基于自适应粒子群算法和支持向量机的控制图模式识别  

Recognition of Control Chart Pattern by Using Adaptive Mutation Particle Swarm Optimization and Support Vector Machine

作　　者：张敏[1] 程文明[1] 

机构地区：[1]西南交通大学机械工程研究所,四川成都610031

出　　处：《工业工程》2012年第5期125-129,共5页Industrial Engineering Journal

基　　金：中央高校基本科研业务费专项资金专题研究项目(2010ZT03);国家自然科学基金资助项目(51175442)

摘　　要：针对目前多品种、复杂化的生产趋势,提出了一种基于自适应变异的粒子群算法(AMPSO)和支持向量机(SVM)的控制图失效模式识别的方法。利用SVM小样本学习能力,设计一对一的SVM多分类器进行控制图模式识别,并利用AMPSO算法优化SVM核函数的参数。通过对10种控制图模式(6种基本模式和4种混合模式)的20维特征仿真数据对该方法进行检验,并通过与BP、SVM、PSO-SVM识别方法的对比分析。仿真试验表明该方法有效提高了控制图模式的识别精度,达到98.14%,而BP仅有75%,为控制图在线实时识别提供了一种可行的途径。Due to the complexity of production processes resulting from multi -item production, effective production control is necessary. For this purpose, an intelligent control chart pattern recognition method is proposed. This method can improve the recognition accuracy by using adaptive mutation particle swarm op- timization （AMPSO） and support vector machine （SVM） classifier. It uses one - against - one SVM multi -class classifier to recognize the control patterns because of its excellent small sample learning. Mean- while, AMPSO is used to optimize the parameters of SVM kernel function. 20 - dimension simulated data sets of ten control chart patterns, including six fundamental patterns and four mix patterns, are used to test the proposed method. Also, it is compared with BP, SVM, and PSO -SVM methods. Simulation results show that the proposed method can get high recognition accuracy, which is up to 98.14%, while it is 75% if BP is applied. This implies that it is a feasible way to recognize control chart pattern in practice.

关 键 词：控制图 模式识别 支持向量机 粒子群 

分 类 号：TH165[机械工程—机械制造及自动化] TH18