An efficient fault diagnosis method combining multi-angle feature expansion and visual image neural networks for solar photovoltaic modules

Fault diagnosis of photovoltaic arrays is a key link to ensure stable operation of photovoltaic systems and improve power generation efficiency, and timely and accurate fault diagnosis becomes particularly important. A photovoltaic fault diagnosis method, termed MA-GCT, is proposed based on multi-angle feature expansion and visual image-based deep learning. Voltage, current, and power signals are enhanced using multi-angle features to enrich fault information. The sequences are subsequently converted into two-dimensional images through gramian angular field (GAF) transformation, enabling the joint representation of temporal dynamics and structural characteristics. A hybrid CNN-transformer architecture is developed to leverage both local feature extraction and global dependency modeling. Experimental validation on multi-class fault scenarios, including simulations and real-fault data, demonstrates classification accuracy, recall, and F1 scores of no less than 96.7 %, 95.0 %, and 95.8 %, respectively. The proposed model consistently achieved a fault diagnosis accuracy exceeding 99 % across both simulation and experimental evaluations, demonstrating its robust ability to extract discriminative features and classify diverse fault types.