Multi-View Travel Time Prediction Based on Electronic Toll Collection Data.

Area of Science:

• Intelligent Transportation Systems
• Traffic Engineering
• Machine Learning

Background:

• Accurate vehicle travel time prediction is crucial for intelligent expressways, aiding management and traveler planning.
• Current methods lack differentiation for vehicle types, road network proximity, and in-depth spatial-temporal analysis.

Purpose of the Study:

• To propose a novel Multi-View Travel Time Prediction (MVPPT) model.
• To address limitations in existing travel time prediction research by incorporating vehicle type differences and spatial-temporal features.

Main Methods:

• Analysis of travel times across different vehicle types to identify key differences.
• Construction of multiple travel time features, including a novel spatial proximity feature.
• Utilizing Convolutional Neural Networks (CNN) for spatial correlation, spatial attention for key information, Bidirectional Long Short-Term Memory (BiLSTM) for temporal correlation, and time attention for temporal information.

Main Results:

• The MVPPT model effectively analyzes travel times for distinct vehicle types.
• Novel spatial proximity features were integrated into the prediction model.
• Experimental results on large-scale real traffic data show the proposed model outperforms state-of-the-art methods.

Conclusions:

• The MVPPT model offers a significant advancement in vehicle travel time prediction for intelligent expressways.
• The model's ability to process multi-view features, including spatial-temporal dynamics and vehicle types, enhances prediction accuracy.
• This research provides a more robust solution for traffic management and personalized travel planning.