有人参与的复杂多智能体网络的协同控制

Hybrid multi-agent systems usually consist of both manned agents and unmanned agents which cooperatively work together. This kind of systems plays an important role in various fields including the national economy, state security, scientific research and the improvement of human life. This kind of multi-agent systems interfered with mankind is called as hybrid intelligent network. This human-machine interactive system is obviously heterogeneous, and at the same time has the new features of hybrid intelligent and strongly coupled human-machine interaction. In the hybrid intelligent network, how to make their respective advantages complementary to each other among different intelligent to achieve coordination and optimization, is still an open question. This calls for new theory and new methodology urgently to guidance the design, development and realization of physical hybrid intelligent network. This project will start with several typical applications of hybrid intelligent network consisting of both manned agents and unmanned agents. Based on the framework of multi-agent system, several important issues will be investigated, including the formation operation, the attitude regulation, human security assurance, maximal covering, task assignment, optimal routing, and rivalry game. Both the theory and methodology of system dynamics, cooperative control, and optimized dispatching will be established for the hybrid intelligent network which exhibits the features of heterogeneity and diverse structure, hybrid intelligent, and strongly coupled human-machine interaction. Numerical simulation, software simulation and physical verification will all be carried out to verify the obtained theory and methodology.

有人个体和无人个体混合组成混杂多智能群体协同工作，这种系统在国民经济、国家安全、科学研究以及改善人类生活等方面有着重大作用。这类有人参与的混杂多智能体群体称为混杂智能网络，该人机混合系统有明显的异构异质的特点同时具有混杂智能和人机强耦合的新特性。在混杂智能网络的协调中差异化的智能如何实现优势互补协调优化，迫切需要建立新的理论与方法，以指导实际混杂智能网络的设计、开发与实现。项目拟从几类典型的有人无人混杂智能网络的应用出发，在多智能体系统的框架下，研究混杂智能网络的编队运行、姿态调节、人类安全保障、最大覆盖、任务分配、最优路由以及博弈对抗等问题；研究建立具有异质异构、混杂智能和人机强耦合特性的混杂智能网络的系统动力学、协调控制与优化调度的理论与方法；并基于数值仿真、软件模拟与实物验证检验所得的理论与方法。

本项目以复杂网络理论研究和多智能体系统理论研究为基础，并从几类典型的多维度智能体系统的实际问题出发，重点研究了以下问题：.（1）混合智能网络的分布式控制协议。研究了异质的领导者、跟随者在不同的模型及在不同网络结构下的一致性问题、二部一致性问题，设计了分布式控制协议实现了多智能群体的协同控制以实现领导者引导下的多组编队控制。.（2）多智能体的协同围捕问题。针对异智混合智能网络的协同围捕问题，引入信用分配机制的强化学习算法提升多智能体围捕-逃逸效率；探讨了引入公平系数的概念来提高狩猎的公平性；针对狩猎过程中动态变化的环境，提出了基于深度确定性策略梯度框架、双向循环神经网络和基于概率奖励的强化学习用于多智能体的追捕和逃避。.（3）研究了智能网络的优化调度。引入了一个称为混合的覆盖-跟踪指标与动态任务分配机制，提出了一种由一队在限定区域内移动智能体来消除多种威胁的协调控制方法；提出采用多智能体强化学习策略来实现网络数据包的路由问题，优化了在固定和时变流量需求下的网络性能。.在国内外发表论文35篇，其中SCI收录24篇，取得发明专利1项，并培养了多名硕士博士生。