Hybrid prediction method for PV power generation based on BiNLSTM framework and similar day clustering

Addressing the strong randomness of photovoltaic (PV) power generation, a Bidirectional Nested Long Short Term Memory (BiNLSTM) prediction model considering weather types is proposed This method initially employs the Pearson correlation coefficient to extract the main meteorological factors influencing PV power generation. Then, the K-means clustering algorithm is used to group historical days with similar weather characteristics, selecting those with strong correlations to construct the training set. Finally, the BiNLSTM model is employed to capture the dependencies between data sequences and execute PV power prediction. In the experimental analysis, the aforementioned method was compared with a model that did not consider similar weather types. Through case studies, it was verified that the proposed method exhibited high accuracy and robustness under various weather conditions, confirming the effectiveness of the proposed model. Compared with traditional photovoltaic power generation forecasting methods, the proposed method effectively improved prediction accuracy.