MPC Performance Assessment And Application Based On Priority Strategy

MPC (Model Predictive Control) is widely applied in industry and always appears in double-layer, one for local steady-state optimization and the other for dynamic control, in practical. MPC controller performance assessment methods based on MVC (Minimum Variance Control) benchmark or LQG (Linear Quadratic Gaussian) benchmark can be used to evaluate the potential control and economic performance, and realize control with the given reasonable minimum constraint back-off value given. At the same time, it provides the dynamic control layer with corresponding optimized control parameters, reducing the fluctuations of process variables and promoting the ability of tracking set-point, so as to improve the system economic benefits. In these methods, different objectives (e.g., safety, product quality, and economics) in general are taken into account in one weighted objective function. The tuning of the weights for different objectives, however, is not a trivial task and still remains an open question. Moreover, due to practical concerns such as safety, product quality, economic, etc., one objective may have total superiority over another. Furthermore, during the MPC steady-state optimizing process, constraints may be divided into hard ones and soft ones. In some extreme case, soft constraints or objective function with a lower priority level would be allowed overflowing or retraction within a certain grade in order to obtain the feasible region or reducing the conservatism. In this work, a prioritized LQG benchmark is developed and applied to the economic performance assessment of MPC with prioritized objectives. The main works are listed as follows:(1) Performance assessment problem of control system with constraints is studied. Take advantage of the flexibility of LMI (Linear Matrix Inequality) in solving LQG problem, a LQG benchmark solving strategy with the variance constraints of input variables and state variables obtained by LMI is proposed. On this basis, a further developed in face of no feasible domain existing, introducing prioritized constraints tuning strategy to ensure LQG benchmark problem with constraints can be solved.(2) The prioritized LQG benchmark is extended to the economic performance assessment of MPC with prioritized objectives. Based on the priority framework, economic performance assessment problems of control system are solved with constrained LQG benchmark using two-phase method. In the feasibility phase, the proposed prioritized LQG benchmark is utilized to secure the feasible domain; then in the optimization phase, the optimal economic performance of control system is calculated with lexicographic multi-objective optimization algorithm.(3) The prioritized LQG benchmark is applied into Control Performance Monitoring/Assessment (CPM/A) field. Based on the priority strategy, a prioritized CPM/A is put up. Diverse performance assessment reports and warnings with different rank are generated according to the different control conditions of different variances. The important variances are monitored first. At the end, these strategies and algorithms are applied into FCC reaction regeneration control system on the FCC simulation platform.