Hybrid Traffic Forecasting Model With Fusion of Multiple Spatial Toll Collection Data and Remote Microwave Sensor Data

In order to forecast the traffic flow more precisely, a novel hybrid model is proposed with multiple sources of traffic data in the spatiotemporal dimension. In the practical application of the proposed model, multiple sources of data are captured and fused from five toll collection gates and one remote microwave sensor based on the correlation analysis. A hybrid model, including the structure of stacked autoencoders and long short-term memory, is used. Stacked autoencoders are used to extract the spatial features. Long short-term memory is used to learn the temporal features. The comparisons of the hybrid model, non-hybrid model, fused data, and non-fused data are provided. The effectiveness of the hybrid model and the fused data demonstrated the best performance. The fused data presented more effective forecast, which encourages that the forecasting model could include more data source to improve the accuracy. Meanwhile, the selection of a suitable model should also be studied for better forecasting result in consideration of difference feature of the data source. The high-accuracy prediction could contribute to further traffic control and prompt the development of the intelligent transport system.