Dynamic Resource Allocation Method Based on Multi-Agent Deep Reinforcement Learning

This paper presents a Dynamic Resource-Target Assignment (Dynamic RTA) method for multi-force joint operations in dynamic battlefield environments, based on Multi-Agent Deep Reinforcement Learning (MADRL) using the Multi-Agent Double DQN (MADDQN) algorithm. The method models enemy threats, platform heterogeneity, and resource units constraints through a Markov Decision Process, integrating global rewards and stepwise penalties. Simulation results show that MADDQN improves cost-effectiveness by 16.06% compared to rule-based methods in an offshore targeted strikes scenario, demonstrating its robustness and efficiency. The research offers a data-driven decision-making framework for multi-domain joint operations, overcoming traditional rule-based limitations.