Research On The Indirect Prediction Method Of Short-term Wind Power Based On CEEMDAN

As a new type of clean energy,wind energy is increasingly concerned by the international community.The study of wind power prediction is an important prerequisite to ensure the safe and stable operation of wind power grid connected system.Accurate wind power prediction can make the power department reasonably arrange the scheduling plan;make the wind farm reasonably arrange the maintenance work for the wind turbine,ensure the normal operation of the unit,improve the utilization rate of the unit,and reduce the operation cost of wind power generation.However,the random fluctuation of the original wind speed signal leads to the large fluctuation of the wind power,which is difficult to predict accurately.It seriously affects the quality of the wind power after the wind power is connected to the grid,and brings a great challenge to the wind power prediction of the wind farm.Therefore,this paper proposes the indirect prediction of short-term wind power,the main research contents are as follows:Firstly,the characteristics of wind power generation are introduced,and the data preprocessing is based on the four wind power data collected by SCADA data acquisition system.In order to ensure the accuracy of the model,DBSCAN algorithm is proposed to eliminate the abnormal points of the data.Then,because it is difficult to improve the prediction accuracy of wind speed due to the random fluctuation characteristics of the original wind speed,this paper uses the self-adaptive noise full set empirical mode decomposition algorithm(CEEMDAN)to decompose the original wind speed into multiple frequency-domain stable subsequences.In order to reduce the calculation,the sample entropy algorithm is used to recombine multiple subsequences.On this basis,BP neural network,RBF neural network,Wavelet neural network and generalized regression neural network are used to predict the wind speed of the new sequence after reorganization.Finally,the final wind speed prediction results are obtained by weighted integration of the prediction results by partial least square regression algorithm.Secondly,the wind power is predicted.In order to improve the prediction accuracy of wind power curve fitting method,polynomial fitting,local weighted polynomial fitting,cubic B-spline and cubic B-spline are respectively used to fit the wind power curve.Through the analysis and comparison of the four fitting results,it can be seen that the fitting accuracy of cubic B-spline is higher.Therefore,it is necessary to the third-order B-spline fitting model is applied to the indirect prediction of wind power.The prediction results show that the method has high accuracy of power prediction.Finally,based on the prediction of wind power points,a nonparametric kernel density estimation method for wind power interval prediction is proposed.According to the distribution of the probability density curve of wind power,the error probability density curves of different regions are obtained,and then the interval prediction of wind power is made by combining nonparametric kernel density estimation.Three evaluation indexes are used to evaluate the interval prediction results,and the effectiveness of the method is verified by comparing the parameter estimation method.