<p>The development of multi-access edge computing and ultrareliable low-latency wireless technology has prompted the evolution of the traditional tightly coupled controller-equipment system into the industrial wireless control system. Such wireless control systems enable on-demand deployment of control tasks in edge or local controllers and are appropriate for flexible intelligent manufacturing. However, unlike public data services, control tasks have strong spatiotemporal correlation features, and traditional offloading schemes are not applicable for such space-time-dependent control tasks. Therefore, a novel double deep graph convolutional Q-network (DDGCQ) model is proposed to optimize the control task scheduling scheme and dynamically allocate networking and computing resources for minimizing the weighted sum of task processing delay and energy consumption. In the proposed model, the graph convolutional and directed graph convolutional networks are integrated with the Q-network to extract the spatiotemporal correlation features of control tasks. The simulation results showed that the proposed DDGCQ model has excellent learning capability for spatiotemporal control tasks. Furthermore, compared with baseline methods, the proposed DDGCQ-based task-scheduling method achieved the minimum weighted cost of latency-energy consumption while maintaining a task execution success rate exceeding 95% under varying environment parameters.</p>