An Improved Proximal Optimization Method for Cloud–Edge–End Collaboration in Railway Surveillance Networks

Rail line safety is the foundation for the operation of high-speed railways (HSRs). With the gradual acceleration of the informatization and intelligent construction of HSRs, mobile edge computing enhances rail line video surveillance by providing intelligent solutions for perimeter intrusion detection, meeting the safety demands of the smart rail era. However, cloud-edge–end collaboration in railway video surveillance faces issues of uneven resource allocation, suboptimal task partitioning and inefficient offloading strategies, resulting in low resource utilization and increased latency. To address these issues, we propose an improved proximal policy optimization algorithm for computation offloading and resource allocation in cloud-edge–end collaboration. We first construct a constrained latency minimization optimization model, and this model is transformed as a semi-Markov decision process model. Subsequently, considering the dynamic characteristics of HSR channel, we designed an improved proximal policy optimization (PPO)-based method for scheduling policy optimization, and incorporating a behavioral cloning network and meta-learning module to enhance the efficiency of experience utilization. Additionally, a composite loss function is introduced to guide stable policy exploration, incorporating clipping optimization and policy entropy regularization to suppress gradient explosion during policy updates, which can improve the efficiency of decision-making. Simulation experiments demonstrate that the proposed algorithm significantly reduces latency and resource occupancy rates, providing theoretical support for cloud-edge–end collaboration in HSR.