In this study, failures in photovoltaic (PV) systems, which have increased in Japan since the 2012 feed-in tariff (FIT) introduction, are diagnosed. The I-V curve method is used at the string level for detecting PV array faults, as measuring at the module level is difficult. This approach extracts characteristic values of failures from the difference data between the string I-V curve and the reference I-V curve. A neural network automatically determines the causes of independent failures with high accuracy (97.6%) by comparing the string I-V curve with the reference I-V curve, which represents normal power generation.
