Follow Datanami:
July 20, 2016

Machine Learning on Track with Rail Trials

Boasting one of the most advanced rail networks in the world, Japan is investigating the use of artificial intelligence and a machine learning technology approach that would attempt to add what promoters call a “train time delay” function intended to provide riders with up-to-date route-planning information.

Fujitsu Ltd. (TYO: 6702) said this week it has begun field trials in Japan for a delay time prediction engine developed with partner SRI International of Menlo Park, Calif. The trial also leverages a route-planning app developed by partner Jorudan Co. that claims about 10 million monthly users.

The partners said the time delay engine learns from past train delay information. Updates on train delays are delivered as a cloud service. The partners said the new function would be added during the field trial to Jorudan’s app, called Norikae Annai.

The prototype function machine is designed to learn from past railway operations data and combine that with train-delay information submitted by users of Norikae Annai in order to make accurate predictions. Delay predictions would then be displayed in real-time on a route search results page based on current railway operations data and user notifications.

Jorudan’s Android-based app was initially designed to provide commuters with estimates on delays, fares and estimated travel times for different forms of public transportation. The service is heavily used when major rail lines are disrupted by accidents or disasters, forcing commuters to look for alternatives such as buses or other rail lines.

During the field trial, Fujitsu said it would provide predicted train delay times using its AI technology along with machine learning tools for the Norikae Annai service.

Based on the results of the Japanese field trial, Fujitsu said it expects to improve the predictive accuracy of the time delay engine. Ultimately, the company said it expects to field a new service in Japan and elsewhere called “Spatiowl” which embeds the time delay prediction engine.

The service leverages large volumes of transit information such as train speeds and locations collected from train and track sensors. These data are combined with Fujitsu’s AI technology called Zinrai in an attempt to increase the accuracy of train delay predictions based on machine learning using accumulated sensor data. The Zenrai AI platform focuses on knowledge processing, sensing and recognition and decision support as well as machine learning techniques that allow the technology to deployed in the field.

The Japanese trial began on Tuesday (July 19) and will extend through the end of September. “Despite lacking experience in rail operations, Fujitsu wanted to provide public transit-related business operators with information that supports users’ choice of routes by learning from past delay information, and to verify its effectiveness in this field trial,” the Japanese company noted in a statement.

Recent items:

Big Data Helps Drive Transportation Planning

Exploring New Arenas For Geospatial Data

Datanami