QoE-Oriented Distributed Mobile Edge Computing Offloading Algorithm Based on D3QN and LSTM

In a Mobile Edge Computing (MEC) environment—characterized by heterogeneous, multi-user, multi-task, and multi-node scenarios—designing efficient, intelligent offloading strategies to enhance system performance and user experience has become a key research focus. This paper proposes a distributed intelligent offloading optimization algorithm based on D3QN (Dueling Double Deep Q-Network). By integrating the dueling network architecture with the double Q-learning mechanism, our approach effectively reduces the risk of Q-value overestimation while improving learning stability and convergence speed. To better capture dynamic fluctuations in edge-node workload, we incorporate an LSTM network that models each node’s computing load state over successive time slots. This temporal awareness enables the agent to make more informed offloading decisions. Each task is characterized by task size, computational density, and maximum tolerable delay; a queue-scheduling module uses the Analytic Hierarchy Process (AHP) to assign priorities, ensuring that urgent tasks are handled first. The reward function defines a Quality of Experience (QoE) metric that jointly considers task delay, energy consumption, urgency, and drop penalties—guiding the agent toward globally optimal scheduling policies. Simulation results demonstrate that the proposed AHP-D3QN-LSTM algorithm outperforms traditional offloading methods across multiple metrics—including QoE, average delay, energy efficiency, task completion rate, and convergence speed—validating its effectiveness and practicality in multi-task MEC scenarios.