超视距空战仿真中的策略识别

Policy Recognitionin Beyond Visual Range Air Combat Simulation

摘要:

针对超视距空战仿真中敌机策略的识别问题,研究了一种基于案例的策略识别方法。该方法通过构建包 含假定的对手任务目标、观测数据、策略等内容的案例库,采用相似度计算选择与新的观测相似的案例库子集,并 计算策略概率分布来识别对手策略。实验证明,相比与常规的基于案例推理方法,在案例中增加敌机任务目标提 高了策略识别准确率,并且在假定敌机目标不正确时,能修复错误假定并进行策略识别,改善了基于案例策略识别 方法的性能。

Abstract:

To recognize the policy of adversary in beyond visual range air combat simulation, a case-based policy recogni? tion method was studied. In this method, a case base containing hypothetical mission goal, observations, and policy of ad? versarial aircraft was constructed. The similarity calculation was used to retrieval a subset of case base and calculate the probability distribution to recognize the adversary’s policy. Experiments showed that compared with the traditional casebased reasoning method, an adversary’s goal improved policy recognition. It also could recognize when its assumptions about the adversary agent’s goal were incorrect, and could correct these assumptions.