A team from the Lawrence Berkeley National Lab at the University of California (UC) is working with the California Department of Transportation (Caltrans) to use high performance computing (HPC) and machine learning to help improve decision making in real time from the agency when traffic incidents occur. .
City traffic roughly follows a periodic pattern associated with the typical “9 to 5” workweek schedule. However, when an accident does occur, traffic patterns are disrupted. Designing accurate traffic flow models for use during incidents is a major challenge for traffic engineers, who must adapt to unforeseen traffic scenarios in real time. The new research was conducted in collaboration with California Partners for Advanced Transportation Technology (PATH), part of UC Berkeley’s Institute for Transportation Studies (ITS), and Connected Corridors, a collaborative program to research, develop and testing an integrated corridor management approach to controlling traffic on major roads in California.
‘)} // ->
‘)} // ->
Caltrans and Connected Corridors are implementing the new system on a trial basis in Los Angeles County as part of the I-210 pilot project. Using real-time data from Southern California partners at the city, county and state level, the goal is to improve Caltrans real-time decision making by executing response plans coordinated with multi-jurisdictional traffic incidents to limit the negative impacts of these events. The first iteration of this system will be deployed in the cities of Arcadia, Duarte, Monrovia and Pasadena in 2020, with plans for future deployments statewide.
The new system uses ‘ensemble learning’, which is the art of combining a diverse set of learners (individual models) to improve, on the fly, the stability and predictive power of the model. Although the concept has been explored by machine learning researchers for some time, the traffic flow model is special due to its temporal characteristic; The traffic flow measurements are correlated over time, as are the prediction results of different individual models. In the Berkeley Lab-Caltrans collaboration, the overall model takes into account the mutual dependence of the submodels and allocates the “voting shares” to balance their individual performance with their co-dependence. The ensemble model also values recent prediction performance more than older historical performance. In the end, the combined model is better than any of the unique models used in the tests in terms of the accuracy and stability of the predictions.
Using data collected from Caltrans sensors on California highways, the project produced new algorithms that resulted in an accurate prediction on a continuous 15-minute basis. The team then validated and integrated the new algorithms using real-time traffic data collected using the Connected Corridors system: a streaming-based real-time transport data hub in which Spark MLlib ( scalable machine learning library) provides machine learning models. that can be used in the overall learning framework offered. The specific implementation of this work was to generate predicted traffic flows at points where detection was present on the highway. This in turn could be used to forecast traffic demands at freeway entrances and traffic flows at freeway exits.
“There are many methods of forecasting traffic flow, and each can be beneficial in the right situation,” said Sherry Li, mathematician in the Computational Research Division (CRD) at Berkeley Lab. “To alleviate the pain of relying on human operators who sometimes blindly trust a particular model, our goal was to integrate multiple models that produce more stable and accurate traffic forecasts. We did this by designing a set learning algorithm that combines different sub-models. “
