Under complex or harsh environmental conditions, single-data modeling approaches for photovoltaic (PV) cells often fall short in terms of accuracy. To overcome this limitation, this study proposes a hybrid Transformer–BiLSTM framework to model photovoltaic (PV) modules, addressing the limitations of traditional single-model approaches. By leveraging the Transformer’s global attention mechanism and BiLSTM (Bidirectional Long Short-Term Memory)’s ability to capture local dependencies, this hybrid model provides enhanced accuracy and generalization for PV module output prediction under various environmental conditions. We construct a multi-type PV module dataset based on real I–V characteristic data from the U.S. National Renewable Energy Laboratory (NREL), applying K-means++ clustering for data preprocessing. Comparative experiments against standalone models (Transformer, BiLSTM, SVM (Support Vector Machine)) and a Transformer–SVM hybrid demonstrate that the proposed model consistently achieves a coefficient of determination (R2) exceeding 0.989 on both training and testing datasets, significantly outperforming standalone Transformer, BiLSTM, SVM, and Transformer–SVM models.
Liu et al. (Thu,) studied this question.