自组织集群系统图形化组合建模方法研究

The behavior analysis and cooperative decision of self-organizing swarm systems are the focus of research on the direction of complex systems. The self-organizing swarm system has the characteristics of individual adaptability, complexity of interaction, and coordination of decision-making. This project develops a graphical composite modeling method for self-organizing swarm systems, on the basis of combining the ontology semantic description and reasoning, the graphical modeling paradigm, the machine learning algorithm and multi-agent modeling technology. It breaks through the key technologies, such as agent description based on semantic Petri nets, multi-agent composite integration based on interaction filtering net, multi-agent collaborative decision-making modeling based on dynamic influence diagram. The purpose is to solve the key scientific problems such as how to guarantee the validity of multi-agent model combination semantics, and how to enhance the flexibility of combination integration and consistency of distributed decision-making. The validity and rationality of the theoretical methods are verified by the empirical research on the typical application problems, such as the formation reconfiguration of unmanned aerial vehicle swarms. This project will provide theoretical approaches for solving complex self-organizing swarm behavior analysis and decision modeling. It has very important theoretical significance and application value to improve the development efficiency of modeling complex self-organizing swarm systems, to enhance the intuition and scalability of complex self-organizing swarm system model, as well as to improve the cooperation and intelligence level of the complex self-organizing swarm decision-making.

自组织集群系统行为分析与协同决策是复杂系统方向的研究重点。本项目针对自组织集群个体的适应性、交互的复杂性和决策的协同性等特征，将本体语义描述与推理、图形化建模范式、机器学习算法与多智能体建模技术相结合，开展自组织集群系统图形化组合建模方法研究，突破基于语义Petri网的智能体模型描述、基于交互过滤网的多智能体组合集成、基于动态影响图的协同决策建模等关键技术，解决如何保证多智能体模型组合语义的有效性、提升组合集成的灵活性以及分布式决策的一致性等关键科学问题。结合无人机集群队形重构等典型应用问题进行实证研究，验证理论方法的有效性与合理性。本项目将为解决复杂自组织集群行为分析与决策建模问题提供理论方法，对提高复杂自组织集群系统的建模效率，增强复杂自组织集群系统模型的直观性与可扩展性，适应复杂自组织集群系统决策的协同化和智能化，具有十分重要的理论意义和应用价值。

建模与仿真技术是研究自组织集群系统行为分析与协同控制的重要途径。本项目紧贴自组织集群系统建模的实际需求，研究自组织集群系统图形化组合建模方法，在可组合智能体模型图形化描述、多智能体组合集成与通信优化、多智能体协同决策图形化建模等方面取得了一系列科研成果：（1）针对传统智能体模型图形化组合能力普遍较弱，以及基于现有图形化描述方法构建的智能体模型缺乏语义描述信息，难以判定模型组合有效性的问题，提出了一种基于语义Petri网的可组合智能体模型图形化描述方法，在传统Petri网范式中引入了复杂分支执行条件，采用输入/输出端口将其封装为描述可组合智能体模型的图形化建模范式，并基于RDF（资源描述框架）描述其语义信息，从而可为正确有效组装智能体模型提供基础支撑。（2）针对现有多智能体模型之间的通信研究多局限于少数具体应用或依靠硬件平台性能提升，并且多智能体之间静态交互结构导致较长的运行时间与资源消耗的问题，提出了基于交互过滤的多智能体组合集成与通信优化方法，构建了基于智能体交互图的组合集成建模范式，实现了基于数据过滤的多智能体模型通信优化方法，从而支持多智能体动态交互演化，满足智能体模型快速迭代优化需求；（3）针对现有机器学习模型缺少智能体状态协同的问题，提出了基于组合影响图的多智能体协同决策建模方法，采用了分解-联合的建模策略，底层根据事先预设的行为规则，由智能体模型计算得到个体状态；上层以个体状态作为随机变量，通过构建动态影响图决策模型计算多个智能体最佳的协同机动决策。（4）结合无人机集群队形重构开展应用研究，开展了空中物资投送场景下无人机编队任务规划仿真与实验验证，对上述理论方法进行验证与改进。上述研究成果为复杂自组织集群系统提供了图形化组合建模方式，可为自组织集群系统的机理研究、设计分析和优化提供相关理论方法支撑。