Traffic flow forecasting in smart cities with deep learning

Traffic flow prediction is a method of processing traffic data. Accurate traffic prediction information can provide a strong basis for traffic management decisions and can also allow drivers to choose more fluent routes for travel, thus avoiding or alleviating traffic congestion. Traditional traffic flow prediction data usually consists of vehicle speed and travel trajectories. Researchers obtain data by arranging traffic sensors at intervals along highways, and these methods have achieved good results when applied to suburban areas and highways. However, urban roads are densely populated and have complex traffic conditions, which are not suitable for the large-scale deployment of sensors to obtain the required traffic data, and therefore existing methods cannot be used for prediction. In response to this problem, we have attempted to use five models: SMA (Simple Moving Average), ARIMA (Autoregressive Integrated Moving Average), LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), and TCN (Temporal Convolutional Network), to predict the traffic flow at four intersections, and have conducted many experiments based on real traffic datasets. The results show that the TCN prediction model achieved the best forecast results, with RMSE, MAP, and MAPE values of 3.9, 2.3, and 16.1, respectively, indicating the predictive accuracy that can be achieved.