Time and Energy-Sensitive End-Edge-Cloud Resource Provisioning Optimization Method for Collaborative Vehicle-Road Systems

With the continuous development of information technology, intelligent transportation system has gradually become the trend of future transportation. However, the increasing number of time-sensitive and computation-intensive applications in intelligent transportation systems has brought severe challenges to resource-limited vehicles. The end-edge-cloud hierarchical computing architecture is an effective means to cope with this challenge. In the collaborative end-edge-cloud vehicle-road system, vehicle users can offload time-sensitive tasks to nearby roadside units to ensure the timing requirement and offload computation-intensive tasks to the cloud to meet their needs of computing power. However, task offloading also leads to additional transmission latency and energy overhead. In addition, tasks may also suffer from errors during transmission, resulting in degraded reliability. Therefore, to ensure the user experience of vehicles in the collaborative end-edge-cloud vehicle-road system, a multi-agent reinforcement learning based resource scheduling scheme is proposed. The scheme makes full use of the end-edge-cloud architecture’s characteristics and adopts the centralized training-decentralized execution framework to construct a deep neural network which decides the optimal offloading and computing resource allocation for tasks and hence optimizes system latency and energy consumption under the reliability constraint. To verify the efficiency of the proposed scheme, a metric named utility value is adopted in the experiment to show the improvement on latency and energy efficiency. Experimental results show that compared with the existing approaches, the utility value increased by our scheme can be up to 221.9%.