Computer model uses AI to detect faults in residential solar


Solar arrays can often experience faults that go undetected for long periods of time, causing generation and revenue losses. 

In particular, residential-scale solar projects lack sensing and instrumentation for performance monitoring and fault detection.

Researchers at the University of Massachusetts Amherst addressed this issue and developed SunDown, a computer model for residential solar fault detection.

The model sidesteps instrumentation and sensors, and instead leverages correlations between the power produced by adjacent panels to detect deviations from a panel’s otherwise expected behavior.

A residential-focused fault detection tool from UMass that relies on AI.

The model can handle concurrent faults in multiple panels and classify the anomaly to determine the probable cause.

SunDown uses two years of solar generation data both from a real home and from a dataset of historic system faults. The result, the researchers said, is a Mean Absolute Percentage Error of 2.98% when predicting per-panel output.

The researchers said the SunDown model has demonstrated an ability to detect and classify faults–including from snow cover, leaves and debris, and electrical failure–with a  99.13% accuracy. It can also detect multiple concurrent faults with a 97.2% accuracy rate.

Two main models are used in SunDown. One is a linear-regression-based model that takes advantage of the high correlation between adjacent panels. The other is a graphical model using what the researchers called “half-sibling regression,” which is based on a recently proposed machine learning technique.

This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: