Short-term photovoltaic power prediction model based on enhanced variational mode decomposition and optimized en-semble neural networks

In order to mitigate the reduced predictive precision of traditional forecasting models under varying weather conditions, this paper proposes a short-term photovoltaic (PV) power prediction model founded on enhanced variational mode decomposition and optimized ensemble neural networks. Initially, key meteorological factors influencing PV power generation are selected through correlation analysis. Subsequently, the Fuzzy C-Means Clustering (FCC) algorithm is adopted to categorize historical data into multiple weather types. The RIME algorithm (RIME) is then applied to optimize the hyperparameters of the Variational Mode Decomposition (VMD) algorithm, decomposing PV power data of different weather types into subseries with distinct frequency characteristics. Thereafter, Convolutional Neural Networks (CNN) are leveraged to extract spatiotemporal features comprehensively, while Bidirectional Long Short-Term Memory (BiLSTM) networks are utilized to capture temporal dependencies within the data sequences. Furthermore, an Attention Mechanism (AT) is incorporated to enhance the influence of highly correlated components. Each PV power sub-series of different weather types is predicted independently, and the Crested Porcupine (CPO) algorithm is introduced to optimize hyperparameters for the model’s forecasting algorithm. Comparative experimental analyses validate that the proposed approach achieves lower prediction errors, demonstrating its effectiveness in enhancing the precision of short-term PV power forecasting.