An Improved Deep Reinforcement Learning-Based Method for Optimal Kill Chain Combination Selection

This paper addresses key challenges in kill chain optimization, including complex state structures, diverse resource constraints, and inefficient strategy generation. To solve these issues, an improved Advantage Actor-Critic algorithm based on deep reinforcement learning is proposed for selecting optimal kill chain combinations. A simulation environment that incorporates resource and time constraints is developed to model distributed combat scenarios. A feature extraction network integrating a gated recurrent unit with a multi-head attention mechanism is designed to enhance the model's ability to capture state-time dependencies and identify key features. The model is trained using the Advantage ActorCritic algorithm to improve policy exploration efficiency and convergence. Experimental results from a combat scenario show that the proposed method outperforms baseline algorithms, including the basic Advantage Actor-Critic algorithm, Proximal Policy Optimization, as well as genetic algorithms in Heuristic algorithms. The results confirm the method's adaptability and practicality, providing a feasible solution for selecting kill chains in complex combat systems.