HCTD3: A Hierarchical Reinforcement Learning Approach for Mixed-Action Resource Allocation in Multi-UAV Cooperative Jamming

Resource allocation for multi-UAV cooperative jamming in modern electronic warfare faces significant challenges due to high-dimensional mixed action spaces, complex constraints, and dynamic environments. To address this, this paper introduces a hierarchical reinforcement learning algorithm, HCTD3. The algorithm mitigates complexity by decomposing the task into two sub-problems: Target selection (discrete actions) and power allocation (continuous actions). It employs the Gumbel-Softmax technique for high-level discrete selection and the Twin Delayed DDPG (TD3) algorithm for low-level continuous optimization, enabling efficient learning in the mixed action space. Experimental results demonstrate that, compared to traditional optimization and single-layer reinforcement learning methods, HCTD3 achieves a 37.2% average improvement in jamming effectiveness and converges 45% faster, showcasing its superior performance.