Lineer öngörü metodları ile rüzgar hızı öngörüsü

Short-term forecasting of wind speed is of great importance to wind turbine operation and efficient energy harvesting. In this thesis, one-step ahead wind speed forecasting is performed. Six approaches based on linear prediction methods are employed for this purpose. The first approach features the autoregressive process (AR) with the model order eight. Model order selection criterias, Akaike information criteria (AIC) and Bayesian information criteria (BIC), are used for optimal model order selection. These information criterias selected the same autoregressive model with an order of eight, which is shown as AR(8). Second approach employs the autoregressive moving average process (ARMA). In this case, AIC and BIC selected the autoregressive moving average model with different order. First model is defined as autoregressive moving average model with an autoregressive order of four and moving average order of three which can be shown as ARMA(4,3) and second model is defined as ARMA(14,13). Third approach features the autoregressive integrated moving average process (ARIMA). In this case, AIC and BIC pointed different model orders once again. In addition the notation of ARMA, an integration process with an order of one is added and shown as ARIMA (30,1,29) and ARIMA(3,1,2). Two different models are performed in this case. On the other hand fourth, fifth and sixth approaches involve employing an exogenous input to the first three approaches. In first case, autoregressive model with an exogenous input, which is denoted as ARX is featured. Depending on the model selection criterias, the order of autoregressive model with an exogenous input is selected as one, which is shown as ARX(1). In the next case, the criterias for model order selection pointed the same model order. Autoregressive order of two and moving average order of one with an exogenous input model, which denoted as ARMAX(2,1) is performed. In third case, AIC and BIC selected the first order integrated autoregressive order of eight and moving average order of seven with an exogenous input which is shown as ARIMAX(8,1,7). By employing these six approaches, one step ahead wind speed forecasting is performed. Wind speed data observed in Bursa-Gemlik location with a time interval of ten minutes. The results are compared using mean absolute error (MAE) and root mean square error (RMSE) as a measure for forecasting quality. The goodness of fit is checked by calculating r-square R^2 statistics. It is found that the AR, ARX, ARMA and ARIMA model is better at predicting the wind speed corresponding the R^2 statistics. MAE, RMSE and R^2 statistics also show that ARX model is the best for forecasting one step ahead wind speed. Moreover, ARMAX model is also good at forecasting wind speed whereas it's lower than AR, ARX, ARMA and ARIMA . Results also show that ARIMAX (8,1,7) model is the worse for forecasting one step ahead wind speed. In order to check the success of criterias for model order selection, various ARIMAX models are analyzed. It can be seen from the results that other ARIMAX models are better than ARIMAX(8,1,7). In other words AIC and BIC is not withstanding selecting the model order of autoregressive integrated moving average models with an exogenous input.