Game-Theoretic Deep Reinforcement Learning for Task Offloading in Heterogeneous MEC Systems

The emergence of new applications has increased the demand of mobile users for edge computing. Therefore, the optimization of base station load balancing and task offloading is particularly important. In this paper, a two-tier heterogeneous mobile edge computing (MEC) system based on Stackelberg Game (SG) is studied. The introduction of SG can improve user association flexibility while reducing base station load. The introduction of network slicing can classify users according to different types, which increases the universality of the proposed framework. To solve the above problem, we map an optimization problem of efficiency, computation delay and computation energy consumption. Due to the non-convexity of the optimization problem, the Multi-agent Deep Deterministic Policy Gradient (MADDPG) algorithm based on SG is proposed in this paper. The gradient descent strategy can converge faster and obtain the optimal solution of user association policy and unload policy. Through simulation, it is verified that the proposed scheme can obtain lower computation delay and computation energy consumption on the basis of solving the problem of base station overload.