Wind Power Prediction Technology Based On Extreme Learning Machine

With the continuous development of wind power technology,the proportion of wind power generation in Chinese power system is becoming larger and larger.But the volatility and intermittent of wind energy always seriously affect the stability of the power system.A more accurate wind farm power prediction can scientifically and timely manage wind power by adjusting the scheduling plan of power system to minimize the negative impact of wind instability on the power grid.In this paper,the existing wind power prediction methods are depicted and an improved prediction model is proposedFirst of all,the parameters of wind farm and all of the key factors of wind power were analyzed.According to the actual wind farm data,the wind speed and direction were analyzed,and the relationship among the wind speed,wind direction,temperature and wind power was studied.The model can provide a theoretical basis for the input variables of wind power and accurately forecast the wind power in the future.Secondly,the wavelet threshold de-noising method and principle were studied.The original wind speed sequence of the actual wind farm is decomposed by wavelet transform.And in order to further distinguish the useful signal and noise signal,the second wavelet decomposition to the high frequency part was explored.Comparing different threshold de-noising methods by de-noising evaluation standard and choosing the best one to deal with the high frequency part of the second wavelet decomposition,which can minimize the negative impact of instability of wind speed series.Thirdly,considering the traditional kernel extreme learning machine,which is based on batch training and its matrix inversion is extremely complicated,an improved kernel extreme learning machine is proposed and applied to actual wind farm power prediction.The engineering example simulation experiment data proves that the prediction accuracy of the improved kernel extreme learning machine is much better than the traditional one.Above all,a more accurate forecasting model is established combing wavelet threshold de-noising method and improved learning machine and the effectiveness of the new forecasting model is verified by the actual wind farm experiment data.The engineering example simulation experiment demonstrates that the model improves the accuracy of wind power forecasting.During the internship,I participated in the development of a wind power prediction software,which is depicted finally.And the extracted method is evaluated using this software.