A Deep Spatial-Temporal Model Based on Discrete Wavelet Transform for Traffic Flow Forecasting

The widespread of the COVID-19 has seriously affected the normal order of transportation. To reduce the spread of the epidemic, many countries have issued policies to restrict resident travel, resulting in high dispersion and irregularity in traffic flow data. To overcome the influence of data discreteness on traffic flow forecasting, this study adopts discrete wavelet transform to split the traffic flow data into discrete quantities, changing trends and discrete baselines. Due to the lack of analysis of the hierarchy and direction among nodes in graph structure, this study proposes a tree convolution to analyze the spatial correlation of the traffic network. To solve the problem of high discreteness of traffic flow data, this study proposes a discrete forecasting module, which is used to convert discrete quantities separated by discrete wavelet transform into high-dimensional discrete features. Then, the predicted traffic data is segmented using discrete wavelet transform, and the newly segmented traffic trends and discrete baselines are then subjected to inverse discrete wavelet transform with the discrete model predicted by the discrete forecasting module to obtain the final traffic flow forecasting result. In comparative experiments, this study is evaluated against established high-level baselines. Our model demonstrates an average improvement of 23.25 % across the selected performance metrics, and it exhibits exceptional performance in long-term predictions.