A robust rule-based method for detecting and classifying underperformance in photovoltaic systems using inverter data

Abstract

This study presents a practical and scalable rule-based methodology for detecting and classifying underperformance in photovoltaic (PV) systems using only inverter data from the alternating current (AC) side. Motivated by the need for reliable, low-cost underperformance detection in distributed PV systems, the proposed approach eliminates reliance on high-resolution direct current (DC) side measurements or complex sensor infrastructure. A suite of algorithms was developed to identify and classify common underperformance events, including generation clipping, inverter tripping, recurring anomalies, and seasonal or daily yield reductions. The method was validated using real-world data from 1089 PV systems (2213 inverter monitors) throughout Australia, representing eight major inverter brands. A subset of 807 industry-labelled fault instances was used for performance benchmarking. The results demonstrated high classification accuracy for underperformance events (92% and 88% for our two defined cases), while highlighting areas for refinement in detecting more ambiguous cases such as generation clipping (56%). This work addresses a critical gap in current performance monitoring practices, offering a robust, low-intervention solution for PV fleet operators seeking to improve reliability, fault response, and economic performance at scale.