Automatic frequency restoration reserve market prediction: Methodology and comparison of various approaches

Abstract

In continental Europe, automatic Frequency Restoration Reserve (aFRR) is the second fastest control reserve market. Due to the complex auction design, market entrance barriers for new players are high and the market is dominated by few operators of conventional power plants. However, a rising share of renewable technologies requires their integration into this market in order to assure future grid stability. Due to the high market complexity, operators and traders of such technologies are currently lacking a tool to estimate earning potentials. Both a market prediction methodology as well as a bidding strategy are required to estimate the earning potentials and to participate in the aFRR market. To encourage participation of new technologies, this paper first provides a detailed market description and then presents a market prediction methodology for estimating revenue potentials and to assist in creating bidding strategies for auction participation. For any potential bid, the acceptance probability within the auction is derived. Both statistical and machine learning based models are used for predicting key market quantities. A model comparison reveals a steadier and usually better performance of statistical models. Exogenous data sources such as weather, electrical loads or market data did not improve the prediction performance.