An Efficient SAC Algorithm with Adaptive Mixture Model for Multiple Resource Allocation in MEC

Abstract

In mobile edge computing, multiple heterogeneous tasks require different trade-offs between communication and computing resource, making joint offloading and resource allocation challenging. We propose an adaptive parameterized mixture model to characterize task-source composition and demand heterogeneity. Based on this model, we formulate a joint optimization problem for task offloading, transmit power, edge computing resources, and bandwidth allocation. To reduce the policy search complexity, a dimension-reduced Soft Actor-Critic algorithm with analytical bandwidth allocation is developed. In the proposed algorithm, the policy learns offloading decision, power control, and computing-resource allocation, while the bandwidth subproblem is transformed into a convex optimization problem and solved through an analytical solution. Simulation results show that, compared with the end-to-end SAC baseline, the proposed algorithm improves the average reward by 24.1% and reduces policy-network complexity by $\mathbf{1 2. 8 \%}$, demonstrating better performance with lower learning complexity.