A Deep Reinforcement Learning Approach with Heuristic Optimization for Resource-Efficient Task Offloading in Mobile Edge Computing
Yanggai Liu, Zeyu Hou, Ke Lin, Lin Li · 2024
Mobile edge computing (MEC) environments are characterized by limited computational resources, posing a significant challenge to achieving optimal task allocation and processing. To address this challenge, this study proposes a novel Hybrid Deep Reinforcement Learning approach (HARL) that integrates heuristic algorithms with deep reinforcement learning. HARL exhibits swift adaptation to diverse environmental states, enabling more precise resource allocation and task offloading decisions. The algorithm optimizes the hyperparameters of the Deep Q-Network (DQN) using a genetic algorithm. This enhances the exploration capabilities of DQN within the search space, allowing it to potentially discover superior strategies and improve system performance and efficiency. Experimental results demonstrate that the HARL method can achieve a significant reduction in energy consumption exceeding 5.11% compared to traditional reinforcement learning algorithms.