Study On The Structure - Activity Relationship Model Of Heavy Metal Deposited By Modified Wheat Straw Based On Machine Learning

With the rapid development of modern industry, the emissions of wastewater containing heavy metal ions increased dramatically, resulting in serious pollution of surface water. The traditional methods of heavy metal wastewater treatment have been unable to take into account the economic and practical effect, so in recent years, the researchers turned to the study of biological materials, such as spent grains, application in heavy metal wastewater treatment. Because of their natural affinity for metal ions, we can use them which contain functional groups or the introduction of a specific group(i.e. modified method) by chemical reaction, to adsorb and purify the heavy metalions in the wastewater.Quantitative Structure Activity/Property Relationship is to use machine learning methods,such as partial least squares(PLS), support vector regress machine(SVR), etc., build the mathematical model to describe the relationship between the biological material molecular structure of a molecule and the biological activity, also use the model to predict the activity of molecules, and explain the mechanism from the point view of descriptors.The relationship between the molecular structure and various physical and chemical properties and biological activity of modified spent grains is very complex, it maybe linear or non-linear relationship between the two. This paper will discuss it through two different methods, feature selection and feature extraction, and eventually reach the purposes of conversion from high dimensional space to low. One is based on different screening strategies and evaluation mechanisms, and the other is based on the conversion of different spaces.The followings are the main work of this paper:1 This paper is a comprehensive interdisciplinary research topic, which need to be converted from the environmental chemistry to machine learning, and to use machine learning methods to mine and process for large data sets.2 Introducing the machine learning methods on the modification experiments of modified spent grains, modeling based on the existing experimental data and give simulation and prediction. In the face of short of the experimental data, the matrix recovery method is successful to extend the amount of data and have been validated on real world chemical experiment.3 During the feature extraction process, the introduction of new algorithm based on the mean of exponentially weighted K-NN distance of the given data to select the neighborhoodsize and the intrinsic dimension automatically in manifold learning. Through this method, it will be more effective to achieve high dimensional data dimensionality reduction with manifold learning methods such as ISOMAP, LLE, etc.4 Two widely used modeling tools SVR and PLS are used to model for descriptors,based on feature extraction and feature selection, and modification of process conditions and adsorption conditions, implement the best matches of functional groups, as well as the optimum condition selection and prediction.