Incremental Learning With SVR Algorithm And Its Application In The Simulation Of Calcium Carbonate Crystallization

The effects of the carbon dioxide concentration in atmosphere on the composition of seawater and marine life living environment,has currently causing widespread concerns.In the study of seawater crystallization problems,the analysis method based on traditional chemistry experiment is rather cumbersome,and the operations are complex,which spends a greater cost.So,how to use computer technology to make simulations for the problems in chemical industry has become a hot research topic.This paper studies the incremental learning algorithm of the support vector regression and its application to the simulation of the reaction process of calcium carbonate crystallization under weak alkaline condition.The main work includes the followings.(1)Study the incremental learning method,statistical learning theory,support vector machine and the core knowledge of calcium carbonate crystallization problems.Analyze the data of chemical experiments of calcium carbonate crystallization,and explore the method how to apply the incremental learning process to the ?-support vector regression algorithm.Complete the theoretical derivation for the combination of incremental learning method and support vector regression algorithm.(2)Based on the incremental learning algorithm of support vector machine,proposed by Syed,use density-selection factors to optimize training samples,and utilize dual weight punishment method to enhance the role of support vector in incremental learning process so as to weaken the errors due to the samples discarded.Finally,this paper proposes the Sample Selectivity Dual Incremental Learning with Support Vector Regression(SSDISVR),and also validates the feasibility and effectiveness of this algorithm through the continuous learning of different samples on UCI data set.And by the comparative experiments with classic SVR and SVM algorithm,it shows that this method shortens the average training time,and the algorithm accuracy has been some improvement.(3)Employ the proposed improved algorithm to the simulation of the process of calcium carbonate reaction crystallization under weak alkaline condition,implement corresponding simulation experiments by using experimental data obtained in different temperatures,concentrations,magnesium / calcium ratios,the results of fitting analysis proved that the dual incremental support vector regression algorithm based on sample-selection has better overall performance.Meanwhile,in order to facilitate the practical application,this paper develops simulation software for the calcium carbonate crystallization.