Task-Offloading Strategy Based on Performance Prediction in Vehicular Edge Computing

In vehicular edge computing, network performance and computing resources dynamically change, vehicles should find the optimal strategy for offloading their tasks to servers achieve a rapid service. this paper, we address multi-layered vehicle edge-computing framework, where each can choose one of three strategies task offloading. For best performance, propose prediction-based task-offloading scheme vehicles, in which deep-learning model is designed predict result (success/failure) service delay, then predicted with successful minimum delay chosen as final strategy. proposed model, an automatic feature-generation based on CNN capture intersection features generate new features, avoiding instability caused by manually features. The simulation results demonstrate that part has important impact prediction accuracy, higher Area Under Curve (AUC) than other methods. Compared SVM- MLP-based methods, average failure rate decreased 21.2% 6.3%, respectively. It be seen our effectively deal dynamic changes resources.