Photovoltaic power generation prediction based on FCM-CNN-LSTM-ATTN

To improve the short-term PV power forecasting accuracy and the model's stability under variable weather conditions, this paper proposes a combined prediction model, FCM-CNN-LSTM-Attn, which integrates Fuzzy C-Means clustering (FCM), Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), and Attention mechanism (Attn). First, the FCM algorithm is used to perform soft clustering on input data based on feature membership, dividing the dataset into three weather types: sunny, cloudy, and rainy, and constructing dedicated prediction sub-models to effectively decouple the interference of weather fluctuations. Then, the CNN network is employed to extract the deep spatial relationships between meteorological features and output, achieving feature dimensionality reduction and key information extraction. Subsequently, the LSTM network captures both short-term and long-term temporal dependencies in the output sequence. Finally, the Attention mechanism (Attn) is introduced to dynamically weight historical features, focusing on key information while suppressing noise. Experiments results indicate that the proposed model performs excellently under different weather scenarios and significantly reduces prediction errors.