Background and objectives

One of the leading wind farm companies in Northern Europe aims to differentiate by better control over its imbalance costs. This cost item is increasingly relevant for them due to increasing penetration of wind energy in general and their objective to be an operator for wind farms they don’t own. Their current imbalance costs amount to roughly 2.5% of revenues and are expected to increase with higher wind power penetration. The key driver for reducing imbalance cost is the accuracy of the prediction of the intra-day power production by the wind farms. The objective of this Proof of Concept was to discover if there is a machine learning / predictive analytics methodology which could reduce the intra-day imbalance in terms of MWh by 10% or more.


Our approach

Together with the power prediction team of our client we developed software to predict both the day-ahead power predictions and the intra-day power predictions, i.e. hour-ahead. The data we used consisted of weather data, forecast and actual, and power production data of the wind farms over a period of roughly 5 years. Data on the wind farms included angle of rotation of the turbines and blades, which is of particular interest in case the turbines are not fitted with a closed loop adjustment of these angles. The weather data were data on a grid of 150km by 150km covering more than one of our client’s wind farms. Based on the type of data that are used for these predictive models, the scalability of the software is expected to be high, which is of high importance to our client so that the software can be rolled out to other wind farms, on-shore and off-shore in their portfolio.




For the wind farms we considered, the accuracy of the power produced is improved with 45% and a reduction in imbalance costs of just over 50%. This financial benefit goes straight to the bottom line without any required changes in policies, processes or organisation.