Intent-Driven Numerology Selection and Resource Scheduling for UAV-aided RAN Slicing via Diffusion enhanced SAC

To address the misalignment between UAV RAN slicing configurations and user service intents, this paper proposes an intent-driven numerology selection and resource allocation algorithm named DiSAC-HER, which integrates the hindsight experience replay (HER) mechanism into the diffusion-enhanced Soft Actor-Critic (SAC) framework. Firstly, targeting the dynamic networking and variable user service intents of UAV base station scenarios, the resource management problem for UAV-mounted base stations is formulated as a mixed-integer nonlinear programming model to realize the joint optimization of numerology parameters and bandwidth resources on dual time scales. Secondly, to solve the challenges of sparse rewards, low sample utilization, and poor policy convergence stability in the reinforcement learning training process under intent constraints, a diffusion model is embedded into the traditional SAC framework to optimize policy representation and accurately fit the optimal resource scheduling strategy in complex and dynamic network environments. Furthermore, the hindsight experience replay mechanism is adopted to resample and relabel sparse failed samples, mining valid experience from invalid samples and further improving the training efficiency and environmental adaptability of the algorithm in complex UAV networking scenarios. Finally, simulation results verify the performance of the proposed algorithm. The results demonstrate that the proposed DiSAC-HER algorithm can effectively match user service intents and improve network slicing response speed, achieving superior resource allocation accuracy and environmental robustness compared with baseline algorithms in dynamic UAV access network scenarios.