基于KPCA-MLR的烟气轮机非线性故障预测  

Nonlinear fault prognosis for stack gas turbine machine based on KPCA-MLR

作　　者：陈默[1] 马洁[1] 

机构地区：[1]北京信息科技大学自动化学院,北京100192

出　　处：《北京信息科技大学学报（自然科学版）》2013年第1期27-32,共6页Journal of Beijing Information Science and Technology University

基　　金：国家自然科学基金面上项目(61273173;51275052);北京市自然科学基金面上项目(4122029)

摘　　要：针对烟气轮机振动信号的非线性、非平稳特性,提出了一种核函数主元分析(KPCA,Kernel Principal Component Analysis)和多层递阶(MLR,Multi-Level Recursive)预测模型相结合的烟气轮机非线性故障预测方法。首先,采用非线性数据模型KPCA对烟气轮机运行状态进行故障检测;然后,采用非线性预测模型MLR分别对故障检测指标T2和SPE统计值进行趋势预测。仿真结果表明,KPCA完全适合于烟气轮机的故障检测,具有处理非线性、非平稳数据的能力,并且MLR模型预测精度较高。Aiming at solving the non-stationary and nonlinear problems of the vibration signal of stack gas turbine, a method is proposed by integrating the KPCA （Kernel Principal Component Analysis） and MLR （ Multi-Level Recursive） to predict the nonlinear fault of the stack gas turbine. The nonlinear data model KPCA is adopted to detect the fault of the state of a stack gas turbine firstly. Then the nonlinear predicted model MLR is used to predict the tendency of the detecting index T2 and SPE statistics. The simulation results show： KPCA is completly suitable for the fault prognosis of stack gas turbine and is capable of dealing with nonlinear and non-stationary data; MLR prediction accuracy is relatively high.

关 键 词：烟气轮机 核函数主元分析 MLR预测模型 故障预测 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]