Comparison of Different Wind Time Series Simulation Methods

Abstract

The assessment of power system reliability under increasing penetration of wind power requires long-term wind data that is not available or does not exist and hence must be simulated. In this research, autoregressive models (AR) ranging from 1st order to 12th order and Markov-switching autoregressive models (MS-AR) ranging from MS(2)-AR(2) to MS(5)-AR(5) are used for wind simulation using 10-minutes wind speed data from NREL for years 2004 and 2005. Simulation results are compared between models, across different seasons, and different data lengths. Consistent with the literature, we find that AR models can efficiently replicate the autocorrelation function (ACF) but not the probability distribution function (PDF) observed in the original data. MS-AR models perform better than AR models in terms of both ACF and PDF and their performance improves with the increasing number of states in the Markov Chain.