Kolumbus: Building a predictive mapping solution

About Kolumbus

Kolumbus is the public transport administration in Rogaland, Norway, serving a population of around 500,000 people.

Industries: Government
Location: Norway

About Computas

Based in Norway, Computas delivers services and solutions for IT work processes and collaboration. The company is one of the largest Google Cloud Platform Partners in the Nordics, with core competencies in systems development, architecture, integration, big data, and machine learning.

With Google Cloud Machine Learning, Kolumbus created a tool to predict the future locations of its vehicles and help customers plan their journeys more effectively.

Google Cloud Platform Results

  • Helps transform Kolumbus from a public transport administrator to a mobility platform provider
  • Keeps costs down with scalable storage and flexible pricing
  • Highly-managed services allow Kolumbus to concentrate on core objectives instead of spending time on IT infrastructure

Learns from over 800K+ recorded trips with 125M data points

Serving the county of Rogaland in Norway, including the city of Stavanger, Kolumbus oversees public transport for a population of 500,000. A publicly-funded body, Kolumbus does not just plan and organize the boat and bus services in Rogaland but sees itself as a mobility provider, aiming to keep private car traffic at zero growth.

“For this project, we needed a platform that could handle a large amount of machine learning and artificial intelligence. For us, Google Cloud Platform is highly competitive and possibly the best platform in this area.”

Audun M. Solheim, Head of Strategy and Development, Kolumbus

As part of its drive to improve service for its customers, Kolumbus developed a real-time map showing the locations of all of its buses and boats. Looking to enhance the map even further, the company decided to turn it into a “time machine” showing customers where buses and boats would be tomorrow or next week at any given time, using predictive analytics and machine learning. To do that, Kolumbus turned to Google Cloud Platform.

“For this project, we needed a platform that could handle a large amount of machine learning and artificial intelligence,” says Audun M. Solheim, Head of Strategy and Development at Kolumbus. “For us, Google Cloud Platform is highly competitive and possibly the best platform in this area.”

Scalable machine learning

As well as overseeing Rogaland’s bus and boat services, Kolumbus has ambitious environmental and innovation goals. In 2016, this led to the launch of a real-time mapping tool, using Google Maps, to let customers keep track of public transport across the region. With each one of its 400 vehicles transmitting data every few seconds, Kolumbus began to amass a huge store of information on its schedules and operations, adding several gigabytes every day.

“With all the data we’d collected before, we could travel back in time to see where all the buses were and how they were doing,” says Audun. “So we thought, why can’t we go the other way and see what will happen in the future?” This new system would require a powerful machine learning platform, huge amounts of storage, and the ability to scale up with ease and affordability. In mid 2017, Kolumbus ran a tender for the project.

“Google Cloud Machine Learning Engine was essential for us. After we trained our TensorFlow models, we could deploy to the cloud without any hassles. It’s much cheaper and requires less maintenance than deploying locally.”

Ture Friese, Lead Developer, Computas

In August, cloud technology specialist Computas won the tender with its proposal to build the new “time machine” with Google Cloud Platform. The new system takes in real-time data from Kolumbus’ vehicle sensors and schedule information, and processes it with Google Cloud Dataflow, Google BigQuery, and Google Cloud SQL. “After processing, we are left with 125 million rows of data,” says Simen Selseng, Knowledge Engineer at Computas. “Each row contains information about a specific bus at that moment in time.”

This dataset is held more securely in Google BigQuery, along with the 2 terabytes of historic real-time data Kolumbus has already gathered. With several gigabytes of data added everyday, a cloud-based storage solution allows Kolumbus to expand quickly without sinking costs into expensive hardware. From here, the data flows into machine learning models built with TensorFlow and operated on Google Cloud Machine Learning Engine. The TensorFlow models predict which vehicles are operating at any given time and how far ahead of or behind schedule they are. Computas built an API with Flask and Google App Engine, which allows the Kolumbus online mapping tool to access the prediction data and display the vehicle locations for any given date and time.

“Google Cloud Machine Learning Engine was essential for us,” says Ture Friese, Lead Developer at Computas. “After we trained our TensorFlow models, we could deploy to the cloud without any hassles. It’s much cheaper and requires less maintenance than deploying locally.”

Low maintenance, high capacity

With its new predictive mapping system, Kolumbus makes planning journeys across Rogaland simple. By typing in a date and time on the map website, within seconds customers can see a continuously updated map of where Rogaland’s buses and boats are predicted to be. If a bus route is usually delayed by traffic at a particular time of day, for instance, then that is reflected in the map and the customer can choose a better route.

Thanks to the ease of use and scalability, Google Cloud Platform can take in data from the 85,000 journeys made every day, add it to the existing 2 terabytes of historical data, and make its predictions without sinking costs into infrastructure or expanding IT resources. “At Kolumbus, we know what we want, but we don’t necessarily have the technical staff to build it,” says Audun. “With Computas we have a partner that’s an expert in this technology, so we can focus on the business needs and our customers.”

“Customer expectations are getting higher every day as technology improves. If we want to stay relevant we need to be evolving all the time and stay on the cutting edge. Computas and Google help us to do that.”

Audun Solheim, Head of Strategy and Development, Kolumbus

With the first phase of the project complete, the Kolumbus mapping solution can predict vehicle positions based on the historical mapping data. By the end of January, Computas and Kolumbus are on schedule to deliver an even more accurate version of the “time machine,” which incorporates external data such as weather and holiday events into the prediction models.

For Kolumbus, the “time machine” is a solid foundation with which to build future projects such as a fleet of autonomous buses and better deviation information to passengers. In addition, Kolumbus’ collaboration with Computas and Google is a key component in its shift from a public transport administrator to a mobility service provider.

“Customer expectations are getting higher every day as technology improves. If we want to stay relevant we need to be evolving all the time and stay on the cutting edge. Computas and Google help us to do that,” says Audun.

About Kolumbus

Kolumbus is the public transport administration in Rogaland, Norway, serving a population of around 500,000 people.

Industries: Government
Location: Norway

About Computas

Based in Norway, Computas delivers services and solutions for IT work processes and collaboration. The company is one of the largest Google Cloud Platform Partners in the Nordics, with core competencies in systems development, architecture, integration, big data, and machine learning.

Google Cloud Platform logo

12 Months FREE TRIAL

Try Kubernetes Engine, BigQuery, and other Cloud Platform products with $300 in free credit and 12 months.

TRY IT FREE
Google Cloud Platform logo

12 Months FREE TRIAL

Try Kubernetes Engine, BigQuery, and other Cloud Platform products with $300 in free credit and 12 months.

TRY IT FREE