Energy-efficient task offloading in user-centric UAV-MEC networks: A discrete soft actor–critic approach with multi-AP cooperation

The integration of unmanned aerial vehicles (UAVs) with mobile edge computing (MEC) in user-centric cell-free (UCCF) networks offers significant potential for enhancing computational capabilities but also creates challenges for energy-efficient operation. This paper addresses these challenges by proposing an energy-efficient sequential offloading framework for UCCF-assisted UAV-MEC systems. The original resource allocation problem couples AP selection with discrete transmit-power control; to make the problem tractable, we decompose it into two coordinated stages. First, a lightweight signal-to-interference-plus-noise ratio (SINR)-ranking heuristic constructs the user-centric AP cluster. Second, given the selected topology, a user-centric discrete soft actor–critic (UC-DSAC) agent learns the discrete transmit-power policy to minimize long-term UAV energy consumption. Extensive simulations demonstrate that the proposed framework achieves up to 15% energy savings compared with state-of-the-art baselines while maintaining competitive rate and delay performance across varying UAV altitudes, AP densities, and preset cluster-size limits. The results support the effectiveness of this sequential AP-selection and power-control framework in balancing energy efficiency with quality-of-service requirements in dynamic UAV-MEC environments.