Interference resource allocation method based on constrained reinforcement learning

In modern manned-unmanned cooperative flight formations confronting advanced ground-based radar networks, traditional UAV swarm jamming strategies rely heavily on centralized command from fixed nodes. This centralization, while optimizing decision quality through global situational awareness, also introduces significant vulnerabilities. Dependence on a single node makes the system prone to failure in high-intensity and uncertain environments. Additionally, UAV platform constraints, such as limited power, sensing range, and communication bandwidth, challenge the feasibility of centralized information processing across the entire network. To address these issues, this study presents a decentralized decision-making framework, incorporating an adaptive cluster jamming resource allocation strategy based on collaborative multi-agent reinforcement learning. The Actor-Critic (AC) algorithm is employed to enable autonomous behavior analysis of UAV swarms via policy networks, while a value network is utilized to evaluate and guide policy iterations. The simulation results showed an interference mode allocation result of 20 drones interfering with 3 radars, verifying the possibility of this strategy working in complex dynamic environments.