Dynamic Task Allocation of Edge–Cloud System for Low-Altitude Economy via Multiagent Actor–Critic-Queuing Computational Framework

The rapid development of low-altitude economy has driven edge-cloud systems to develop resource-optimal task management strategies. The key challenge lies in optimizing computing resource allocation to guarantee real-time responsiveness and feedback. Therefore, to address the task allocation problems within the edge-cloud system for low-altitude economy, we propose a computing framework named multi-agent actor-critic-queuing (MA2CQ). Leveraging queueing theory and a multi-agent actor-critic architecture, MA2CQ optimizes resource efficiency by minimizing long-term latency and maximizing throughput, formalized via the Witte performance metric. The framework employs rapid parameter iteration updates through the policy gradient reinforcement learning algorithm, enabling real-time scheduling and quick adaptability. Furthermore, we mathematically prove the convergence of the algorithm. The experimental results validate the effectiveness of the designed algorithm.