复杂任务通常动态出现在系统中且需要多个智能体合作执行，而联盟是实现合作的 重要方式之一。使用传统的指挥控制方法寻找一个联盟时间复杂度较高，所以智能体如 何自主组建联盟是提高多智能体系统任务完成率的关键。考虑到复杂任务的动态性、智 能体资源的异构性以及在不完全信息下智能体对彼此决策的不确定性，智能体之间如何 开展自主合作面临挑战。本文以联盟博弈理论和自组织理论为支撑，分析各智能体之间 的博弈关系和合作问题，将应对复杂任务的问题转化为基于联盟博弈的异构智能体自组 织联盟组建问题，研究智能体自组织合作并完成任务的过程，具体而言： 首先，考虑任务的复杂性和以及智能体资源的有限性，自组织过程中必然涉及智能 体之间的合作。本文重点以合作联盟组建的方式研究智能体的自组织行为，并使用博弈 论来分析智能体之间的冲突与合作。 其次，针对通信是智能体开展自组织合作的前提，引入了社交网络作为智能体进行 信息交互的通信网络。基于联盟博弈理论将智能体自组织联盟组建问题建模为联盟形成 博弈模型，综合考虑可能影响联盟组建的各项成本，设计了联盟特征函数和个体效用函 数，并赋予资源三种状态以减少重叠联盟带来的复杂性。考虑智能体作为理性的参与人， 设计了就资源贡献和个体预期效用为前提的协商机制。基于势博弈理论证明了上述偏好 顺序下所形成的联盟结构最终收敛到纳什稳定。 最后，针对上述设计的协商机制和联盟特征函数，提出了基于协商机制的自组织联 盟组建算法，以支持动态任务的组建，并证明了该算法得到的最终联盟结构也是收敛到 纳什稳定。同时对所提出的算法设计了仿真对比实验，实验结果也验证了所提出算法的 在系统效用、任务完成率以及资源利用率上均优于对比算法。

Complex tasks usually appear dynamically in the system and require multiple agents to cooperate, and coalition is one of the important ways to achieve cooperation. Using the traditional command and control method to find a coalition has high time complexity, so how to form a coalition independently is the key to improve the task completion rate of multi-agent system. Considering the dynamics of complex tasks, the heterogeneity of agent resources and the uncertainty of agent's decision-making under incomplete information, how to carry out autonomous cooperation between agents faces challenges. Supported by the theory of coalition game and self-organization, this paper analyzes the game relationship and cooperation between agents. The problem of dealing with complex tasks is transformed into the problem of self-organizing coalition formation of heterogeneous agents based on coalition game, and the process of agent self-organizing cooperation and completing tasks is studied. The main research results of this paper are as follows. Firstly, considering the complexity of the task and the limited resources of the agent, the cooperation between the agent s is inevitably involved in the self-organization process. This paper focuses on the self-organizing behavior of agents in the form of cooperative coalitions, and uses game theory to analyze conflicts and cooperation between agents. Secondly, regarding communication as a prerequisite for self-organizing cooperation among agents, social network is introduced as the communication network for information interaction among agents. Based on the coalition game theory, the problem of agent self-organizing coalition formation is modeled as a coalition formation game model. Considering the costs that may affect the coalition formation, the coalition characteristic function and individual utility function are designed, and three states of resources are given to reduce the complexity caused by overlapping coalitions. Considering agents as rational participants, a negotiation mechanism based on resource contribution and individual expected utility is proposed, which mainly comes from the bargaining of game theory. Based on the potential game theory, it is proved that the coalition structure formed under the above preference order eventually converges to Nash stability. Finally, aiming at the above proposed negotiation mechanism and coalition characteristic function, a self-organizing coalition formation algorithm based on negotiation mechanism is proposed to support the formation of dynamic tasks, and it is proved that the final coalition structure obtained by the algorithm also converges to Nash stability. At the same time, simulation comparative experiments are designed for the proposed algorithm. The experimental results also verify that the proposed algorithm is superior to the comparison algorithm in system utility, task completion rate and resource utilization.