Performance Assessment And Monitoring Of MPC With Model-Plant Mismatch

Model predictive controller has been widely used in the industry processes,such as, refinery and chemical factory. The performance of control system will directly affect the level of safety and economic benefits. The process model plays an important role in the model predictive control. The study of performance assessment and monitoring of MPC with model-plant mismatch is very important. In this paper, dynamic matrix control (DMC) arithmetic is adopted. The studies of performance assessment based on cross-correlation analysis and a data-based eigenvalue analysis and hypothesis testing methods are simulated on Wood-Berry tower process.First, in this paper, the theory based on cross-correlation analysis is applied to the performance assessment of DMC with model-plant mismatch. In order to make dithering signal of each channel independently in multi-variable system, introduce an input dithering signal as the new control variables, then eliminate the correlation between the dithering signals from the impact of predictive bias. Analyze the cross-correlation between excitation signal of manipulated variable and prediction error to determine the mismatch of transfer function matrix. Simulation graphs of cross-correlation coefficients distribution show the results of performance assessment with model-plant mismatch.Next, a data-based eigenvalue analysis is adopted for performance assessment and monitoring of DMC with mismatch. Generalized eigenvalue analysis is used to extract corresponding eigenvectors based on the outputs data of benchmark period and monitored period. A statistical inference method is further developed for the generalized eigenvalues and the corresponding confidence intervals are derived from asymptotic statistics. The covariance-based performance indices within the isolated worse and better performance subspaces are then derived to assess the performance degradation or improvement. Also, Simulation graphs of confidence interval distribution show the results of performance assessment and monitoring with mismatch.Last, a hypothesis testing methodology is used for the performance assessment and monitoring of MPC with mismatch according to closed-loop operating data. The problem of model-plant mismatch is transformed into the effectiveness of the model. Also apply hypothesis testing to determine whether presence or absence of the mismatch. Simulation graphs of t-statistic distribution in the confidence interval show the performance assessment information of the mismatch.Three methods mentioned above have been all applied to the Wood-Berry tower process, successfully. The mismatches, such as gain mismatch, time constant mismatch and delay mismatch, are all analyzed in each method. These simulation results show the feasibility and effectiveness of the methods above.