To address the complexity of beyond-visual-range (BVR) air combat and the limitations of deep reinforcement learning, including slow convergence, weak generalization, and parameter sensitivity, this paper proposes a State-Event-Condition-Action (SECA)-driven autonomous decision-making framework with a timeline (T-SECA). T-SECA integrates finite-state machines with Event-Condition-Action rules, partitions BVR engagements into four timeline phases, and establishes a three-layer architecture comprising radar perception, tactical decision-making, and game confrontation. A 25-rule expert system supports phased decisions and dynamic responses across the engagement cycle, enabling interpretable and real-time tactical reasoning. Simulation studies in eight adversarial scenarios against two representative expert systems achieve win rates above 60% across all attack directions, indicating superior stability and effectiveness. Additional experiments demonstrate robustness under electromagnetic interference, sustaining approximately 60% win rates even under severe jamming, and show promising generalization in multi-aircraft confrontations, with 2v2 and 3v3 win rates remaining above 60%. These results demonstrate that T-SECA provides an interpretable and robust framework for autonomous BVR decision-making.
Wang et al. (Tue,) studied this question.