【摘要】 在空战决策过程中，如何有效识别关键状态并提升智能体在这些状态下的决策能力，已成为强化学习算法提升空战决策性能的重要研究方向。针对空战决策中的智能体问题，提出一种基于深度强化学习的动态切换策略框架，旨在提升智能体在复杂空战环境中的决策质量。研究通过表示学习和聚类分析技术对高维状态空间进行降维与分类，以识别关键状态：在非关键状态下，使用深度强化学习算法进行决策；在关键状态下，采用逆向动力学模型生成目标状态的对应动作序列，并利用平行仿真策略在多个仿真环境中执行动作序列，快速逼近目标状态。仿真结束后，通过优势值评估选择最优决策路径。实验结果表明，该方法能够提升智能体在关键状态下的决策能力，从而为复杂空战环境中的智能决策提供新的解决方案。更多还原

【Abstract】 In air combat decision-making, effectively identifying the key states and improving the decision-making ability of intelligent bodies in these states is the key research direction of reinforcement learning algorithms. In this paper, a dynamic strategy switching framework built by deep reinforcement learning was proposed for the intelligent body problem in air combat decision-making, aiming at increasing the decision-making quality of the intelligent body in the complex environment. This study identified critical states using dimensionality reduction and classification of highdimensional state space through representation learning and cluster analysis techniques in the non-critical state, a deep reinforcement learning algorithm was employed for decision-making; in the critical state, an inverse dynamics model was adopted to generates the target state’s corresponding action sequence and a parallel simulation strategy was utilized to execute the action sequence in multiple simulation environments to approximate the target state rapidly. At the end of the simulation, the optimal decision path was determined by advantage value evaluation. The experimental results show that the method can improve the decision-making ability of the intelligent body in critical states, providing a new solution for intelligent decision-making in complex air combat environments.更多还原