Short-Term Traffic Flow Prediction Method with Multiple Factors and Deep Learning

The growing complexity of the transport system has led to a surge in traffic pressure, accidents and a deteriorating traffic environment. Real-time and accurate traffic flow forecasting capabilities are required to achieve timely traffic guidance, improve road safety and travel efficiency, and enhance the traffic environment. One of the reasons for the difficulty in predicting traffic flows is that traffic conditions are highly susceptible to a variety of factors such as weather and traffic accidents, so there is a need for a method of predicting traffic flows that incorporates a variety of influences. This paper proposes that weather-related data, traffic incident effects and time-series data form a multidimensional factor to accurately predict short-time traffic flows. This paper combines the advantages of Long Short-Term Memory Network(LSTM) in multidimensional data processing with the powerful past and future information extraction capability of Bi-directional Gated Recurrent Unit(Bi-GRU), while introducing an attention mechanism to better capture the impact of different moments on the prediction results, and proposes a Multidimensional Factor Fusion Network(MFFN). The model is able to take full account of multiple factor influences and multiplex data to provide highly accurate forecasting results.