Solution Prediction And Knowledge Graph Construction Based On Hybrid Machine Learning Algorithm

Solubility is an important physical property.As an essential evaluation index in the fields of chemical material production and drug research and development,solubility plays an important role in ensuring the availability of industrial process substances and is an important factor in realizing the control and optimization of industrial process.The solubility is affected by many factors,and the prediction of solubility by experimental measurement and traditional physical model cannot meet the requirements of modern industrial production.With the development of machine learning,the related research of dissolution prediction has a new direction.From the perspectives of external factors and internal factors,this paper predicts the solubility of carbon dioxide in polymers and the solubility of compounds in water by using different machine learning models,and builds the knowledge graph of solubility-related content.The main research work is as follows:(1)Taking temperature and pressure as the influencing conditions of the solubity,the traditional particle swarm optimization algorithm was improved by adaptive weight,and the improved APSO was used to optimize the regularization parameters and kernel function tuning parameters of the least square support vector machine(LSSVM).The APSO-LSSVM model was established to predict the solubility of carbon dioxide in 8 different polymers and compared with the traditional machine learning model.The results show that APSO-LSSVM model has good predictive ability.(2)With the molecular fingerprint of compounds as the influencing factor of solubility,the Cuckoo algorithm(CS)was used to optimize the superparameters of Light GBM,and the CS-Lightg BM model was established to predict the water solubility of 2446 organic compounds.Compared with five different machine learning models(RF,GBDT,XGBoost,Light GBM,BO-Light GBM),the results show that CS-Light GBM model has better performance in predicting the water solubility of compounds.(3)Proposed the concept of solubility knowledge graph,elaborated the construction process of solubility knowledge graph,and prospected the future development of solubility knowledge graph.The research work in this paper has made a certain contribution to the application of machine learning technology in the field of dissolution prediction,and opened up a way for the embedding of knowledge map in the field of dissolution prediction,and injected more possibilities for chemical material processing,drug research and development,biological information and other fields.