Adaptive Task Scheduling in Heterogeneous Cloud-Edge Collaboration Systems via Improved A3C Framework

Cloud-edge collaboration systems have been extensively adopted in latency-sensitive applications. However, achieving efficient task scheduling in heterogeneous cloud-edge systems remains a significant challenge due to dynamic environments and complex resource variability. To address the high failure rates, long response times, and limited adaptability of existing algorithms, we propose a novel task scheduling method based on an improved Asynchronous Advantage Actor-Critic (A3C) framework. Specifically, the traditional A3C architecture is augmented with a Residual Recurrent Neural Network (R2N2) to effectively capture the stochastic dynamics of computational tasks and the variability of heterogeneous cloud-edge resources. Additionally, a dual-stage attention mechanism is introduced to explicitly model the interactions among task characteristics, time urgency, and resource states. Experimental results demonstrate that, compared with baseline methods, the proposed approach reduces average response time and energy consumption by 9% and 10%, respectively, while lowering the task failure rate by 12%, thereby improving overall scheduling performance under stringent constraints.