基于反事实的可解释空战关键节点确定方法OA

The core of the post-combat analysis of air combat is to achieve rapid assessment and tactical optimization through the analysis of critical nodes.However,existing studies have not fully considered the causal relationship between situational features and combat outcomes,and their results lack interpretability.To address this issue,an interpretable counterfactual-based method for determining critical nodes in air combat is proposed.First,counterfactual analysis is used to compute the global importance scores of features for feature selection,thereby providing weight coefficients for situational assessment.Next,an improved random forest model,integrated with the counterfactual framework,is applied for real-time prediction,exploring the threshold ranges of joint features.Finally,critical nodes are identified as either moments when there is a significant change in the situation and a shift in the predicted label,or when the prediction is"victory"but the friendly forces are at a disadvantage.By incorporating maneuver recognition and considering human factors such as pilot decision-making,the model's interpretability and accuracy are enhanced.Simulation experiments demonstrate that the proposed method effectively identifies critical nodes and provides a more realistic analytical perspective.