摘要：

工艺参数间的非线性耦合关系，给生产过程的状态识别带来了很大的困难。为此，引入新的核映射准则，利用梯度优化方法选取核参数，并采用核Fisher方法进行降维处理，实现对生产状态在可视平面上的逐层多故障分类，完成对当前生产过程的状态诊断。利用TE数据进行实验验证，结果表明，与核主成分分析方法相比，该方法可以得到更加准确的诊断结果。

关键词: 核函数, 核Fisher判别, 生产过程诊断, 质量分类, 核参数选取

Abstract:

Because of the nonlinear coupling relationship among process parameters, it is difficult for condition recognition of production process. A new criterion of kernel mapping is proposed, and the gradient optimum algorithm is used to select kernel parameter. Kernel Fisher discriminant analysis is applied to reduce dimension for realizing the multi-classification in the visual plane with layer by layer analysis method. The judgment on the current state can be realized. The Tennessee Eastman(TE) process data are used for model validation, the proposed method has more satisfactory performance on condition diagnosis, compared with kernel principal component analysis.

Key words: kernel function, kernel Fisher discrimination, diagnosis of production process, quality classification, selection of kernel parameters

中图分类号: 

• TP277