Research On The Adsorption Model Of Metal Pollutants Based On Machine Learning

In recent years,the potential value for pollutant treatment has become increasingly prominent with the research and rise of machine learning.More and more scientists have turned their attention to the application of machine learning methods in pollutant removal to establish model with good performance and greatly reduce the calculation cost.Machine learning methods provide a cost-effective tool for optimizing chemical reactions,accelerating molecular design,and other important applications.The study introduces two representative adsorbent materials,graphene and vanillin,and evaluates the four machine learning methods of random forest,extreme random forest,extreme gradient boosting,and light gradient boosting from multiple dimensions.The aim is to explore adsorption energy prediction model suitable for removal of metal pollutants.The specific research content is as follows:(1)In terms of data acquisition and preprocessing,the computational chemistry software Gaussian is used to establish the geometric configuration of graphene oxide and vanillin for adsorbing metal ions.They were optimized based on density functional theory to obtain data such as adsorption energy and structural parameters,which were used for subsequent modeling.And based on the principle of singular value decomposition to fill in missing values.(2)In terms of feature selection,selected features and obtain their importance rankings by training a gradient bossting machine.It is found that 9 features such as ion radius,Millikan charge,and boiling point have the most significant effects on the adsorption energy.The work explains the important factors affecting the reaction performance of graphene oxide and vanillin to remove metal contaminants from a theoretical level.(3)In the establishment and comparison of prediction models,four widely used ensemble learning methods are used to train prediction models,including random forest,extreme random trees,extreme gradient boosting and light gradient boosting.The models were compared and analyzed from the three dimensions of goodness of fit coefficient R2,mean square error MSE and running time.The optimal adsorption energy prediction model was obtained.In order to improve the prediction accuracy of the machine learning model and reduce the risk of overfitting,the grid search method is selected for hyperparameter tuning,and the training set and the validation set are divided through cross-validation,which greatly reduces the risk of overfitting.(4)In terms of the applicability of the prediction model,the trained model is extended to vanillin adsorption system for metal ions,which further shows that the previous extreme random forest model is more suitable for pollutant removal system with oxygen-containing functional groups.In general,the research combines the advantages of machine learning to save experimental process and time.Based on less time and operating cost,it has obtained a more accurate adsorption energy,which can bring new ideas to the research of metal ion removal performance.It has a certain reference value for the experimental and theoretical research on the prevention and control of metal pollution.