An approach for Offloading Divisible Tasks Using Double Deep Reinforcement Learning in Mobile Edge Computing Environment

Mobile Edge Computing (MEC) revolutionizes computing by decentralizing resources nearer to end-users, facilitating efficient task offloading to MEC servers, and addressing latency and network congestion. To tackle security challenges, we introduce a novel double deep reinforcement learning strategy for divisible task offloading in MEC setups. Our approach involves assessing offloading security levels based on task-source distances, creating a unique MEC state framework, and implementing dynamic task division for parallel execution across multiple nodes. By modeling task offloading through Markov Decision Process (MDP), we optimize Quality of Physical Experience (QoPE), considering time delays, energy usage, and security concerns. The proposed SQ-PER algorithm, integrating a self-adaptive Q-network with prioritized experience replay based on Double Deep Q-Network (DDQN), boosts learning efficiency and stability. Simulation outcomes underscore substantial reductions in time delay, task energy consumption, and offloading security risks achieved with the SQ-PER algorithm.