Photovoltaic Power Prediction Model Based on Dual Domain Decomposition and Dual Network Structure

In order to improve the accuracy of Photovoltaic (PV) power prediction, a model considering signal decomposition technique and hybrid deep learning is proposed. First, a dualdomain decomposition model based on maximal overlap discrete wavelet transform (MODWT) and complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is proposed to preprocess the data. Secondly, these components are correlated and analyzed into temporal and stochastic components, where the temporal component is input into the long short-term memory network for training, and the stochastic component with several types of environmental factors is jointly input into the convolutional neural networks (CNN) for training, and then the component predictions of the dual network are reconstructed to obtain the point prediction results. The examples show that the proposed model for the point prediction session is optimal among all models of the setup, and the average absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE) are improved by 21.75%, 15.84%, and 21.11%, respectively, compared to the dual decomposition single mesh model, which is the best performer among the compared models.