A Bidirectional Gated Recurrent Unit and Temporal Convolutional Network With a Self-Attention Mechanism to Improve Traffic Flow Prediction Performance

An effective temporal modeling approach is crucial for improving traffic flow prediction accuracy. Traditional traffic flow prediction methods have certain limitations in capturing long-term dependencies and enhancing computational efficiency. This is especially true when dealing with long-sequence data, where the prediction accuracy of these methods often falls short of expectations. This paper proposes a hybrid model that combines the Bidirectional Gated Recurrent Unit (BiGRU) and Temporal Convolutional Network (TCN) with a Self-Attention (SA) mechanism to enhance prediction performance. The BiGRU captures bidirectional temporal dependencies, while the TCN enhances training efficiency and models long-sequence dependencies through parallel computation. The self-attention mechanism further improves the model’s ability to capture long-term dependencies, enhancing overall prediction performance. The model is validated on several real-world traffic datasets and its performance is compared with traditional methods. Results show that the BiGRU-TCN-SA model reduces Mean Absolute Error (MAE) by 30.04%, 23.09%, 20.84%, 14.94%, and 11.57%, compared to LSTM, TCN, LSTM-TCN, BiLSTM-TCN, and BiGRU-TCN models, respectively. Root Mean Square Error (RMSE) is reduced by 20.91%, 15.39%, 11.99%, 9.27%, and 6.24%, respectively. We further validate the BiGRU-TCN-SA model using Akaike Information Criterion (AIC) and Schwarz’s Bayesian Information Criterion (SBIC) on ablation experiments, the proposed model achieves the lowest AIC (34.35) and SBIC (121.63) values compared to baseline models, demonstrating the superior performance of the BiGRU-TCN-SA model in traffic flow prediction tasks.