Research on OTA Task Scheduling and Adaptive Fault-Tolerance Algorithm Under Cloud-Edge Collaboration

Over-the-air (OTA) updates in edge computing systems face practical challenges due to unstable network conditions and heterogeneous node capacities. To address this, we propose a task scheduling framework that integrates Deep Q-Network (DQN) reinforcement learning with a genetic algorithm. The model was tested with 120 OTA tasks across 50 industrial edge nodes. Results show that the proposed method reduces average scheduling latency by 23.9% and energy use by 18.5% compared to static baseline methods. Under network delays up to 300 ms, the task success rate remained at 99.2%, significantly outperforming FIFO and fixed-priority schedulers by 27.6%. The load distribution, measured by the coefficient of variation (COV), improved from 0.42 to 0.17. This indicates better task balancing among nodes. The framework adapts to fluctuating network conditions and provides a reliable solution for industrial and vehicle-mounted systems. However, long-term deployment effects and scalability in real-world environments require further investigation.