Physics-informed multi-step real-time conflict-based vehicle safety prediction.

Area of Science:

• Roadway safety
• Traffic engineering
• Artificial intelligence

Background:

• Real-time vehicle safety prediction is crucial for proactive collision avoidance.
• Current methods may lack accuracy in predicting complex traffic scenarios.
• Drivers need advance warnings to implement evasive actions.

Purpose of the Study:

• To develop a physics-informed, multi-step, real-time conflict-based vehicle safety prediction model.
• To enhance roadway safety by providing timely and accurate conflict risk assessments.
• To enable drivers to take appropriate evasive actions based on predicted risks.

Main Methods:

• Combining traffic shockwave properties (physics-informed) with deep learning features.
• Developing a multi-step prediction model for future vehicle safety indicators.
• Utilizing a customized loss function to prioritize high-risk events.

Main Results:

• The proposed model demonstrates superior prediction accuracy compared to benchmark methods.
• Numerical experiments validate the model's effectiveness in real-time safety prediction.
• Sensitivity analysis provides insights for practical parameter selection.

Conclusions:

• The physics-informed deep learning model significantly enhances real-time vehicle safety prediction.
• The model's ability to predict conflict risk levels aids in mitigating potential collisions.
• This approach offers a promising direction for intelligent transportation systems and safer roadways.