An interpretable machine learning-based model for prediction of THMs and identification of key factors in water supply networks

XU Xiao-yan; HUO Ran-ran; LI Hui-ping; ZHOU Bai-qin; DU Zhen-qi; PANG Wei-hai; ZHOU Ting; DONG Xiao-chen; LI Ya-xian

PDF(1746 KB)

China Environmental Science ›› 2026, Vol. 46 ›› Issue (1) : 208-218.

An interpretable machine learning-based model for prediction of THMs and identification of key factors in water supply networks

XU Xiao-yan¹, HUO Ran-ran¹, LI Hui-ping², ZHOU Bai-qin³, DU Zhen-qi⁴, PANG Wei-hai², ZHOU Ting¹, DONG Xiao-chen¹, LI Ya-xian¹

Author information +

History +

Abstract

A one-year study was carried out to investigate the changes in water quality parameters such as water temperature, residual chlorine, turbidity, pH, UV₂₅₄, dissolved organic carbon (DOC) and trihalomethanes(THMs) along the main pipe of a regional water supply system in Suzhou (2km sampling interval). The results of spatial scale analysis showed that the concentration of THMs at the end of the water supply pipe network at 8~10km was at a relatively high level; the results of time scale analysis showed that the concentration of THMs at the end of the water supply pipe network reaches the highest peak in June, with the concentration of 39.06 μg/L. Four typical machine learning algorithms were selected to construct the prediction model of THMs in the pipe network, and the interpretable gradient boosting tree (GBDT) algorithm was able to effectively avoid model overfitting, and the R² of the four THMs prediction accuracies reached 0.839, 0.906, 0.836, and 0.935, respectively. The further application of Shapley's additional explanation (SHAP) to the global analysis of the GBDT model revealed that the water temperature, UV₂₅₄, DOC, and the distance of the water supply were positively correlated with the generation of THMs, while the residual chlorine showed a significant negative correlation with the THMs. The local analysis of SHAP showed that residual chlorine concentration in the water supply network lower than 0.45, 0.55 and 0.55mg/L can significantly reduce the influence weights of residual chlorine on TCM, BDCM and DBCM. Based on the data-driven model constructed in this study, the generation concentration of THMs can be efficiently predicted by environmental factors, and the residual chlorine dosage can be controlled in real time to prevent the over-generation of THMs in summer or at the end of the pipe network.

Key words

water supply network / residual chlorine / distance to water supply / THMs / machine learning / SHAP analysis

Cite this article

EndNote

Ris (Procite)

Bibtex

Download Citations

XU Xiao-yan, HUO Ran-ran, LI Hui-ping, ZHOU Bai-qin, DU Zhen-qi, PANG Wei-hai, ZHOU Ting, DONG Xiao-chen, LI Ya-xian. An interpretable machine learning-based model for prediction of THMs and identification of key factors in water supply networks[J]. China Environmental Science. 2026, 46(1): 208-218

References

[1] 张金松,卢小艳.饮用水消毒工艺及副产物控制技术发展[J]. 给水排水, 2016,52(9):1-3. Zhang J S, Lu X Y. Drinking water disinfection process and by- product control technology development[J]. Water Supply and Drainage, 2016,52(9):1-3.
[2] 陈超,王玉,谢宇煊,等.饮用水水质标准升级带来的消毒副产物挑战与对策[J]. 给水排水, 2024,60(9):7-14. Chen C, Wang Y, Xie Y X, et al. Challenges and countermeasures of disinfection by-products brought by upgrading drinking water quality standards[J]. Water Supply and Drainage, 2024,60(9):7-14.
[3] 李惠平.纳滤膜在高品质饮用水处理中的应用研究[D]. 兰州:兰州交通大学, 2020. Li H P. Research on the application of nanofiltration membrane in high quality drinking water treatment[D]. Lanzhou: Lanzhou Jiaotong University, 2020.
[4] 任洋洋,钱海雷,裴赛峰,等.应用中断时间序列方法评估干预措施对陆家嘴水厂出厂水三卤甲烷的影响[J]. 环境与职业医学, 2024, 41(4):420-424. Ren Y Y, Qian H L, Pei S F, et al. Application of an interrupted time series approach to assess the effects of interventions on trihalomethanes in Lujiazui water plant effluent[J]. Environmental and Occupational Medicine, 2024,41(4):420-424.
[5] 王荀,周玉潇,韦慧慰,等.深圳市龙岗区生活饮用水中三卤甲烷健康风险评估[J]. 华南预防医学, 2024,50(7):594-598. Wang X, Zhou Y X, Wei H W, et al. Health risk assessment of trihalomethanes in drinking water in Longgang District, Shenzhen[J]. South China Preventive Medicine, 2024,50(7):594-598.
[6] Brown D, Bridgeman J, West J R.Predicting chlorine decay and THM formation in water supply systems[J]. Reviews in Environmental Science & Bio/technology, 2011,10(1):79-99.
[7] Huang R, Ma C, Ma J,et al.Machine learning in natural and engineered water systems[J]. Water Research, 2021,(21):117666.
[8] Bowden G J, Nixon J B, Dandy G C, et al. Forecasting chlorine residuals in a water distribution system using a general regression neural network[J]. Mathematical & Computer Modelling, 2003,44(5/6):469-484.
[9] Li R A, Mcdonald J A, Sathasivan A, et al. A multivariate Bayesian network analysis of water quality factors influencing trihalomethanes formation in drinking water distribution systems[J]. Water Research, 2021,(190):116712.
[10] Bowden G J, Nixon J B, Dandy G C, et al. Forecasting chlorine residuals in a water distribution system using a general regression neural network[J]. Mathematical & Computer Modelling, 2003,44(5/6):469-484.
[11] Mahya P, Johannes Fürnkranz. An empirical comparison of interpretable models to post-hoc explanations[J]. AI, 2023,4(2):426-436.
[12] Li L B, Qiao J D, Yu G, et al. Interpretable tree-based ensemble model for predicting beach water quality, Water Research, 2022,(211):118078.
[13] Yuan X Z, Suvarna M, Low S, et al. Applied machine learning for prediction of CO₂ Adsorption on biomass waste-derived porous carbons, Environmental Science&Technology, 2021,(55):11925–11936.
[14] 于林堂,陈咚咚,陶翠翠,等.基于可解释性机器学习的芬顿工艺降解有机污染物速率的研究[J]. 中国环境科学, 2025,45(8):4294-4302. Yu L T, Chen D D, Tao C C, et al. Research on the degradation rate of organic pollutants in fenton process based on explainable machine learning[J]. China Environmental Science, 2025,45(8):4294-4302.
[15] 张煦,黄霞,张晓彤,等.基于可解释机器学习方法的水库总磷预测研究[J]. 水生态学杂志, 2025-05-23:1-14. Zhang X, Huang X, Zhang X T, et al. Prediction of total phosphorus in reservoir based on interpretable machine learning methods[J]. Journal of Hydroecology, 2025-05-23:1-14.
[16] GB5750-2023生活饮用水标准检验方法[S]. GB5750-2023 Standard examination methods for drinking water[S].
[17] Li H P, Zhou B Q, Xu X Y, et al. The insightful water quality analysis and predictive model establishment via machine learning in dual- source drinking water distribution system[J]. Environmental Research, 2024,(250):118474.
[18] 陈雨桐.集成学习算法之随机森林与梯度提升决策树的分析比较[J]. 电脑知识与技术, 2021,17(15):32-34. Chen Y T. Analyzing and comparing random forest and gradient boosting decision tree of integrated learning algorithms[J]. Computer Knowledge and Technology, 2021,17(15):32-34.
[19] 周杰英,贺鹏飞,邱荣发,等.融合随机森林和梯度提升树的入侵检测研究[J]. 软件学报, 2021,32(10):3254-3265. Zhou J Y, He P F, Qiu R F, et al. Research on intrusion detection by Fusing Random Forest and Gradient Boosting Tree[J]. Journal of Software, 2021,32(10):3254-3265.
[20] Baptista M L, Goebel K, Henriques E M P. Relation between prognostics predictor evaluation metrics and local interpretability SHAP values[J]. Artificial Intelligence, 2022,(306):103667.
[21] DB32/T 4786-2024城镇供水服务质量标准[S]. DB32/T 4786-2024 Quality standards for urban water supply services[S].
[22] 徐鹏,李忠群,程战利,等.水源水消毒副产物生成势与UV₂₅₄/DOC的相关性研究[J]. 环境科学学报, 2018,38(8):3021-3026. Xu P, Li Z Q, Cheng Z L, et al. Correlation study between the generation potential of disinfection by-products and UV254/DOC in water source water[J]. Journal of Environmental Science, 2018,38(8):3021-3026.
[23] Lin H, Dai Q, Zheng L, et al. Radial basis function artificial neural network able to accurately predict disinfection by-product levels in tap water: Taking haloacetic acids as a case study[J]. Chemosphere, 2020,248:125999.
[24] 张心悦,周克梅,刘煜,等.供水管材对管壁生物膜特性和水质稳定性的影响[J]. 中国给水排水, 2022,38(9):44-51. Zhang X Y, Zhou K M, Liu Y, et al. Effect of Water Supply Pipe Material on Biofilm Characteristics and Water Quality Stability[J]. China Water & Wastewater, 2022,38(9):44-51.
[25] Li H, Huo R, Xu X, et al. Applied gradient boosting decision tree algorithms for accurate prediction of trihalomethanes: A case study in dual-sources drinking water distribution system in metropolitan[J]. Journal of Water Process Engineering, 2023,(56):104416.
[26] GB 5749-2022生活饮用水卫生标准[S]. GB 5749-2022 Hygienic standard for drinking water[S].
[27] 韦运明,吕竹,黄昭铭.城市供水管网余氯影响因素研究[J]. 供水技术, 2022,16(6):10-14. Wei Y M, Lv Z, Huang Z M. Study on the influencing factors of residual chlorine in urban water supply network[J]. Water Supply Technology, 2022,16(6):10-14.
[28] 季京宣.H市双水厂管网供水分界带定位及水质改善研究[D]. 杭州:浙江大学, 2022. Ji J X, H. Research on the localization of water supply demarcation zone and water quality improvement of dual waterworks pipe network in H city[D]. Hangzhou: Zhejiang University, 2022.
[29] 卢小艳,蔡广强,刘丽君,等.南方某河流原水中消毒副产物前体物的季节性变化[J]. 中国给水排水, 2015,31(11):51-55. Lu X Y, Cai G Q, Liu L J, et al. Seasonal variation of disinfection by-product precursors in raw water of a southern river[J]. China Water Supply and Drainage, 2015,31(11):51-55.
[30] 邹琳,周圣东,顾新春,等.加氯对不同工艺出水消毒副产物生成的研究[J]. 中国给水排水, 2019,35(5):1-7. Zou L, Zhou S D, Gu X C, et al. Study of chlorination on the generation of disinfection by-products in different process effluents[J]. China Water Supply and Drainage, 2019,35(5):1-7.
[31] Mohamed M, Mounir F, Ahmed E A. Machine learning methods for predicting argania spinosa crop yield and leaf area index: A combined drought index approach from multisource remote sensing data[J]. AgriEngineering, 2024,6(3):2283-2305.
[32] 郑震,冷东梅,郑莺,等.山仔水库环境因子与藻类群落生长的相互作用机制探究及应对建议[J]. 灌溉排水学报, 2023,42(11):131-139. Zheng Z, Leng D M, Zheng Y, et al. Exploration of the interaction mechanism between environmental factors and algal community growth in Shanzai Reservoir and suggestions for response[J]. Journal of Irrigation and Drainage, 2023,42(11):131-139.
[33] 成佩,孟勇.集成学习方法的应用与比较[J]. 统计与决策, 2023, 39(23):28-33. Cheng P, Meng Y. Application and comparison of integrated learning methods[J]. Statistics and Decision Making, 2023,39(23):28-33.
[34] 郭衍昊,窦杰,向子林,等.基于优化负样本采样策略的梯度提升决策树与随机森林的汶川同震滑坡易发性评价[J]. 地质科技通报, 2024,43(3):251-265. Guo Y H, Du J, Xiang Z L, et al. Evaluation of Wenchuan co-seismic landslide susceptibility based on gradient boosting decision tree and random forest with optimized negative sample sampling strategy[J]. Geoscience and Technology Bulletin, 2024,43(3):251-265.
[35] 王晋宇,陈玲瑚,赵辰,等.饮用水中三卤甲烷生成影响因素的初步研究[J]. 净水技术, 2009,28(6):30-34. Wang J Y, Chen L H, Zhao C, et al. Primary investigation on influencing factors of trihalomethanes generation in drinking water, water purification technology, 2009,28(6):30–34.
[36] 王玲,幸薇洁,张翔飞,等.成都市龙泉驿区生活饮用水中消毒副产物含量及其影响因素分析[J]. 预防医学情报杂志, 2025,41(6):813-818. Wang L, Xing W J, Zhang X F, et al. Analysis of disinfection by-products in drinking water and their influencing factors in Longquanyi District, Chengdu[J]. Journal of Preventive Medicine Intelligence, 2025,41(6):813-818.
[37] Dong F L, Pang Z, Yu J Q, et al. Spatio-temporal variability of halogenated disinfection by-products in a large-scale two-source water distribution system with enhanced chlorination[J]. Journal of Hazardous Materials, 2022,423(Feb.5 Pt.A):127113.1-127113.11.
[38] Fang C, Yang W Y, Lu N N, et al. Alkaline chlorination of drinking water: A trade-off between genotoxicity control and trihalomethane formation[J]. Water Research: A Journal of the International Water Association, 2023,246(Nov.1):1.1-1.10.