摘要
针对在二维空间中,由变量均值偏移量较小但协方差不变所导致地故障类重叠,并因此降低故障识别率的问题,提出一种基于主成分分析(PCA)修整和线性判别分析(LDA)的新方法(PLDA).该方法通过减弱不同故障类的重叠主成分对分类的影响,达到提升LDA故障识别率的目的.对24种具有代表性故障组合的模拟样本运用10次10倍交叉验证,试验结果表明PLDA算法的平均故障识别率为94.6%,远高于传统的核化LDA算法和LDA算法的60.0%和61.9%.
In two space dimensions, faults tend to overlap each other as a result of a small shift of mean of variables with the same covariance. Moreover, overlapping could reduce the rate of identifying faults. To overcome this problem, a new algorithm, called PLDA, was proposed by integrating PCA-shaping and LDA. This algorithm declined the influences of overlapping principal-components of different fault classes on classification in order to increase the rate of LDA. The result of a validation of 24 kinds of representative simulation combinations by using 10-times and 10-fold cross validations reveals the average recognizing rate of fault is 94.6 %, compared to rates of 60.0% and 61.9% of KLDA and LDA respectively.
出处
《系统工程学报》
CSCD
北大核心
2012年第5期712-718,共7页
Journal of Systems Engineering
基金
国家自然科学基金资助项目(70931004
71002105)
关键词
PCA修整
LDA
重叠故障识别
变量均值小偏移
PCA-shaping
LDA
identification of overlapping faults
small shift of mean of variable
