Ultra-short-term Photovoltaic Power Forecasting Based on Secondary Decomposition and BiGRU
For the forecasting of ultra-short-term photovoltaic power,a hybrid prediction model based on complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN),variational mode de-composition(VMD)and bidirectional gated recurrent unit(BiGRU)was developed.The photovoltaic power generation signal was decomposed using CEEMDAN,and the decomposed signals were clustered and reconstructed using sample entropy and the K-means methods.Then,the VMD was applied for the secondary decomposition of complex signals to mitigate signal non-stationarity.The decomposed signal components were employed as inputs for training,validation and prediction in the BiGRU model.Subse-quently,the predicted results from each signal component were linearly combined to obtain the final fore-casting results.Results show that the hybrid model outperforms single models,confirming the effective-ness of the model.By comparing the forecasting performances under typical weather conditions and various evaluation metrics,the generality of the proposed method was validated.
万方数据
维普
