Two-stage Transfer Learning Short-term Wind Power Prediction Based on Two-dimensional Wind Speed Correction and Multiple Integration
In order to improve the power prediction accuracy of newly-invested grid-connected wind farms with insuffi-cient data,a method,which is named for two-stage transfer learning short-term wind power prediction,based on two-dimensional wind speed correction and multiple integration is proposed.Firstly,at data enhancement stage,the wind measurement from weather stations before connection to grid is used.The spatio-temporal correlation of wind farms is taken into consideration,and time series features are constructed and scenes are matched.The forecast wind speeds are preliminarily corrected in both temporal and spatial dimensions.Then,the preliminary corrected results are reconstructed as the input of the next integrated learning,and a multiple integrated learning model is constructed to correct forecast wind speed again.Finally,at power prediction stage,the forecast power is obtained by GRU based on the data correction.The results show that the proposed method can be adopted to reduce the root mean square error of the forecast wind speed by 1.038 m/s and improve the accuracy of power prediction by 4.718%.The research can provide a reference for new-ly-invested grid-connected wind farms power prediction.
power predictionwind speed correctionintegrated learningtransfer learningspatio-temporal correlationnewly grid-connected wind farms
万方数据
维普
