Photovoltaic Fault Detection Using SE-MobileNet: A Lightweight Channel Attention Network

Engineering notes

Engineering notes will be added by the Power for Solar editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为光伏效率、钙钛矿太阳能电池、储能技术、太阳能热利用、BIPV、并网技术等高价值论文补充中文说明。

Original abstract

The global energy sector is rapidly transitioning from fossil fuels to renewable sources, with solar photovoltaic (PV) systems emerging as a leading solution for sustainable electricity generation. However, maintaining the cleanliness and structural integrity of PV panels is crucial for optimal operation. Therefore, early and accurate fault detection is essential to ensure maintained efficiency and system reliability. This study introduces a lightweight deep learning framework, termed SE-MobileNet, for diagnosing PV surface faults such as dust and bird droppings, as well as physical and electrical damage. The proposed model incorporates a Squeeze-and-Excitation (SE) block into a MobileNet backbone to enable channel-wise feature recalibration without imposing a significant computational burden. Experimental results demonstrate that this adaptive channel-wise refinement yields a notable improvement in classification performance, increasing the macro F1-score from 0.776 to 0.837, while outperforming a range of benchmark convolutional and transformer-based models. Furthermore, the model achieves a favorable trade-off between accuracy and computational complexity, with an inference time of 12.73 ms and 78.54 FPS, enabling practical deployment under near real-time conditions in PV monitoring systems.

Links and sources

Comments