具有混杂特性的多智能体系统的协同演化算法研究

Collaboration of intelligent groups under dynamical network environment plays a significant role in national defense construction, scientific exploration and people's life. In views of hybrid characteristics on structures and dynamics of biotic population and artificial intelligent groups, this project will be devoted to investigating cooperative evolution algorithms of multi-agent systems with hybridity. Combining practical problems of cooperative hunting of hybrid human-robot groups, main contents of this project include: constructing several hybrid evolutionary models of human-robot multi-agent systems in light of differences of function, attribution and interaction among different agents, and revealing the hybridity of multi-agent; studying cooperative evolution patterns of intelligent groups under specific hunting tasks, proposing some state-event driven schemes by incorporating parameter changes and structure switches, and designing relevant cooperative evolution algorithms with detailed convergence analysis and computational complexity analysis; analyzing several typical uncertainties of networked agents in hunting, proposing a cooperation-performance index corresponding to the uncertainty and hunting behavior, designing adaptive cooperative algorithms adjusted by desired performance index, and discussing the derived index to reveal the influence of uncertainty on hunting performance. In summary, this project will propose new theories and methods of evolution analysis and algorithm design for multi-agent systems with hybridity, and will also provide technical guidance for some practical applications of multi-agent systems with hybridity, like human-robot collaboration, intelligent transportation and supervision, and production of complex chemical metallurgy.

动态网络环境下智能群体协作模式在国防建设、科学探索与人类生活中发挥着重要作用。项目针对生物种群和人工智能群体的模式结构混杂性和动态行为混杂性等特点，提出具有混杂特性的多智能体系统的协同演化算法研究。研究内容包括结合实际有人/无人机群体协作围捕的复杂问题，分析不同智能体的功能、属性和交互模式的差异性，揭示智能群体的混杂特性，建立若干典型的人机混杂性演化结构；基于围捕任务分析智能群体的协同演化模式，提出综合参数智能变化和结构动态切换的状态事件触发机制，设计相应的协同演化算法并分析其收敛性和计算复杂性；联合不确定性与围捕任务提出相关的协同性能指标，设计基于性能自适应调节的协同演化算法，揭示不确定性对协同性能的影响。项目旨在建立具有混杂特性的多智能体协同演化分析与算法设计的新理论与方法，为人机混合协作、智能交通与监管、复杂化工冶金生产等混杂性多智能体系统的开发设计及应用提供理论与技术指导。

随着计算机和网络技术的飞速发展，人们所面对的机器系统愈发复杂。如飞行器控制系统和复杂化工冶金体系等复杂系统呈现混杂特性，不仅涉及连续时间演化过程，还存在一定的离散调度如跳变、脉冲和切换等环节。项目结合生物和人工智能群体的模式结构混杂性和动态行为混杂性等特点，探索了混杂性多智能体系统的协同演化机理和控制实现问题。针对实际有人/无人机群体协作围捕的复杂任务，研究不同智能体的功能、属性和交互模式的差异性，建立了若干典型的人机混杂性多智能体模型；研究围捕过程中智能群体的协作模式，提出了综合参数智能变化和结构动态切换的状态事件触发机制，并设计状态事件驱动的协同演化算法；针对不确定环境下智能体适应性差，设计了基于性能自适应调节的协同控制算法，研究揭示了模型和通信的不确定性对协作性能的影响。项目的研究成果对于认识混杂性智能群体的演化规律，改造（控制设计）复杂智能群体的行为具有重要的理论和实际意义。项目共发表SCI收录论文27篇，发表在Automatica、Neurocomputing、IEEE Transactions on Industrial Electronics和IEEE Transactions on Neural Networks and Learning Systems等杂志；出版英文著作1部《Introduction to Hybrid Intelligent Networks》；出国合作交流2次；培养毕业博士4名。