上海交通大学学报(自然版) ›› 2018, Vol. 52 ›› Issue (10): 1324-1332.doi: 10.16183/j.cnki.jsjtu.2018.10.022

• 学报(中文) • 上一篇    下一篇

基于性能触发的双层结构模型预测控制

王琳,邹媛媛,李少远   

  1. 上海交通大学 自动化系;系统控制与信息处理教育部重点实验室, 上海 200240
  • 通讯作者: 李少远(1965-),男,教授,电话(Tel.): 13818686162;E-mail: syli@sjtu.edu.cn.
  • 作者简介:王琳(1993-),女,陕西省宝鸡市人,硕士生,主要从事模型预测控制研究.
  • 基金资助:
    国家自然科学基金(61590924,61773162,61833012),上海市自然科学基金(18ZR1420000)资助项目

Two-Layered Model Predictive Control with Performance-Based Trigger

WANG Lin,ZOU Yuanyuan,LI Shaoyuan   

  1. Department of Automation; Key Laboratory of System Control and Information Processing of Ministry of Education, Shanghai Jiao Tong University, Shanghai 200240, China

摘要: 双层控制结构广泛应用于工业过程控制,针对实际系统中不确定性导致的性能下降和控制不可行的问题,提出一种事件触发的动态实时优化(Dynamic Real-Time Optimization, D-RTO)与模型预测控制(Model Predictive Control, MPC)双层结构.上层采用经济模型预测控制(Economic MPC, EMPC)对目标函数进行优化,计算最优参考轨迹并传给下层;下层采用MPC使系统跟踪上层轨迹;通过构建基于性能指标的事件触发条件来及时补偿不确定性造成的系统性能损失,无需等到D-RTO的下一次优化.当实际系统性能指标与上层优化的性能指标之差超出阈值时,需要重新求解EMPC,并基于当前状态更新参考轨迹.在此基础上,进一步分析了事件触发的D-RTO与MPC双层控制结构的可行性和闭环稳定性.数值模拟实验验证了方法的有效性.

关键词: 模型预测控制, 双层控制结构, 动态实时优化, 事件触发, 不确定性

Abstract: In this study, we propose a two-layered control framework integrating event-triggered dynamic real-time optimization (D-RTO) and model predictive control (MPC), addressing the issue of infeasibility and performance loss caused by uncertainties. The upper layer utilizes economic MPC (EMPC) to minimize an economic cost function, calculating reference trajectories. The lower layer utilizes MPC to steer the plant to track the reference trajectories. A triggering criterion based on cost function is proposed to compensate for performance loss in time, without waiting for the next D-RTO optimization. When the deviation between these two layers’ cost function exceeds a pre-set threshold, the EMPC is triggered to excute optimization and update the reference trajectory based on the current system states. Feasibility and closed-loop stability of the proposed two-layered control framework have also been proved. Simulations demonstrate its effectiveness.

Key words: model predictive control (MPC), two-layered control framework, dynamic real-time optimization (D-RTO), event-triggered, uncertainty

中图分类号: