The Decomposition–Extraction–Fusion Framework for Robust Short‐Term Photovoltaic Power Forecasting Under Chaotic Uncertainty

Accurate photovoltaic (PV) power prediction is of great significance to the stable operation of new power systems. However, PV power generation is influenced by meteorological conditions and geographical location, leading to high uncertainty and fluctuations to precise prediction. To effectively capture the complex patterns of PV power generation, this study proposes an intelligent hybrid prediction model. First, an analysis of intrinsic and extrinsic uncertainties of the PV power generation system is conducted to gain a comprehensive understanding of the physical mechanisms underlying power generation. Second, the adaptive empirical wavelet transform (EWT) is employed to decompose the PV power sequence, extracting rich features. Then, considering the volatility, uncertainty, and complexity of those decomposed components, the TOPSIS and the attention mechanism are applied to allocate the feature fusion and associate weights, respectively. Finally, the local features of the data are extracted using the convolutional neural network (CNN), and the dynamic weights of the attention mechanism are used to focus on the key features. These features are then used as the input of the bidirectional long short‐term memory (BiLSTM) network. The whale optimization algorithm (WOA) is introduced to adaptively optimize the hyperparameters, thereby improving the training efficiency and accuracy of the model.