Modelling and fitting of the wind data using different time series models and investigating the relared applications of fitted data. Urla and RisØ cases

Abstract

This thesis is prepared as an outcome of Energy Engineering Master of Science program at IZTECH. Main purpose of this study is to investigate the possible ways of estimating the evolution of wind speed in Turkey, which is useful in predicting the wind power generation. Wind Energy has recently been recognized as one of the most promising renewable energy sources in the world. Despite its high potential, one major problem is that it is an intermittent energy source which follows, in general, statistically a quite noisy evolution with large variability and difficulty in forecasting. Standard time series models have been employed to forecast the wind speed in the literature (such as ARIMA, ARMA). The majority of these, however, are based on a univariate modelling. This is likely to create a significant loss in forecast accuracy as the important dynamics of wind such as ambient temperature, absolute pressure, wind direction and humidity are ignored. So, aim of the present study is to incorporate these factors in a multivariate VAR setting and estimate the wind speed in 4 different locations around Urla City (nearby Izmir-Turkey) by employing hourly data between June-2000 and October-2001. To provide a benchmark, I also compare estimations from VAR with the predictions from ARIMA and SARIMA models. The results indicate two important conclusions. First, it has been shown that all models provide an accurate estimate of wind speed. Second, multivariate VAR and SARIMA is clearly shown to outperform the ARIMA model by improving the wind speed predictions and producing less forecast errors. Thus, these models are demonstrated to be helpful in estimating the wind power generation as well.