基于KPCA残差方向梯度的故障检测方法及应用

发布时间：2018-11-18 18:15

【摘要】：针对核主元分析(KPCA)在应用过程中非线性映射不存在原像、故障变量无法辨识、工程应用困难等问题,提出了一种改进的KPCA残差方向梯度故障检测方法。利用主元统计量和残差统计量的偏微分之间存在着相关性这一性质,对与主元统计量相关的格拉姆矩阵偏微分中间计算过程进行优化,提出一种新的KPCA残差方向梯度算法,在此基础上结合统计量形成系统故障检测的新方法。非线性系统仿真表明,改进的KPCA残差方向梯度法不仅具有较优的故障变量辨识能力,还极大地减小了计算量,缩短了计算时间。大型热力系统的应用进一步表明,无论对于单故障和多故障的情况,方法均具有较好的故障检测能力,并且不存在残差污染,易于工程实现。
[Abstract]:A modified KPCA residual direction gradient fault detection method is proposed to solve the problems of nonlinearity mapping of kernel principal component analysis (KPCA) in the application process, failure variable identification and engineering application difficulties. By using the property of correlation between principal component statistics and partial differential of residual statistics, the Gramm matrix partial differential intermediate calculation process related to principal component statistics is optimized, and a new KPCA residual direction gradient algorithm is proposed. On the basis of this, a new method of system fault detection is formed by combining statistics. The simulation of nonlinear system shows that the improved KPCA residual direction gradient method not only has better identification ability of fault variables, but also greatly reduces the computational complexity and time. The application of large scale thermal system further shows that the method has better fault detection capability and no residual pollution, which is easy to be realized in engineering.
【作者单位】：南京工程学院电力仿真与控制工程中心;东南大学能源热转换及其过程测控教育部重点实验室;华北电力科学研究院有限责任公司;
【基金】：国家自然科学基金(51176030) 南京工程学院引进人才科研启动基金(YKJ201445)项目资助
【分类号】：TP277

【相似文献】