A gated Recurrent unit considering spatial correlation for short-term traffic volume forecasting in highway tunnels

Growing emphasis on low-carbon targets and energy conservation has significantly increased demand for efficient ventilation systems and energy optimization in highway tunnels, necessitating more accurate traffic forecasting. This study investigated spatiotemporal patterns of traffic flow within tunnels and at adjacent monitoring points, specifically considering unique tunnel ventilation requirements. Recognizing distinct traffic characteristics during different periods, we proposed enriching the dataset with comprehensive public holiday records to improve forecasting accuracy during these critical intervals. Through spatial correlation analysis, we established the quantitative relationship between inter-node distances and optimal sliding time steps that maximize spatial dependency, along with a methodological framework for estimating these temporal parameters. A comparative evaluation of prevalent deep learning architectures identified the Gated Recurrent Unit (GRU) as the most suitable baseline. Building on spatial correlation insights, we developed the Considering Spatial Correlation (CSC) method, which employs adaptive attention mechanisms to dynamically weight critical sliding time steps, enabling focused extraction of spatially influential features. By integrating CSC with GRU, we constructed the CSC-GRU model. Experimental comparisons with state-of-the-art baselines, including the Graph Neural Networks-Long Short-Term Memory (GNN-LSTM) model, demonstrate CSC-GRU model's superior performance, achieving a 9.9 % reduction in Mean Absolute Error (MAE) and a 13.5 % improvement in Symmetric Mean Absolute Percentage Error (SMAPE) over the best-performing baseline. Notably, the forecast curves exhibit enhanced responsiveness to abrupt traffic fluctuations during peak periods. This research provides a robust framework for optimizing traffic forecasting and energy management in tunnel ventilation systems, offering practical implications for sustainable infrastructure development.