Support Vector Regression Based On Genetic Algorithm And Its Application In Phase Separation Procedure With Salts

In the research of chemical separation, phase separation procedure with salts is a research focus now and it is recognized as an effective method.Currently, the theory is not yet mature and there are many difficulties in the process of development. Therefore, this paper constructed the model of multi-output support vector regression based on Genetic algorithm and applied it to therelevant prediction of phase separation procedure with salts. Following work has been done:Firstly,as an interdisciplinary research,through understanding the basic theoretic-al knowledgeof chemical separation and focusing on the methods of azeotropic organics/water mixtures’ separation,for the lackingdata of phase separation procedure with salts, this paper used computer simulation to research the liquid-liquid equilibrium of organics-water-inorganic salt system,thenmulti-output support vector regression algorithm was proposedto predict the content of organics and water after salting-out effect.Secondly,multi-output support vector regression algorithm was studied. This paper used genetic algorithm to optimize the parameters of MSVR. It realizes the automatic selection of parameters and avoids the randomness and subjectivity, so the GA-MSVR model was proposed.Related experimental design and validationare done by using simulated data of multidimensional function.Thirdly,based on theanalysis ofphase separation procedure with salts, a design idea was formed and the paper developed an application which focus on calculating separation effect.In different experimental environment, it can obtain the content of organics and water when the two phase reached equilibrium, which achieve a simulation of separation effect.Finally,in order to reduce the blindness and cost of chemical experiment, a forecasting software forphase separation procedure with saltswas developed.GA-MSVR is applied to the separation of azeotropic organics/water mixtureswhich provides a method of computer simulation for this issue.Relevant experiments were done and a promising result was got.In comparison with the multiple single output support vector regression algorithm, GA-MSVR has better prediction accuracy because it considers the relation betweenthe multiple outputs.