Fault Classification in Photovoltaic Arrays Using Convolutional Neural Networks with L2 Regularization

Fault detection in photovoltaic (PV) arrays is of paramount importance, because the output power is maintained at its highest value and the efficiency of the system is prolonged. In the past, there were issues of output accuracy of the detection method in complex situations such as PS and high impedance faults. This paper proposes a method for diagnosing PV failures based on the combination of Hybrid Transformer- Convolutional Neural Network (CNN) Architecture framework that is used for the purpose. Proposed method implies an extension of transformers that is capable to capture long-range dependencies of images, and CNNs to extract spatial features. This is because it performs in a higher way which is to achieve better fault classification performance. A contrast with the extant techniques that deal with limited few scenarios of faults, we have faced up to five different faulty scenarios, PS included and have devised the maximum power point tracking (MPPT). The results of this experimental work proved that the implemented hybrid method is much better than the previous methods regarding fault detection, and succeeded in having the accuracy of 88.4% compared to the other methods; and it is a promising one for betterment of the fault diagnosis in the photovoltaic systems.