Weather-independent MPP prediction of PV plants using on-site measured dark IV string curves and deep learning

The dark IV string curve method is capable of detecting some faults on PV modules on-site without the need for specific weather-conditions or being time-consuming. This work improves this technique by developing a neural network to predict the MPP of the whole string using only the on-site measured dark IV string curve as input. After briefly describing the main points of the preprocessing and the training, the results of the model are discussed and illustrated graphically. The model is capable of predicting the MPP of a string with a mean deviation of only 2.39% on a test-set.