Prediction of distributed photovoltaic power based on data interpolation with SE-CNN-BiGRU fusion model

Aiming at the shortcomings of traditional PV power prediction methods, which are difficult to balance the accuracy and robustness. A neural network model including missing value interpolation based on Copula function and Pearson coefficient, and combining convolutional neural network (CNN) of SE channel attention mechanism with bidirectional gated recurrent unit (BiGRU) is proposed, which can realize the accurate prediction of short-term PV power. The method firstly filters out the main meteorological factors affecting PV power through correlation calculation; then uses missing value interpolation based on Copula function and Pearson coefficient; next fully extracts the spatiotemporal features of the main meteorological parameters and the historical power through the CNN incorporating SE; after that captures the dependencies between data sequences by using the BiGRU; and finally validates the prediction by evaluating the metrics results. The comparative experimental and simulation results show that the method has low prediction error and can realize the accurate prediction of short-term PV power at the day- ahead minute level.