Generating synthetic wind data

NIWA has created synthetic, multi-year, 10-minute wind datasets at 15 wind farm sites across New Zealand. These will be used by the Electricity Commission to model the impact of wind farms on the national grid.

Problem

Wind data at ten-minute to hourly time scales is an important input into modeling the performance of wind farms and their impact on the national electricity system. However, relatively little ten-minute wind data at actual wind farm or proposed wind farm sites is publicly available in New Zealand. In the absence of actual observed data, simulated or “synthetic” data can be used, providing it is well calibrated against real data and it preserves the meteorologically significant spatial relationships between wind farms. This strict requirement is necessary to produce realistic results for applications such as:

creating generation scenarios during the passage of frontal systems
calculating the capacity contribution of wind
estimating the contribution of wind during different seasons.

Within this project (sponsored by the Electricity Commission) NIWA joined forces with MetService to fill the gap and create just such a dataset.

Solution

The joint NIWA-MetService^TM solution was to use the several years of archived 12 km gridded hourly NWP (Numerical Weather Prediction) winds over the whole of New Zealand that exist at both institutions as a basis for the long term synthetic data. These NWP winds have inter-site correlations and temporal variability (on a synoptic time-scale) that are known to match the observations well.

By developing a robust statistical relationship between these hourly NWP winds and hourly speeds observed at hub-height at wind farms it was possible to produce an hourly synthetic wind dataset which preserved the statistical properties of the hourly observed data. To then obtain a ten-minute synthetic dataset with all the desired properties of the ten-minute wind farm observations, realistic ten-minute fluctuations in wind speeds for these wind farm sites were then superimposed on the hourly time-series. While some refinements to this approach were necessary, it formed the basis for the successful solution and the creation of a synthetic wind dataset with all the required physical properties.