Spatial Extension of Road Traffic Sensor Data with Artificial Neural Networks

Area of Science:

• Transportation Engineering
• Artificial Intelligence
• Network Analysis

Background:

• Accurate traffic flow estimation is crucial for efficient transportation network management.
• Monitoring all road links is often infeasible due to cost and infrastructure limitations.
• Existing methods may not fully leverage available data for comprehensive traffic insights.

Purpose of the Study:

• To develop and evaluate a method for estimating traffic flows on unmonitored road segments.
• To utilize data from monitored links to infer conditions on non-monitored segments.
• To assess the effectiveness of artificial neural networks (ANNs) for this traffic estimation task.

Main Methods:

• Implementation and testing of single-layer feed-forward artificial neural networks (ANNs).
• Supervised learning approach using datasets generated via traffic simulation techniques.
• Varying ANN configurations, including the number of neurons and dataset generation methods.

Main Results:

• The proposed ANN method achieved very good results when travel demand patterns were known and used for training.
• Promising results were obtained even when travel demand patterns were not explicitly considered.
• ANNs with fewer neurons demonstrated higher effectiveness compared to those with more neurons for this specific problem.

Conclusions:

• Artificial neural networks offer a viable approach for estimating traffic flows on unmonitored road links.
• The accuracy of the estimation is significantly influenced by the inclusion of travel demand patterns in the training data.
• Optimizing ANN complexity, specifically reducing the number of neurons, can enhance performance in traffic flow prediction.