ZHUJing-yi,HUANGYuan-xi,YIQi-hang,et al.Prediction of Reactivity between Potassium Permanganate and Organic Pollutants Based on Machine Learning[J].China Water & Wastewater,2024,40(11):41-48.
基于机器学习的高锰酸钾与有机污染物反应活性预测
- Title:
- Prediction of Reactivity between Potassium Permanganate and Organic Pollutants Based on Machine Learning
- Keywords:
- potassium permanganate; organic pollutant; reaction rate constant; machine learning; random forest (RF) algorithm; SHAP
- 摘要:
- 高锰酸钾（KMnO4）氧化能力强、性质稳定，是自来水厂处理微污染原水和突发污染事件的常见应急氧化药剂，高锰酸钾与有机污染物的反应速率常数（k值）是衡量其与有机污染物反应活性的重要参数。然而有机污染物种类繁多，单纯依靠试验方法或理论计算获得大量k值费时费力且成本高。为此，收集了574种有机污染物的数据，以分子指纹（MFs）和pH作为输入特征，利用4种算法（随机森林RF、神经网络NN、极致梯度提升XGBoost和支持向量机SVM），分别建立了预测k值的机器学习（ML）模型，发现4个模型均有良好的稳健性和预测能力，其中RF模型的预测性能最好（R2测试集=0.818，RMSE测试集=0.476）。采用SHAP方法对RF模型进行了模型解释，发现供电子基团、pH等对k值的预测影响最大，表明模型正确学习了k与污染物结构之间的关系。同时，计算了RF模型的应用域（AD）以验证其适用范围。该模型可准确、便捷地获得k值，有助于了解高锰酸钾与有机污染物的反应活性。
- Abstract:
- Potassium permanganate (KMnO4) has strong oxidation ability and stable properties, and is a common emergency oxidation agent for treating micro-polluted raw water and sudden pollution in waterworks. The reaction rate constant (k value) of potassium permanganate to organic pollutants is an important parameter to evaluate its reactivity with organic pollutants. However, there are many kinds of organic pollutants, and it is time?consuming and costly to determine a large number of k values simply by means of experiments or theoretical calculations. This paper collected a total of 574 organic pollutants, and employed four algorithms, such as random forest (RF), neural network (NN), extreme gradient boosting (XGBoost) and support vector machine (SVM), with molecular fingerprints (MFs) and pH serving as input features to establish machine learning (ML) models for predicting k values. The four models all had good robustness and prediction ability, among which the RF model demonstrated the best prediction performance (R2test=0.818, RMSEtest=0.476). The SHAP (SHapley Additive exPlanations) method was utilized to analyze the RF model, revealing that electron-donating groups and pH exerted the most significant influence on the prediction of k value. This indicated that the model correctly learned the relationship between k values and the chemical structures of organic pollutants. Simultaneously, the application domain (AD) of the RF model was calculated to validate its suitability. The developed model can accurately and conveniently determine the k values, aiding in understanding the reactivity of potassium permanganate with organic pollutants.
相似文献/References:
[1]郭钦,庞素艳,姜成春,等.KMnO4氧化降解2,4-二氯酚的产物及机理推测[J].中国给水排水,2018,34(21):53.
GUO Qin,PANG Su yan,JIANG Cheng chun,et al.Degradation of 2,4-Dichlorophenol by Potassium Permanganate: Oxidation Products and Tentative Mechanisms[J].China Water & Wastewater,2018,34(11):53.
[2]邓俊杰,劳志雄,陈明良,等.强化老旧水厂常规处理工艺优化提升出水水质[J].中国给水排水,2021,37(2):85.
DENG Jun-jie,LAO Zhi-xiong,CHEN Ming-liang,et al.Optimizing and Improving Effluent Quality by Strengthening Conventional Water Treatment Process in an Aging Waterworks[J].China Water & Wastewater,2021,37(11):85.
[3]李思莹,贾学斌,张军.污水厂污泥堆肥有机污染物降解及土地利用生态风险[J].中国给水排水,2023,39(4):18.
LISi-ying,JIAXue-bin,ZHANGJun.Research Progress on Degradation of Organic Pollutants in Sewage Sludge Composting and Ecological Risk of Land Application[J].China Water & Wastewater,2023,39(11):18.
[4]胡凯,边玛拉珍,王威,等.聚氯乙烯和磺胺嘧啶复合污染对活性污泥的影响[J].中国给水排水,2024,40(3):79.
HUKai,BIANMALa-zhen,WANGWei,et al.Effect of Polyvinyl Chloride and Sulfadiazine Combined Pollution on Activated Sludge[J].China Water & Wastewater,2024,40(11):79.