Photovoltaic power prediction model based on EMD-KPCA-GRU

Abstract

With the aim of enhancing the accuracy of PV power forecasting, a PV power prediction model has been presented based on the EMD-PKCA-GRU neural network. Firstly, the sequence of four meteorological factors constraining the output power of photovoltaic power generation is decomposed using the empirical model decomposition method. The nonstationary nature of the meteorological factor sequences is reduced. Next, feature extraction is conducted from decomposed sequences using kernel principal component analysis (KPCA) methods. This eliminates the redundancy of the original sequence and reduces the dimensionality of the model input. Finally, the Gated Recurrent Unit (GRU) is established, using the GRU to predict PV power generation. The validation was conducted using the 2021 data from the Trina 1B power station in the DKASC dataset from Australia. The outcome shows that the prediction accuracy is higher than the traditional model.