Digital Twin Construction and Resource Allocation on Internet of Vehicles

Aiming at the problems of data synchronization and resource limitation in smart driving on Internet of Vehicles (IoV), we investigate the construction of digital twins (DTs) and resource allocation on IoVs with the aim of achieving the optimal system performance. DT is a key technology in the future of autonomous driving. By digitizing physical components, it can effectively predict, analyze, and optimize various services of intelligent driving. First, we propose an accurate DT construction scheme (SL-ADTC) based on swarm learning (SL), which uses distributed learning algorithms to predict the queuing waiting time at relay nodes and the update frequency of DTs to solve data deviation and desynchronization for constructing DTs. Second, to address the problem of insufficient local computational resources of the task vehicles, a DT-based task division cooperative processing (DT-TDCP) mechanism is proposed. Then, considering the contradiction between task delay and system energy consumption, a joint optimization problem of computational offloading decision and resource allocation is proposed to minimize the total system cost under the premise of guaranteeing the high accuracy of DTs. Regarding the problem that the state space dimension is too high in reinforcement learning and the algorithm has difficulty in convergence, a DT-assisted multiagent classification proximal policy optimization (MACPPO) algorithm is proposed. Numerical results show that the proposed schemes outperform several benchmark schemes in terms of DTs construction accuracy, service latency, and system energy consumption.