Ultra-short-term Wind Power Prediction Based On Combined Model

Due to the pollution and non-renewability of fossil energy,new energy generation has received great attention.Wind energy has a wide range of sources and abundant resources,and has developed rapidly due to its mature technology.However,due to the intermittent and fluctuating wind power,it poses a huge challenge to the stable operation of the power grid.Therefore,real-time and effective prediction of wind power generation is very important for the development of wind power and the stable operation of the power grid.This paper does the following research on wind power forecasting:First,from the perspective of the original data of the wind farm,the basic characteristics of wind power are analyzed.In view of the large amount of abnormal data in the wind power data,this paper proposes a method for cleaning abnormal data based on the LOF algorithm,which makes data more reliable.Secondly,based on the non-linear characteristics of wind power data,this paper uses the least squares support vector machine as the prediction method.Decomposing the original data by MEEMD reduces the noise in the original data,reduces the prediction time,and improves the prediction accuracy.Finally,the LSSVM model has limited processing power for non-stationary components of wind power.This paper proposes a method to predict ultra-short-term wind power based on the combined model of LSSVM and ARIMA.The data is decomposed into low-frequency and high-frequency signals using MEEMD.In this paper,the error value is sent to a vector machine for learning,and the wind power prediction and error prediction are combined to further improve the prediction accuracy.Combined with the historical actual output data of a wind farm,a variety of models are compared,and the MEEMD-ARIMA-LSSVM combined prediction error correction model proposed in this paper has higher prediction accuracy from different angles and can adapt to the higher requirements for prediction accuracy case.