机器学习预测有机水污染物光催化降解速率  

Machine learning based prediction the photocatalytic degradation rate of organic water pollutants

作　　者：朱炜[1] 王嘉伟 张梦源 杨旭东 宋振阳 李庆[1] ZHU Wei;WANG Jiawei;ZHANG Mengyuan;YANG Xudong;SONG Zhenyang;LI Qing(School of Environmental and Chemical Engineering,Xi’an Polytechnic University,Xi’an 710048,China)

机构地区：[1]西安工程大学环境与化学工程学院,陕西西安710048

出　　处：《纺织高校基础科学学报》2024年第1期26-33,共8页Basic Sciences Journal of Textile Universities

基　　金：陕西省自然科学基础研究面上项目(2022JM-092);西安市科技计划先进制造业技术攻关项目(21XJZZ0015);西安市碑林区应用技术研发项目(GX2212)。

摘　　要：为了预测有机污染物的光催化降解速率,探究污染物分子结构与其降解速率之间的构效关系,设计了一种基于分子指纹的机器学习模型。该模型使用81种有机污染物的523条记录作为模型数据,将污染物MACCS分子指纹与5种实验条件(辐照度、温度、催化剂用量、污染物初始浓度和pH值)作为输入特征,采用10种机器学习算法进行建模。结果显示LightGBM算法性能最佳(R~2=0.909 4)。利用沙普利加法解释(Shapley additive explanations, SHAP)框架评估了各输入特征对光催化降解速率的贡献程度,探讨了各输入特征影响光催化降解速率的具体原因。分析表明,在光催化降解中污染物本身结构特征是影响光催化降解速率的主要原因。而且结构中含有卤素原子、N原子和不饱和碳的污染物分子降解速率最快,而结构中含有醚键或羰基的污染物分子降解速率最慢。To predict the photocatalytic degradation rate of organic pollutants,and investigate the structure-activity relationship between pollutant molecular structures and degradation rate,a machine learning model based on molecular fingerprints was developed.The model utilized a dataset comprising 523 records of 81 organic pollutants,with MACCS molecular fingerprints of pollutants and five experimental conditions(irradiance,temperature,catalyst dosage,initial pollutant concentration,and pH value)serving as input features.Ten machine learning algorithms were employed for modeling purposes.The results demonstrated that the LightGBM algorithm exhibited the highest performance(R 2=0.9094).Additionally,the contribution of each input feature to the degradation rate was evaluated using the SHAP framework,to elucidate the specific factors influencing the degradation rate.The analysis revealed that the inherent structural characteristics of the pollutants played a pivotal role in determining the rate.Moreover,the pollutants containing halogen atoms,nitrogen atoms,and unsaturated carbon within their molecular structures exhibited the fastest degradation rate,whereas those containing ether bonds or carbonyl groups exhibited the slowest degradation rate.

关 键 词：有机污染物 光催化 分子指纹 构效关系 机器学习 轻量级梯度提升 

分 类 号：O644.1[理学—物理化学] TP181[理学—化学]