A Lightweight, Transferable, and Self-Adaptive Framework for Intelligent DC Arc-Fault Detection in Photovoltaic Systems
A new AI framework hits 99.99% accuracy and 0% false-trip rate for detecting dangerous electrical arcs in solar panels.
A research team has published a paper on arXiv detailing a novel AI framework that solves a critical safety problem in residential solar power. The proposed LD-framework is designed for intelligent DC arc-fault detection, a major fire hazard in photovoltaic (PV) systems. Conventional Arc-Fault Circuit Interrupters (AFCIs) struggle with reliability due to signal noise from inverters, hardware differences, and environmental changes. This new approach uses a three-part system: LD-Spec for efficient on-device spectral analysis, LD-Align to adapt the model across different inverter brands, and LD-Adapt, a cloud-edge update mechanism that allows the system to learn from new, unseen operating conditions over time.
Extensive testing demonstrates the framework's exceptional performance. Trained and validated on over 53,000 labeled samples, it achieved near-perfect metrics, including a 0.9999 accuracy and 0.9996 F1-score. Crucially, it maintained a 0% false-trip rate during scenarios that typically cause nuisance trips, like inverter startup or load switching. The system also showed remarkable transferability, adapting to new hardware with only 0.5% to 1% of labeled data from the target device. In field adaptation tests, it recovered detection precision from a low of 21% up to 95% when faced with previously unencountered conditions, proving its long-term viability and robustness for real-world deployment.
- Achieved 0.9999 accuracy and 0% false-trip rate across 53,000+ test samples, including tricky scenarios like inverter startup.
- Uses a three-part AI system (LD-Spec, LD-Align, LD-Adapt) for on-device inference, cross-hardware transfer, and long-term cloud-edge updates.
- Demonstrated efficient adaptation, requiring only 0.5%-1% of new target data to work on different hardware and recovering precision from 21% to 95% in the field.
Why It Matters
This directly addresses a major safety and reliability barrier, enabling wider, safer adoption of residential solar power systems.