Influence of Noise on Prediction of Power of Wind Power Plant

A sequence of wind farm electric powers can be viewed as a time series. Usually, a time series consists of four basic components: trend, cyclical, seasonal component and random component. The first three components of the stochastic process can be estimated. However, the random component cannot be predicted effectively. The aim of this paper is to analyse how significantly the random component affects the effectiveness of short-term forecasts of the output power. The research is based on an analysis of data measured with a period of ten minutes during 50 days. For short-term predictions of electric power two different approaches were used. The first approach is a method of an exponential smoothing - one of many time series methods by which the future electric power of a wind power plant is calculated. The second approach is based on an analysis of previous data. The method estimates future power at a given time shift period, if several last powers w(1), w(2), ..., w(n) were obtained. Both approaches were used on real data and on data smoothed by moving averages. Applying of the moving averages removes the random component from a time series. Thus, the comparison of predictions on smoothed and non-smoothed data demonstrate how significant the impact of random component on power estimation is.