Online Traffic Flow Prediction for Edge Computing-Enhanced Autonomous and Connected Vehicles

Abstract

The development of edge computing based autonomous and connected vehicles (CAVs) provides a very promising solution for the construction of intelligent transportation system. Unfortunately, the existing methods are difficult to predict traffic flow accurately in such case due to not only the dynamic nature of the CAVs but also the considerable amount of missing traffic flow data. Based on this, we propose an online rolling traffic flow prediction method to provide support for the CAVs can be seen into practice. The new matrix factorization techniques proposed can learn the low-dimensional embeddings in the online setting and impute missing ones simultaneously. Moreover, instead of directly predicting the high-dimensional traffic flow data, a standard vector autoregressive (VAR) process is employed on low-dimensional embeddings to predict future values. Further, a multidimensional Cadzow method is proposed to solve the coefficient matrices of VAR efficiently even if there is noise. The simulation results on two real datasets show the applicability of the proposed method to online traffic flow prediction for edge computing-enhanced CAVs.