Key data set selection algorithm based on PLS regression in industrial process

Abstract

In this paper, based on the traditional Partial Least Squares(PLS) algorithm, a new way to select the most effective data sets for the PLS regression is proposed. The reason why we apply this new approach is that it could maintain or even surpass the original performance of control and diagnosis of a certain process while keep the data sets as less as possible to enhance the conciseness. The most significant advantage of the proposed data set selection method is that it identifies the data sets with the most typical characteristics of the group of data, excluding less informative data. Based on the ordinary PLS algorithm and with the improvement of conciseness, a better performance on prediction and fitting could be achieved. To achieve these goals, the ordinary and the improved PLS algorithm are introduced, after which a new method of data set selection is proposed. Specifically, the enhanced effectiveness of the proposed approach could be revealed by the simulation results of a numerical case.