Architecture for Monitoring PV Degradation in Real Time from EL Images

We developed an architecture to monitor the degradation rate of PV (Photovoltaic) in real time from EL(Electroluminance) images. In recent years, technology has been developed to measure EL images outdoors, but it has been difficult to accurately grasp performance degradation from EL images. First, we extracted the luminance distribution from the EL image of the PV module that underwent an accelerated test, and created an input layer for machine learning. Second, in the training of a neural network that isolates EL images of PV modules before and after the output drops rapidly, we found a high correlation between the output and degradation of the hidden layer, and created an approximate equation. By applying the luminance distribution of the measured EL image to the learned parameters and applying the output of the calculated hidden layer to an approximate formula, we succeeded in calculating the power generation performance instantaneously.