The University of Massachusetts Amherst
University of Massachusetts Amherst

Search Google Appliance


Validity of stationary probabilistic models for wind speed records of varying duration

TitleValidity of stationary probabilistic models for wind speed records of varying duration
Publication TypeJournal Article
Year of Publication2014
AuthorsArwade SR, Gioffre M
JournalRenewable Energy
Start Page74
Date Published09/2014
KeywordsMean wind speed, Stationarity, Stochastic process, Wind energy

A method for assessing the degree of non-stationarity in annual wind speed records is presented. The method uses quantitative tests on the wind speed records to assess the length of the period over which an assumption of stationarity in the wind record can be considered to provide reasonable engineering accuracy. The tests evaluate stationarity in second moment properties and marginal distribution. Numerical examples are provided for three offshore sites along the Atlantic Coast of the United States–one off of Virginia, and two off of Maine. The examples illustrate that an assumption of stationarity over a period of one week is largely justified, but that such an assumption over periods of one month is certainly not. Assuming stationarity over a period of a week can lead to errors in model values of the second moment properties of 2%–3% whereas the assumption applied to a monthlong period can lead to error greater than 10%. Examination of the persistence of marginal distribution reveals that, although true stationarity in marginal distribution persists for a few days at most, there exist two ‘seasons’, winter and summer, during which the marginal distribution remains relatively consistent, with rapid changes in marginal distribution occurring near the beginning and ends of these seasons. Results are found to be largely consistent across the three sites investigated as numerical examples. The methods and results presented here may be useful to those investigating the potential for offshore wind energy development using stochastic process theory to study wind speed or power production since stationary stochastic models provide simpler and more accessible predictions of quantities such as probabilities of exceedance of threshold values, upcrossing rates, and residence times.