Transformer-based travel time estimation method for plateau and mountainous environments.

Area of Science:

• Intelligent Transportation Systems
• Machine Learning for Geospatial Analysis
• Wilderness Navigation Technology

Background:

• Current travel time estimation (TTE) predominantly focuses on urban settings, leaving a gap in wilderness area applications.
• Plateau and mountainous terrains present unique challenges for accurate TTE due to complex topography and variable conditions.

Purpose of the Study:

• To develop and evaluate a novel Transformer-based model for accurate travel time estimation (TTE) in wilderness environments.
• To enhance TTE model generalization for plateau and mountainous regions using meta-learning.

Main Methods:

• Developed a Transformer-based model integrating positional encoding and multi-head self-attention for TTE.
• Employed a meta-learning strategy to improve model generalization across diverse wilderness terrains.
• Utilized terrain-weather and spatio-temporal features from datasets in western China.

Main Results:

• The proposed Transformer model achieved a 14.89% improvement in MAPE for plateau environments and 12.20% for mountainous environments.
• Outperformed the MetaTTE-GRU model, a leading urban TTE method, in wilderness settings.
• Demonstrated superior accuracy compared to existing LSTM-based estimation models.

Conclusions:

• The Transformer-based model offers a robust and accurate solution for travel time estimation in complex wilderness environments.
• The meta-learning strategy enhances the model's applicability across varied and challenging geographical topographies.
• This research advances intelligent driving systems by extending TTE capabilities beyond urban areas.