遗传优化支持向量机在失火故障诊断中的应用

摘要

故障样本的缺乏严重制约智能故障诊断的发展,支持向量机算法的提出有效地解决了小样本学习问题.然而支持向量机算法中两个参数惩罚因子C和核参数γ对故障样本的准确识别起着决定性作用.针对参数较难选择问题,采用遗传算法对支持向量机中的两个参数进行全局寻优.把汽车在典型故障下尾气中各气体的体积分数作为训练样本,样本经过主成份分析实现降维和去相关.用处理过的样本和最优参数建立基于支持向量机的多元分类器模型,进行故障类别诊断.使用LIBSVM工具箱进行仿真,结果表明经遗传算法优化后的支持向量机对于小样本故障诊断有很高的准确率.%The development of intelligent fault diagnosis is restricted by the shortage of the fault sample, in this paper, a new method-support vector machine is proposed, and then Genetic Algorithms is used to optimize the important parameters-penalty factor C and kernel parameter y. In this paper, the engine fault diagnosis is regarded as training samples. The training samples realized dimension reduction and decorrelation by principal component analysis. Support vector machine model of engine fault diagnosis is based on worked samples and optimized parameters. The simulation is realized by LIBSVM. The result shows that the support vector machine has a very high accuracy.