关键词: 人机协同, 强化学习, 可解释性, 智能空战, 意图识别

Abstract: Manned-unmanned aerial vehicle (UAV) cooperation represents a critical operational paradigm for future air combat, where deep reinforcement learning serves as a key enabling technology. However, the "black-box nature" of deep reinforcement learn-ing renders the learned strategies difficult to interpret and trust, making explainable deep reinforcement learning essential for achieving intelligent cooperative air combat. This paper proposes a Bayesian-Shapley framework-based explainable deep rein-forcement learning method, which enables interpretable modeling and verification of the decision-making process, thereby as-sisting pilots in understanding UAV decision logic. The proposed approach first constructs a decision intent analysis framework for cooperative missions using dynamic Bayesian networks, capable of identifying critical decision nodes in trajectory segments. Subsequently, it employs the Shapley value-based contribution assessment algorithm to achieve state-level quantitative analysis of decision rationale at key nodes. Finally, by reconstructing the state input space of the deep reinforcement learning model, the method significantly enhances model interpretability and trustworthiness while maintaining original policy performance, with the effectiveness of the explanatory results validated through state space ablation simulations.

Key words: Human machine collaboration, Deep reinforcement Learning, Interpretability, Intelligent air combat, Intention identification