基于博弈论的对地观测卫星系统分布式优化理论与方法研究

Originating from the actual requirements of collaboration in the distributed earth observing satellite system (DEOSS), this project aims to investigate game theoretic distributed optimization theory and methods, using the tools of distributed learning, inhomogeneous Markov chain theory, population dynamics and so on. Firstly, by in-depth analysis on the DEOSS operational process, we specify the cooperative elements in mission planning, resource allocation, and data transmission, based on which global optimization models and game theoretic coordination models are established. For distributed mission planning with uncertainty, we propose a time variant spatial adaptive play based learning algorithm to better balance computational efficiency and global optimality. To realize online coordination and allocation of continuous observing resource, we propose a distributed coordination algorithm by taking advantage of local population dynamics. Finally, we employ the concept of minimum weighted vertex cover and construct a new data delaying mechanism. For this, a limited memory best response based learning algorithm is proposed to achieve the cooperative selection of delaying nodes in large-scale networked systems. Our research can not only help enrich the current distributed optimization theory, improve the coordination capability of distributed satellite systems, lower earth observing costs, but can also provide theoretical and technological support for the construction of our own intelligent and autonomous distributed earth observing systems.

本项目从分布式对地观测卫星系统的协作需求出发，以分布式学习、非齐次马尔可夫链、群体动力学等理论为工具，对基于博弈论的分布式协同优化理论与方法进行研究。深入分析观测卫星系统运行过程，明确任务规划、资源分配及数据传输中的协作问题及要素，形成全局决策与优化模型，进而建立基于网络博弈的分布式优化框架，并证明其有效性；提出基于时变空间自适应游戏的学习算法，解决计算效率与最优性间的平衡问题，实现面向不确定性的分布式任务规划；提出基于局部群体动力学的协调策略，解决面向连续观测资源的实时分配与协调；构建基于网络最小权顶点覆盖的数据传输框架，提出基于有限记忆的分布式学习算法，实现大规模网络系统中通信节点的协同选取。本课题的研究工作不仅有助于丰富分布式优化理论现有成果、提高分布式卫星系统的协作能力、降低对地观测任务成本，同时也可为构建我国未来智能型、自主型的空基分布式观测体系提供理论支撑。

本项目从分布式对地观测卫星系统的协作需求出发，以势博弈、群体动力学等为理论工具，对基于博弈论的分布式协同优化理论及方法进行了研究。分析了观测卫星系统运行过程及协作要素，建立了全局决策模型与基于网络博弈的分布式优化框架，提出了基于时变空间自适应游戏的学习算法，解决了计算效率与最优性间的平衡问题，实现了不依赖全局信息的分布式任务规划；提出了基于局部群体动力学的协调策略，解决了连续观测资源的实时分配与协调问题；建立了基于最小权值顶点覆盖的数据传输框架，设计了基于有限记忆的分布式学习算法，完成了大规模网络系统中通信节点的协同选取。发表SCI论文7篇，国际会议论文6篇，授权发明专利1项。研究工作有助于丰富分布式优化现有理论成果、提高分布式卫星系统协作能力、降低对地观测任务成本，同时也可为构建我国未来智能型、自主型的空基分布式观测体系提供理论支撑。