Nonlinear System Identification in Frequent and Infrequent Operating Points for Nonlinear Model Predictive Control

Abstract

This paper studies identification of a process in both frequent and infrequent operating points to design a nonlinear model predictive controller. Although, many of industrial processes normally work around an operating point, however they should seldom work in some infrequent points as well. In this case, due to low ratio of data points, identification of the processes based on all data set results in poor identification of the infrequent operating points. To resolve this problem, in this paper, at the first step, a data clustering strategy is used to group the data in different operating points. Since the ratio of infrequent to frequent data points is extremely low, the strategy used is the fuzzy Gath-Geva clustering methodology. Then, at the second step, a new approach has been proposed to compromise performance of identification of the nonlinear model for frequent and infrequent operating points. It is shown that if the ratio of data associated with frequent operating point to data of infrequent operating point is appropriately selected, the performance of the model remains satisfactory in the frequent operating point while the performance in the infrequent operating point is significantly improved as well. The proposed method gives an interval for appropriate ratio of data set in the highly nonlinear pH neutralization process.

DOI: http://dx.doi.org/10.5755/j01.itc.42.1.997