Vehicle repacking strategy and enhanced asynchronous advantage actor-critic for multi-objective task scheduling and orchestration in cloud–edge vehicular networks

Vehicular cloud networks (VCNs) enhance task scheduling and improve driving experiences within vehicle networks. VCNs demonstrate greater resource efficiency; however, they experience significant challenges due to prolonged response times referred to as end-to-end delay (EED). Due to the extended duration required by VCNs, vehicular edge networks (VENs) have been introduced as a viable solution for applications necessitating quick responses. VENs face challenges, including vehicle disparities and limitations in computing and storage capacities. Existing heuristic solutions demonstrate inadequate performance and insufficient robustness in addressing these issues. This study presents a network system integrating edge and cloud computing with an enhanced asynchronous advantage actor-critic (A3C) deep reinforcement learning (DRL) methodology. The system organizes and assigns tasks based on available resources and possible task modifications. The architecture relies fundamentally on the vehicle repacking (VR) method, reorganizing vehicle locations within the network to mitigate compute and storage resource overload. The system optimizes resource utilization and equitable job distribution through appropriate car readjustment, reducing traffic and improving performance. Our neural network architecture classifies incoming applications as delay-sensitive or compute-sensitive based on their requirements, thereby improving scheduling efficiency. Simulation results indicate that our solution is effective and outperforms non-DRL and DRL-based approaches. The model demonstrates a reduction in end-to-end delay by 12.04%, an improvement in resource utilization by 49.58%, an enhancement in service quality by 58.76%, a decrease in packet losses by 46.33%, and a reduction in overhead by 71.94%. These results underscore the significant advantages of the VR strategy for VENs’ performance.