Post-disaster spectrum resource shortages and user mobility issues are the main challenges to the effective application of drone communications in emergency housing. This paper establishes a UAV-based D2D communication network system model to simulate post-disaster emergency housing scenarios. The communication link instability caused by frequent user mobility in emergency housing scenarios is transformed into the minimum perturbation problem in graph theory, and an interference model is established. To address load imbalance and link congestion in post-disaster emergency housing scenarios, a DDPG deep reinforcement learning method is developed by combining DQN and DPG methods, and a dynamic routing strategy based on DDPG is designed. This strategy can dynamically and adaptively adjust the routing overhead of wireless links based on the network status in the emergency housing scenario, thereby ensuring stable communication in the emergency housing scenario. Under the DDPG-based dynamic routing strategy, the average jitter of the routing algorithm remains below 3.0 ms, and the node death time is delayed (138 s), demonstrating superior stability and energy balancing capabilities.
