Back to Blog

Improve your ML with EQ

Hugo Birkelund
Archived blog post. This blog post has been transferred from our previous blogging platform. Links and images may not work as intended.

Machine learning (ML) and artificial intelligence (AI) present some fascinating modeling approaches. However, even with ML and hyper-fast computing, the old saying "garbage in, garbage out" or simply GIGO holds true. EQ provides structured and curated data for modeling. Our database is a treasure trove for model developers.

EQ = self-service

Help yourself to data. EQ’s business is to provide the best possible curated historical data and aligned forecasts, empowering you to excel in your job with minimum waste of time and energy. If you don't already have an account, get your access here.

Looking for something that is not there

Throwing in more data will not help you (much) if what you are looking for is not there. I have tried to illustrate the problem with two pictures. In the left picture, I have simply removed information and added noise, analog to a what you might see in a dataset.


Thanks to Fiona for guarding my Parrot Zik headset.

You can easily observe the problem in a picture. However, in data analysis, you can, at best, anticipate that something is missing. It could be too big times steps, poor spatial resolution, misaligned input used for model fitting and forecasting, or good old fashioned out-layers and missing values. More data might add to the problem and certainly waste your time.

A checklist

A checklist before you hit enter "estimate model" might save you a lot of time:

  1. Check your input data. At EQ, we do this as a profession. For instance, we provide curated synthetic data which we believe is a far superior alternative to actuals generated by TSOs.
  2. Avoid settling for raw weather data. Failing to align the output generated by weather models with real-life observations creates nasty errors and gut-wrenching frustrations. I know, from personal experience.
  3. Consider adding contextual information. Capacity changes, holiday effects, REMIT data, and more input add depth to the analysis. Failing to add context might contaminate your model and drain it for explanatory power.
  4. Look out for "known" non-linearity. A simple trick is, e.g. to use transformed data. Regardless of how advanced your model is, this simple trick might limit the data requirement tremendously, and improve the forecasts properties.

Take vitamins, not medicine

EQ provides a vast set of curated historical weather-driven fundamentals and indices. We believe they are perfect input for modeling and forecasting. Further, we provide all data in 15 minutes resolution, which is likely to become the power market’s preferred resolution. Using the correct data granularity from the get-go hugely simplifies modeling.

Contact me to discuss the best solutions for your team. We may have your data requirement covered, faster and simpler than you can imagine.

...I almost forgot

You do not have to download EQ’s data. Connect your developer tool or models directly into EQ’s timeseries database. It saves you time and simplifies your life.


EQ generates approximately 4,5 million data points per price areas every day. It makes sense to leave the data in our custody.

More from the Blog

Spot prices increase as Central Western European hydropower production hits 25 year low

Eylert Ellefsen
Eylert Ellefsen

This April has seen a notably cold spell of weather sweep across Europe. Temperatures have been 4 or 5 degrees Celsius below normal in most areas, whilst spot prices for the past month have delivered 5-7 €/MWh higher than expected in several areas. In this blog post, we study how the Central Western European (CWE) hydropower system developed during this cold spell and compared this situation to available historical statistics.

Read Story

Delayed spring thaw inflow for Sweden – detailed modelling with SMHI Hydro GWh

Eylert Ellefsen
Eylert Ellefsen

The temperature outlooks until the first week of May are much lower than normal in the Nordic region. Read on to see the likely consequences for inflow- and snow conditions in SE1 and SE2.

Read Story

Germany avoided a supply squeeze this winter, but what next?

Eylert Ellefsen
Eylert Ellefsen

When Germany closed down 4.8 GW hard coal plants this winter, the impact could have been severe, but the average output from German coal plants increased year-on-year due to a positive clean dark spread. That escape from a capacity squeeze will be harder to avoid next year as more capacity will closed down.

Read Story

Ready to try Energy Quantified?

No payment or credit card required.
Would you rather like a personal demo? Book a demo