Research on Traffic Flow Forecasting Based on Deep Learning

Traffic flow forecasting (TFF) is the key technology of intelligent transportation systems, plays a vital role in intelligent transportation and has attracted the attention of researchers worldwide. Forecast methods and models based on deep learning (DL) are the current research hotspots in this field. On the basis of the main methods and models of TFF, this paper mainly reviews the related research on TFF based on DL. First, from the perspective of scientometrics, the researchers, countries, and institutions of TFF based on DL are counted, and the cocitation network of keywords, journals, and authors is analysed. Then, TFF methods based on DL are reviewed from three aspects: time series, space-time, and spatiotemporal graphs. This paper focuses on research on forecast methods based on spatiotemporal graphs, clarifies the research trends in this field from the aspects of graph spatiotemporal networks, graph autoencoders, and graph attention networks, and summarizes the structure and characteristics of different forecast models. Finally, from the aspects of applied research and model research, the follow-up research issues, challenges, and future research directions in this field are discussed.