Wind Farm Generation Forecast Based on Componential Sparse Bayesian Learning

Abstract

Probabilistic forecast is different from expectation forecast by the capability of forecasting the distribution of random variables.Based on the componential sparse Bayesian learning,this paper proposes a novel method to forecast the short-term wind farm generation.With this method,a time series of wind farm generation is decomposed into trend component and disturbance components by discrete wavelet decomposition Mallat algorithm.The trend component is then forecasted according to its strong correlation with wind speed and its self-correlation property,while the disturbance components,which are more stationary,are forecasted according to their self-correlation property.A sparse Bayesian learning method is used to establish the forecasting model to give probabilistic forecasts to trend component,disturbance components,and as well as the total wind farm generation.Several learning machines are set up to fulfill a multi-step probabilistic forecast.Case study shows the effectiveness of the proposed method by continuous 7 200 times forecasting tests for a given actual wind farm.