Fault‐Specific Diagnostic Regions Based on Normalized DC Voltage and Current Indices for Photovoltaic Systems Using Routine Inverter and Environmental Data

ABSTRACT The growing deployment of photovoltaic (PV) systems has increased the need for practical data‐driven approaches to support reliable operation and maintenance (O&M). However, direct current (DC) voltage and current are highly sensitive to environmental and operating conditions, while many diagnostic approaches require infrared (IR) or electroluminescence (EL) imaging, current–voltage (I–V) curve tracing, or additional sensors. This study defines fault‐specific diagnostic regions based on normalized DC voltage and current indices for PV systems using routine inverter and environmental data. Fault‐induced maximum power point (MPP) shifts were analyzed using measured I–V curves, and normalized voltage and current indices were derived from inverter‐collected DC voltage and current under normal and emulated fault conditions. The diagnostic regions characterized normal and fault conditions at string and array levels, and classifier‐based evaluation confirmed their separability. The neural network achieved the best performance, with 99.9% validation accuracy and a 99.8% F1‐score. The method was validated on a 3‐kW PV system and applied to operational data from a 387‐kW PV plant. In the field system, actual faults, including multiple blown string fuses, were identified. After fault correction, the DC performance ratio increased by approximately 40%. These results support PV system monitoring and maintenance without additional measurement equipment.