Traffic flow prediction method based on EWT-XLSTM network

With the continuous advancement of urbanization, intercity expressway traffic flow in China has been increasing steadily. In expressway operation and management, section-level traffic flow is a critical metric for making informed and scientific decisions. However, current prediction methods often lack sufficient accuracy. A novel traffic flow prediction method is proposed based on Empirical Wavelet Transform (EWT) and the Extended Long Short-Term Memory (XLSTM) network. Firstly, historical traffic flow data are decomposed using EWT to reduce noise and extract latent trend patterns. Secondly, the decomposed sequences are transformed into labeled datasets using a sliding window approach. Finally, the data is subsequently trained and predicted using the XLSTM model. The proposed method is validated on real-world expressway traffic data from Zhejiang Province of China. The evaluation results demonstrate favorable performance, with a Root Mean Square Error (RMSE) of 7.65, a Mean Absolute Percentage Error (MAPE) of 0.11, and a Mean Absolute Error (MAE) of 5.59, indicating the method’s strong applicability and effectiveness.