Abstract: Performance monitoring of model predictive control(MPC)technology has received much attention in both academic and industrial circles.For model-based control technique,model accuracy is a key factor to ensure its performance.So model-plant mismatch(MPM)is important in the procedure of control performance monitoring.In recent years,the multi-step prediction error approach for MPC performance potential has been developed without requiring the knowledge of plant model and interactor matrix.However,since the MPC controller has strong robustness to MPM and other uncertainties,the multi-step prediction error approach is not sensitive to performance potential change in MPC controller which is caused by MPM.Considering that model prediction residuals of MPC controller could effectively reflect the information of MPM,an improved approach based on model prediction residuals was developed to resolve this problem.Meanwhile,a performance benchmark and real-time performance monitoring strategy based on the improved approach was defined to monitor MPC control performance.Theoretical analysis and experimental results showed that the improved approach not only possessed the advantages of the original approach,but also could better detect the change in MPC controller performance potential by MPM.Simulation example in the continuous stirred tank heater(CSTH)system illustrated the feasibility and effectiveness of this approach.

Key words: model predictive control, multi-step prediction error, model-plant mismatch, model prediction residual, real-time performance monitoring

摘要： 由于模型预测控制器对模型失配等不确定因素具有较强的鲁棒性,因此现有的多步预测误差方法不能及时显著地检测到由模型失配导致的MPC控制器性能潜能的变化。针对上述问题,提出一种改进的多步预测误差方法和实时性能监控策略。考虑到MPC控制器的模型预测残差能有效反映模型失配等信息,利用预测残差对现有多步预测误差方法进行改进,改进的方法能够更好地检测由模型失配引起的MPC控制器性能潜能的改变。在连续搅拌槽加热器(continuous stirred tank heater,CSTH)系统上的仿真实验验证了该方法的可行性与有效性。

关键词: 模型预测控制, 多步预测误差, 模型失配, 模型预测残差, 实时性能监控