Short-Term Traffic Flow Prediction Model Based on Temporal Convolutional Network and Bidirectional Gated Recurrent Unit

Abstract Short-term traffic flow prediction constitutes a core component of intelligent transportation systems, and it plays a critical role in alleviating traffic congestion, optimizing road-network operation and management, and improving urban travel efficiency. This paper proposes a novel short-term traffic flow prediction model that combines a temporal convolutional network with a bidirectional gated recurrent unit. To ensure the validity and reliability of subsequent modeling, three outlier detection approaches are utilized to preprocess raw traffic flow data. Specifically, the temporal convolutional network module is adopted to extract fine-grained short-range local features from traffic flow sequences and enhance feature information exploitation. The extracted spatiotemporal features are subsequently input into the BiGRU module to capture long-term temporal dependencies embedded in traffic flow data. Owing to this dual-module architecture, the model can fully excavate complex intrinsic correlations within traffic flow samples and attain favorable prediction accuracy. Moreover, an improved gray-wolf optimizer is developed to tune the hyperparameters of the combination model. This strategy effectively overcomes the drawbacks of traditional manual hyperparameter tuning, such as heavy computational burden and unstable model performance. To verify the effectiveness and generalization capability of the presented model, comprehensive experiments are carried out on two public traffic flow datasets that differ in geographical origin, sampling interval, and sample size. Comparative experimental results show that the proposed model yields lower prediction errors and better overall performance metrics, confirming its considerable practical application potential for real-world traffic prediction tasks.