Case Study: Kolumbus predicts vehicle locations and improves journey planning with Google Cloud Platform

Kolumbus Building a predictive mapping solution

Kolumbus, the public-transport administrator for Norway’s Rogaland region (about 500,000 people), wanted to turn its existing real-time vehicle map into a predictive “time machine” so customers could see where buses and boats will be at any future date and time. The project required handling massive streaming data (400 vehicles sending updates every few seconds, several GB/day) and a scalable machine-learning platform to model historical patterns and operational variability.

Computas built the solution on Google Cloud Platform, using Dataflow, BigQuery (storing 2 TB and 125M rows), Cloud SQL, TensorFlow on Cloud ML Engine, and an App Engine/Flask API to power the map. The system learns from 800K+ recorded trips and 125M data points, processes ~85,000 journeys daily, and delivers fast, date/time-specific vehicle predictions—improving trip planning, keeping infrastructure costs down through cloud scalability, and helping Kolumbus evolve from a transport administrator into a mobility platform provider.

Kolumbus

Audun M. Solheim

Head of Strategy and Development