Checkpoint data-driven GCN-GRU vehicle trajectory and traffic flow prediction

Area of Science:

• Urban planning and traffic management
• Data science and artificial intelligence
• Transportation engineering

Background:

• Massive traffic data necessitates advanced short-term traffic analysis for urban road networks.
• Accurate vehicle trajectory and traffic flow prediction are crucial for traffic management, including path planning and congestion warnings.
• Existing methods often rely on GPS data, which has limitations in coverage and data weight.

Purpose of the Study:

• To propose a novel method for accurate urban traffic flow prediction using traffic checkpoint data.
• To leverage the broad coverage, high reliability, and lighter weight of checkpoint data for traffic analysis.
• To improve the accuracy of vehicle trajectory prediction for enhanced traffic flow forecasting.

Main Methods:

• A checkpoint data-driven approach was used, transforming data into time-series vehicle trajectories.
• A Graph Convolutional Neural Network (GCN) was employed to learn spatial characteristics from checkpoint data and spatial relationships.
• A Gated Recurrent Unit (GRU) was integrated with GCN to capture spatiotemporal correlations, forming a GCN-GRU model for trajectory prediction.
• Traffic flow was forecasted using the output of the GCN-GRU trajectory prediction model.

Main Results:

• The GCN-GRU model demonstrated superior performance compared to single models (GCN, GRU, BiGRU, BiLSTM) in vehicle trajectory prediction.
• The GCN-GRU model achieved significant reductions in Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).
• The forecasted traffic flow using the GCN-GRU model showed a Mean Absolute Percentage Error (MAPE) of 0.18, confirming its reliability.

Conclusions:

• The proposed GCN-GRU model effectively extracts spatiotemporal features from checkpoint data for accurate vehicle trajectory prediction.
• This method provides a reliable approach for short-term traffic flow forecasting in urban road networks.
• Utilizing traffic checkpoint data with advanced deep learning models offers a promising alternative to GPS-based traffic analysis.