Dual-Timescale Joint Optimization of Edge Caching, Task Offloading, and Resource Allocation Based Deep-Reinforcement-Learning

This paper studies dual-timescale joint optimization of edge caching, task offloading, and resource allocation for bidirectional computation tasks in cloud-edge-end network. In the considered network, task execution depends on both local input data and remote input data, while cache updating and online scheduling evolve at different timescales. Accordingly, the optimization objective is constructed as the long-term average utility, which jointly captures task delay and user-side energy consumption under cache, communication, and computation resource constraints. To solve it, a joint decision optimization algorithm is developed, in which improved PSO is used for long-timescale cache placement and DDPG is adopted for short-timescale task offloading and resource allocation. Simulation results show that the proposed algorithm achieves good convergence behavior, remains stable under different user scales, and outperforms the baseline strategy in terms of converged reward.