模型预测控制器权重参数整定非线性规划法  

Nonlinear programming method for tuning weight parameters of model predictive controller

作　　者：冯泽民 李乔 谭陆西[2,3,4] 董立春 FENG Zemin;LI Qiao;TAN Luxi;DONG Lichun(College of Safety Engineering,Chongqing University of Science and Technology,Chongqing 401331,P.R.China;School of Chemistry and Chemical Engineering,Chongqing University,Chongqing 400044,P.R.China;National-Municipal Joint Engineering Laboratory for Chemical Process Intensification and Reaction,Chongqing University,Chongqing 400044,P.R.China;Key Laboratory of Low-grade Energy Utilization Technologies&Systems,Ministry of Education,Chongqing University,Chongqing 400044,P.R.China)

机构地区：[1]重庆科技学院安全工程学院,重庆401331 [2]重庆大学化学化工学院,重庆400044 [3]重庆大学化工过程强化与反应国家地方联合工程实验室,重庆400044 [4]重庆大学低品位能源利用技术及系统教育部重点实验室,重庆400044

出　　处：《重庆大学学报》2022年第4期111-121,154,共12页Journal of Chongqing University

基　　金：国家自然科学基金项目(21776025)。

摘　　要：模型预测控制(MPC)权重参数的整定是其取得良好控制性能的关键。针对基于双层结构多目标优化的MPC权重参数整定方法存在求解过程较慢、耗时较长的问题,提出了一种非线性规划整定方法。该方法将MPC权重参数整定中每个时间采样点的MPC子优化问题等价为外层MPC权重参数整定优化问题的最优KKT(Karush-Kuhn-Tucker)条件,将MPC权重参数整定的双层多目标优化问题转化为单层非线性规划问题。仿真案例表明,基于单层非线性规划整定方法的MPC控制性能优于或近似于基于双层多目标优化整定方法的MPC控制性能;而且基于单层非线性规划的整定方法能够快速获得MPC权重参数,时间成本由基于多目标优化整定方法所需的1.0~1.5 h缩短到10~90 s。The tuning of the weight parameters on the input and output variables can significantly affect the performance of a model predictive controller(MPC)to achieve a good closed-loop dynamic response.However,the currently available approaches based on the bi-layer multi-objective optimization(MOO)for tuning MPC weight parameters are computation-consuming.In this study,a new tuning algorithm is proposed,which converts the bi-layer MOO-based approach into a single-layer nonlinear programming(NLP)problem by treating the sub-optimization problem of MPC in the lower layer as the optimal KKT(Karush-Kuhn-Tucker)condition of the optimization in the upper layer,so as to reduce the computational cost.The simulation results demonstrate that the MPC tuned by NLP method shows similar or even better performance than the MPC tuned by MOO-based method.Moreover,by using the NLP tuning method,the computational time of the MPC tuning can be significantly reduced from a range of 1.0 h to 1.5 h for the MOO-based tuning method to a range of 5 s to 90 s.

关 键 词：模型预测控制 动力学模拟 非线性规划 多目标优化 参数整定 

分 类 号：TQ021.8[化学工程]