The short-term wind power forecast based on phase-space reconstruction and neural networks

The short-term forecast of the wind power of a wind farm is of great significance for the security and stability of a grid-connected generation system. An accurate forecast may reduce the spinning reserve of a grid while providing reliable references for operation dispatch of a wind farm. In order to improve the accuracy of short-term forecasts, introducing the phase-space reconstruction technique of the chaos theory, this paper was established forecasting models by reconstructing the historical time-series data of the wind power of a single unit based on the dynamical properties of chaos sequences, choosing the best delay time with the mutual information method, determining the best combination of embedding dimensions with the Cao algorithm, as well as utilizing the Elman recurrent neural network and others like the BP. As comparative case analysis shows, the forecasts of Elman model are more accurate than that of the others, exhibiting a positive prospect of utilizing this combined model of phase-space reconstruction and neural network in wind power forecasting of a single unit.