Linearity Decomposition-Based Cointegration Analysis for Nonlinear and Nonstationary Process Performance Assessment

The assessment of operating performance plays an important role in guaranteeing high efficiency, low cost, large economic benefit, etc., for modern industry under routine operating conditions. However, the nonlinear and nonstationary properties that widely exist in real industrial processes, make it a great challenge to obtain accurate assessment of operating performance. To solve the problem of performance assessment for nonstationary processes with nonlinearly cointegrated relationships, the linearity decomposition based cointegration analysis (CA) is proposed through multi-objective optimization in the present work. First, the nonstationary critical-to-performance variables are selected to remove the redundant information and improve the accuracy for assessment. Second, the selected nonstationary critical-to-performance variables are then decomposed into a set of local sub-blocks. Each sub-block contains a group of linearly cointegrated variables, based on which a two-layer multiblock assessing model can be established for nonlinear and nonstationary process using CA. Bayesian inference based criterion is adopted to indicate the operating performance for online assessment. Finally, the feasibility and efficacy are illustrated via a numerical example and a pulverizing process of a real thermal power plant.