Short-Term Traffic Flow Prediction for Urban Expressways by Combining Variational Mode Decomposition and Deep Learning

Precise short-term traffic flow prediction plays a vital role in modern intelligent transportation systems. In recent years, Long Short-Term Memory (LSTM) neural networks have emerged as a go-to solution for predicting traffic patterns. This paper combines data decomposition with deep learning and designs corresponding deep learning models for prediction based on the characteristics of different IMF components. The predicted results of each component are then weighted and fused using a genetic algorithm based on energy ratio. A Variational Mode Decomposition–LSTM with Weighted Fusion Prediction Model (VMD-LSTM-WFP) Model is proposed. Taking traffic flow data from a road section in D city, China, as an example, prediction analysis is conducted. Experimental results show that, on average, the VMD-LSTM-WFP model demonstrated a reduction of 59.42% in RMSE and 58.43% in MAE, alongside a 26.97% improvement in the R² metric, when compared to the other models. These results demonstrate the effectiveness of the VMD-based decomposition and deep learning methods in improving prediction performance and provide valuable references for future intelligent transportation management.