Developing a Neural-Kalman Filtering Approach for Estimating Traffic Stream Density Using Probe Vehicle Data

Area of Science:

• Transportation Engineering
• Traffic Flow Theory
• Data Science

Background:

• Accurate traffic vehicle counts are crucial for intelligent transportation systems.
• Traditional methods often struggle with real-time data variability and noise.
• Probe vehicle data offers a promising but challenging source for traffic estimation.

Purpose of the Study:

• To develop a novel model for estimating vehicle counts on signalized roads.
• To improve the accuracy and reliability of traffic data using advanced filtering techniques.
• To explore the integration of neural networks for enhanced estimation performance.

Main Methods:

• Utilized the adaptive Kalman filter (AKF) for real-time traffic vehicle count estimation.
• Incorporated real-time probe vehicle data into the AKF model.
• Developed a hybrid adaptive Kalman filter neural network (AKFNN) approach.

Main Results:

• The AKF reduced prediction error by up to 29% compared to the traditional Kalman filter.
• The AKFNN approach further improved accuracy by up to 26% over the AKF.
• The AKF model's sensitivity to initial conditions was identified, highlighting the need for proper parameter selection.

Conclusions:

• The proposed adaptive Kalman filter (AKF) offers superior accuracy for traffic vehicle counts over traditional Kalman filters.
• The adaptive Kalman filter neural network (AKFNN) provides the highest accuracy by effectively estimating market penetration rates.
• Optimizing initial conditions is key to maximizing the performance of the AKF model.