Design And Analysis Of The Performance Guaranteed Aggregation Based Model Predictive Controllers

Model Predictive Control (MPC), inborn with the modern industry, furnishes the control strategy by the online optimization, which manages the system constrains explicitly under the guarantees of the performance of controllers. However, because the optimization problems of MPC are solved on line, the heavy computation burden of solving the optimal problems restricts the application of MPC, especially in the large-scale industrial systems with a great mount of constrains. To apply the common model complexity reducing strategy in Dynamic Large-scale Systems, the aggregation strategy, to the MPC, the aggregated MPC is proposed. The number of optimal variables, as well as the online computation cost, in aggregated MPC is reduced by the linear mapping in the form of the aggregation matrix. Nevertheless, the introducing of aggregation leads to the decreasing of optimization degrees and that makes the performance of controllers unwarrantable.General framework of aggregation strategy covers most of current methods of optimal variable reduction. Taking the aggregation MPC described by this framework as the model, this paper first introduces the equivalent aggregation strategies for unconstrained MPC and terminal zero constrained MPC. Because the analysis form of optimal solutions to the MPC for the systems with constrains cannot be obtained, the quasi-equivalent aggregation method is studied in this paper. The design of performance improvement in quasi-equivalent aggregation strategy is analyzed via simulation and its effect is proved. An algorithm for choosing the parameter in quasi-equivalent aggregation strategy is proposed in this paper base on the simulation and the analysis. According to this algorithm, the aggregated MPC, which satisfies the performance requirement in the whole state space, is obtained by design according to its performance at several representative initial points. The application of this algorithm to the Continuous Stirred Tank Reactor (CSTR) system also approves its effect.In the end, the choice of initial points is discussed via theoretical analysis. The relation between control inputs of MPC and initial points'position in the state space is also discussed.