Research On Data-Driven-Based Wind Power Feature Extraction And Power Prediction Method

To achieve the goal of carbon peaking and carbon neutrality,the installed capacity of wind power has increased significantly in recent years,and the uncertainty of wind power brings great challenges to the safe and stable operation of the power system.In order to reduce the abandonment of wind power and reduce the operation risk of the power grid,it is necessary to improve the accuracy of wind power prediction.Therefore,aiming at the ultra-short-term and day-ahead prediction of wind power,wind power prediction methods under different data conditions are proposed in this thesis.1)Under the condition that only one-dimension wind power data is provided,the core task of wind power prediction is to accurately extract,fit and predict the wind power change trend from the power data with intermittent,volatile and random features.Therefore,an ultra-short-term wind power prediction method based on time-frequency feature extraction is proposed.Firstly,the variational mode decomposition(VMD)algorithm is used to decompose the wind power into multiple intrinsic mode functions(IMF)with different frequencies,so that different features of wind power can be processed separately to avoid mutual interference among various features.Then,the time-frequency features of each IMF are analyzed,and prediction models are respectively established for each IMF based on the analysis results.Finally,the result of wind power prediction is obtained by summing and reconstructing the prediction results of each IMF.The proposed method is verified based on the real dataset of a wind farm in East China.The results show that the method can effectively extract the timefrequency features of wind power and reduce the ultra-short-term prediction errors of wind power.2)If the data of adjacent observation points can be obtained in addition to the wind farm’s own data,the spatio-temporal correlation between adjacent multiple points can be used to overcome the inherent drawbacks of the prediction method based on the time continuity of wind power.Under this data condition,the core task of wind power prediction is to accurately extract the spatio-temporal features from the spatio-temporal data of adjacent multiple points,and provide future information for the prediction task in the downstream with the data that has occurred in the upstream.Therefore,an ultrashort-term wind power prediction method based on spatio-temporal feature extraction of multiple points is proposed.Firstly,the wind speed features at observation points are expanded,and the correlation between power of wind farm and wind speed at multiple points is evaluated.Then,the wind power and wind speed data are constructed into a spatio-temporal feature matrix,and input into the combined prediction model based on the convolutional neural network(CNN)and bidirectional gated recurrent unit(BGRU)to realize ultra-short-term prediction of wind power.The proposed method is verified based on the real dataset of a wind farm,provided by the National Renewable Energy Laboratory(NREL).The results show that the method can effectively extract the spatiotemporal features of multiple points and reduce the ultra-short-term prediction errors of wind power.3)For wind power day-ahead prediction,due to the uncertainty of wind power,the prediction method that only relying on the historical trend of wind power for extrapolation is difficult to ensure the accuracy and reliability of the day-ahead prediction.Therefore,it is necessary to introduce numerical weather prediction(NWP)data to provide future information for the day-ahead prediction.However,the lateral errors in time and the longitudinal errors in amplitude between the NWP data and the wind farm measured data will directly affect the accuracy of the wind power prediction results.Therefore,a day-ahead wind power prediction method based on NWP wind speed error correction is proposed.Firstly,the lateral errors in time and the longitudinal errors in amplitude between the NWP data and the wind farm measured data are analyzed,and the NWP errors are corrected by using time series correlation test and residual channel attention network(RCAN).Then,combined with the corrected NWP data and the real-time data of the wind farm,a day-ahead wind power prediction model based on BGRU is established.The proposed method is verified based on the real dataset of a wind farm in East China.The results show that the method can effectively correct the NWP wind speed error and further reduce the day-ahead prediction errors of wind power.There are 29 figures,19 tables and 82 references in this thesis.