面向计算机生成兵力的意图识别方法研究

Computer generated forces are playing important roles in military simulation systems in operation analyzing, training and testing & evalutation, and the critical issue of creating CGFs is modeling human behaviors which deserves more efforts. Intention Recognition(IR) is a key step in the process of CGFs’ cognition, the CGFs with ability of IR can make counter decisions & plans and have human-like cognition mechanism. Our research will begin from constructing the IR framework in CGFs’ cognition, and the modeling methods of IR based on statistical relational learning and decision/planning theory are discussed. Our proposal aims to make CGFs can recogize different operation intentions under complex conditions such as multi-agent cooperation, observation missing, changeable enemies’ intentions and lacking training data. The main research content includes: analyzing the path plan, tactical strategy and cooperation relationships of recognizing objects; IR modeling methods based on logical hidden Markov model and decision/plan theory; abstracting domain knowledge in simulation scenarios and representing them formally; experiments design and evaluation indices of the models. The results of our research will improve the cognition level and decsion ability of CGFs significantly and make contributions to increasing reality of simulation & efficiency of computing and reducing cost of developing simulation systems.

CGF在作战分析、人员训练以及装备测试评估等作战仿真系统中发挥着重要作用，研究CGF的难点是人类认知行为的建模。意图识别是CGF认知过程中的关键环节，具备意图识别能力的CGF能够进行针对性的任务规划和决策，其认知机理与人类更加接近。本项目以CGF认知过程中的意图识别框架为切入点，研究了基于统计关系学习和决策/规划理论的意图识别建模方法，使CGF在多智能体协同、观测数据不完整、对手意图可变、训练数据不足等复杂条件下能够识别多种作战意图。主要研究内容包括：分析识别对象的路径规划、战术策略和协同关系；基于逻辑隐马尔可夫模型和决策/规划理论的意图识别建模方法；仿真场景中的领域知识提取与形式化表示；模型实验设计与评价指标。本项目的研究成果将显著提高CGF的认知水平和决策能力，对于提高仿真的真实性和模型运行效率、降低仿真系统的开发成本具有重要意义。

计算机生成兵力（Computer Generated Force，CGF）是作战仿真领域的关键前沿技术之一。为解决现有仿真系统中CGF对抗能力不足、行为表现不真实的问题，必须从机理上提高CGF的认知水平。意图识别是一类重要的认知行为，具备意图识别能力的CGF能够像人类一样“知己知彼”，即利用获取的情报信息判断对手意图，并为制定有效对策提供依据。本项目首先介绍了CGF意图识别的研究背景、研究意义以及理论研究和应用现状，在此基础上分析了意图识别理论与自动规划、参数估计和不确定性推理等相关理论的关系，并构建了意图识别的一般性研究框架。在该框架下，意图识别问题被分解为三个部分：a）问题的形式化描述；b）行为参数获取；c）意图推理。项目将意图识别的研究框架与具体应用相结合，针对四种作战仿真中的典型场景，分别提出了相应的意图识别建模理论和推理算法。成功解决了面向计算机生成兵力的一系列意图识别问题。