Impact of turbine- and farm-scale flow effects on wind farm performance

Thesis

When wind turbines are placed together in a farm, they produce less power than in isolation. Behind every turbine is a low speed wake which can reduce the power output of downstream turbines. These turbine-wake interactions depend highly on the wind direction and turbine layout. The interaction between the atmosphere and the farm as a whole also causes power losses. Understanding and predicting these power losses is vital for the offshore wind energy sector.

To better understand farm performance, we propose new measures of 'turbine-scale loss' and 'farm-scale loss'. The 'turbine-scale loss' quantifies the power loss due to internal flow interactions only (i.e. turbine-wake interactions). The 'farm-scale loss' gives the loss due to the farm-atmosphere interaction. We conduct a combined theoretical and computational study to show that, for large offshore wind farms, the power losses on the farm-scale is typically more than twice as large as on the turbine-scale. This is contrary to the current consensus that turbine-wake interactions are the most important effects for wind farms.

Traditionally, power losses in farms are classified into 'wake losses' and 'farm blockage losses'. However, we show that neither loss is well correlated to the overall wind farm efficiency. In contrast, the new 'farm-scale loss' is shown to be strongly correlated with the overall farm efficiency. These new metrics give a novel and clear understanding of the aerodynamics influencing wind farm efficiency.

We also develop new low-cost models to predict the turbine- and farm-scale losses in wind farms. To predict turbine-scale losses, we develop statistical emulators to predict the turbine-wake interactions as a function of the turbine layout. The statistical emulators are trained using a multi-fidelity Gaussian Process on high-fidelity large-eddy simulations and low-fidelity wake model predictions. The trained emulators predict the farm 'internal' turbine thrust coefficient with a mean absolute error of less than 1% compared to the high-fidelity simulations.

To predict farm-scale losses, we derive a simple analytical model to predict the farm-atmosphere interactions. The model is derived using a quasi-one-dimensional control volume analysis. An initial comparison with high-fidelity wind farm simulations shows a satisfactory agreement in the prediction of farm-scale losses. These two new models could be used to predict the performance of large wind farms at a very low computational cost, and to optimise the design of future farms.

Authors: Kirby, A

• DOI: 10.5287/ora-aeepgpvo8
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: wind energy; fluid mechanics; engineering; machine learning
• Subjects: Wind power; Fluid mechanics; Computational fluid dynamics; Machine learning