摘要
为了解决多工况、非线性工业过程的故障检测问题,在基于先验知识的基础之上提出了基于多核主元分析方法(Multiple-Kernel Principal Component Analysis,KPCA)的故障检测办法。首先收集每个工况稳态过程的历史正常数据,直接建立子KPCA模型求得各自的控制限,其次收集工况间的过渡过程的历史正常数据,采取加权平均法求其控制限,最后对过程的故障数据进行检测。以田纳西-伊斯曼过程(Tennessee-Eastman Process,TEP)与MATLAB结合进行仿真实验。仿真结果表明,与单工况、非线性过程进行相比,该方法更为快速、准确。
In order to solve the problem of fault monitoring on multiple-condition and nonlinear industrial process, a fault detection method based on KPCA (Multiple-Kernel Principal Component Analysis) is proposed based on a prior knowledge. Firstly, the historical data of each normal steady-state process operating conditions are collected and then subsidiary KPCA model is created directly to obtain each respective control limit. The historical data of transition process between two conditions is collected later to seek its control limit by using weighted average method. At last, the obtained control limits are applied to monitor failure process. Simulation experiment is carried out based on Tennessee-Eastman process (TEP) combined with MATLB. Compared with the single condition and nonlinear process, the results show that this method works faster and more accurately.
出处
《北京信息科技大学学报(自然科学版)》
2016年第6期32-36,共5页
Journal of Beijing Information Science and Technology University
基金
国家自然科学基金资助项目(61273173)
关键词
多工况
非线性
多核主元分析方法
子KPCA模型
加权平均法
multiple-condition
nonlinear
multiple-kernel principal component analysis method
subsidiary KPCA model
weighted average method