Adaptive clustering of production state based on kernel entropy component analysis

Abstract

Listen

Translate article

In case of the unknown production quality information, the clustering method with process data is used to acquire the production status. Feature extraction is an important factor to ensure the accurate rate of clustering. As a common non-linear feature extraction method, kernel principal component analysis uses the variance as the information metric. Because the variance is not always effective in some cases, the Renyi entropy is used as the information metric to extract feature in this paper. And then an adaptive clustering method based on the maximization of the difference between within-class and between-class scatter of angular distance is proposed. Simulation data, Tennessee Eastman process data and hot strip rolling process data are used for model validation. As a result, the proposed method has better performance on feature extraction, compared with kernel principal component analysis.