From proof of concept to container applications delivering online predictions as streaming data

This talk will show you how we deployed our machine learning models as container applications, and the most important takeaways from the process.

Statnett is the operator of the electrical transmission grid in Norway. One of the responsibilities of the balancing operators in Statnett is maintaining balance between power consumption and power production. In order to balance production and consumption, the operators need a reliable prediction of the power consumption.

It only took three weeks to come up with an offline power consumption prediction that outperformed the software Statnett currently purchases, however, the big bulk of the work has been turning the trial and error style code from the proof of concept phase into a service that can provide predictions at any time, with automated pipelines for training and deployment.

We have created APIs for our machine learning algorithms, and run them as container applications. The data sources are real time data from our streaming platform, and our resulting predictions are published on the streaming platform.

In this talk, we will present our key learning points from the development and deployment process leading to production ready software.