【作者】 王磊；

【导师】 王维平；

【作者基本信息】 国防科学技术大学 ， 控制科学与工程， 2010， 博士

【摘要】 巡航导弹协同攻击是一种重要的协同作战样式，随着巡航导弹攻击对象从单目标、多目标到体系目标演变，其相应的打击方式也发生了重要的变化。针对不同的攻击目标，弹与弹之间的协同，以至弹群与弹群之间的协同，都成为在对抗条件下完成复杂使命的必要手段。协调多枚巡航导弹的作战从而提高对具备体系特征的目标的打击效能，具有十分重要的意义。巡航导弹协同攻击体系目标战术任务规划问题(MMSCA-SoST)由于对抗过程的复杂性，传统的基于专家的、基于案例的以及基于运筹学的方法在求解该类问题时存在诸多困难。本文将仿真引入MMSCA-SoST问题求解，把作战效能评估看作知识生成的过程，把仿真优化看作利用知识求解复杂军事问题的活动，并将二者结合起来进行研究。在人机结合的思想指导下，用认知心理学、创新计算和数据挖掘的相关理论方法来丰富仿真优化这种复杂军事问题的求解模式。针对国内外相关技术在本文研究背景下的不足，论文主要研究内容和成果如下：(1)建立了MMSCA-SoST问题模型，提出了问题求解框架。定义了多弹协同攻击体系目标战术任务规划的相关概念，用对其中的关键元素进行数学建模。将MMSCA-SoST归结为一个集目标选择、调度、指派和参数优化为一体的多约束、强耦合的复合组合优化与决策问题。针对仿真求解的需求，建立了协同攻击效能仿真模型基本框架，以此作为建立仿真系统的基础。分析了问题的复杂性以及仿真引入带来的复杂性。在全面总结基于仿真的复杂问题求解模式，对已有的模式进行对比的基础上，提出了“仿真+数据挖掘+基于认知行为的智能算法”的新模式。分析了多弹协同攻击体系目标效能产生机理，基于效能产生结构提出了MMSCA-SoST分层求解框架，将其分解并抽象为弹群使命优化和弹群行动优化两个子问题，相应地将总体目标和约束条件分散到各子问题中，降低了问题的求解难度。(2)提出了一种模拟人类创造性思维问题求解过程和行为的智能算法——认知演化算法(CEA)。该算法以知识为核心，将问题求解看成一个基于知识的创造性思维过程，对发散思维、收敛思维、记忆、执行、学习和价值测度六个关键模块进行建模，充分发挥了知识演化和基于知识的创造性思维技巧在问题求解中的作用。通过一个扩展的路径优化问题分析了CEA各参数对算法性能的影响，并将CEA与经典的智能算法进行了比较。实验结果表明，针对计算密集型和知识密集型优化问题，该算法能够以较少的目标评价次数得到问题的较优解。(3)建立了巡航导弹协同攻击两层效能知识模型，并分别提出了建立两层知识模型所需的仿真数据挖掘算法。针对弹群使命层知识的表示，提出了基于贝叶斯网络(BN)的概率规则表示方法：针对基于仿真数据的概率规则构建方法，重点研究了非确定先验信息条件下的BN结构学习算法，提出了一种基于改进最小描述长度测度和模拟退火的结构学习算法，有效地克服了传统方法无法有效地利用不确定先验信息的缺点。提出了基于BN的弹群行动层知识模型——BOEM，给出了BOEM的模型结构和建模流程。为了快速产生BOEM建模所需数据，提出了基于概率规则元模型的行动层仿真(PRMS)，该方法能够在不过多损失仿真精度的条件下，有效地提高仿真效率。为提高BOEM的推理能力，对建模过程中的连续变量离散化问题进行研究，提出了基于推理信息量的BN连续变量离散化方法。(4)在两层效能知识模型的基础上，利用CEA解决MMSCA-SoST问题，提出了协同攻击分层认知演化模型及其实现算法。具体包括：在MMSCA-SoST问题分层求解框架所定义的求解逻辑流程基础上，将CEA与效能知识模型进行集成，具体实现了其中每个组件的行为，形成基于知识的协同攻击分层认知演化模型，这是一个可计算的模型。提出了弹群使命优化和弹群行动优化两个子问题的具体求解算法。(5)以潜射反舰导弹协同打击水面舰艇编队为例，应用所提出的方法对问题求解，验证了方法的可行性和有效性。更多还原

【Abstract】 Cooperative attacking of cruise missiles is an important style of cooperative operation. The target of cruise missiles engagement has evolved from single target, multiple targets to system of systems target. In this condition, cooperation among single missile or even among missiles groups becomes a significant manner to finish complex missions. How to coordinate operations of multiple missiles to maximize the combat effectiveness has become a crucial problem with great theoretical significance and practical value.As the engagement process becomes more complex, the traditional methods, i.e. experts based method, case based method and operational research theory based method, are difficult to address the need of tactical mission planning for Multiple Missiles Static Cooperative Attacking System of Systems Target (MMSCA-SoST). This dissertation will take the operational effectiveness evaluation as a knowledge generation process and the simulation optimization as a knowledge based problem solving activity, and to combine the study. In the thought of human-computer cooperation, take cognitive psychology, creative computation and data mining as complementary to simulation optimization theory which is a complex military problem solving model. Considering the actual background of cruise missile cooperative attacking, and aiming at the defects of existing research results, studies made in the dissertation are as follows:(1) The basic mat hematic model for MMSCA-SoST problem and the general solving framework are presented. The concepts relevant to MMSCA-SoST are defined and the mathematical model of MMSCA-SoST is established. MMSCA-SoST is a combination optimization problem with multi-constraints and strong coupling which comprised of four sub problems, i.e. target selection problem, scheduling problem, assignment problem and parameters optimization problem. The basic framework of simulation models is established as the foundation of simulation systems construction to fulfill the needs of simulation based problem solving. The complexity of MMSCA-SoST itself and the complexity induced by simulation are analyzed. A novel simulation based problem solving model, i.e. "simulation + data mining + cognition behaviors based intelligent algorithm" , is presented. According to the effectiveness analysis of cooperative operation of multiple missiles, a hierarchical MMSCA-SoST solving framework based on the effectiveness generation model is presented in which MMSCA-SoST is divided into two sub problems, i.e. missiles group mission planning problem and missiles group operation planning problem. The framework can prevents us from bogging down by resolving the over-complicated model.(2) Cognition Evolutionary Algorithm (CEA) towarding simulation optimization is put forward. Inspired by the process of human creative thinking, cognition evolutionary algorithm, a novel intelligent algorithm based on cognition science and computational creativity, is proposed, which simulates the creative thinking based problem solving process and behaviors. The algorithm is comprised of six components, i.e. divergent thinking, convergent thinking, memory, and execution, learning and value measures. Taking the problem-solving as knowledge based creative thinking process, knowledge evolution and knowledge based creative thinking skills play important roles in the algorithm. We explore the impact of the parameters of cognition evolutionary algorithm through an extended path optimization problem. The results show that the novel algorithm can reduce object score evaluation times for solving optimization problems which are compute-intensive and knowledge-intensive compared to other classic intelligent algorithms.(3) The two-level effectiveness knowledge models of cruise missiles cooperative attacking as well as the corresponding simulation data mining algorithms are presented. Missiles group mission level knowledge representation based on Bayesian Networks (BN) is proposed. A Structure learning method of Bayesian network is presented to solve the problem of structure learning with uncertain prior information. An improved MDL measure named SMDL is proposed to fuse the prior information in learning process. Simulated annealing method is used to solve the problem. The experimental results on the Asia network show’ that the proposed algorithm is more accurate than classical ones with fewer samples. A BN based knowledge model, i.e. BOEM, is proposed to model missiles group operation level knowledge. The specification and the modeling process of BOEM is detail studied. The concept of probability rule surrogate model based simulation is given to rapidly generate data samples for BOEM construction. Continuous variables discretization, the key component of BOEM learning algorithm, is studied, and a discretization algorithm based on reasoning information is proposed.(4) Cooperative attacking hierarchical cognition evolutionary computing model (CHCEM) and its algorithms are presented to resolve MMSCA-SoST based on two-level effectiveness knowledge model and CEA. The behaviors of components in hierarchical MMSCA-SoST solving framework are implemented based on the integration of CEA and effectiveness knowledge following the logic process defined in the same framework. CHCEM which is a computable model is established. The algorithms for resolving MGMP and MGOP are putting forward.(5) The process of the proposed method is given and the validity is proven with a case study of submarine-launched anti-ship missiles against surface fleet.更多还原