Skip to main content

High performance computing for wind energy

Periodic Reporting for period 1 - HPCWE (High performance computing for wind energy)

Reporting period: 2019-06-01 to 2020-05-31

As techniques for the manufacture of large wind turbines (with rotor diameter over 150 m) and offshore installation become mature, wind as a clean and renewable alternative to fossil fuels has become an increasingly important contributor to the energy portfolio. Annual energy production by wind turbines has reached around 5% of worldwide electricity usage, and 10.4% in the EU, at the end of 2016. The EU’s current target is to reach 31% by 2030. Brazil, the collaborator of HPCWE, has the 9th largest wind capacity in the world and is experiencing over 10% annual growth with nearly 500 wind farms already deployed.

The goal of HPCWE is to address key challenges for wind energy technologies to benefit from state-of-the-art HPC by exploring academic and industrial collaborations between Europe and Brazil. The HPCWE consortium consists of 11 partners including 2 HPC centers, 7 top academic institutes and 2 leading enterprises from six European countries and Brazil, with proven track records on HPC hardware, wind energy, numerical methods, software developments and industrial applications. HPCWE aims at establishing an EU-Brazil network, coordinating the action of universities, companies and consultancies with complementary expertise, to build and test beyond-state-of-the-art HPC strategies for the numerical simulation of wind flow in wind energy exploitation. Four objectives have been subsequently identified:

Objective 1: Efficient use of HPC resources in the simulation of flow around a wind turbine across all relevant operating conditions.

Objective 2: Accurate scale-integration in wind energy beyond state-of-the-art

Objective 3: Improved I/O for adjoint-based optimization

Objective 4: To build new partnerships between EU and Brazil for the application of HPC in the wind energy sector.

Towards the end of the project, HPCWE will deliver a step change in the application of HPC on wind flow simulations and will reshape almost every stage of wind energy exploitation. The success of HPCWE will also bridge the current discipline boundaries between the underlying scientific areas (fundamental fluid physics, wind energy, high-order numerical methods and atmosphere science, etc.).
The key achievements corresponding to the four objectives are listed below.

1. Efficient use of HPC resources in the simulation of flow around a wind turbine across all relevant operating conditions:
1.1 Novel algorithms have been developed and implemented for accurate simulation for flow over turbines as well as Fluid-Structure interaction (FSI) algorithm to capture the highly non-linear interaction between rotor and wind.
1.2 Highly-resolved Fluid and structure coupling represented by flow around a sectional/quasi-3D for flexible turbine blade has been demonstrated using software package Nektar++/SHARPy developed at ICL during this project.
1.3 High performance LES code UTwente-LES with advanced sub-grid and wind turbine models has been tested for highly-resolved aerodynamic predictions of wind turbines in atmospheric boundary layer flow.

2. Accurate scale-integration in wind energy beyond state-of-the-art:
2.1 At the meso scale, various Weather Research and Forecasting (WRF) Models have been tuned to lift the efficiency and accuracy of simulations.
2.2 A non-supervised classification algorithm (e.g. k-means) is under test to classify the large number of inflow conditions extracted from WRF into a small number of groups by coupling EWP-WRF with the CFD code Code_Saturne.
2.3 A novel scheme is developed to obtain a hybrid solution merging low-fidelity and continuous WRF data and high-fidelity sparse experimental measurements and has been applied and tested in the North Sea area.
2.4 To investigate how uncertainty develops in this scale integration process, the Dakota software framework has been applied based on wind turbine flow simulated from several codes.

3. Improved I/O for adjoint-based optimization:
3.1 Modal decomposition techniques, including wavelet transformation, complex orthogonal modal decomposition and dynamic modal decomposition, have been applied for data reduction in optimization related with fluid flow.
3.2 An image-based empirical wavelet transform technique has been developed and applied to several cases to identify key fluid physics and further data reduction.
3.3 State-of-the-art hardware and software tools such as I/O burst buffer and the Cray DataWarp are under tests for the reduction of I/O time.

4. Building new partnerships between EU and Brazil for the application of HPC in the wind energy sector:
4.1 Several new partnerships between European and Brazilian organizations in wind energy and HPC techniques have been established based on those existing before the execution of this project.
4.2 The European industrial partners of HPCWE are working with Brazilian wind farm developer Casa dos Ventos on a Brazilian site, and the European academic partners on wind resource assessments.
4.3 The European and Brazilian partners of HPCWE have co-organized a workshop and another two will follow in the next year.
Progress beyond the stat of the art:

i) Highly-resolved Fluid-structure coupling represented by flow around a sectional/quasi-3D for flexible turbine blade has been demonstrated. Such simulations require unsteady flow modelling capable of capturing the time dependent large scale forcing that is coupled to long wavelength structural vibrations (see figure 1).

ii) A high-fidelity and high-efficiency scale integration scheme for wind resource assessment is developed. This scheme combines mesoscale (~1000 km) weather simulation and microscale (1-2 km) simulation of fluid flow around a wind farm (see figure 2).

iii) A novel scheme is developed to obtain a hybrid solution merging low-fidelity and continuous WRF data and high-fidelity sparse experimental measurements and has been applied and tested in the North Sea area. This data fusion has been demonstrated to be effective in both spatial and temporal extrapolation of the field measurements using numerical data. A journal publication is submitted (see figure 3).


The developed techniques will be applicable for other disciplines using computational methods and other users as well, including industry. Even though only problems from CFD containing structures with different length- and time-scales will be studied here, the findings will be similarly applicable to other disciplines that use HPC to generate large amounts of raw data. The expected impacts include

1. Release of open source frameworks for explicitly-coupled fluid-structure modelling of blades and advanced wind turbine models.

2. The scale integration technique offering an efficient approach for wind resource assessment with confidence.

3. Reduction of the I/O load in the optimization loop by flow field interpolation/reconstruction, data filtering, etc.

4. Acceleration of both I/O and computation by exploring the use of I/O burst buffer, the Cray DataWarp and non-volatile memory.

5. Enhanced and sustainable partnership and identification of future cooperation between European and Brazilian partners.

6. Improved industrial cooperation and the exchange of knowledge and best engineering practices in wind energy between Europe and Brazil.