- New method flags solar plant underperformance using only inverter data from the alternating current side.
- Validation on more than 2000 systems shows above 90 percent accuracy for major underperformance events.
- Approach offers a low cost pathway to scale monitoring of distributed solar assets without extra field sensors.
A new study from Australia titled ‘A robust rule-based method for detecting and classifying underperformance in photovoltaic systems using inverter data,’ has introduced a practical rule based method to detect and classify underperformance in photovoltaic systems using only inverter data from the alternating current side. The approach is aimed at asset owners and operators of distributed solar who need reliable fault detection without investing in complex direct current measurement infrastructure.
The methodology uses a suite of algorithms to pick up common performance issues such as generation clipping, inverter tripping, zero generation, recurring anomalies and seasonal or daily yield drops. It does this by analysing patterns in the alternating current power output rather than relying on high resolution direct current side voltage and current measurements.
Researchers validated the approach on real world data from more than 1000 solar systems and over 2200 inverter monitors installed across Australia. The dataset covered eight major inverter brands and included more than 800 fault events that had been labelled by industry to benchmark the performance of the algorithms.
The results showed strong classification accuracy for underperformance events, reaching about 92 percent for major underperformance and 88 percent for minor events. Accuracy was lower, at around 56 percent, for more ambiguous phenomena such as generation clipping, pointing to opportunities to refine how these events are defined and detected in practice.
By working with the data that inverters already provide, the method offers a simple, low cost and easily deployable alternative to more data intensive monitoring solutions. It is particularly relevant for medium scale and distributed portfolios where retrofitting detailed sensor packages is not commercially attractive.
The study notes that limited ground truth labelling remains a constraint, but the industry verified subset still provides meaningful validation of the approach. Future work will focus on tuning thresholds, reducing false alarms and incorporating complementary information such as event logs to strengthen fault diagnosis.
For African solar markets that are rapidly expanding grid tied and behind the meter installations, such low intervention monitoring tools could support better fleet reliability and economic performance. By enabling early detection of major and minor underperformance without additional hardware, the methodology can help asset managers protect energy yields and improve long term system safety across diverse operating environments.
Author: Bryan Groenendaal
GoodWe has an extensive distribution network across the African continent.
For enquires email: sales.africa@goodwe.com
