Strategic bidding in pay-as-bid power reserve markets: A machine learning approach

The shift towards decentralized and renewable energy systems adds complexity to electricity grid management, where balancing economic and technical needs is critical. The Ancillary Services Market (ASM) – also referred to as power reserve market – plays a crucial role in ensuring grid stability by managing balancing reserves. However, its outcomes remain difficult to predict for participants due to technical prioritization by system operators. In this paper, we propose a machine learning approach to model pay-as-bid ASM and derive an optimal bidding strategy for market participants. Specifically, we develop a probabilistic classification pipeline to estimate the acceptance probability of individual bids based on price and other explanatory variables including bid characteristics and indicators of network conditions. Additionally, we employ eXplainable AI (XAI) techniques to interpret the underlying machine learning system and provide insights into actual market clearing mechanisms. We then show how this bid acceptance probability estimator can be used by market participants to optimally price their bids and maximize expected marginal profit. Applied to the Italian ASM over the 2018–2022 period, the results show that bid acceptance probabilities are accurately estimated and that the proposed bidding strategy enables market players to significantly increase their profits. Furthermore, the analysis partially reveals the opaque clearing apparatus of the ASM, identifying a systematic bias in favor of gas-fired generation technologies. The findings of this study are not only of interest to market participants but also to regulatory authorities.