作战任务序列智能生成方式及其态势表征方法

Intelligent Generation Mode of Course of Action and Its Situation Characterization Method

作战任务序列(course of action,COA)智能生成,特别是战役层次COA智能生成,是军事智能决策领域研究的热点问题.COA智能生成领域多使用基于层次任务网络(hierarchy task network,HTN)、简单时序网络(simple temporal network,STN)等知识规则的方法,而很少采用以深度强化学习为代表的数据驱动智能技术.造成这种现象的原因之一是数据驱动智能技术依赖于有效的态势信息输入,而战役层次战场态势的高度复杂,无法直接输入智能模型,需要精心设计的态势抽象与表征模型作为输入来驱动智能模型.解决这一问题的关键是以具体的COA智能生成方式为牵引,根据COA智能生成模型的具体需求来确定态势抽象与表征方法.因此,研究了较为常用的COA生成方式,并进行对比分析,根据战役层次兵棋推演的特点,设计适用于战役层次COA智能生成方式;基于COA智能生成方式的关键信息需求,提出基于作战能力的战场态势的抽象与表征方法;该方法可以为数据驱动的COA智能生成模型提供恰到好处的输入,并在2020年“先知·兵圣”战略战役兵棋对抗演习中得到检验验证.

The intelligent generation of Course of Action(COA),especially in the Campaign-level,is a hotspot issue in the military field of intelligent decision making.At present,the methods based on knowledge rules are mostly used in the COA intelligent generation field,and a data-driven AI method represented by deep reinforcement learning is rarely used.One of the reasons is that data-driven AI methods rely on the valid situation information input.The situation information in the campaign-level operations is highly complex,and cannot be directly input to the AI model.The carefully designed situation abstraction and characterization model are needed to be inputs to drive the AI model.To solve the problem,the key is to take the specific COA intelligent generation mode as the traction,the situation abstraction and characterization method is determined according to the specific requirements of intelligent generation model.Thus,the presently common used COA generation mode is mainly researched,on the basis of COA generation mode based on the capability,according to the features of wargaming in the campaign level,COA intelligent generation mode suitable for campaign level is designed.According to the key capability requirement,the abstraction and characterization method of battlefield situation based on operation capability is proposed.The method can provide the proper input to COA intelligent generation model driven by data and is tested and verified by the strategic and campaign wargaming exercises of 'XianzhioBingsheng' of year 2020.And thus we generate a situation abstraction and expressing methods which is suitable for the special capability needs of the COA autonomous developing methods.This method present an effective way to drive the data-driven AI model.And it has been used in the strategic and campaign level wargaming contest.