基于MW-KECA与变量贡献SVDD的间歇过程故障检测系统

Fault Detection System for Batch Processes Based on MW-KECA Algorithm and SVDD Method of Variable Contribution

针对间歇过程的非线性和时变性特点以及故障易误报的问题,提出了一种将移动窗-核熵成分分析(MW-KECA)故障监测与基于变量贡献的支持向量数据描述(SVDD)故障诊断集合而成的故障检测系统。MW-KECA方法构建局部模型能有效处理数据的时变性,同时保留KECA优秀的非线性处理能力。故障诊断中以各变量对CS统计量-向量间角度关系指标的贡献作为输入数据来构建SVDD分类器,相较于原始数据,故障贡献能够突出同类相似信息和异类差异信息。通过青霉素发酵仿真实验,验证了检测系统在监测准确性与故障识别率上都有良好效果,证明了该检测系统的有效性。

In view of the non-linear and time-varying characteristics of batch processes and the problem that fault diagnosis is prone to false positives,a fault detection system combining the fault monitoring method of moving window-kernel entropy component analysis(MW-KECA) and the fault diagnosis method of support vector data description(SVDD) based on variable contribution is proposed.The local model constructed by MW-KECA method can effectively deal with the time-varying data,while retaining the excellent ability of KECA algorithm to deal with nonlinear data.In fault diagnosis,SVDD fault classifier is constructed with the contribution of each variable to Cauchy-Schwarz(CS) statistics,which is vector angle relation indicator as input data.In fault diagnosis,the contribution of angle statistics between vectors,Cauchy-Schwarz(CS) statistics,of each variable is used as input data to construct SVDD classifier.Compared with the original data,fault contribution data can highlight similar information in the same category and different information in different categories.The penicillin fermentation simulation research shows that the detection system has good effect on monitoring accuracy and fault identification rate,which proves the effectiveness of the detection system.