内模强化学习型模型预测控制及其在人工胰脏上的应用

Internal model control-enhanced learning-type model predictive control: application to artificial pancreas

中文关键词  迭代学习控制  模型预测控制  间接型迭代学习控制  内模控制  人工胰脏  1型糖尿病

英文关键词  iterative learning control (ILC)  model predictive control  indirect ILC  internal model control  artificial pancreas  type 1 diabetes mellitus

中文摘要

      在学习型模型预测控制的框架里, 迭代学习控制器被用来更新模型预测控制器的设定点. 在已经发表的研究成果里, 学习型模型预测控制用到的是比例型的学习率, 这种学习率的学习能力有限, 而且怎样设计学习增益仍然是一个开放性问题. 在本文中, 基于内模控制理论提出的PID型的迭代学习控制器被用来更新模型预测控制器的设定点. 为了方便起见, 本文提出的结合算法可称为内模强化学习型模型预测控制. 本文提出的算法应用在1型糖尿病人的人工胰脏闭环控制上. 仿真结果显示, 本算法对比于比例学习型模型预测控制可以达到更好的收敛性能, 而且对非重复干扰有很好的鲁棒性.

英文摘要

      In the framework of a learning-type model predictive control (L-MPC), an iterative learning control (ILC) is used to update the setpoint for model predictive control (MPC). In the reported studies, the L-MPC usually has a P-type ILC, which has limited learning capability and also how to design its learning gain remains an open problem. A PID-type ILC was proposed to design the learning-type setpoint for MPC based on internal model control (IMC) theory. For convenience, the proposed combination is named IMC-enhanced L-MPC. The proposed method was applied to the closed-loop control of an artificial pancreatic ˉ-cell for type 1 diabetes mellitus (T1DM). The simulation results show that the proposed algorithm can produce superior convergence performance compared with the P-type L-MPC, and also it has excellent robustness to non-repetitive disturbances.