Economic performance assessment of multivariable model predictive control based on linear quadratic Gaussian benchmark
基于线性二次型高斯基准的多变量预测控制技术经济性能评估

Because of the drift in parameter values and the lack of maintenance in the controlled process, the performances of the applied controller gradually deteriorate with time. An economic performance assessment is necessary to give the benefits an evaluation for ensuring system optimal operation status. Since the existing minimum variance control benchmark does not consider restriction conditions of the controller, we propose a two-layer structure model predictive control (MPC) based on the linear quadratic Gaussian (LQG) benchmark, by introducing the weighted input and the weighted output to the upper-layer economic performance index. By solving the LQG problem, we obtain a set of discrete values of the control signal and the output signal for determining their variances and fitting the Pareto optimal surface function. From the formulated propositions, we solve for the optimal economic index and the optimal set-point value. This performance assessment method has been applied to a delayed coking furnace; results show the effectiveness of the proposed approach.