基于自适应变异粒子群优化的SVM在混合气体分析中的应用

针对混合气体定量分析中，支持向量机建模的参数难以确定、红外光谱数据计算量过大以及气体间交叉干扰等问题。提出了一种自适应变异粒子群优化的支持向量机方法，用于建立基于红外光谱的多组分混合气体定量分析模型。混合气体主要由浓度范围在0.5%~8%的CO、3.6%~12.5%的CO2及200×10-6~3270×10-6的C3H8三种组分气体组成。利用粒子群优化算法对支持向量机建模中的参数进行优化选择，并与遗传算法优化的支持向量机作对比。实验表明，采用此方法建模所用时间为39.524 s，遗传算法为26.272 s；针对CO2独立建模的预测结果，粒子群优化算法均方差为0.000123758，遗传算法均方差为2.14952。在建模时间略高的情况下，粒子群优化算法预测结果均方差明显低于遗传算法。

Application of Support Vector Machine Based on Adaptive Mutation Particle Swarm Optimization in Analysis of Gas Mixture

For the difficult in selecting parameter of SVM modeling,the data calculation excessive in infrared spec?troscopy,as well as crosstalk between gases and other issues in the quantitative analysis of mixed gas. A solution of adaptive mutation particle swarm optimization support vector machine was proposed. It was to establish the models of a multi-component mixture gases quantitative analysis based on infrared spectroscopy. Multi-component mixture gases are composed of CO,with the concentration range from 0.5%to 8%;CO2,with the concentration range from 3.6%to 12.5%;C3H8,with the concentration range from 200×10-6 to 3270×10-6.Use the Particle swarm optimization algorithm to optimize select the parameters in support vector machine modeling,and compare the support vector machine modeling parameters in genetic algorithm optimization. Experiments show that it takes 39.524 s for modeling and it takes 26.272 s with genetic algorithm;for the predict results of CO2 in independent modeling,the variance of PSO algorithms is 0.000 123 758,the variance of genetic algorithms is 2.149 52. In the case of modeling time slightly higher,the predict results were sig?nificantly lower than the variance of the genetic algorithm.