CAJ | 학술논문

为了诊断多元控制图发出的报警信号是由哪一个或者哪些变量组合发生均值偏移引起的,提出了基于粒子群优化(PSO)算法的支持向量机(SVM)多元控制图均值偏移诊断模型.模型中使用2T 控制图对多元过程进行控制,在假设过程方差-协方差矩阵保持不变的前提下,根据不同的均值偏移模式,产生 SVM 训练数据集和测试数据集,用 PSO 对 SVM 的参数进行优化,最终得到优化的 SVM 模型.结果表明,基于粒子群优化算法的支持向量机模型(SVM-PSO)比基于SVM和基于神经网络(ANN)模型的分类能力更强,分类准确率超过85%.
위료진단다원공제도발출적보경신호시유나일개혹자나사변량조합발생균치편이인기적,제출료기우입자군우화(PSO)산법적지지향량궤(SVM)다원공제도균치편이진단모형.모형중사용2T 공제도대다원과정진행공제,재가설과정방차-협방차구진보지불변적전제하,근거불동적균치편이모식,산생 SVM 훈련수거집화측시수거집,용 PSO 대 SVM 적삼수진행우화,최종득도우화적 SVM 모형.결과표명,기우입자군우화산법적지지향량궤모형(SVM-PSO)비기우SVM화기우신경망락(ANN)모형적분류능력경강,분류준학솔초과85%.
To detect the variable shifts causing out-of-control signals in multivariate control chart, this paper proposes a model of support vector machine (SVM) monitoring the mean shifts of multivariate control charts based on particle swarm optimization (PSO) algorithm. Under the assumptions that the variance matrix is constant, T2 control chart is used to monitor the multivariate process. Based on different mean shift patterns, the sample data are generated. Finally, the optimized model is attained after the parameters of SVM are optimized using PSO. The simulation comparative studies show that the classification ability of the proposed SVM-PSO method outperforms that of the SVM based model and artificial neural network (ANN) based model. The classification rate of the proposed method is higher than 85%.