Intel HPC Cluster Boosts Natural Power’s Wind Modeling Efficiency

How do you calculate success? For Natural Power, success recently came in the form of a 160-node supercomputer capable of running some of the most advanced wind flow models ever seen. See case study.

According to the Global Wind Energy Council (GWEC), the renewable wind energy industry set a new record for annual installations in 2014, with global wind generating capacity increasing by 51,477 MW to 369,553 MW. If this trend continues, the GWEC forecasts that wind power could supply between eight and 12 percent of worldwide electricity by 2020.

Yet wind power forecasting requires a complex series of calculations and analysis of potential performance. When planning new installations, investors, utilities, and turbine manufacturers all need and accurate picture of predicted energy yield. They also need to know how it will impact local populations and wildlife, and whether it will deliver a return on the financial investment necessary for set up, security, and ongoing maintenance.

That’s where Natural Power’s expertise in wind modeling comes into play. “Wind power has enormous potential, but there are multiple variables involved at each installation that will affect yield and, consequently, the financial viability of both new and existing sites,” said Claude Abiven, senior technical manager for Natural Power.

Wind Modeling: Forecasting the Future

Natural Power gleans wind speed measurements over the course of a year to forecast potential power production. The company then uses a number of algorithms to calculate likely output.

“Simple linear wind-flow prediction models can run on a laptop computer,” said Abiven. “But we also need models that are a lot more computationally intensive when the farm is located in a complex or forested terrain, or in the presence of complex atmospheric phenomena such as sea breezes, for example. In these circumstances, linear wind-flow prediction models generally provide a much less accurate picture.

Temperature fluctuations and climate change are an important factor that impacts the accuracy of preexisting wind power forecasting models. To address the issue, Natural Power teamed up with the University of Porto in Portugal, and Renewable Energy Systems, to develop a more advanced computational fluid dynamics (CFD) model.

The new model predicts complex thermal flows at micro scales and, thanks to coupling with meso-scale modeling (short-term forecasting), delivers more accurate wind speed projections than other industry models. This also results in increased accuracy when predicting turbulence intensity and wind shear.

Optimized for HPC: Harnessing Wind Data

With a more robust modeling system, Natural Power quickly realized the need to expand beyond its existing server cluster. “With so many more equations to consider, running the model on our existing cluster would result in either a loss of precision, or would take about 60 days to produce results,” said Abiven.

Natural Power turned to Intel to optimize its code for HPC. Following a demonstration by Intel, Natural Power upgraded to SuperMicro supercomputers powered by the Intel Xeon processor E5-2660 v3. This enabled them to pull in the computing strength of 160 compute nodes; a whopping tenfold increase in power over the previous configuration.

Natural Power’s internal tests show that individual cores run 1.7 times faster. This enables it to run both its standard and coupled CFD models at wind farm scale; producing results in less than a week. As a result, calculations that used to take in excess of two months are now completed in days.

“No customer wants to wait 60 days for modeling outcomes,” said Abiven. “With the use of wind power continuing to expand worldwide, and customer demand increasing all the time, our Intel Xeon processor powered supercomputer is set to play a central role in our ongoing success.”

Read the full story in this new case study.