Wind power forecasting and error analysis using the autoregressive moving average modeling

This paper presents a method for wind power forecasting and studies the relationship between the accuracy of the forecast and wind power variability. Actual wind power measurement data is applied to model an auto regressive moving average (ARMA) process. Burg and Shanks algorithms are then utilized to determine the model coefficients. Variability, accuracy and measured error in forecasts generated by the model are used to asses the data and the quality of the forecast. The model is shown to have good accuracy in forecasts within one hour and declines in accuracy further ahead in time. Drawing comparisons between forecasts generated for cases of differing data variability, the aggregate power generation of a group of wind farms is shown to have better accuracy in forecasts than the single wind farm.