Comparison of Different Wind Speed Prediction Models for Wind Power Application

Abstract

In this paper, the capability of prediction models is compared for wind speed forecast at different time horizons (i.e. very-short term, short-term, medium term and long term horizons) with the aim of determining their prediction accuracy. The models include: Persistence, second order Markov chain, autoregressive moving average (ARMA) and Weibull models. The models have applications in the areas of electricity market clearing, regulation actions and maintenance scheduling to achieve optimal operating cost. The data used for the study consist of ten-minute average wind speeds for Alexander Bay region of South Africa. Statistical measure and error measures were employed for model validation. The key result reveals that the autoregressive model is best suited for very short and long term wind speed prediction while second order Markov chain is the most appropriate model for short and medium term prediction. Persistence model appears to be the least accurate of all the models for all time horizons.