基于事件驱动的模型预测控制策略研究

Model predictive control (MPC) has been widely applied in various process industries. It is well known that MPC is capable of tackling the control problems in multiple-input-multiple-output systems with constraints. However, a considerable amount of computation and communication are involved due to the fashion of receding horizon optimization, which limits the wider use of MPC. To this end, this project will systematically investigate MPC approaches that are based on event-driven mechanisms. The designs of event-driven MPC strategies will be carried out according to several key steps, including formulating the optimal control problem (OCP), sampling the system information, as well as solving the optimization problem. This project will mainly consider the event-driven prediction horizon and the event-driven terminal constraint with respect to the formulation of OCP, which aim to reduce the complexity of the optimization problem to be solved online. Besides, two classes of event-driven strategies will be proposed corresponding to the sampling and the optimization procedures, i.e., the strategies involving the time-driven sampling as well as the event-driven optimization, and the strategies where both sampling and optimization are driven through event mechanisms. These two classes of strategies are designed in order to further reduce the cost caused by computations and communications. Moreover, feasibility analysis and numerical verifications will be carried out in this project, which help to provide guidance to the implementation of event-driven MPC in practical applications.

模型预测控制(MPC)广泛应用于过程工业，虽然MPC擅长处理多输入多输出的受约束控制问题，但滚动优化的执行方式常常需要较大的计算量和通信量，这也限制了MPC的进一步使用和推广。针对这个问题，本项目拟研究基于事件驱动的MPC策略，针对MPC执行时的多个环节提出相应的事件驱动型设计方案。这些环节主要包括优化问题的表述以及滚动优化所涉及的信息采样和在线优化等。其中，针对优化问题的表述，项目拟提出事件驱动型预测时长和事件驱动型终端约束这两种设计方案，其目的是降低优化问题的复杂度以减少在线计算量；针对滚动优化中的信息采样和在线优化，重点研究采样由时间驱动而优化由事件驱动的策略，以及采样和优化均由事件驱动的策略，旨在进一步减少在线执行所需的计算量和通信量。此外，项目将对所提出的各类事件驱动型MPC方案进行有效性分析和仿真验证，为其应用与推广提供理论支持。

模型预测控制（MPC) 正受到越来越多应用领域的关注。其特点是能处理多人多出且有系统约束的控制问题。然而，MPC滚动优化的特点決定了其计算负担较大，因此需要考虑在策咯上进行改进以适应更广的应用场景。为此，本项目对事件触发型 MPC 策略进行了调研与设计。项目针对RMPC 策咯中的成本两数、优化过程及信息采样三个环节分别进行探讨，分析了各个环节进行事件驱动MPC 策略设计的可行性。针对成本两数的策咯设计主要针对预测步长较长的情况，为此离线设计虚拟终端约束集，将长步长MPC 问题转化为短步长 MPC 问题。针对优化过程的事件驱动型策略设计主要通过采用新型成本函数得到实现，以闭环稳定性作为触发条件设计依据，执行起来较为便利，在该方式下 MPC 优化问题的求解只在某些特定时刻进行，极大地减少了计算量。针对信息采样的设计则在减少计算量的基础上进一步对通信量进行考虑，通过虚拟终端约束集与离线设计相结合，该方法可以实现在线运行时对于驱动事件的自触发，适合于通信带宽受限的应用场景。仿真表明上述方法的可行性，与传统方法比计算量/通信量得到显著减少。此外，调研了预测控制在电力电子、电机控制方面的适用性，并对事件驱动型 MPC 策咯在相关方向上的应用进行了探索。项目发表论文21篇，其中第一作者/通讯作者论文9篇，包括Automatia、IEEE Transactions on Circuits and Systems I: Regular Papers, Drones等期刊，授权发明专利3项。