基于提升小波和递推LSSVM的实时故障诊断方法  

Real-time fault diagnosis approach based on lifting wavelet and recursive LSSVM

作　　者：杨青[1,2] 田枫[2] 王大志[1,3] 吴东升[1,2] 王安娜[3] 

机构地区：[1]长春理工大学光电工程学院,长春130022 [2]沈阳理工大学信息科学与工程学院,沈阳110159 [3]东北大学信息科学与工程学院,沈阳110004

出　　处：《仪器仪表学报》2011年第3期596-602,共7页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(61050006)资助项目

摘　　要：提出了一种基于提升小波(LW)与递推最小二乘支持向量机(RLSSVM)相集合的实时故障诊断方法(LW-RLSSVM)。该方法首先通过提升小波变换对数据实时去噪,再通过实时算法训练最小二乘支持向量机分类器。由于采用了递推算法,节省了存储空间和运算时间,同时增加了诊断模型的适应性。为验证所提方法的有效性,将LW-RLSSVM应用于TE过程和青霉素发酵过程。实验结果表明,LW-RLSSVM集合方法能有效实现实时故障诊断,在诊断速度和适应性方面,优于基于第一代小波与LSSVM相集合(W-LSSVM)的故障诊断方法;在诊断精度等方面,该方法优于LSSVM、RLSSVM等方法。An ensemble fault diagnosis method based on lifting wavelet（LW） and recursive least squares support vector machine（RLSSVM）,called LW-RLSSVM,is proposed to realize real-time fault diagnosis for complex industrial processes.Firstly,data are denoised by lifting wavelet transform in real time,then least squares support vector machine classifier is trained using real-time algorithm.With recursive algorithm,storage space is saved and computing time is shortened,while the adaptability of diagnostic model is increased.To validate the performance and effectiveness of the proposed scheme,LW-RLSSVM was applied to TE process and penicillin fermentation process.Experiment results show that the LW-RLSSVM algorithm can achieve real time fault diagnosis;it is superior to conventional ensemble approach based on traditional wavelet and LSSVM（W-LSSVM） in diagnosis rate and adaptability,and is superior to LSSVM and RLSSVM algorithms in diagnosis precision.

关 键 词：实时故障诊断 提升小波 递推SVM 集合故障诊断 LW-RLSSVM 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置]