Study On Wind Power Forecasting Of Wind Farm Based On Statistical Methods

Wind power has a rapid development and growing proportion of the power grid. But wind power has the disadvantages of intermittence and randomicity, which will bring challenge to the safety, stabilization, scheduling of power grid. Wind power prediction is an effective approach for these problems. Based on several statistical methods for wind power forecasting are discussed in this paper.Wind power forecast model of every10min for30wind turbines which using SVM regression(SVM-reg), BP Network(BP) and Adapting Partial Least Square Regression (APLSR) repectively were developed based on actual wind power recorded data and wind speed, wind direction, atmospheric temperature, relative humidity and atmospheric pressure values which predicted by WRF at the hub height during Jan to Dec.,2008. For assessing the forecasting effects of these three wind power forecasting model, forecasting experiments during Jan to Dec.,2009were carried out by the same way. The results show that:the SVM-reg prediction is better than the other two methods, and if the BP and APLSR prediction ineffective, the SVM-reg forecasting can have a better accuracy. The SVM-reg forecasting results and the actual wind power recorded data’s correlation coefficient ranged from0.71-0.82, normalized root mean square error ranged from9.8%to16.5%, and normalized mean absolute error was between5.4%and10.5%. Theirs full year’s correlation coefficient was0.79, normalized root mean square error was13.3%, and normalized mean absolute was8.3%.A wind power ensemble forecasting method was developed using a linear ensemble forecasting method to integrate SVM-reg, BP and APLSR predicted results in the end. In this ensemble method use the minimum root mean square error as the objective function to distribute the weight coefficient. The test results show: the prediction accuracy rate of ensemble forecasting method is slightly higher. Compared to the SVM-reg results, the annual normalized root mean square error of0.4%, the annual normalized mean absolute of0.2%.