Adaptive Information Synchronization and Prediction for Digital Twin-Enabled Autonomous Driving

While adopting Digital Twin (DT) systems keeps a growing trend in autonomous vehicle (AV) applications, ensuring efficient and accurate information synchronization between the physical entity layer and DT layer remains a major challenge. Solely using a fixed synchronization policy often suffers from high communication overhead, increased latency and low resource utilization efficiency in variable environments. Additionally, relying on a universal prediction model is hard to maintain prediction accuracy across varying traffic scenarios. In this paper, we propose an adaptive information synchronization and prediction model switching method. Our method can adaptively adjust the DT-physical layer synchronization frequency based on the similarity estimation between the two layers, achieved by switching between short-term and long-term prediction methods dynamically to better adapt to time-varying traffic characteristics, thereby improving the accuracy of the prediction results. Concurrently, the DT-enabled system utilizes the predicted vehicle state information to assist AVs in making overtaking decisions, optimizing traffic flow and improving road safety. Experimental evaluations show that our scheme effectively balances accuracy and resource efficiency in synchronization of DT, demonstrating its suitability for variable and resource-constrained vehicle network environments.