Reinforcement Learning for Adaptive Sampling: Managing Network Resources in Cloud-Based Cyber-Physical Systems

This paper presents a novel approach to optimizing sampling frequency in cloud-based cyber-physical systems (CPS) using reinforcement learning (RL). We address the critical challenge of balancing control performance with network resource utilization in scenarios where the controller is deployed in the cloud and subject to variable network conditions. Our proposed solution employs a Proximal Policy Optimization (PPO) algorithm to dynamically select from discrete sampling frequencies based on system state and network conditions. Using a vehicle speed control system as a case study, we demonstrate that our RL-based approach significantly improves control performance while efficiently managing network resources. Results show smoother system response and reduced network congestion compared to random frequency selection strategies.