The Short-Term Photovoltaic Power Prediction Model Based on FCM-BLS

In recent years, the application of deep learning in photovoltaic (PV) power forecasting has demonstrated high prediction accuracy, but it is also accompanied by limitations such as long training times. To address these challenges, this paper proposes a novel short-term PV power prediction model based on Fuzzy C-Means (FCM) and the Broad Learning System (BLS). First, a self-organizing map is employed to finely cluster PV power data at different time points, extracting power fluctuation characteristics under various meteorological conditions. Then, a multi-step prediction model based on BLS is constructed. By expanding the number of nodes along the network's width and training the neural network through the pseudo-inverse of matrices, the model achieves a strong capability for high-dimensional data fitting while enhancing computational efficiency. Finally, simulation analyses are conducted using real-world PV power data, and the proposed model is compared with conventional forecasting methods, demonstrating its significant advantages in both prediction accuracy and training efficiency.