Fault Identification System for Photovoltaic Panels with Artificial Intelligence

This research presents the design and simulation of a neural network-based fault identification system for a photovoltaic panel. The system allows detecting mismatch and degradation faults caused by humidity, which are equivalent to the increase or decrease of the internal series resistance of the panel respectively, thus preventing damages that could limit its performance and lifetime. Mismatch failures are caused by the occurrence of hot spots, while panel exposure in humid environments causes failures due to moisture degradation. A photovoltaic panel is modeled using the parameters provided by its manufacturer. A series resistance estimator based on the recursive least square's method with a forgetting factor and upper and lower confidence intervals is proposed. Fault identification is performed using a multilayer perceptron neural network with supervised training. Inputs to the network are irradiance and estimated series resistance value. Outputs are: normal operation, failure due to mismatch and failure due to moisture degradation. The estimator is evaluated for various scenarios, including normal and failures operation. In addition, it is subjected to different solar irradiance profiles based on real data. The estimator demonstrates good performance, correctly identifying all evaluated operating points.