Short term power forecasting of a wind farm based on atomic sparse decomposition theory

Abstract

The wind power data have very strong nonlinearity and non-stationarity, but the traditional method mainly focuses on the nonlinear problem of the wind power data and doesn't analysis the non-stationary problem. This paper proposed the combining method of atomic sparse decomposition and artificial neural network (ANN) to research the short-term forecasting of the wind power. Firstly, wind power data samples were decomposed into non-orthogonal atom sequences and residual sequences. Then ANN was used to model and predict the residual sequences, and the atom sequences adopt the adaptive prediction. Finally, the forecasting results were stacked and reconstructured. The generation power of an actual wind farm was forecasted by this method. The results show that the combining method of atomic sparse decomposition and ANN can reduce non-stationary behavior of the signal, produce sparser decomposition effect and better predict the variation tendency of the wind power.