Optimal control of wind farms to maximize power is a challenging task since the wake interaction between the turbines is a highly nonlinear phenomenon. In recent years the field of Reinforcement Learning has made great contributions to nonlinear control problems and has been successfully applied to control and optimization in 2D laminar flows. In this work, Reinforcement Learning is applied to wind farm control for the first time to the authors’ best knowledge. To demonstrate the optimization abilities of the newly developed framework, parameters of an already existing control strategy, the helix approach, are tuned to optimize the total power production of a small wind farm. This also includes an extension of the helix approach to multiple turbines. Furthermore, it is attempted to develop novel control strategies based on the control of the generator torque. The results are analysed and difficulties in the setup in regards to Reinforcement Learning are discussed. The tuned helix approach yields a total power increase of 6.8% on average for the investigated case, while the generator torque controller does not yield an increase in total power. Finally, an alternative setup is proposed to improve the design of the problem.

Two of the major limitations facing the adoption of large-eddy simulation (LES) to the industry today are a lack of validation against full-scale measurements and the high computational cost. The lattice Boltzmann method is an approach to conduct LES that is suitable for parallelization on graphics processing units, leading to reduction in energy-to-solution by multiple orders of magnitude compared to Navier-Stokes solvers. We validate the lattice Boltzmann solver VirtualFluids against the measurements published in the SWiFT benchmark and the results obtained with LES by the participants in the benchmark. We compare inflow, turbine response and wake quantities and show that our method yields similar results. While the other LES methods vary in the required energy by one order of magnitude, our methodology is always about one to two orders of magnitude more efficient. The benchmark allows for a comparison to a large number of models, however, the scale of the turbine is not representative of modern turbines and therefore important challenges of modern turbines, such as blade deflection, could not be validated.

Deep convolutional neural networks are a promising machine learning approach for computationally efficient predictions of flow fields. In this work we present a simple modelling framework for the prediction of the time-averaged three-dimensional flow field of wind turbine wakes. The proposed model requires the mean inflow upstream of the turbine, aerodynamic data of the turbine and the tip-speed ratio as input data. The output comprises all three mean velocity components as well as the turbulence intensity. The model is trained with the flow statistics of 900 actuator line large-eddy simulations of a single turbine in various inflow and operating conditions. The model is found to accurately predict the characteristic features of the wake flow. The overall accuracy and efficiency of the model render it as a promising approach for future wind turbine wake predictions.

The use of graphics processing units (GPUs) has facilitated unprecedented performance gains for computational fluids dynamics in recent years. In many industries this has enabled the integration of large-eddy simulation (LES) in the engineering practice. Flow modelling in the wind industry though still primarily relies on models with significantly lower fidelity. This paper seeks to investigate the reasons why wind energy applications of LES are still an exception in the industrial practice. On that account, we present a survey among industry experts on the matter. The survey shows that the large runtimes and computational costs of LES are still seen as a main obstacle. However, other reasons such as a lack of expertise and user experience, the need for more validation, and lacking trust in the potential benefits of LES reveal that computational efficiency is not the only concern. Lastly, we present an exemplary simulation of a generic offshore wind farm using a GPU-resident Lattice Boltzmann LES framework. The example shows that the runtime requirements stated by a large part of the respondents can already now be fulfilled with reasonable hardware effort.