基于学习算法的混杂多智能体系统协作与优化研究

Multi-agent systems with hybridity are ubiquitous in nature and society. Theories of multi-agent coordination and optimization provide powerful guidance for real design of engineering systems such as unmanned systems and power systems. Motivated by practical surveillance and strike problems of mixed human-vehicle systems in dynamic unknown environments, this project focuses on the diversity of intelligence, attribution and decision ability of human agent and vehicle agent, and studies coordination and optimization of hybrid multi-agent systems by resorting to learning-based algorithms. The main objective of this project is to create a new theory and technology that allowing for network systems to cooperate and optimize in dynamic and hybrid unknown environments. The main contents include: studying structural evolving characterizations of coupled human-vehicle systems, proposing desirable learning algorithms, and modeling multi-agent systems with typical hybridity; analyzing learning and evolving features of hybrid agents, designing adaptive learning-based coordinated control algorithms by employing neural networks such that cooperation problems of coupled human-vehicle systems in a dynamic unknown environment can be solved; synthesizing adaptive dynamic programming and hybrid control methods, and designing adaptive hybrid optimization algorithms such that optimization problems regarding time-cost, communication and energy consumption can also be addressed for unknown human-vehicle systems. This project will bring a new theory for network science and an advanced technology for applications like mixed human-vehicle operations and supervised unmanned vehicle detection in the real-world.

具有混杂特性的多智能体系统普遍存于自然演化与社会实践中，其协作与优化理论为无人系统、能源系统等实际工程系统的设计与应用提供了强大的理论指导。本项目从动态未知环境下人机混合系统侦察打击的实际问题出发，针对有人主体与无人主体在智能、属性与决策能力等方面的差异性，开展基于神经网络学习算法的混杂多智能体系统协作与优化研究，旨在运用神经网络学习方法建立新的多智能体系统理论与技术。研究人机协作群体的结构演化特性，提出合理的学习算法，建立典型的混杂多智能体系统模型；分析混杂多智能体系统的学习与演化特征，基于神经网络设计自适应学习协调控制算法，解决动态未知环境下人机混合群体的协同问题；融合自适应动态规划方法与混杂控制技术，设计自适应学习优化算法，解决动态未知环境下人机混合群体时间、通信或资源相关的性能优化问题。通过研究，将为人机混合作战、有人监督的无人机探测等实际应用提供有力的技术支撑。

四年来，项目组成员按计划开展了该项目的研究，完成了计划预期研究任务，取得了预期的研究成果。该项目从人机混合协同任务出发，运用复杂网络、多智能体系统、神经网络学习及混杂系统等理论与技术，建立了混杂多智能体系统的学习建模、协调控制与动态规划理论和方法。项目研究解决了动态未知环境下人机混合群体的协作与优化问题，具体包括针对实际人机混合群体的跟踪、编队、打击等复杂协作式决策与优化任务。通过对现实中典型人机混合多主体网络分析归纳建立了混杂多智能体系统模型，分析了系统的混杂智能学习特性，探讨了混杂多智能体系统基于学习方法的协作演化模式与自适应动态规划机理；针对人机混合多主体群的队形控制、目标跟踪、最优覆盖、包含控制等协调控制与优化调度问题，提出了异质异构、混杂智能和强约束协同控制与优化算法，并分析算法的收敛性与性能；针对异构异质混杂集群围捕问题，提出了基于深度神经网络的强化学习算法，实现了动态未知环境下混合智能群体的分布式协作与优化。最后，利用数值仿真、软件模拟和实物验证验证了所得结果的有效性。.本项目组成员在国内外刊物和会议上发表论文共34篇，其中SCI源刊25篇，比如IEEE Transactions on Industrial Informatics，IEEE Transactions on Fuzzy Systems，IEEE Transactions on Industrial Electronics，IEEE Transactions on Neural Networks and Learning Systems，IEEE Transactions on Systems, Man, and Cybernetics: Systems，IEEE Transactions on Cybernetics，IEEE Transactions on Circuits and Systems I: Regular Papers和IEEE Transactions on Network Science and Engineering等。