PO040

WakeBlaster - Development of a Real-Time Wind Farm Simulator

Wolfgang Schlez, Philip Bradstock, Staffan Lindahl, Michael Tinning
ProPlanEn, Bristol, UK

Abstract

Most wake models currently in use by the industry are based on methods that have been developed more than 30 years ago. The research community on the other hand applied complex numerical solutions to the Navier Stokes Equations (NS). Solvers include Direct, Large Eddy and Detached Eddy methods. Other solutions are based on complex general purpose RANS solvers. Accurate modelling of the wakes in large offshore wind farms is computationally demanding and using complex models is performance intensive. Consequently hardly any of these models have made their way into day to day industrial application.
Industrial application demands a cost efficient model of the relevant physical processes matching in accuracy and parameterisation the experimental information available from validation cases. Such a model makes use of both the latest advances in understanding of the physics and computational hardware. The new model we developed in the WakeBlaster project simulates with high accuracy the flow in a wind farm in real-time. Real-time indicates here that the model runs one flow scenario for a large wind farm in less than 10 minutes.
We show that the model developed is matching in simple cases the accuracy of 2D Reynolds Averaged Navier Stokes Equations (RANS) models of the Ainslie family of models and standard analytical models of the PARK model family, which are used as baseline to represent the state-of-the-art. The new model surpasses in more complex validation cases the baseline models and is implementing a 3D-RANS solver, specialised in application in wind energy.

Method

A 3D RANS solver has been developed using a mesh and domain tailored specifically to the flow inside a wind farm. An eddy viscosity turbulence closure model is parameterised with turbine data, flow and ambient flow conditions for the wind farm. The model explicitly models interaction between multiple wakes and between wakes and the boundary layer. The added detail allows to improve the accuracy over those of current models while the design makes use of high performance computing on general purpose graphic processor units GP-GPU. The new model allows using cloud infrastructure for increased performance or scaling for application in large wind farms. The model is calibrated and validated against operational data from several wind farms.

Results

The new wake model has been validated against operational wind farm data. We show improved accuracy if compared to standard models and improved performance compared to research models. The model is able to calculate wake deficit profiles and locations based on SCADA information and parameters typically available to a major consultant, wind farm developer or wind farm operator. The wake information of the entire wind farm is able to be calculated in a ten minute period given appropriate hardware.

Conclusions

We describe the WakeBlaster project and the approach taken to develop a real-time wake model. The model is advancing the state of the art in industrial wake modelling by providing a finely tuned elevated level of model accuracy that can be supported by data from field measurements, while only using readily available parameters.The tool makes use of modern developments in high performance computing on GP-GPUs and also using cloud-based systems

Objectives

We explain the challenge of developing a new wake model and how problems were overcome in the process to quantify and reduce uncertainty in current wake models. Major challenges include separating wake model uncertainty from flow model uncertainty; deriving adequate validation cases; developing a new model with level of detail that is adequate to capture the underlying physical effects, possessing an accuracy to match the measurements with reasonable computational cost.