Enhanced DDPG algorithm for latency and energy-efficient task scheduling in MEC systems

This paper presents an improved Deep Deterministic Policy Gradient (DDPG) algorithm for task scheduling in Mobile Edge Computing (MEC) systems, focusing on reducing latency and energy consumption. The enhanced DDPG incorporates a soft update mechanism for target network parameters and an experience replay buffer to improve stability and training efficiency. A composite reward function, considering both latency and energy consumption, is designed and compared in linear and nonlinear forms. Experimental results demonstrate that the nonlinear reward function achieves faster convergence and better system performance. The improved DDPG outperforms traditional algorithms like Shortest Job First (SJF), Round Robin (RR), and Earliest Deadline First (EDF) in terms of latency and energy consumption across different task loads. At a task load of 10, the improved DDPG achieved a latency of 250 ms and energy consumption of 965 J, significantly lower than other algorithms. Ablation experiments further confirm the positive impact of each enhancement, showing that the combination of soft updates and experience replay buffers significantly reduces both latency and energy consumption. The study concludes that the improved DDPG algorithm provides an effective solution for task scheduling in MEC systems, offering robustness and adaptability in complex network environments. Future research will focus on real-world deployment and further optimization of the algorithm for diverse MEC scenarios.