自动化学报2025，Vol.51Issue(6)：1248-1260,13.DOI:10.16383/j.aas.c240468
基于异构图神经网络的可解释兵棋态势预测方法
An Interpretable Wargame Situation Prediction Method Based on Heterogeneous Graph Neural Networks
摘要
Abstract
In complex and changeable modern wargame simulations,accurate battlefield situation prediction and in-terpretation are crucial for high-quality decision-making.To address the challenges of difficult expression of com-plex situations and insufficient model interpretability in wargame deduction,this paper proposes an interpretable wargame prediction model WarGraph based on heterogeneous graph neural networks.The model consists of three modules:Multi-relational graph modeling,temporal analysis,and interpretable prediction.We first combine replay data with prior knowledge to construct a multi-relational heterogeneous graph,effectively modeling the intricate re-lationships between the environment and the operators.This enables capturing the complex interactions between combat units and the environment,realizing the representation of complex deduction situations.Then by lever-aging Transformer-based temporal analysis,we dynamically track the overall situation evolution and use attention mechanisms to identify key decision-making moments.This model can not only accurately predict the outcome of battles in wargame replays,but also the introduction of the attention mechanism enables a better explanation of the key factors in decision-making.Using replay data of 108 matches from the"MiaoSuan·ZhiSheng"wargame platform in 2021,the results show that the proposed model achieves a prediction accuracy of up to 90.91％,about 9.09％higher than the baseline models.Visualization of the attention coefficients demonstrates that the model captures critical moments in the decision-making process,which further validates its interpretability.关键词
兵棋推演/态势预测/图神经网络/可解释分析/深度学习Key words
Wargame deduction/situation prediction/graph neural network/interpretable analysis/deep learning引用本文复制引用
陈露,尚家兴,刘大江,张玉芳,倪晚成..基于异构图神经网络的可解释兵棋态势预测方法[J].自动化学报,2025,51(6):1248-1260,13.基金项目
中国科学院自动化研究所复杂系统认知与决策实验室开放基金(CASIA-KFKT-10)资助Supported by Open Fund of the Laboratory of Cognition and Decision Making for Complex Systems,Institute of Automation,Chinese Academy of Sciences(CASIA-KFKT-10) （CASIA-KFKT-10）
