Short-term photovoltaic power prediction based on dual decomposition with TCN-Informer-xLSTM

As renewable energy generation is increasingly integrated into power grids worldwide, the random nature of renewable energy output poses significant challenges to the stability of power systems. Therefore, it is essential to accurately predict the output of renewable energy sources. In this paper, a dual decomposition algorithm based on variational mode decomposition (VMD) and improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) is proposed to decompose the original photovoltaic power sequence and combine the entropy values of the subsequences to obtain the predicted sequences for the high frequency and low frequency components. Then, different prediction models are used for the high-frequency and low-frequency sequences to predict the photovoltaic outputs, where the Temporal Convolutional Networks (TCN)-Informer model is used for the high-frequency component and the xLSTM model is used for the low-frequency component, and finally, the RIME algorithm is applied to find the optimization of the hyperparameters. The results of simulation analysis show that the quadratic decomposition method proposed in this paper significantly improves the prediction accuracy of photovoltaic sequences and reduces the computational complexity