【作者】 慈元卓；

【导师】 谭跃进；

【作者基本信息】 国防科学技术大学 ， 管理科学与工程， 2008， 博士

【摘要】 多星对地移动目标搜索问题具有重大的军事意义和民用价值,在态势评估、精确打击引导、海洋搜救和海关缉私等方面具有广阔的应用前景。采用多星对地移动目标进行搜索大大增加了卫星任务计划编制的难度,必须借助任务规划技术才能较好地管理和分配卫星资源。目前面向移动目标搜索的多星任务规划问题研究仍处于探索阶段,在理论研究和实际应用中存在许多亟待解决的问题。基于上述背景,在总结分析国内外相关工作的基础上,论文针对面向移动目标搜索的多星任务规划问题展开研究,涉及的内容包括卫星任务规划问题形式化描述、问题模型设计及优化算法设计等方面,主要的工作及创新点如下:(1)建立了面向移动目标搜索的多星任务规划问题的形式化描述,提出了面向移动目标搜索的多星任务规划的两种模式深入分析卫星成像工作原理和卫星对移动目标搜索活动的业务流程,建立了移动目标及任务区域的形式化描述,并提出了卫星搜索候选行动集合的构造方法。在此基础上,对面向移动目标搜索的多星任务规划问题进行了阐述,分析了其中的约束条件和不确定性因素,并提出了两种规划模式:面向移动目标搜索的多星离线任务规划模式和面向移动目标搜索的多星在线规划模式。通过问题描述和规划模式划分,能够把握问题的主要矛盾和次要矛盾,为问题的建模和算法设计奠定了基础。(2)建立了基于部分可观马尔可夫决策过程(POMDP)的面向移动目标搜索的多星离线任务规划模型针对离线任务规划模式,在无侦察信息反馈条件下建立了未发现目标假设条件,依据该假设条件提出了移动目标分布离线更新和转移的方法,对未来时刻目标的分布进行估计,获得目标分布的先验概率。在目标分布先验概率的基础上,采用不确定序贯决策求解框架——部分可观马尔可夫决策过程(POMDP)建立了多星离线任务规划模型。模型以最大化累计探测概率为目标优化卫星的搜索活动,特点是在动态不确定环境下具有较强的适应性,求解的最优策略能够针对不同的初始先验分布制定卫星的最优搜索计划。(3)提出了大地坐标系下目标运动预测算法在线任务规划模式中,针对运动参数未知的移动目标,首先在平面笛卡尔坐标系下对目标的运动性质进行分析,推导出一种基于高斯分布的目标转移概率密度函数;然后在三维笛卡尔坐标系内进行扩展,得到地球表面上目标运动转移概率的数学描述;最后借助坐标转换原理和曲面积分方法,得到了大地坐标系下基于高斯分布的目标转移概率计算方法。该算法能够解决地面或者海面上移动目标的运动预测问题,并且避免了频繁的坐标转换,提高了计算的效率,同时在先验信息稀少、样本点不足的条件下也有较好的性能。(4)建立了基于预测模型控制(MPC)的面向移动目标搜索的多星在线滚动任务规划模型针对在线任务规划模式,根据侦察信息的反馈,采用预测模型控制理论和方法,建立了多星在线任务规划的闭环控制策略。该闭环中包括相互作用的目标预测模型和多星任务规划模型,其中目标预测模型利用大地坐标系下目标运动预测算法针对单先验点和双先验点模式,可以获得目标在未来时刻分布的先验概率;多星任务规划模型采用滚动优化的理念,将目标预测模型获取的先验概率作为反馈输入到后续的滚动窗口,在这些滚动窗口中以探测概率收益和搜索信息收益为优化目标,建立周期性滚动任务规划模型。通过闭环反馈控制机制和滚动窗口内反复进行的局部优化,达到在线规划卫星的目的,从而实现在整个任务时域上问题的求解。更多还原

【Abstract】 Moving target search by multi-satellite is of great military and civilian value, which will be applied in situation assessment, strike navigation, maritime rescues and anti-smuggling. Since the complexity in making plans for moving target search, manual and single mode are no longer fit for the requirements. Mission planning is able to manage and allocate the satellite resources optimally. Till now multi-satellite mission planning for moving target search has been still in the initial stage with many open problems in theory and practice to be tackled.This dissertation focuses on multi-satellite mission planning for moving target search, which includes formal problem description, planning mode, mathematic models, optimal algorithms and etc. The main work and contribution are as follows:(1)The formal problem description and two planning modes are presented. Following through the analysis of satellite imaging procedure and operation flow of moving target search, the method for optional action set and the quantitive modeling of moving target and task region are reached. Based on above, the search problem, the constraints and uncertainty are analyzed. The on-line planning mode and off-line planning mode are also proposed. The formal description and the planning modes are the key to design mathematic models and algorithms.(2)The POMDP model for multi-satellite off-line planning is presented. For off-line planning with no detection hypothesis, the method for moving target off-line update and transition is proposed to acquire target prior distribution probability. With the help of the prior distribution probability, the multi-satellite POMDP mission planning model is designed to maximize the detection probability sum. The model is adaptive to uncertainty which can transform different initial prior distribution probability into satellite search operation plans.(3)Motion prediction algorithm in geodetic coordinate systems is proposed. Motion prediction in geodetic coordinate systems is well discussed. Firstly, moving target motion in plane Cartesian coordinate systems is analyzed and a Gaussian distribution of target transition probability density function (PDF) is deduced; Secondly, the PDF is extended to denote target transition probability in 3-D Cartesian coordinate; Thirdly, the method for computing the target transition probability in geodetic coordinate systems based on Gaussian distribution is deduced by coordinate systems transformation and curved surface integral method. The Motion prediction algorithm is able to avoid frequent coordinate transformation. It is still effective even with little prior information and sample points.(4)The MPC model for multi-satellite on-line planning is presented. For on-line planning, the close-loop control architecture is acquired based on MPC. In the close-loop control architecture, the target prediction model and multi-satellite mission planning model are interactional. The target prediction model is utilized to get the target prior distribution in future by motion prediction algorithm. With the help of prior distribution, the multi-satellite mission planning model is to maximize the detection reward and information reward in each rolling window. The prediction model and the planning model are the key to predict the state of the system and to control optimally. By the iterative rolling optimization and state feedback in close-loop architecture, the global solution is obtained.更多还原