Dynamic Task Offloading and Migration Algorithms for Rail Transit Based on Mobility Prediction

With intelligence becoming a key trend in the evolution of high-speed railway (HSR) systems, the demand for computational resources from various onboard devices continues to grow. While cloud computing enhances the computational capacity of onboard equipment, it also introduces long transmission latency and exacerbates core network congestion. Edge computing has emerged as a promising solution to reduce transmission latency and relieve the burden on core networks by offloading computation-intensive tasks to edge servers located near the data source. However, the high mobility of trains introduces substantial challenges. Offloading tasks may be disrupted when a train exits the coverage area of its serving edge server, leading to task failures and service interruptions. Moreover, the short execution window of tasks and the dynamic variability of available computational resources further increase the risk of task re-execution. To address these challenges, this paper proposes a mobility-aware dynamic task offloading and migration framework tailored for high-speed rail transit environments. Specifically, a dynamic feature attention mechanism (DFAM) is employed to predict future train trajectories and estimate the time at which the train will depart the current edge server's coverage zone. Using the predicted trajectory, a predictive deep deterministic policy gradient (P-DDPG) algorithm is developed, integrating computational resource awareness and a dynamic noise adjustment strategy to enable rapid and adaptive offloading and migration decisions. Simulation results demonstrate that the proposed framework significantly reduces task execution latency and improves offloading reliability compared to conventional schemes.