CASIA OpenIR  > 毕业生  > 硕士学位论文
人机对抗中位置估计及其应用
潘毅
学位类型工程硕士
导师杨一平 ; 倪晚成
2018-05
学位授予单位中国科学院研究生院
学位授予地点北京
关键词信息素模型 位置估计 决策支持 作战推演 兵棋
其他摘要       人机对抗是人工智能的炼金石。由于机器博弈算法的发展和计算机硬件的更新换代,人工智能在人机对抗上取得了许多进展。不完全信息下的人机对抗问题,由于更为贴近实际,更为接近通用人工智能,近年来受到了越来越多的关注。
兵棋推演作为对现实战争的模拟游戏,是研究不完全信息下机器博弈问题的良好平台。在推演中,敌我双方尽量隐蔽自己的意图,秘密调兵遣将,以期给对手突然的打击。由于侦查、地形、伪装等因素限制,推演者无法实时准确掌握敌方作战单位的所处位置,是不完全信息的具体体现。位置信息的缺失,给指挥员判断敌方意图,正确评估作战威胁和战场态势造成了很大的困难。
       本文以陆军战术兵棋推演作为模拟对抗环境,针对陆军对抗这一不完全信息条件下的指挥决策中对敌方作战单位的位置估计问题,提出了一种用于表示敌方位置意图的信息素模型,并提出了基于该信息素模型的敌方作战单元位置估计算法,进一步将位置估计的结果应用于战场威胁评估,帮助计算机理解战场态势信息。本文的主要研究工作与创新点有:
  1. 归纳、分析陆军战术兵棋模拟对抗中的决策关键要素,提出了一种表征敌方位置意图的信息素模型。该模型使用信息素向量和敌方意图向量分别表征地图单元格的特性和敌方作战单位的意图:信息素向量将作战决策相关的因素建模为对作战单元散发着吸引或排斥力的“信息素”,地图上的各个位置在一定范围内散发这些信息素,不同位置散发信息素的量大小不同;通过对陆军战术兵棋的特征分析,本文使用一个8维信息素向量表示地图上每个位置的属性和特征,并给出了各项信息素的计算方法。敌方意图向量将作战单位的主观意图建模为作战单位对各种信息素的吸引或排斥的程度,形成一个对应8种信息素的8维作战单位意图向量。
  2. 基于上述信息素模型,本文提出了一种敌方位置估计算法。该算法以用信息素向量和作战单位意图向量共同计算的“信息素效用”作为评价指标。首先根据兵棋机动规则和视野信息,计算敌方作战单位在每一回合的可达范围,及可达范围中的隐蔽点;然后使用用信息素效用对可达范围里的各个位置进行评价,最终取效用最大的位置为对敌方作战单位当前的可能位置。经实验验证,该算法模型合理,计算快速。对于具有历史观察信息的敌方作战单位,top3位置估计结果准确率为73%,若容忍一个位置的偏移误差,估计准确率高于90%;对于完全没有历史观察信息的敌方作战单位的top3位置估计结果,若容忍一个位置的偏移误差,估计准确率大于40%,能够指出敌方作战单位的大致位置和方向。数据表明,本文提出的敌方作战单位的位置估计算法能够为己方提供有效的参考信息。
  3. 在本文位置估计成果基础上,提出了结合位置估计的战场威胁态势评估计算方法,对不结合位置估计和结合位置估计两种情况,分别进行了的威胁态势进行了评估,并与实际数据集中的威胁态势进行对比。对比结果表明,如不结合位置估计,在不完全信息下的威胁评估与实际威胁严重偏离;结合位置估计的威胁评估结果与实际威胁相近,有效减少了不完全信息对战场态势判断的影响。故本文使用结合位置估计的威胁评估,以估计值代替当前不可见作战单位所造成的威胁,能够更深层次的挖掘战场信息,更准确地表征战场态势。
  4. 基于信息素模型建模的决策因素分析成果,设计并实现了一款可视化的陆军战术辅助决策工具软件。该工具可以分析、显示地图中的多种关键信息,辅助推演者/指挥员制定、调整、优化作战计划。
;     Human-computer is the testbed for artificial intelligence. With the development of machine game algorithm and computer hardware, Artificial intelligence has made many progress in human-computer confrontation. The problem of human-computer under incomplete information is closer to reality and general actificial intelligence, it has attracted more and more attention in recent year.
    War-Game is a good platform for studying human-computer problem under incompleteinformation, which is the simulation of real war. War is the most typical problem of incomplete information game. Both sides try to conceal their intentions and deploy forces underhandedly, which aims to launch a death fight upon opponent. The lcak of the information of enemy is the main effect of incomplete information. Due to the limitations of investigation, terrain and camouflage, it’s impossible to know the accurate location of enemy’s units all the time, which affect the commander’s judgement of enemy’s intentions and the assessment of combat situation.
    This papers proposes a pheromone-based model and relevant method for enemy’s location and prediction in an army tactics War-Game system. The estimation results are further applied for the battlefield threat assessmen which is helpful to the mining of battlefield imformation.
The main works and results and innovation points in this paper are as follow:
  1. This paper analyses and summarize the key factors for moving a combat unit to a location and proposes a pheromone-based model to represent each unit’s intention. This model use a pherpmone vector and an intention vector to represent each locations property and each combat units’ intention. Pheromone vector model the key element related with war as pheromone which would attract or repell combat unit. Each location would emit all kinds of pheromone within a certain range. Based on the analysis of characteristics of army tactical War-Game, this paper use a 8-dimension vector to model each locations’ property which weight represent the amount of corresponding pheromone. Each enemy’s units’ intenion is model as the attracting or repelling degree for each pheromone, which is formed as an intention vector corresponding to 8 kind of pheromone.
  2. This paper proposes an algorthm to estimate enemy’s units’ location in every round. This algorithm calculates enemy’s moving range in each round with the visual information. Then the pheromone-based model is used to assess each location of moving range. The location which attracts enemy’s unit most may be the accurate location. The experiments results show that this algorithm is feasible and pratical. The accuracy of top3 locations estimation with historical observation information is 73%. If one offset is toleranted, the accuracy is over 90%. The accuracy of top3 locations estimation without historical observation information over 40%, if one offset is toleranted, which still can point out the general position and direction of the enemy units. The data shows that the estimation results are accurate enough to support for decision-making.
  3. This paper proposes a threat assessment method based on the results of location estimation. The threat assessment result with or without location estimation are invidually compared with the actual threat in the actual data set. The comparation shows that the assessment without location estimation is deviate from actual threat largely and the assessment based on location estimation is very similar with accurate threat. The comparasion shows that the assessment based on location estimation can represent the threat generated by invisible enemy’s units, which is helpful for mining of battlefield information deeply.
This paper develops a tool for map analysis based on the research of pheromone-based model which can visualize the information of war. This tool can analyse and display a variety of key factors of battleground, which can help commader make, adjust and optimize operational plan.
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/20985
专题毕业生_硕士学位论文
作者单位中科院自动化研究所
推荐引用方式
GB/T 7714
潘毅. 人机对抗中位置估计及其应用[D]. 北京. 中国科学院研究生院,2018.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
人机对抗中位置估计及其应用.pdf(3346KB)学位论文 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[潘毅]的文章
百度学术
百度学术中相似的文章
[潘毅]的文章
必应学术
必应学术中相似的文章
[潘毅]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。