【作者】 霍霄华；

【导师】 沈林成；

【作者基本信息】 国防科学技术大学 ， 控制科学与工程， 2007， 博士

【摘要】 多UCAV(Unmanned Combat Aerial Vehicle)协同作战将成为未来战争的一种重要方式。研究多UCAV协同任务规划建模与优化方法,为UCAV制定出协同的任务计划和飞行航迹,使多UCAV协同系统整体作战效能优于各UCAV独立作战效能的总和,是发挥多UCAV协同作战优势的关键,具有重要理论价值与现实意义。解决该问题的核心就是对多UCAV协同任务规划问题进行合理的建模和求解,以及适应环境动态变化的在线规划。论文借鉴模型预测控制(Model Predictive Control,MPC)思想,从问题建模、求解算法和滚动优化方法三个方面开展研究,主要工作及创新点如下:(1)建立了多UCAV动态协同任务规划的基本数学模型。在深入分析多UCAV协同执行作战任务的独特性基础上,提出基于MPC的多UCAV协同任务规划建模方法。以包含敌方目标、威胁以及我方UCAV在内的复杂大系统作为被控对象,以UCAV任务计划与航迹计划作为控制输入,通过提取系统关键状态及其变化规律,建立了描述系统动态变化过程的预测模型和反映控制输入性能的优化模型。为了便于求解,根据不同时间粒度对基本数学模型进行递阶分解,将多UCAV协同任务规划问题分解为任务调度和航迹规划两个子问题,分别求解UCAV任务计划和航迹计划。进一步分析了这两个问题的组合复杂性,提出了分层求解方案。仿真实验表明,能否描述多UCAV协同执行任务过程的动态性是影响模型可用性的重要因素;本文建立的多UCAV动态协同任务规划模型能够描述这一特性,有利于提高控制效果。(2)分析了粒子群优化算法(Particle Swarm Optimization,PSO)求解离散组合优化问题的两种设计原则,针对任务调度和航迹规划问题的排序和无序组合特性,提出了求解排序组合优化问题的连续空间离散PSO(Continuous Space DiscretePSO,CDPSO)算法和无序组合优化问题的离散空间离散PSO(Discrete Space DiscretePSO,DDPSO)算法。在CDPSO算法中,使用实数粒子编码方式,充分利用粒子位置的整数和小数部分,将离散问题映射到相对较小的连续粒子位置空间。提出了动态子群和双精度高斯位置扰动策略,使得粒子群能够较均匀地在问题空间搜索,避免陷入局部极值,在保持传统PSO算法快速收敛的同时,加强了算法局部搜索能力。DDPSO算法的核心是将PSO算法映射到离散问题空间,在经典PSO算法的基本框架下,通过重新定义粒子位置和速度在离散空间内的表示形式、加减法和数乘计算等运算法则,形成了一种全新的PSO算法模式。针对两类典型组合优化问题的实验结果表明,两种算法能够快速、稳定收敛到最优或次优解,搜索效率高。(3)提出了基于任务滚动窗口与SA-CDPSO(Simulated Annealing CDPSO)算法的任务调度方法。针对在线任务调度问题的实时性要求,扩展MPC传统滚动优化方法,提出基于任务轴的滚动窗口优化方法,通过在线进行的优化计算与滚动实施,快速响应环境变化。研究了适合于任务调度的滚动窗口更新、滚动机制、局部优化问题构造方法。针对任务调度问题约束条件多、搜索空间不连续的特点,提出SA-CDPSO混合算法。使用CDPSO算法进行全局搜索,再根据CDPSO产生的初始解构造邻域,利用CDPSO的快速收敛特性和SA的局部搜索能力,确保算法能够快速收敛到最优解。基于正交实验法构造了具有代表性的测试问题实例,对SA-CDPSO算法和滚动窗口方法进行全面实验验证和分析。对于不同测试问题实例,SA-CDPSO算法都表现出稳定、优良的搜索能力;任务滚动窗口优化方法能适应于各种环境变化。(4)提出了基于双精度滚动窗口与AI-DDPSO(Artifical Immune DDPSO)算法的航迹规划方法。针对在线航迹规划的特殊性,提出异步双精度滚动窗口策略,分别在两个信息粒度不同的窗口内进行局部精细航迹规划和全局粗略航迹规划,并按照不同方式采用不同频率推进窗口。重点研究了求解精细航迹的AI-DDPSO混合算法,结合UCAV航迹特有的最小直飞距离、最大转弯角、爬升/俯冲角、以及最小调整角度等约束,设计了基于角度偏差的粒子编码方式,确保航迹满足UCAV飞行约束。在DDPSO算法基础上,基于人工免疫(Artifical Immune,AI)原理,设计了粒子逃逸算子和群体排斥算子,通过检测粒子个体和群体浓度改变粒子运动轨迹,以保持个体和群体的多样性。实验表明,该方法可充分利用当前新信息,快速规划出较优的飞行航迹。更多还原

【Abstract】 Using multiple UCAVs(Unmanned Combat Aerial Vehicle) fight cooperatively will be an important manner to perform a military mission in the future. Studying on the multi-UCAV cooperative mission planning modeling and optimizing methods is the key to take full advantage of multi-UCAV combating cooperatively, which is of significant theoretical value and great practical value. The problem focus on planning elegant task plan and path plan for each UCAV to make the integrated effects of multi-UCAV system combating cooperatively exceed the total effects of each UCAV combating separately. Properly modeling and solving, online planning for dynamic environment are the keys to solve this problem. Based on the model predict control(MPC) theory, this dissertation studies the problem model, algorithms and online optimizing methods for multi-UCAV cooperative mission planning. The main work and contributions are as follows:(1) A basic mathematic model for multi-UCAV dynamic cooperative mission planning is presented. Following thorough analysis on the characteristics of multi-UCAV performing combat missions, a modeling method based on MPC is proposed. Setting the complex system includes targets, threats and UCAVs as the controlled object, setting the task plan and path plan of UCAVs’ as the controlling inputs, a state-predict model reflects system’s dynamics and an optimization model reflects the control inputs performance are presented, by extracting the primary states and their transformation of the whole system. By decomposing the basic mathematic model according to different time granularity, the multi-UCAV cooperative mission planning problem is devided into two subproblems for solving the task plan and path plan respectively, which are called task scheduling and path planning. Then with the analysis on the combination of each subproblem, a logic process for solving them is presented. Simulations show that characterizing the dynamics in the process of multi-UCAV completing mission is an important performance factor of planning model. And the model of multi-UCAV dynamic cooperative mission planning problem presented in this paper is beneficial to improve the effectiveness of control. (2) Two different approaches for particle swarm optimization(PSO) algorithm solving discrete combinatorial optimizing problems(DCOP) are proposed. On analyzing the different combinatorial characters of task scheduling and path planning problems, two discrete PSO algorithms are presented. One is continuous space discrete PSO(CDPSO) for DCOP with scheduling, the other is discrete space discrete PSO(DDPSO) for DCOP without scheduling. In CDPSO algorithm, both the integer and decimal fraction of a particle’s position are utilized simultaneously through real-number coding, so the discrete problem is mapped into a smaller continuous particle position space. Dynamic sub-swarms strategy and double precision gauss disturbances strategy make particles searching the problem space comprehensively and breaking away from the local extreme. Thus the algorithm performance in local searching is improved while keeping the convergence ability of traditional PSO. The key of the DDPSO is mapping the PSO algorithm into discrete problem space based on the framework of classical PSO algorithm. A totally new algorithm pattern is proposed by redefining the particles’ position, velocity and their operation rules. The tests results for two typical DCOP show that those algorithms can both find the best or hypo-best result quickly and steadily with high efficiency.(3) An online task scheduling method based on rolling window and SA-CDPSO (Simulated Annealing CDPSO) algorithm is presented. Since task scheduling online needs to respond realtime, a rolling window optimizing approach is proposed by expanding traditional rolling horizon approach of MPC. The approach can respond very quickly for environment changes by optimizing and rolling online. Rolling rules, rolling window updating and local optimizing problem formatting approaches fit for task scheduling problem are studied. A SA-CDPSO hybrid algorithm is proposed for the task scheduling problem for its discontinuous problem space caused by various constraints of the problem. A result in the global problem space is found using CDPSO algorithm, and then the neighborhood around the result is constructed and searching in the neighborhood is carried out using SA algorithm. This hybrid technique ensures algorithm converging at the best result quickly. A set of test instances which reflects different mathematical properties of the problem are constructed via orthogonal design method for verifying the performance of the SA-CDPSO algorithm and rolling window method. Experiments show that the SA- CDPSO algorithm can solve all of the instances effectively with steady and excellent performance, which proves the rolling window method is adaptive to varieties in environment.(4) An online path planning method based on double precision rolling windows and AI-DDPSO (Artifical Immune DDPSO) algorithm is presented. A double precision rolling windows strategy for online path planning is proposed. Local elaborate path and global cursory path are planned respectively in different window with different information granularity. Two windows roll on different frequency with different mode. An AI-DDPSO hybrid algorithm is proposed for elaborate path planning. An angle err based particle coding method is presented, taking into account the restrictions in UCAV path planning, including minimal straight flying distance, maximal turning angle, climbing-diving angle and minimal adjusting angle. Escape operator and repulsion operator based on artifical immune(AI) theory are proposed, which can keep the diversity of particle swarm by checking the density of particle individuals and swarm to change particles tracks. This method can utilize current information sufficiently and get an optimum path for UCAV quickly.更多还原