基于机器学习的高锰酸钾与有机污染物反应活性预测  

Prediction of Reactivity between Potassium Permanganate and Organic Pollutants Based on Machine Learning

作　　者：朱静怡 黄苑曦 衣启航 伍洋涛 卜令君 周石庆 ZHU Jing‑yi;HUANG Yuan‑xi;YI Qi‑hang;WU Yang‑tao;BU Ling‑jun;ZHOU Shi‑qing(Hunan Engineering Research Center of Water Security Technology and Application,Changsha 410082,China;Hunan University Design and Research Institute Co.Ltd.,Changsha 410082,China)

机构地区：[1]水安全保障技术及应用湖南省工程研究中心,湖南长沙410082 [2]湖南大学设计研究院有限公司,湖南长沙410082

出　　处：《中国给水排水》2024年第11期41-48,共8页China Water & Wastewater

基　　金：国家自然科学基金资助项目(52100007)。

摘　　要：高锰酸钾(KMnO_(4))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(R测试集2=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。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.

关 键 词：高锰酸钾 有机污染物 反应速率常数 机器学习 随机森林算法 SHAP 

分 类 号：TU991[建筑科学—市政工程]