航空学报 > 2022, Vol. 43 Issue (S1): 726896-726896   doi: 10.7527/S1000-6893.2022.26896

基于证据网络因果分析的空中目标意图识别

张瑜, 邓鑫洋, 李明炟, 李新宇, 蒋雯   

  1. 1. 西北工业大学 电子信息学院,西安 710129
  • 收稿日期:2022-01-06 修回日期:2022-01-25 发布日期:2022-03-04
  • 通讯作者: 蒋雯,E-mail:jiangwen@nwpu.edu.cn E-mail:jiangwen@nwpu.edu.cn
  • 基金资助:
    国家自然科学基金 (62173272)

Air target intention recognition based on evidence-network causal analysis

ZHANG Yu, DENG Xinyang, LI Mingda, LI Xinyu, JIANG Wen   

  1. 1. School of Electronics and Information, Northwestern Polytechnical University, Xi'an 710129, China
  • Received:2022-01-06 Revised:2022-01-25 Published:2022-03-04
  • Supported by:
    National Natural Science Foundation of China (62173272)

摘要: 目标意图及时准确的估计对战场指挥决策具有重要的影响。考虑到战场环境复杂多变,其检测数据、信息、知识具有较高的随机性和认知不确定性,本文面向目标意图知识推理,基于结构因果模型,利用证据网络在不确定性处理方面的推理优势,提出了一种基于证据网络因果效应分析的空中目标意图识别方法。该方法通过d-分离检验构建兼容信息特性与意图多样性的因果证据网络模型,使模型同时考虑领域知识和数据间的条件独立关系,并将认知不确定性内嵌于推理引擎中,推理实现不确定战场环境下空中目标意图估计。此外,基于提出的因果证据网络模型和干预算子计算目标意图与属性之间的因果效应,挖掘目标意图估计蕴含的因果逻辑关系。仿真分析表明:本文提出的方法能够同时处理随机不确定性与认知不确定性,克服了传统意图识别方法缺乏因果分析的缺陷,是对目标意图和战场态势要素的深层次认知,实现了意图识别的“由果溯因”和“由因识果”。

关键词: 意图估计, 不确定信息, d-分离, 因果证据网络, 结构因果模型, 因果效应分析

Abstract: Timely and accurate estimation of target intention plays an important role in battlefield command decision-making. Considering the complex battlefield environment, the high randomness and ambiguity of test data, information and knowledge, this paper proposes an air target intention recognition method based on evidence-network causal effect analysis. A causal evidence network model that is compatible with information characteristics and intention diversity is constructed by the d-separation test. The model takes into account the conditional independent relationship between domain knowledge and data, and embeds cognitive uncertainty in the inference engine to achieve intention estimation of aerial targets in uncertain battlefield environment. In addition, based on the proposed causal evidence network model and intervention operator, causal effects are calculated and mined among target intentions and attributes. Simulation analysis shows that the proposed method can deal with both random and cognitive uncertainties, overcoming the defect of lacking causal analysis of traditional methods. With the proposed method, deep cognition of target intention and battlefield situation elements and “effect traceability” and “effect recognition” of intention recognition can be realized.

Key words: intention estimation, uncertain information, d-separation, causal evidence network, structural causal model, causal effect analysis

中图分类号: