论文部分内容阅读
针对核主元分析(KPCA)中的复杂运算和故障分离问题,提出一种基于免疫核主元分析(immune-KPCA,IKPCA)的故障诊断方法。该方法使用小波变换技术对数据进行预处理,然后利用基于克隆选择原理的免疫算法对建模数据进行压缩,提取特征样本建立核矩阵以降低运算复杂程度。在IKPCA监控统计量检测到故障后,基于灵敏度分析思想构造贡献图分离故障变量。在连续搅拌反应釜(CSTR)仿真过程上的应用结果表明,本文提出的方法能够显著地降低核矩阵的计算量,比传统的PCA、KPCA方法更有效地检测过程故障,而且能够正确地识别故障变量。
Aiming at the complex operation and fault separation in kernel principal component analysis (KPCA), a fault diagnosis method based on immune-KPCA (IKPCA) is proposed. This method uses wavelet transform to preprocess the data, and then uses the immune algorithm based on the principle of clonal selection to compress the modeling data and extract the characteristic samples to establish the nuclear matrix to reduce the computational complexity. After the IKPCA monitoring statistics detect a fault, construct a contribution graph to separate the fault variables based on the sensitivity analysis. The application results of the CSTR simulations show that the proposed method can significantly reduce the computational complexity of the nuclear matrix, detect the process faults more effectively than the traditional PCA and KPCA methods, and can correctly identify the faults variable.