Online Complex Nonlinear Industrial Process Operating Optimality Assessment Using Modified Robust Total Kernel Partial M-regression
Fei Chu,Wei Dai,Jian Shen,Xiaoping Ma,Fuli Wang
DOI: https://doi.org/10.1016/j.cjche.2017.06.019
IF: 3.8
2018-01-01
Chinese Journal of Chemical Engineering
Abstract:Although industrial processes often perform perfectly under design conditions, they may deviate from the optimal operating point owing to parameters drift, environmental disturbances, etc. Thus, it is necessary to develop efficacious strategies or procedure to assess the process performance online. In this paper, we explore the issue of operating optimality assessment for complex industrial processes based on performance-similarity considering nonlinearities and outliers simultaneously, and a general enforced online performance assessment framework is proposed. In the offline part, a new and modified total robust kernel projection to latent structures algorithm, T-KPRM, is proposed and used to evaluate the complex nonlinear industrial process, which can effectively extract the optimal-index-related process variation information from process data and establish assessment models for each performance grades overcoming the effects of outlier. In the online part, the online assessment results can be obtained by calculating the similarity between the online data from a sliding window and each of the performance grades. Furthermore, in order to improve the accuracy of online assessment, we propose an online assessment strategy taking account of the effects of noise and process uncertainties. The Euclidean distance between the sliding data window and the optimal evaluation level is employed to measure the contribution rates of variables, which indicate the possible reason for the non-optimal operating performance. The proposed framework is tested on a real industrial case: dense medium coal preparation process, and the results shows the efficiency of the proposed method comparing to the existing method.