An Evaluation of Hourly Average Wind-Speed Estimation Techniques

Wind-speed time series are often used in computer models to estimate the performance of a wind turbine or an array of wind turbines. In such modeling applications, it is desirable to have complete data for an annual period. However, equipment failure or extreme weather events can result in incomplete data. For certain models to run properly and achieve the best results, it is necessary to estimate the missing data. The estimated wind-speed values should replicate, as closely as possible, the time-scale properties and statistical characteristics of the missing data. One complete year of data from three sites in different wind regimes was used to test three different data estimation techniques by creating a sufficient number of gaps (sequences of missing values) of various durations to simulate annual data recovery rates of 90% and 80%. Then three modeling techniques were used to estimate the missing data. The accuracy of the estimated data was determined by comparing the statistical characteristics (average, standard deviation, and root-mean-square error) of the estimated data to those derived from the corresponding original data values. The estimated data were compared to the original wind-speed data, and also used as input to the Wind Power Simulator (WIPS) model[1]. The model results from the time series with the estimated data were compared to those from the original, complete time series. The results showed that all three estimation techniques did quite well in replicating the data in the original time series and each worked best at one of the three sites. Errors in the estimated average wind speed were rather small, mostly within ±5%. The errors in theoretical energy from the WIPS model for the entire year of data were mostly on the order of 2% or less. Errors tended to increase with gap length but there was no significant difference in results for the two simulated recovery rates.