Joint Estimation-Control-Scheduling for Wireless Networked Control Systems via Deep Reinforcement Learning

Wireless networked control systems (WNCSs) are increasingly adopted in industrial cyber-physical systems due to their flexibility and cost-effectiveness. However, limited radio resources and harsh industrial environments degrade the control performance of WNCSs. In this paper, we propose a joint estimation-control-scheduling (JECS) method based on deep reinforcement learning (DRL) to enhance the control performance of WNCSs. To this end, we first design a deep learning-based estimation method and a DRL-based control policy for individual subsystems to mitigate the effects of packet loss and maintain stability. Building upon the trained estimation and control models, we further develop a centralized scheduling policy with a Transformer-based architecture for a whole WNCS. The novelty of the scheduling policy lies in transforming discrete scheduling actions into continuous scores to prioritize subsystems, enabling scalable and efficient resource allocation under varying numbers of subsystems and network resource availability. Simulation results demonstrate that the proposed JECS method significantly outweighs baseline methods in terms of control performance and system capacity.