Towards Data-Driven Vehicle Estimation for Signalised Intersections in a Partially Connected Environment.

Area of Science:

• Traffic Engineering
• Intelligent Transportation Systems
• Data Science

Background:

• Connected vehicles (CVs) offer potential for advanced traffic control, but infrastructure-based sensing may become obsolete.
• A key challenge in transitioning to a fully connected environment is managing incomplete data from mixed vehicle types.

Purpose of the Study:

• To develop data-driven methods for estimating vehicles approaching signalized intersections using partial information from CVs.
• To address the challenge of unknown connected vehicle penetration rates and their impact on traffic data.

Main Methods:

• Development of machine learning models to capture complex, nonlinear relationships between CV data and non-connected vehicle counts.
• Utilizing easily collectible real-world data for model training.
• Employing synthetic data from calibrated microscopic simulations when real data is insufficient.

Main Results:

• Machine learning models demonstrate high accuracy in estimating non-connected vehicles despite varying CV penetration rates.
• Training with synthetic data shows comparable performance to using only real data.
• The proposed methods are robust and perform well under simulated real-world conditions.

Conclusions:

• Data-driven estimation methods are effective for predicting vehicle presence at intersections with partial CV data.
• The approach is adaptable, utilizing both real and simulated data for robust model training.
• These methods show significant promise for near-future intelligent transportation system applications.