Neuro-fuzzy networks for short-term wind power forecasting

Abstract

This paper presents a statistical model based on a hybrid computational intelligence technique that merging neural networks and fuzzy logic for wind power forecasting. A mesoscale NWP model is used to forecast meteorological variables at a reference point of a wind farm for the next 36 hours at half-hour intervals. The output of the NWP model, together with measured data form SCADA and wind tower, is processed by the proposed model to accurately forecast the wind power of each wind turbine in the wind farm. The network architecture and the training algorithm are introduced. The forecasting approach is applied for the wind power forecasting of a real wind farm located in China. The root mean square errors (RMSE) between the forecasted wind power and actual wind power are less than 20%. From the forecasting results obtained, we conclude: The trained neuro-fuzzy networks are powerful for modeling the wind farm and forecasting the wind power. Due to the adaptability of neuro-fuzzy networks, the proposed approach can be integrated into an on-line wind power forecasting system that automatically be tuned during operation.