基于深度学习的预警探测体系探测效能智能评估方法研究

The effectiveness evaluation of complex system for the warning and detection SoS (system of systems) has been a difficult problem. Traditional method of non-big data, the current method of big data (including complex network) are incompetent, and an effective intelligent evaluation method is needed to explore the breakthrough. Deep learning emerged as a new field of artificial intelligence research, which has a deep structure that reflects non-linear expression, multi-level learning, autonomous extraction, historical "memory", breaking through traditional techniques, providing a valuable opportunity for the study of complex systems assessment issues. Basing on the big data generated by computer war gaming system, this paper studies the intelligent evaluation method of warning and detection SoS with deep learning. Through the study of characterization method which reflects the characteristics of the early warning detection system and adapts to input of deep learning, it constructs data sample for system detection effectiveness evaluation. Through the design of deep learning network, it can extract and analyze key effectiveness indicators of warning and detection system autonomically. By constructing the performance-related knowledge graph of warning detection SoS basing on the deep learning evaluation network, it describes and analyzes emergence mechanism of SoS detection effectiveness, formulating a complete set of intelligent assessment methods, achieving effective assessment, solving the bottleneck problem of previous method, such as being subjective, unscientific, difficult to analyze and explain complex mechanism, SoS evaluation characterization is too difficult to understand, supporting intelligent combat command, auxiliary decision-making research, promoting development of our military command information system. The research and expected results of the project will be of universal reference value for the research of complex system analysis and evaluation theory and technology, complexity mechanism analysis and so on. It is of important practical significance to improve the combat command and joint training level of our army and enhance the national defense and early warning and detection capability.

对于预警体系这种复杂系统的效能评估一直是个难题，亟需探索突破。深度学习突破了传统技术而具备了深层结构，体现出非线性表达、多层学习、自主提取、历史“记忆”等优势，为研究复杂系统评估问题提供了宝贵契机。本项目基于计算机兵棋推演系统这个实验平台，以推演产生的大数据为基础，针对基于深度学习的预警探测体系探测效能智能评估方法展开研究，通过研究预警探测体系表征方法构建出评估数据样本集；通过设计深度学习网络，实现体系探测效能关键指标自主提取；通过构建基于深度学习评估网络的预警探测体系探测效能相关知识图谱，描述分析其涌现性机理；解决传统非大数据、大数据（包括复杂网络）方法主观性强，复杂性机理难以解释，评估表征维度过高难以理解等瓶颈问题，实现有效的智能评估。预期成果对于复杂系统评估理论和技术、复杂性机理分析等问题的研究都具有普遍的借鉴价值，对于提高我军作战指挥水平，提升我国预警探测能力也具有重要的现实意义。

随着人工智能技术快速发展，战争形态不断演进，基于人工智能的体系效能智能评估显得越发重要。预警探测体系是战争中对抗双方作战力量中不可或缺的重要部分，其在战争中的作用愈发重要。预警探测体系探测效能评估是指挥员分析战场形势的一个重要环节，也是指挥员指挥决策的先决条件，对于指引战争的胜利至关重要。而如何对预警探测体系探测效能进行智能化评估是一个必须解决的重难点问题。对此，本项目展开了深入研究。.经过三年来的努力，本项目取得了较为丰硕的成果。着眼智能化的体系效能评估这个问题，设计了预警探测体系及运行对抗特征指标，构建了基于复杂网络的预警探测体系运行对抗模型及评估指标，通过建立对抗推演数据采集管理机制、提出面向预警探测效能评估的数据预处理方法、设计数据样本集构建方案，实现了面向预警探测效能评估的数据样本集构建。在此基础上，提出了基于SAE的预警探测任务执行效果评估方法、基于LSTM的预警探测体系效能评估方法，并进一步提出了面向预警探测体系探测效能评估的SAE+LSTM模型构建方法，实现了基于复合深度学习的预警探测体系探测效能智能评估，同时，从立足深度学习模型，以及构建贝叶斯知识图谱等两个方面，探索分析了预警探测体系探测效能涌现性机理。本项目还搭建了面向智能化体系效能评估的3套可支持预警探测体系对抗推演的软件系统和相应的仿真实验环境，并依托这些系统环境验证了本项目所构建模型和所提出算法的有效性。.本项目成果可为预警探测体系探测效能智能评估提供切实、可行的评估度量技术，也可为实际作战提供有价值的仿真评估依据。本项目开展的工作为项目成果今后切实能在实际作战、演习演练、教学培训以及进阶科研中取得实际效益奠定了基础。总体来看，本项目保质保量完成了计划中的研究内容，达到了项目预期目标，展现出了项目的研究价值和现实意义。