Combined Machine Learning Approaches for Energy Balancing and Trading in Austria

Abstract

The energy markets are subject to the volatile difference between power generation and consumption. The required balancing is accomplished by trading and production adjustments within and among regional subdivisions called control areas. This implies that short- or medium-term forecasts of the control area imbalance are a crucial asset for traders and stakeholders. The objective of this work is to investigate the prediction of this delta using different types of available data sources. To extract relevant input features for prediction, we applied statistical analysis, clustering, and transformations to the data. We then reduced the problem to an ordinal classification task and employed different machine learning approaches. To assess the quality of the forecasts, we observed the theoretical profit of a combined simulated trading strategy. Our results show a positive outcome and indicate a viable approach to provide helpful insights and decision support to stakeholders.