Machine Learning Proxies Integrating Wake Effects in Offshore Wind Generation for Adequacy Studies

Abstract

Considering the increasing proportion of offshore wind generation in the energy mix, it becomes essential to properly account for aerodynamic effects that impact the power extracted from the wind. Indeed, due to computational matters, offshore wind energy is currently modelled in a very simple and approximate way in adequacy studies, neglecting important factors such as wake effects. Hence, in this paper, data-driven proxy models are developed to learn the complex relation between free flow wind information and the resulting aggregated output power of wind farms. Those Machine Learning-based models are used as fast and reliable surrogates of wake models, embedding their ability to describe the wind behaviour, but with much lower computational times. These models are then included in an adequacy study built upon sequential Monte-Carlo simulations. The collected results are compared with those obtained with the current simplified modelling approach for offshore generation. We observe the importance of accurately representing intra-park aerodynamic effects since reliability indices can be significantly underestimated when using the simplified modelling.