respectively. The ANN model based on PCA feature extraction can effectively mitigate the problem of incomplete LCA data and provides an efficient and reliable method for predicting the environmental impacts of emerging contaminants.

Abstract:Nanofiltration technology offers an effective solution for the separation and resource recovery of heavy metals in industrial wastewater. However, the physicochemical properties of nanofiltration membranes suitable for heavy metal separation remain unclear, making it difficult to prepare high-performance nanofiltration membranes for heavy metal separation. This paper developed a machine learning prediction model incorporating membrane characteristics, operating parameters, and solution properties. Through a comparison of ten models, the XGBoost model yielded R² values of 0.98 and 0.92 for membrane flux and heavy metal rejection rate, respectively. The SHAP analysis results revealed that water contact angle, molecular weight cut-off, and membrane surface potential were the key determinants of performance. Enhancing hydrophilicity, moderately increasing the molecular weight cut-off, and imparting more positive charges on the membrane surface collectively enhanced membrane flux. In contrast, heavy metal rejection rate was governed by the combined effects of size exclusion and electrostatic repulsion. Based on the high-accuracy prediction model, an NSGA-Ⅱ bi-objective optimization algorithm was constructed. The results show that within the ranges of water contact angle from 60° to 66°, molecular weight cut-off from 200 to 412 u, and membrane surface potential from 47 to 73 mV, nanofiltration membranes can achieve efficient treatment of 10 kinds of heavy metal ions, with rejection rate exceeding 99% and membrane flux ranging from 25 to 30 L/(m²·h). The nanofiltration membrane optimization framework developed in this paper provides both theoretical and technical support for the development and application of high-performance nanofiltration materials.

Abstract:Nanobubbles (NBs) exhibit great application potential in water treatment owing to their large specific surface area, highly negative Zeta potential, and extended stability. However, the stability of the NBs is affected by the coupled interaction of multiple factors such as gas type, pressure, flow rate, medium, pH, and temperature. Conventional empirical trial-and-error approaches are inefficient, difficult to reproduce, and non-generalizable. To overcome these limitations, this paper established a data-driven inverse design framework combining interpretable machine learning and experimental validation to reveal key laws of NBs′ stability and propose transferable preparation strategies. A random forest model was optimized, achieving a precision of 0.816, a recall of 0.814, and an F₁ score of 0.810 on the test set. Model feature analysis indicated that ozone, carbonate or mildly alkaline media, and suitable pressure and temperature were the main factors improving the stability. Guided by inverse prescriptions, NBs retained strong stability after 21 days of storage, and in the system coupled with persulfate, the removal rate of methylene blue by NBs reached (96.8±2.3)%, outperforming the control group. Long-term performance tests further showed that the NBs/PMS system achieved about 51% TOC mineralization of natural organic matter (NOM) within 30 days, demonstrating sustained oxidative capacity under complex matrices. Energy consumption evaluation indicated that, for 90% methylene blue removal, the single-treatment energy consumption of the NBs/PMS system was approximately 35.7% of that required by the conventional UV/PMS process, corresponding to a reduction of about 64.3%. This result highlights its advantages in energy efficiency and cost control. This study not only provides design principles for regulating the stability and theoretically reveals the laws of multi-factor synergistic effect but also verifies its application potentials in terms of long-term stability and engineering energy efficiency. It is of great significance for promoting the large-scale application of NB technology and the industrialization of green water treatment.

Abstract:Eutrophication is one of the most challenging environmental problems facing surface water at present, and existing studies mostly focus on regional-scale analysis. To reveal the eutrophication patterns and their complex driving mechanisms in large-scale river basins, this paper evaluated and classified the eutrophication status of river basins across China and explored the potential driving factors affecting the changes in eutrophication status by combining machine learning and the explainable artificial intelligence method of SHapley Additive exPlanations (SHAP). The results show that eutrophication is widespread in most river basins, and the eutrophication problem in eastern, northeastern, and central China is more severe than that in western regions, exhibiting significant spatial-temporal clustering characteristics; total phosphorus and total nitrogen are key water quality factors, with contributions of 20.6% and 75.4%, respectively, and both have a significant positive spatial correlation with river basin eutrophication; the Light Gradient Boosting Machine (LightGBM) model performs best among all models, with an accuracy of 91%. SHAP analysis reveals that compared with anthropogenic factors, natural driving factors play a more important role in explaining the overall changes in the eutrophication status of river basins, among which topographic conditions control the accumulation and loss of nutrients, while precipitation plays a regulatory role. However, with the aggravation of eutrophication, the impact of anthropogenic factors such as population density, gross domestic product, and wastewater treatment capacity gradually becomes prominent. If economic development is accompanied by the improvement of infrastructure, eutrophication can be mitigated to a certain extent. Given the synergistic effect of internal nutrient loads and external river basin characteristics on eutrophication, it is recommended to implement precise management and control based on spatial zoning and classification to effectively address this environmental challenge. This study deepens the theoretical understanding of the driving mechanism of river basin-scale eutrophication by combining artificial intelligence methods and provides quantitative and interpretable scientific evidence for water environment governance.

Abstract:To support the refined management of water ecology in cold-region river basins and clarify the differential driving characteristics of water quality and antibiotics on microbial communities in winter and summer, as well as the concentration effect inflection points of key environmental factors, this paper focused on a typical cold-region river basin, integrated 132 valid samples (68 in summer and 64 in winter), and constructed an association dataset containing eight water quality indicators, 11 antibiotics, and microbial communities. This paper applied the Random Forest (RF) model to analyze the relationship between environmental factors and community diversity, combined SHapley Additive exPlanations (SHAP) interpretable analysis to identify key driving factors, their concentration effect inflection points, and multivariate interactions, and analyzed the seasonal differences in community function potential through Canonical Correspondence Analysis (CCA). The results indicate that the environmental and community characteristics of this cold-region river basin exhibit significant seasonal differences; the dissolved oxygen (DO, mean value of 10.32 mg/L) in winter is significantly higher than that in summer (6.95 mg/L); the mass concentration fluctuation of antibiotics such as norfloxacin (NOR, mean value of 156.76 ng/L) in summer is much higher than that in winter (1.23 ng/L), and the community diversity and functional potential show significant seasonal differences; TN is the key factor driving the changes in Shannon index in both winter and summer, while PD_whole_tree index is dominated by macrolide and tetracycline antibiotics in summer and by DO and quinolone antibiotics in winter, indicating that the driving factors of microbial community diversity exhibit seasonal specificity; SHAP analysis further identifies the potential concentration effect inflection points of key factors such as DO (7.41 mg/L in summer, 10.26 mg/L in winter), and TN (1.01 mg/L in winter), preliminarily revealing their concentration ranges that exert promoting or inhibitory effects on microbial community diversity; there are synergistic effects between multiple antibiotics and DO in summer and between TN and oxytetracycline (OTC) in winter, and the multi-factor combined pressures in complex environments have a superimposed impact on microbial communities; the community information storage and processing functions in winter are dominated by DO and temperature, while the cellular processes and metabolic functions in summer are regulated by multiple antibiotics, forming a seasonal differentiation pattern of "water quality dominance in winter and antibiotic regulation in summer". Methodologically, this paper integrated machine learning and interpretable analysis and revealed the seasonal response patterns of microbial communities to combined environmental pressures in cold-region river basins. The proposed concentration effect inflection points provide a scientific reference for the season-specific ecological risk management of river basins.

Abstract:Accurate estimation of municipal solid waste (MSW) generation and identification of its key associative characteristics are important prerequisites for achieving refined global urban environmental management and differentiated governance. However, the scarcity of long-time-series data globally, especially in developing countries, and the insufficient capability of traditional models in resolving non-linear associative characteristics make it difficult for existing studies to accurately reconstruct the historical generation scale and evolutionary laws of global MSW. To this end, this paper established an intelligent estimation and cluster analysis framework for global MSW generation based on XGBoost and K-Means++ algorithms. The results indicate that XGBoost exhibits excellent generalization capability, with a coefficient of determination (R²) of 0.91±0.07 for MSW generation prediction in unknown countries; the comparison of cross-validation strategies confirms that the model successfully captures the intrinsic associations of socio-economic characteristics for MSW generation prediction, rather than merely relying on time trends, showing good spatial generalization reliability. SHAP interpretation analysis proves that population scale and economic volume are the core associative factors affecting MSW generation and reveals a unique "N-shaped" non-linear associative trajectory between GDP (in terms of purchasing power parity, PPP) per capita and generation, breaking the limitations of traditional linear extrapolation models. The estimation results indicate that the total global MSW generation has climbed from (1.29±0.04) billion t in 1991 to (2.14±0.16) billion t in 2021, with an average annual growth of 1.7%. Cluster analysis identifies four typical evolutionary patterns among global countries: agricultural subsistence, emerging industrial, mature stable, and high-consumption affluent patterns, among which emerging industrial countries are the main source of global MSW increment. The constructed prediction framework possesses good universal applicability and can provide a scientific basis for global solid waste′s refined management, regionally differentiated policy formulation, and infrastructure planning.

Abstract:In practical engineering, the sludge line of wastewater reclamation plants faces challenges including high parameter sensitivity, drift nonlinearity, and significant time lag, which introduce considerable uncertainty into the operation and scheduling of the sludge line and restrict the synergistic capacity of pollution and carbon reduction. Conventional mechanism-driven models require a large amount of high-quality and well-annotated input data, making it difficult to predict sludge line parameters accurately and efficiently. To enhance the synergy of pollution and carbon reduction for the sludge line in wastewater reclamation plants, this paper established a prediction model based on the multi-head self-attention mechanism (Transformer) using a hybrid modeling strategy of "empirical model baseline and residual prediction", based on the operational data of the water line and sludge line of a wastewater reclamation plant in Beijing. Specifically, this paper first generated a baseline through linear fitting of an empirical model and then utilized the Transformer model to accurately predict the residual. Taking the inlet and outlet water quality indexes and operating parameters of the water line as inputs, the model simultaneously predicted the dewatered sludge output and sludge moisture content and compared its performance with LightGBM, block recurrent neural network BlockRNN (LSTM) and temporal convolutional network (TCN) models. The results indicate that relying on the global temporal correlation capture capability of the Transformer model based on multi-head self-attention mechanism, it can effectively adapt to the nonlinear fluctuation characteristics of sludge line operating parameters and achieve the optimal comprehensive performance: R² of dewatered sludge output prediction reaches 0.778 8, and that of sludge moisture content prediction reaches 0.679, both of which are significantly higher than those of the comparative models. Meanwhile, the backtesting mechanism can improve the operational stability of the model, continuously enhance its prediction accuracy, and thus adapt to the long-term time series prediction scenarios in practical engineering. This model can realize high-precision prediction of key parameters in the sludge line, provide reliable technical support for scenarios such as real-time regulation of the sludge line process, precise chemical dosing, and intelligent equipment scheduling in wastewater reclamation plants, and facilitate the synergistic governance of pollution and carbon reduction for the sludge line.

Abstract:Membrane fouling seriously restricts the application of ultrafiltration (UF) technology in municipal wastewater treatment. Secondary effluent organic matter (EfOM), characterized by its complex composition, is considered the primary cause of membrane fouling. However, there is no method to effectively identify the key components of membrane fouling. To address this, this paper proposed a key membrane fouling factor identification framework based on the tabular prior-data fitted network (TabPFN) deep learning model by using SHapley Additive exPlanations (SHAP) method. Water quality and membrane fouling data were collected from a modified ceramic ultrafiltration membrane (MCUM) during a two-hour operation cycle. UV₂₅₄, dissolved organic carbon (DOC), fluorescence regional integration values, and fluorescence component intensities (F_max) were selected as input variables to predict normalized specific flux, reversible resistance, and irreversible resistance. The results show that the TabPFN model achieves high-precision prediction without hyperparameter tuning (R²=0.97 for normalized specific flux), outperforming conventional machine learning models in terms of overall performance. Further improvement is obtained by introducing a post-hoc ensembling (PHE) strategy, increasing the prediction performance of R² to 0.98. SHAP analysis indicates that tryptophan-like, fulvic acid-like, and humic acid-like substances have predominant influences on the reduction of membrane flux and the formation of irreversible resistance and reversible resistance, respectively. PDPs results further reveal significant nonlinear synergistic effects among these components, especially the synergistic enhancement effect of the fulvic acid-like and aromatic substances in the irreversible fouling process. This study reveals the key components of secondary EfOM responsible for membrane fouling, providing theoretical support for membrane fouling control strategy optimization and the understanding of fouling mechanisms.

Abstract:To address the problems that models easily fall into local optima and have insufficient accuracy in the high-dimensional nonlinear spectral inversion of soil organic matter (SOM), this paper proposed an inversion method of SOM based on a Support Vector Machine (SVM) optimized by an improved goose swarm optimization (IGOOSE) algorithm. Taking soil samples from typical farm areas in Heilongjiang Province as the research object, this paper used the Mahalanobis distance method to remove abnormal samples and screened the optimal spectral preprocessing method based on partial least squares regression (PLSR) modeling results; by constructing full-band SVM, random forest (RF), and extreme learning machine (ELM) models under the optimal preprocessing condition, this paper determined the optimal basic inversion model; furthermore, this paper used the successive projection algorithm (SPA), random frog feature importance selection (RFROG), and competitive adaptive reweighted sampling (CARS) to conduct spectral feature dimensionality reduction and selected the features obtained by the optimal dimensionality reduction method as the model input. In the process of model parameter optimization, by dynamically adjusting the population size and integrating the Lévy flight strategy, random perturbation mechanism, and hybrid neighborhood search method, this paper conducted a three-stage adaptive improvement on the basic goose swarm optimization (GOOSE) algorithm to construct the IGOOSE algorithm. After verifying its effectiveness via standard test functions, this paper applied IGOOSE to SVM parameter optimization and compared it with the original GOOSE algorithm to identify the inversion model of SOM with better performance. The results show that the accuracy of the PLSR model constructed after first derivative (D1) preprocessing is significantly superior to other preprocessing methods; the optimal basic model constructed based on full-band D1 data is SVM, and its five-fold cross-validation average coefficient of determination (R²) reaches 0.762 4; CARS is the optimal spectral feature dimensionality reduction method, and the five-fold cross-validation average R² of the CARS-SVM model constructed based on its selected features is 0.753 7; IGOOSE-SVM is significantly superior to GOOSE-SVM in terms of generalization ability and prediction stability, and the R² of the training set and test set increase by 0.106 6 and 0.041 1, respectively, while the root mean square error (RMSE) decreases by 2.541 3 g/kg and 0.528 7 g/kg, respectively. Finally, the R²of the training set and test set reach 0.925 2 and 0.841 6, respectively, and the RMSE is 4.418 9 g/kg and 7.040 7 g/kg, respectively. The research indicates that the proposed improvement strategy effectively mitigates the modeling difficulties caused by high-dimensional nonlinear spectral data. The IGOOSE-SVM model provides a reliable approach for achieving high-precision inversion of SOM and provides a theoretical basis for the application and improvement of optimization algorithms in the field of spectral modeling.

Abstract:The sulfate reduction-sulfur autotrophic denitrification-nitrification (SANI) process is a novel biological wastewater treatment technology for nitrogen removal, which has advantages such as low sludge yield, high temperature tolerance, and salt tolerance. However, the complex microbial communities within this process are extremely sensitive to operational parameters, leading to easy fluctuation of system performance, which urgently requires reliable modeling tools for optimization. Traditional mechanistic models are cumbersome to construct and have limited universality, while machine learning methods are often limited by small sample size, data imbalance, and insufficient interpretability. To this end, this paper proposed a new data-driven method integrating generative adversarial network (GAN) and staged machine learning modeling (including XGBoost, Artificial Neural Network, and Support Vector Machine algorithms), constructed two-stage prediction models for anaerobic sulfate reduction and anoxic sulfur autotrophic denitrification according to the biochemical mechanism of the process, and integrated acclimation and operation data to expand the applicability of the models. The results show that the root mean square error (RMSE) of the model for the prediction of COD and nitrogen removal rates are as low as 6.8% and 5.6%, respectively, exhibiting excellent prediction accuracy. The interpretability analysis further clarifies the key environmental factors and optimal control ranges of the process: influent COD of 340-500 mg/L, influent sulfate (as S) of 300-450 mg/L, anaerobic effluent sulfide (as S) of 50-100 mg/L, anaerobic working volume of 5-6.5 L, and hydraulic retention time of 6-8 h. This paper provides a reliable tool for the precise regulation of the SANI process and further constructs a transferable and scalable integration paradigm of "data augmentation, staged modeling, and mechanism interpretation", which provides a brand-new methodological foundation for the intelligent optimization of complex biological wastewater treatment processes.

Abstract:To enhance the operational efficiency of wastewater treatment plants, this paper took effluent total nitrogen as the primary control indicator, comparing the predictive capabilities of mechanistic and machine learning models. Based on model construction, the improvement effects of parameter optimization on operational efficiency (effluent total nitrogen and energy consumption) were further investigated. First, the response surface methodology was coupled with the activated sludge model to establish a quadratic polynomial response surface for optimizing three key process parameters: dissolved oxygen in the aerobic zone, internal sludge recirculation ratio, and external sludge recirculation ratio. This optimization aimed to achieve the optimal effluent total nitrogen. In parallel, six classical machine learning models such as random forest model, extreme gradient boosting model, and light gradient boosting machine model were evaluated and selected. Furthermore, a multi-objective optimization framework was developed by coupling the Twin Delayed Deep Deterministic Policy Gradient Algorithm with a random forest model, synergistically optimizing effluent total nitrogen and energy consumption. The results demonstrate that the response surface methodology coupled with the activated sludge model achieves a high goodness of fit (R²=0.944 5). Under optimal conditions, the removal efficiency of effluent total nitrogen can reach 89.14%, with the optimal effluent total nitrogen concentration of 3.533 mg/L, representing a 45.48% reduction compared with that of actual wastewater treatment plants. Among the machine learning prediction models, the random forest model exhibits the best predictive performance (R²=0.79,E_MR=7.5%). After applying the multi-objective optimization control strategy processed by a Twin Delayed Deep Deterministic Policy Gradient Algorithm based on reinforcement learning, the average effluent total nitrogen concentration is 6.20 mg/L, 4.32% lower than that of actual wastewater treatment plants, while aeration and pumping energy consumption decrease by 33.12%, and the operating cost index is reduced by 25.16%. In the short term, the activated sludge model achieves superior performance for the single objective of total nitrogen removal. However, the advantage of the Twin Delayed Deep Deterministic Policy Gradient Algorithm optimization lies in its multi-objective optimization, delivering significant improvements in controlling energy consumption, cost, and stability while maintaining effluent total nitrogen compliance. In the long run, this approach provides superior overall treatment results. In conclusion, the Twin Delayed Deep Deterministic Policy Gradient Algorithm coupled with random forest model enables effective multi-objective optimization control and offers both theoretical and methodological support for the intelligent operation of wastewater treatment plants.

Abstract:Ensuring a high-quality, safe, and reliable drinking water supply is a necessary guarantee for sustainable socio-economic development. Disinfection is an indispensable process in drinking water treatment. The disinfection process must effectively inactivate pathogens, strictly control disinfectant residuals and byproduct generation, and achieve collaborative control of biological and chemical risks. The effect of drinking water disinfection is influenced by multiple factors including environmental conditions, water quality background, disinfection methods, and standard requirements, making its efficacy control particularly difficult. Consequently, the disinfection process exhibits higher complexity, placing higher demands on the refined and dynamic regulation of operation. In recent years, the rapid development of artificial intelligence (AI) technology has provided new ideas for the optimization of drinking water disinfection processes and the control of derived risks. This paper systematically elaborated on the application of AI-driven multi-source data fusion in improving the efficacy of disinfection processes. By taking key links such as precursor risk identification, disinfection efficacy assessment, byproduct prediction, and water quality safety screening as objects, this paper systematically reviewed the technical methods for applying dynamic learning algorithms in improving microbial inactivation effects, controlling disinfection byproduct generation, and rapidly screening toxicological risks and proposed strategies and development directions for the application of AI technology in disinfection processes. With the continuous development of AI technology and the wide application of big data, AI will play a more important role in the optimal application of disinfection processes, ensuring the biological and chemical safety of water quality and coping with complex water quality problems.

Abstract:Membrane technology, possessing advantages such as high-efficiency separation, simple operation, and high integration, is a key technology for promoting the quality improvement and efficiency enhancement of drinking water. However, membrane technology is still limited by the problems of insufficient intrinsic performance of membrane materials, difficulty in dynamic regulation of membrane units, and complex operation and maintenance of membrane systems, which constrain the effectiveness of its practical engineering applications. Through the high coupling of data and intelligent analysis, artificial intelligence can empower reverse design of materials, precise regulation of processes, and intelligent scheduling of systems. Promoting the deep integration of artificial intelligence and membrane technology is of great significance for achieving the leapfrog development of membrane technology. This paper reviewed the coupling characteristics and development pathways of artificial intelligence at the three levels of membrane material, membrane unit, and membrane system and emphatically discussed the application potential of artificial intelligence in high-performance membrane material design, process energy conservation and consumption reduction and water quality safety assurance, and whole-process system evaluation and multi-objective intelligent management and control. At present, the application of artificial intelligence in membrane filtration technology for water treatment still faces problems such as sparse key data, insufficient model prediction accuracy and generalization ability, and high integration and deployment complexity. It is urgent to construct cross-scenario standardized datasets, develop data-mechanism fusion reinforcement models, and establish modular and lightweight edge intelligence frameworks, so as to provide theoretical and technical references for the research and application of artificial intelligence-driven membrane filtration technology for water treatment.

Abstract:The water environmental risks of emerging contaminants (ECs) pose a severe challenge to conventional water treatment technologies. Advanced oxidation processes (AOPs) rely on the potent oxidative efficacy of reactive oxygen species (ROS). However, their vast variety, complex reaction pathways, and transient existence cause significant bottlenecks for traditional experimental methods in systematically elucidating the generation laws and action mechanisms of ROS, as well as the structure-activity relationships of high-efficiency catalysts. As a powerful data mining and pattern recognition tool, machine learning (ML) is fundamentally changing the research paradigm of AOPs, providing unprecedented solutions for the precise analysis and targeted design of this complex system. This paper systematically summarized the three core applications of ML in AOP research, namely, the prediction and optimization of reaction processes and ROS kinetics, the deep analysis of complex reaction pathways and microscopic mechanisms, and the intelligent screening and rational design of high-performance catalysts. The content covers radical pathways such as hydroxyl radicals (HO·) and sulfate radicals (SO-₄·), as well as non-radical systems such as singlet oxygen (¹O₂) and high-valent metal species. The results indicate that by integrating multi-source data and physicochemical models, ML not only achieves the accurate prediction and optimization of the performance of AOPs but also demonstrates strong capabilities in revealing implicit structure-activity relationships, discovering new descriptors, and predicting novel catalysts, promoting the transition of this field from "trial-and-error method" to "rational design". The focus of breakthroughs in this field will be on developing ML models with stronger interpretability and integrated physical mechanisms and establishing standardized databases to ultimately realize the intelligent and customized design and application of AOPs.

Abstract:To address the high costs of environmental pollutant detection and the reliance on animal experiments, quantitative structure-activity relationship (QSAR) models have been widely used in predicting the molecular activity and ecotoxicity of typical environmental pollutants (such as persistent organic pollutants and endocrine-disrupting chemicals). However, traditional QSAR methods often suffer from the "activity cliff" and the scarcity of small-sample data when identifying characteristic pollutants in complex environmental samples, resulting in insufficient generalization capability of the models. In view of these issues, this paper systematically reviewed the mainstream training paradigms of deep learning in environmental molecular modeling, including strategies such as transfer learning, pre-training and fine-tuning, and self-supervised learning, and it discussed their theoretical foundations and technical implementations in feature extraction, model robustness, and cross-task generalization. Literature analysis and method comparison show that compared with traditional machine learning methods, these training paradigms demonstrate stronger generalization capability, cross-task adaptability, and identification advantages for structural features of unknown compounds under small-sample conditions across different types of self-constructed environmental datasets. This paper further proposes a molecular modeling workflow suitable for environmental pollutant modeling and risk assessment and summarizes the key points of the model evaluation system, providing a systematic pathway for constructing highly generalizable and interpretable environmental deep learning models. This research can provide structured technical guidelines for model development in the environmental field and a reference framework for model comparison, promote the systematic evaluation of different deep learning training strategies, and offer support for the industry to establish more unified and reproducible intelligent modeling standards, thereby promoting the development of environmental pollutant screening and risk assessment toward data-driven and intelligent directions.