Wind power forecasting: averages better than sums?

The aim of this study is to identify a class of models appropriate to describe the wind power production in some wind farms and to provide reliable forecasts in a 72-hour horizon. Wind farms hosts several turbines, the activity of which determines the total power at a point in time. Their functioning depends on several reasons: climatic conditions, electrical and/or mechanical brakes which are uncontrolled or controlled, as in the case of maintenance. Therefore, both the uncertainty related to the the climate (wind speed, direction, air density etc.) and to the functioning of turbines affect the total power generation in the wind farm. These sources of variability are modeled using a time series approach, focusing on average power production per turbine rather than on the overall wind farm power production. The advantage of our approach is to reduce the high variability of the total power, separating the climatic component to the turbines operation-related one. Several models are proposed, consisting of dynamic systems of equations.