Posters

Abstract

Wake effects have a significant impact on wind farm production and assets lifetime. Regions of slow, highly turbulent wind lead to a power reduction and an increased fatigue for the turbines within or at the edge of the wake. Wind farm control (WFC) is an efficient way to mitigate the production losses, for example using wake steering which consists in voluntarily misaligning upstream turbines in order to redirect their wakes away from downstream turbines. Optimising the misalignment of each turbine throughout the farm can increase the overall wind farm production. Wind farm flow control is most beneficial when wake effects are large, and is therefore of interest for large wind farms. To address this, we have developed an accelerated algorithm that allows to take up to 500 turbines into account in a wake steering optimisation with a very good accuracy and computation times compatible with real-time control and energy yield assessment. Also, as large offshore wind farms often have neighbouring farms, we have developed a methodology capable of efficiently taking the wake of surrounding farms into account. The new algorithm processes turbines sequentially in  descending order following the wind propagation. While it does not guarantee to find the optimal solution, it converges toward an excellent candidate, as it follows the physical process of wake formation, and runs much faster than conventional algorithms like COBYQA [1], featuring a linear dependency on the number of turbines. The resulting misalignment setpoints are close to the optimal solution and the difference in terms of production gain is negligible. The wake simulator used is Farmshadow, a software developed by IFP Energies nouvelles.  Farmshadow has been optimized for fast computation of the production, taking the misalignment and turbulence into account. It is also capable to accurately predict inter-farm wakes, based on analytical wake models [2]. The algorithms were tested on large offshore farms arranged in clusters, and the impact of neighbouring farms was taken into account with almost no additional computational costs. Optimisation was performed on different subsets of the clusters with the total number of turbines ranging from 30 to 350, and up to 150 neighbouring turbines.   [1] T. M. Ragonneau, Model-based derivative-free optimization methods and software, 2023, https: //arxiv.org/abs/2210.12018, https://arxiv.org/abs/2210.12018. [2] F. Blondel and M. Cathelain, “An alternative form of the super-Gaussian wind turbine wake model,” Wind Energy Science, 2020.