粒子群优化SVM在气体定量分析中的应用  

Application of Support Vector Machine Optimized by Particle Swarm in Quantitative Analysis of Gas

作　　者：郑尧军[1] 陈红岩[2] 冯勇[1] 陈开考[1] 曲健[2] 

机构地区：[1]浙江经济职业技术学院汽车技术学院,杭州310018 [2]中国计量学院机电工程学院,杭州310018

出　　处：《传感技术学报》2016年第7期1121-1126,共6页Chinese Journal of Sensors and Actuators

摘　　要：针对机动车尾气排放CO气体的定量分析中,支持向量机建模的参数难以确定、光谱数据计算量过大等问题,提出了一种自适应变异粒子群优化的支持向量机方法,对浓度范围在0.5%~8%的20组不同浓度的CO气体进行定量分析。通过对汽车尾气中CO气体的初始数据进行优化,再将优化的核函数带入支持向量机进行浓度的回归分析,将结果与BP神经网络模型回归效果作对比,实验表明：粒子群寻优得到的最优参数c=39.315 2,g=0.178 55;BP神经网络的适应度值在迭代60次之后趋于稳定,SVM建模时间约为BP网络的1/30,且SVM预测精度明显高于BP网络。相比与BP网络,SVM更适合处理气体定量分析问题。For the problems of the quantitative analysis of vehicle exhaust emissions of CO gas,it is difficult to determine parameters of SVM modeling,calculate excessive data in infrared spectroscopy,and other issues. A solution of support vector machine of adaptive and mutate particle swarm optimization was proposed. 20 different groups of CO gas which concentration range from 0.5% to 8% was analyzed. According to this method,the spectrum data of CO in vehicle exhaust is optimized. The kernel function was used in SVM to analysis the concentration. Then compare the effect with the result received with the BP neural network model. The result shows that the best parameter in PSO is c=39.315 2 and g=0.178 55,the fitness of BP neural network became stable after 60 iterations,the time of modeling by SVM was about 1/30 of BP modeling,and the prediction accuracy of SVM is significantly higher than BP. Compared with BP network,SVM is more suitable for processing quantitative analysis of gas.

关 键 词：传感器应用 支持向量机 粒子群优化 BP神经网络 遗传算法 

分 类 号：TH744[机械工程—光学工程]