20.06.2023 | Adrian's assignment in Malmö

Enriching views and insights

Adrian Meier, Trainee Machine Learning Engineering, joined the Physical Optimization & Structuring desk in Malmö for his second assignment. He gives us an in-depth insight into his daily life in Sweden.

 

Arriving in January, the warm, inclusive welcome from both the team and people of Malmö far outweighed the weather’s attempts to send me back south.

The Axpo SE team consists of around 15 people, most of them titled Originator, Controller or Portfolio / Operations Manager. They moved to their new, modern office space at the buildings top floor just last year. While 9 is the number you select by elevator, taking the stairs, you only need to climb eight floors to reach the same destination.

Navigating through the Nordics

Skåne, in the southern part of Sweden, is flat. This makes biking in Malmö easy (most bikes here have no more than three gears – for windy days). Yet, the flatness complicates orientation. There are no natural vantage points to provide an overview of the city, and most of the buildings are a similar height. The few exceptions, like the skyscraper Turning Torso, can’t be spotted from within the maze of the inner city.

On the first morning I cycled to work, it was dark, cold, raining, and windy. Following the crowd of undeterred bikers on the excellent bike lanes, I took a wrong turn. Finally admitting defeat, I had to take off my soaking gloves and navigate using GPS. Since then, I’ve cycled to work every day – and in every type of weather. 

The great view from the office

Aside from a great place to work, the office provides a fantastic viewpoint: city on one side, the sea and Øresund Bridge connecting to Copenhagen, and sunsets on the other side.

My mission at the Nordic Physical Desk

To cite our wiki: "The Nordic Physical Desk trades short term physical power and manages and develops the physical electricity value chain in the Nordic and Baltic countries."

My mission here is diverse and multifaceted:

  • To learn about physical power trading, including gaining an understanding of the market drivers and considerations between intra-day, day ahead and imbalance markets
  • To further the cooperation between the Advanced Analytics team (my first rotation within the traineeship) and the Physical Desk.
  • To apply my computer/data science background to further the automation / optimization or to provide insights from data.

The physical desk (presumably like most front desks) operates at high speed. From day to day, the priorities can change dramatically – either to exploit the newest opportunity in the markets, or to prevent disaster from an unexpected change. Speed of development is valued more than maintainability, reusability or even correctness to some degree. For the most part we'd rather run a simple solution that covers 80% of the cases, than to spend extra time developing a more general, complex approach. By the time the general solution would be ready, the window of opportunity might have already closed.

This contrasts with my experience in the Advanced Analytics team, where we followed a very fundamental, long-term approach to build generalised solutions. To be one step removed from the front line is a luxury I was unaware of at the time.

A variety of data-driven tasks

My tasks within the first four months in the Nordics included developing access points to various data sources. There’s a wealth of grid, market, weather and other data to consider when trading physical power. Many of the inputs are forecasts, and better forecasts allow for better decisions.

To better our forecasts, I helped revise a forecast of market metrics, and benchmarked energy production, consumption, and price forecasts from different providers.

Adrian Meier, Trainee Machine Learning Engineering at Axpo

I also worked on estimating the potential to offer frequency containment reserves (FCR) for specific wind farms. The idea is to provide automatic curtailment of windfarms in real-time, based on the grid frequency, to help stabilise the grid in cases of over-production or under-demand. For a wind park, we consider the maximal downregulation to be the real-time production. To be paid for your service, the FCR regulating bid must be placed in advance. However, due to the imperfect nature of wind power production forecasts, there is always a risk of the bid being greater than the actual production. When that happens, the failure to downregulate results in a penalty. I viewed this as an optimisation problem with a risk parameter: for a target risk (i.e. failure to downregulate no more than one hour per month) what is the maximum volume of downregulating power we can bid?

Finally, I had the chance to visit the Oslo office for a week, to investigate a systematic market pattern. There, like everywhere in the Nordics, I was told: "You have to come back in summer!"

Until then, I'll enjoy the mild Scandinavian spring.

More articles for you

Show all

Energy market

Price swings amid volatile weather, market speculation, and geopolitical risks

European Energy Markets Monthly, October 2024

Read more

Renewable energy

Green Deal Team: Keeping the EU on track for climate neutrality in 2050?

EU energy & climate policy: 2024 - 2029

Read more

Energy market

No room for experimenting with the WACC

Notwendigen Netzausbau nicht gefährden

Read more

People

Expert insights on Energy’s next decade

Axpo Netherlands: the first 10 years

Read more