Local spatial and temporal relation discovery model based on attention mechanism for traffic forecasting.

Area of Science:

• Intelligent Transportation Systems (ITS)
• Deep Learning
• Traffic Forecasting

Background:

• Accurate traffic prediction is crucial for ITS, but existing deep learning models struggle with asynchronous spatio-temporal correlations and historical data impact.
• Current methods lack interpretability regarding explicit spatial-temporal relationships in traffic networks.

Purpose of the Study:

• To enhance traffic prediction accuracy in intelligent transportation systems.
• To develop a model that extracts comprehensive and explainable spatial-temporal relevance in traffic networks.
• To address limitations in modeling asynchronous dependencies and historical data influence.

Main Methods:

• Proposed a novel attention-based local spatial and temporal relation discovery (ALSTRD) model.
• Implemented feature representation learning for latent traffic information.
• Utilized a local attention mechanism for asynchronous historical data dependencies.
• Employed an attention network and Pearson Correlation Coefficient for detailed influence extraction.

Main Results:

• The ALSTRD model achieved significant improvements in prediction accuracy over baseline methods.
• Demonstrated effectiveness in extracting fine-grained correlations and dynamic associations in historical traffic data.
• Showcased improved ability to elucidate spatio-temporal correlations, offering more robust explanations.

Conclusions:

• The proposed ALSTRD model offers superior traffic forecasting accuracy by effectively modeling complex spatio-temporal dynamics.
• The integration of attention mechanisms and Pearson Correlation Coefficient enhances model interpretability.
• This approach advances intelligent transportation systems through more accurate and explainable traffic predictions.