摘要：

Fisher判据分析(FDA)是工业过程故障诊断的常用降维算法，但其无法提取数据中的非线性关系。基于核的FDA(KFDA)是对FDA的非线性推广，但在FDA和KFDA小样本情况下都会面临矩阵的病态问题，正则化技术是解决该问题的有效途径。推导一种转化为方程组求解并表示成对偶形式的核正则化FDA算法，给出其用于故障诊断的流程，并针对TE过程的故障数据进行仿真。

关键词: 故障诊断, 正则化, Fisher判据分析, 核方法

Abstract:

Fisher Discriminant Analysis(FDA) is a widely used dimensionality reduction technique for fault diagnosis in industry process. Kernel FDA(KFDA) is developed to solve nonlinear problem. Small sample size(3S) problem is arisen in both FDA and KFDA. Regularized FDA is an effective solution for 3S problem. A novel kernel form of regularized FDA which transforms it to equation solving problem and expresses in the dual form is obtained and implementation procedure for fault diagnosis is given. Experimental results on the Tennessee Eastman(TE) process show that the algorithm is valid, and regularized KFDA outperforms FDA for 3S problem.

Key words: fault diagnosis, regularization, Fisher Discriminant Analysis(FDA), kernel method

中图分类号: 

• TP277