Multiscale analysis and deep learning-based prediction of partial shading for TBC and TOPCon PV modules

In practical applications, partial shading significantly reduces the power generation of photovoltaic modules, leading to considerable energy losses and hotspot effects. While the superior shading tolerance of back-contact (TBC) cells is well-recognized, its quantitative impact on the actual energy yield of photovoltaic systems remains insufficiently explored. This study bridges this gap by combining multiscale simulations with experimental measurements to systematically evaluate tunnel oxide passivating contact (TOPCon) and TBC modules. PVsyst simulations reveal that although TBC modules achieve a marginally higher annual energy yield (0.05%–0.46%) under partial shading compared to TOPCon, both technologies exhibit comparable performance under large-area shading conditions—a finding corroborated by field tests showing aligned I–V characteristics in such scenarios. However, TBC demonstrates clear advantages in small-area shading, reducing power loss by 4.88% in residential installations. PVmismatch simulations further validate the rules of I–V curve deformations under shading. Furthermore, we propose a Bidirectional Residual Variational Autoencoder (BD-ResVAE) model, which enables accurate bidirectional prediction between shading patterns and I–V curves, achieving a structural similarity index of 0.87–0.93, an inception score of 1.01–3.33 and an average mean square error of 0.002, providing a cost-effective solution for real-time PV system monitoring.