Development and Evaluation of an Automatic Failure Detection System for O&M of PV Portfolios

Tools that help automate operation and maintenance (O&M) processes are fundamental to expanding the ability of service providers to meet the demand of the growing number of PV systems. Within maintenance processes, the daily routine of online monitoring is a repetitive and time-consuming activity that relies heavily on manual analysis by specialized personnel. Automating this routine is challenging and requires a reliable automatic fault detection (AFD) system. This study proposes an AFD system consisting of nine algorithms primarily grounded in experts' knowledge that analyze different aspects of the PV system for abnormal behavior. The algorithms were individually developed, tested and validated with the help of 5 years of data, including tickets created by manual monitoring, from 80 rooftop PV systems. The complete AFD system was then tested by monitoring 170 systems for one month, where it correctly detected 99% of the problems and issued less than 12% of false alerts. It also identified problems (13.2% of the total of actual faults) that passed unnoticed by conventional monitoring. The proposed AFD system provides a promising solution for automatizing part of the online monitoring routine and helps operators expand their capacity to monitor more systems.