基于自适应动态规划理论的事件触发机制与控制律一体化设计

With the networked, distributed and intelligent development of the complex control system, the influence of constrained computation ability and limited communication resources on control system performance become increasingly prominent. Therefore, optimization of data perception and intelligent control strategy attracted increasing attention. Improvement of data perception efficiency based on the event-triggered mechanism (ETM) and optimization of decision process by adaptive dynamic programming (ADP) theory is playing an important role. However, existing ETM designs depend on not only the exact model of systems but also the controller with predetermined structure and parameters. The separated design of ETM and controller severely restricts its application. This project aims to propose a co-design scheme to implement “efficient precise perception” and “intelligent optimal decision” based on data-driven approach and the combination of ETM with ADP theory, in order to achieve efficient use of communication resources comprehensive optimization of control performance. The details of this project are as follows..1. This project will propose a synchronous ETM based on the data-driven Q-learning algorithm;.2. Based on the state observer design, an asynchronous ETM will be developed;.3. For the coordinate control problem of multi-agent systems, a distributed self-triggered mechanism will be presented. .This project will provide new research and theoretical methods for control and optimization of large-scale complex dynamical systems, and has great theoretical significance and broad application prospect.

随着复杂控制系统的网络化、分布化和智能化发展，计算能力和通信资源受限对控制系统性能的影响日渐突出，优化系统信息感知和智能决策过程成为研究热点。以事件触发机制（ETM）提高感知效率，以自适应动态规划（ADP）优化决策过程成为重要的研究方向。然而已有的ETM设计方法不仅依赖被控对象的精确模型，也需要预先给定控制器的结构与参数，感知和决策的分离式设计严重限制着其应用。本项目采用“高效精确感知”和“智能优化决策”的一体化设计思想，以数据驱动为主要形式，将ETM和ADP有机结合，实现资源的高效利用和控制性能的全面优化。研究内容包括：.1.基于数据驱动Q学习方法的同步式ETM设计方法；.2.基于状态观测器理论的异步式ETM设计方法；.3.面向多智能体系统协同控制的分布式自触发控制器设计方法。.本项目的研究成果将为大规模复杂系统的控制和优化提供新的研究模式和理论方法，具有重要的理论意义和广阔的应用前景。

近年来，随着人工智能理论与应用的快速发展，强化学习/自适应动态规划方法在智能决策方面卓越的性能受到了广泛的关注。本项目针对复杂非线性动力学系统的调度、优化与控制等关键科学问题，将自适应动态规划方法与事件触发采样机制进行有机结合，为实现高性能自学习优化控制奠定理论基础。首先，研究了数据驱动控制性能在线优化方法，避免了优化控制算法对于系统模型先验信息的依赖，进而实现了数据驱动无模型在线优化与控制算法，相关成果发表于：IEEE TNNLS、IEEE TSMC、IEEE TCYB等国内外重要学术期刊。其次，针对通信约束条件下动态系统控制与优化问题，提出了静态事件触发控制律和动态事件触发控制律设计方法，建立了完善的事件触发控制器设计和性能优化的理论体系，为复杂动态系统的通信资源优化利用与控制性能优化设计提供了新的方法和技术，相关成果发表于：IEEE TNNLS、IEEE TCYB等国内外重要学术期刊。此外，研究了面向输入输出约束的控制与优化方法，综合考虑了外部干扰、系统不确定性和输入限制影响的非线性鲁棒控制器设计，相关成果发表于：IEEE TNNLS、IJRNC等国内外重要学术期刊。总体而言，基于本项目参与人员共计发表论文28篇，包括20篇SCI论文和8篇EI论文，指导学生获得了2项学术奖励，培养研究生5名，顺利完成了项目目标和指标。