The wind energy sector relies on CFD modelling to make predictions about power production and maintenance scheduling requirements for onshore and offshore wind farms. CFD requires specialist software and High Performance Computing (HPC) hardware to deliver engineering data - typically sets of simulations from a range of wind directions with different environmental conditions. For each simulation, the wind speed is used to determine the power production from each wind turbine and hence the whole array.
The methods used to date have generally not included wake interactions because of the increased model complexity and the (historical) costs of large scale simulations. Instead, empirical correction factors are applied to the raw wind speed data produced by the CFD (without wakes) in order to estimate the performance with wakes. With increased wind farm size, wake effects are becoming significant, so simple correction factors no longer work as well. The industry recognises that the continued use of basic CFD introduces uncertainty and risk into the projections for power production and maintenance. The challenge is to produce high fidelity CFD results within a short timescale (days) at an acceptable price.
SWEPT2 (May 2015 to April 2018) was a collaborative project between SSE, DNV-GL, The Centre for Modelling & Simulation (CFMS), Zenotech, ORE Catapult, STFC and the universities of Surrey, Strathclyde, Bristol and Imperial College London. The project developed a new CFD capability for wind turbine modelling, with validation and benchmarking on high performance computers. When building a complex digital product, not only is there a need to push the boundary of simulation to increase confidence but also remove barriers to deeper understanding.
The new capability is freely available for academic use and can be used commercially under license. The project included a blind test against SCADA data for the SSE Greater Gabbard wind farm. A significant point to highlight, CFD simulations were accurate to 2.4% without any special tuning. In addition, raw CFD outputs from the SWEPT2 programme have been used by the growing Artificial Intelligence (AI) and machine learning community in the UK (i) using raw data from the Surrey EnFlo wind tunnel, compared against CFD, and (ii) with high fidelity CFD data for the Horns Rev wind farm to evaluate a suite of machine learning algorithms. The technology developed is applicable across multiple sectors, with direct benefit to those in aerospace, automotive, construction and energy production.
Fig: power output predicted in a blind test for the SSE Greater Gabbard wind farm.