Dynamic Offloading of Complex Tasks Based on Generative Adversarial Q-Networks in Edge Computing

In edge computing environment, dynamic offloading complex tasks is a key issue that affects the overall system performance and user experience. In cloud-edge-end collaborative systems, efficiently offloading tasks with dependencies and data sensitivity, while maximizing resource allocation and quality of service, is a challenging problem for edge service providers.To address this, we propose an optimization method based on the Generative Adversarial Q Network (GAQN).By incorporating Generative Adversarial Networks (GAN), we enhance Q-value estimation, improving the accuracy and stability of the policy. Specifically, we model complex task offloading in edge computing as a Markov Decision Process (MDP). Then,We design and implement GAQN, combining a generative adversarial network with a dueling deep Q network (Dueling DQN) to optimize task scheduling. Experimental results demonstrate that GAQN exhibits good stability and convergence with proper training, yielding optimal Q values. In terms of service quality, GAQN effectively reduces total system delay and energy consumption. Additionally, GAQN adapts to different network environments and efficiently handles task dependencies and data sensitivity, ensuring data privacy and smooth task execution.