Abstract The inherent variability and stochastic nature of photovoltaic power(PV) generation pose substantial challenges to ensuring grid stability. As the level of PV integration into the grid continues to rise, accurately predicting its power output becomes increasingly critical. This study presents a new PV power prediction model utilizing the density-based spatial clustering of applications with noise(DBSCAN)-bidirectional long short-term memory(BiLSTM)-Transformer framework. The DBSCAN clustering algorithm is applied to analyze historical power data, categorizing it into three distinct groups corresponding to different weather conditions. Then, the BiLSTM-Transformer architecture is employed to develop a power output prediction model tailored for the three weather scenarios. Experimental findings demonstrate that the proposed DBSCAN-BiLSTM-Transformer PV power prediction model exhibits superior accuracy, enhanced generalization, and increased robustness compared to alternative prediction models.
Liang et al. (Fri,) studied this question.