基于二维风速修正和多重集成的两阶段迁移学习短期风电功率预测

Two-stage Transfer Learning Short-term Wind Power Prediction Based on Two-dimensional Wind Speed Correction and Multiple Integration

为了在数据量不足的情况下增强新投运风电场的功率预测能力,提出了一种基于二维风速修正和多重集成的两阶段迁移学习短期风电功率预测方法。首先,在数据增强阶段,引入投运前气象站的测风数据,基于风电场的时空相关性关系,通过时序特征构建和场景匹配,从时空2个维度对预报风速进行初步修正。然后,对初步修正后的结果进行数据重构,以重构后的数据作为下一次集成的输入,构建多重集成模型对预报风速进行二次修正。最后,在功率预测阶段,基于一阶段的修正结果,通过门控循环单元(gate recurrent unit,GRU)得到预测功率。算例结果表明,所提方法使预报风速的均方根误差降低了1.038m/s,功率预测精度提升了4.718%。论文研究可为新投运风电场的短期功率预测提供参考。

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.