UAV swarm interception strategy based on temporal memory MADDPG

To address the shortcomings of Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm in UAV swarm interception scenarios, namely, their disregard for historical temporal information and reliance solely on current observations, leading to decision-making myopia and suboptimal collaborative interception performance, a Temporal Memory Multi-Agent Deep Deterministic Policy Gradient (TM-MADDPG) algorithm integrating Long Short-Term Memory (LSTM) is proposed. First, integrating LSTM into the actor network enables agents to extract situational evolution features and target motion trends from historical observation sequences, thereby generating forward-looking cooperative interception actions. Second, a sequence experience replay mechanism with corresponding buffers is designed to adapt to temporal decision input requirements. A composite reward function combining dense proximity rewards, sparse success rewards, and behavioral penalties is constructed to guide the UAV swarm in efficiently learning cooperative interception strategies. Finally, experiments were conducted with Blue Team employing intelligent maneuvering strategies across diverse adversarial scenarios. Comparisons with MADDPG and Meta-Weight Multi-Agent Deep Deterministic Policy Gradient (MW-MADDPG) algorithms demonstrated that the TM-MADDPG algorithm exhibits superior effectiveness and robustness in dynamic adversarial environments.