A looking glass into future energy prices with machine learning

Posted 13th April 2021 | 475 words | 3 minutes

We would all love to predict the future and know what energy prices would look like in 10 years, but since we don’t have a time-traveling DeLorean to rely on, the Energy Markets team at Zeigo has been busy training algorithms to predict future energy prices with machine learning.

coverphotoml

Future electricity prices & long-term procurement

A vital step in analysing any long-term power purchase strategy is to assess its costs against market predictions. For contracts with lifetimes spanning multiple years (5+ years), there is not enough liquidity in future contracts to assess market value and so there is a need for long term price forecasts.

So how can we predict what electricity will be worth in 2030?

To predict price beyond a 3–4-year horizon, the most common approach is to use a fundamental model which aims to simulate the physical and economic relationships existing in electricity markets. This type of model often takes inputs such as demand, commodity prices and technology costs. These inputs are then fed into a dispatch model, often used in parallel with other economic models to understand investments in different technologies or the retirement of old power plants.

Fundamental models are widely used and are the basis of analysis provided by many companies and governments for long-term planning. However, they do have certain drawbacks. The first and most important is the complexity of such models which are often built over long periods of time. The second is the difficulty in obtaining detailed data such as power plant databases which are necessary for the dispatch model to run. Finally, these models can sometimes be static or inefficient for testing multiple scenarios rapidly.

Zeigo’s machine learning approach

Our team at Zeigo has been working on creating predictions of future electricity prices using a machine learning-based approach. In contrast with a fundamental model, a machine learning model learns, using historical data, how a multitude of different drivers impact power price and can then make future predictions based on its training. The main strength of this type of modelling is the ability to handle the complexity and non-linearity of electricity price features.

At Zeigo, we are using artificial neural networks to quickly create a range of forward curves based on different scenarios, highly adaptable to many different scenarios regarding future development of price drivers.

Forward curves: no winner takes all

Forward curves can vary significantly and differing opinions on long-term price development are common. This is why it is important to have the ability to look at different predictions and scenarios to make educated risk assessments. As a stand-alone model or in combination with more traditional models, ML-generated curves can give a better understanding of the potential future prices and allow experienced users to tailor predictions based on their own views.

To receive updates on Zeigo’s data insights, industry events and news, please contact info@zeigo.com.