Input model identifying techniques for 48H local wind power forecast

The national energy grids face increasing challenges integrating renewable energies, through the continued installation of highly volatile sources of wind power and photovoltaic. Advanced local forecasting methods are needed for optimal grid integration in energy systems with high levels of renewable energy and for trading activities of single or distributed wind power plants. This paper compares variable reducing and selecting algorithms of data mining for designing a wind power forecasting model up to 48h based on artificial neural networks. The used methods based on multiple correlation coefficients, eigenvalues and entropy. This investigation uses online measurement data of a wind park at mountainous country site in Germany and climatic measurement and forecast data of the German weather service.