Traffic engineers have to grapple with unforeseen scenarios that display unpredictable traffic flow patterns. Predicting these patterns in real time is no doubt a challenge. Accidents cause heavy disruption in urban traffic flows. Scientists, particularly machine learning researchers, show interest in such predictions and have been toying with different models for the same. A team of computer scientists at Lawrence Berkeley National Laboratory have participated in a collaborative program for with California Partners Advanced Transportation Technology (PATH) for managing transportation corridors in California. Moreover, the aim of the program is to test Integrated Corridor Management approach in California Department of Transportation (Caltrans) decision-making in real time in case of unforeseen situations and develop corresponding response plans. The system has been put to test in Los Angeles County.
Machine Learning Technologies help in Real-time Learning
Ensemble learning forms the core of the new system. The model is integrated with novel algorithms. These use traffic data collected from Caltrans sensors on California highways. The Connected Corridors system enabled them to use real-time traffic data. The system they developed comprise scalable machine-learning library. The main motive of the researchers and traffic engineers was to predict traffic demands at various freeway entrances and anticipate flows at freeway exits. The ensemble technique in itself isn’t effective to predict real-time traffic flows. The machine learning technologies they employed enabled them to do just that—constant learning in real-time, including from the past data.
The contributors of the project believe that the Connected Corridors program if implemented well can lead to significant improvements in various traffic management strategies. These improvements notably pertain to traffic flows, traffic delays, and sudden changes caused by unforeseen incidents. The researchers hope to test the concept and deployment of the concept.
Of note, Berkeley Lab’s Laboratory Directed Research and Development (LDRD) program funded the project.