An Australian research team has developed a five-step rules-based method that detects and classifies underperformance in PV systems using only AC-side inverter data. The approach has been validated on more than 1,000 systems and provides a low-cost, low-intervention solution for improving the reliability, fault response and performance of PV systems.
A research group from Australia has developed a new rules-based method for detecting and classifying underperformance in PV systems using only inverter data from the alternating current (AC) side.
The method has been validated using more than 1,000 PV systems across Australia, with more than 2,000 inverter monitors.
“Motivated by the need for reliable, low-cost detection of underperformance in distributed PV systems, the proposed approach eliminates the dependence on high-resolution direct current (DC) measurements or complex sensor infrastructure,” the research group said. “This work addresses a critical gap in current performance monitoring practices and provides a robust, low-intervention solution for PV farm operators looking to improve reliability, fault response and economic performance at scale.”
The new method follows a five-step process.
Step one collects and processes all necessary data from the PV system. Inputs include AC data at five-minute intervals, along with basic metadata such as system size, location, tilt, and azimuth.
Step two compares the expected generation under the prevailing meteorological conditions with the measured generation.
Step three applies daily if-then rules to detect underperformance. Major underperformance occurs if actual production falls below 60% of expected production for at least three consecutive non-cloudy days. A minor underperformance is recorded when production is lower than 80% of expected yield for seven consecutive days. Additional algorithms identify systematic variations between weekdays and weekends that may indicate planned maintenance, as well as seasonal performance issues.
Step four uses algorithms based on five-minute data points to classify the type of underperformance. Specific routines detect generation cuts to zero, non-zero cuts, generation cuts, zero power, recurring underperformance, power flow anomalies, and over-generation.
Step five produces a report summarizing the flagged events, their severity, and the identified error type.
Image: Sydney University of Technology, Solar Energy, CC BY 4.0
The proposed method was demonstrated in a case study of 1,089 PV systems and 2,213 inverter monitors in Australia, with a five-minute resolution over the period from January 2021 to September 14, 2023.
Capacities ranged from small homes with a capacity of less than 10 kW to large installations of more than 50 kW, spanning eight Australian states and territories.
To validate the system, the team uses a list of 807 manually labeled faults from 177 PV locations.
“Results from this dataset showed strong classification accuracy for major and minor underperformance (92% and 88%, respectively), with lower accuracy for more ambiguous categories such as generation clipping (56%),” the group said. “These findings suggest opportunities to refine detection thresholds and better align algorithmic definitions with industry interpretations.”
They further noted that future work would focus on improving threshold tuning, reducing false positives, and incorporating complementary data sources, such as event logs, to increase the robustness of the system. They added that these approaches could ultimately support the development of low-intervention integrated monitoring systems that can ensure sustainable performance and safety for a range of PV installations.
Their findings appeared in “A robust rule-based method for detecting and classifying underperformance in photovoltaic systems using inverter data”, published in Solar energy. Scientists from Australia Sydney University of Technologyenergy resource management company Diagno Energy, and the University of New South Wales contributed to the research.
This content is copyrighted and may not be reused. If you would like to collaborate with us and reuse some of our content, please contact: editors@pv-magazine.com.
Popular content
