Updates and insights

Staying ahead: How energy supplier Ustekveikja Energi uses Montel AI to forecast demand

Edited by: 

Simon White
Content Manager

October 27th, 2022

We sat down with Portfolio Manager at Ustekveikja Energi, Torbjørn Røed Meberg, to understand more about the ways in which they use Montel AI’s demand forecasting product and how it benefits their business.

Having begun in 1992, around the time Norway’s power markets were liberalised, Ustekveikja has now spent 30 years delivering power to homes and businesses across the country. “Of course I cannot tell you too much about what happened at that time”, says Torbjørn “I wasn’t even born then!”

Neither, it must be said, was Montel AI. In fact, artificial intelligence around that time was thought of as being more like the Terminator or Robocop, rather than as a forecasting tool for energy suppliers.

But as a new generation of energy traders have come into the industry over that period, so too have the tools available to them to help address their professional needs.

Having joined Ustekveikja fresh from university, Torbjørn’s degree in engineering & power system analysis made him a natural fit for the company’s physical trading desk. Now a part of the financial trading side of the business, he takes responsibility for managing their energy supply position in real-time.

"

We needed a product which could help us forecast demand for any given settlement period – it’s my job to make sure that our supply position matches our customer’s needs, so it was important from the outset to find a product I trust.

"
- Torbjørn Røed Meberg, Portfolio Manager, Ustekveikja Energi

After speaking with a colleague who had previously worked with Montel, Torbjørn got in touch to find out more about the applications for AI forecasting. At that point the solution had only been used to help predict the output from renewable power plants, but such is the versatility of Artificial Neural Networks (ANNs) any scenario requiring data-based forecasting can be addressed through this specific type of machine learning.

Put simply, ANNs put the raw historical data through multiple layers of mathematical functions. This then produces the forecast when it is combined with weather data - also fed into the model by Montel.

The more interesting point here, however, is that these networks also begin to recognise patterns as they appear in the data, making the outputs more and more accurate each time the model is run. Therefore, the more times the model is run, the networks become more and more familiar with the expected end results.

"

The hit rate we’ve seen so far has been really, really good. With the current market conditions, this is vital to our business, particularly as the cost of imbalance in the market has increased by almost 7 times when compared to the average cost over the past ten years.

"
- Torbjørn Røed Meberg, Portfolio Manager, Ustekveikja Energi

The reason for the success of the load prediction model rests, in truth, on both sides of the agreement. Whilst Montel AI have provided the technical solution, by being able to specify and use the historical demand data used in the model, Ustekveikja can ensure they are receiving predictions they trust.

Looking at the predictions over time, Montel’s data shows that predictions have been running at around 96% accurate.

And it’s not just the money being saved that stands out to Torbjørn, but the practicalities of using the product too. “We send the data the model needs via Montel AI’s Secure File Transfer Protocol and then receive the forecasts back the same way. With the whole thing being automated, it saves so much time compared to creating these forecasts manually. It gives me time to put my focus elsewhere because I am confident in taking a position based on the prediction.”

And what about when something looks wrong? “It is so easy to recalibrate the model. Where a forecast comes back with an unexpected result, it can often be the result of an anomaly in the data. So, we simply identify that item, change it to what we would normally see and run the model again.”

The solution is scalable too, with predictions sent for each of Norway’s five Distribution System Operator zones twice a day.

And all of that sums up why Torbjørn is happy to recommend Montel AI to anyone prepared to listen.

“The best thing about it? I guess it would have to be the way the forecasts improve incrementally over time, but where an automated product is saving you both time and money it’s hard to pick out just one thing!”

You can find out more about Montel AI's forecasting services across renewables, spot markets and demand on our dedicated webpage.