摘要
利用机器学习对已发表文献中的数据进行训练,可以精确且迅速地预测特定吸附剂对水中磷的吸附,避免耗时的测试。文章采集吸附剂的8个物化特征以及4个环境因素,训练所得模型可以精确预测吸附量和吸附容量,R2最大分别为0.997和0.999,RMSE和MAE最小仅为0.001。吸附量主要受制于吸附剂的物理特性,环境因素次之。吸附容量主要受制于吸附剂的物化特性。与现有研究对比发现,将磷在多数吸附剂上的吸附归因于化学因素有待商榷。从模型解释的结果而言应是以物理吸附为主。提高磷的吸附量可从改变环境因素着手,而吸附容量的提升应对吸附剂物理特性的改善予以更多考虑。依据特征重要性排序简化特征输入对吸附量和吸附容量预测效果仍然优异,尤其是对低吸附量和吸附容量的预测精准,R2最大分别为0.970和0.974。该研究以机器学习方法创建和简化了特定吸附剂对磷的吸附量和吸附容量的预测,并解释了不同变量的重要性以及影响,对于快速预测磷的吸附提供了必要补充。
The use of machine learning to train data that collected from the existed database for constructing the predictive framework can accurately and swiftly predict the adsorption performance on phosphate,avoid time-consuming process.In this study,8 physiochemical features of adsorbents and 4 environmental features are collected to construct the predictive model for the prediction of adsorption amount of phosphate and adsorption capacity of the adsorbent.The trained model accurately predicts the adsorption amount and capacity with the maximum R2 of 0.997 and 0.999,and the minimum RMSE and MAE of 0.001,respectively.The adsorption amount is primarily subject to the physical features of adsorbents and the environmental features,while the adsorption capacity is mainly controlled by the physiochemical features of adsorbents.Compared with the current opinion that the adsorption of phosphate is primarily delivered by chemisorption,this study finds that physisorption is the primary contributor to the adsorption mechanism based on the interpretation of the trained model.And thus,the promotion of adsorption amount depends on the improvement of adsorption environment,while the promotion of adsorption capacity should pay more attention on the modification of physical properties of adsorbents.The simplified models with the most influ-ential features as the inputs also deliver the accurate prediction especially for small values of adsorption amount and capacity with the maximum R2 of 0.970 and 0.974 for amount and capacity,respectively.This study develops and simplifies the predic-tive frameworks for adsorption amount and capacity,and the importance of different features are sorted and evaluated,which enlarges the knowledge on the way to quickly predict phosphate adsorption.
作者
蒋佰果
李惠平
周佰勤
庞维海
JIANG Baiguo;LI Huiping;ZHOU Baiqin;PANG Weihai(Weifang Municipal Engineering Design and Research Institute Limited,Weifang 261000,China;Key Laboratory of Yangtze River Water Environment,Ministry of Education,Shanghai 200092,China;School of Environmental Science and Engineering,Tongji University,Shanghai 200092,China;School of Municipal and Environmental Engineering,Harbin Institute of Technology(Shenzhen),Shenzhen 518055,China)
出处
《环境科学与技术》
CAS
CSCD
北大核心
2024年第5期77-89,共13页
Environmental Science & Technology
基金
国家自然科学基金(52270188)。
关键词
磷
吸附
机器学习
模型解释
模型简化
phosphate
adsorption
machine learning
model explanation
model simplification
