A simple hourly wind power simulation for the South-West region of Western Australia using MERRA data
*Subscription may be required
A simple simulator capable of generating synthetic hourly values of wind power was developed for the South West region of Western Australia. The global Modern Era Retrospective Analysis for Research and Applications (MERRA) atmospheric database was used to calibrate the simulation with wind speeds 50m above ground level. Analysis of the MERRA data indicated that the normalised residual of hourly wind speed had a double exponential distribution. A translated square-root transformation function yn=(√(1.96+ ye )−1.4)/0.302 was used to convert this to a normal-like distribution so that autoregressive (AR) time series analysis could be used. There was a significant dependency in this time series on the last three hours, so a third order AR model was used to generate hourly 50m wind speed residuals. The MERRA daily average 50m wind speed was found to have a Weibull-like distribution, so a square root conversion was used on the data to obtain a normal distribution. The time series for this distribution was found to have a significant dependency on the values for the last two days, so a second order AR model was also used in the simulation to generate synthetic time series values for the square root of the daily average wind speed. Seasonal, daily, diurnal, and hourly components were added to generate synthetic time series values of total 50m wind speed. To scale this wind speed to turbine hub height, a time varying wind shear factor model was created and calibrated using measured data at a coastal and an inland site. Standard wind turbine power curves were modified to produce an estimate of wind farm power output from the hub-height wind speed. Comparison with measured grid supervisory control and data acquisition (SCADA) data indicated that the simulation generated conservative power output values. The simulation was compared to two other models: a Weibull distribution model, and an AR model with normally distributed residuals. The statistical fit with the SCADA data was found to be closer than these two models. Spatial correlation using only the MERRA data was found to be higher than the SCADA data, indicating that there is still a further source of variability to be accounted for. Hence the simulation spatial correlation was calibrated to previously reported findings, which were similar to the SCADA data.
|Publication Type:||Journal Article|
|Murdoch Affiliation:||School of Engineering and Information Technology|
|Copyright:||© 2016 Elsevier Ltd.|
|Item Control Page|