基于改进网格搜索法的支持向量机在气体定量分析中的应用

Application of Support Vector Machine Based on Improved Grid Search in Quantitative Analysis of Gas

针对气体定量分析中,支持向量机建模的参数难以确定以及现有的方法历时长等问题,提出了一种改进的网格搜索法,用于建立基于红外光谱的CO2气体定量分析模型。通过对汽车尾气中CO2气体的初始数据进行优化,再将优化的核函数代入支持向量机进行浓度的回归分析。对浓度范围在0.025%~20%的20组不同浓度的CO2气体进行定量分析,并与PSO算法作对比。实验表明,采用改进的网格搜索法获得的最佳参数c=0.25,g=2.828 4,PSO获得的最佳参数c=18.302 1,g=0.01,所用时间比PSO算法节省约5倍。预测结果误差在5%以内,符合国家对尾气排放的相关标准。

According to the difficult in selecting parameter of SVM when modeling on the gas quantitative analysis, and existing methods need long time, SVM optimized by improved grid search method was proposed to built an infra- red spectrum quantitative analysis of gas. According to this method, the spectrum data of COg in vehicle exhaust is optimized. The kernel function leads SVM and calcu-late the concentration. By using improved grid search and PSO to make the contr-ast, quantitatively analyzed 20 different concentrations of CO2 in the concentration range between 0.025%~20%. The experiment results show that this method gets c = 0.25,g = 2.828 4,PSO gets c = 18.302 1, g=0.01 ,the time of modeling by improved grid search was reduced to one fifth of that of PSO optimization. And the prediction error is less than 5%, in line with national standar-ds for exhaust emissions.