Work

From Correlation to Prediction: Electroluminescence-based Indicators and Composite Health Scoring for Photovoltaic Power Loss Prediction in a 204-Panel Fleet

Year: 2026
Type: article
Abstract: Electroluminescence (EL) imaging is an established non-destructive technique for detecting cell cracks, hot spots, and edge damage in crystalline-silicon photovoltaic (PV) modules, but most fleet-scale studies stop at correlation analysis without testing out-of-sample predictive power.This study analyses a 204-panel, 63.24 kWp rooftop PV fleet and develops a composite Health Score combining Power Performance, Crack Severity, and Thermal Risk dimensions with an edge-damage penalty as an operational prioritisation tool.Pearson/Spearman association, 5-fold crossvalidated Random Forest and linear-regression models, and a five-scenario weight-perturbation analysis separate association from prediction and test the Score's ranking stability.Crack count showed the strongest association with power loss (r = 0.96) and the highest cross-validated predictive performance (R² = 0.92), while crack area and hot-spot area showed weak, sometimes inverse, associations (r = -0.12 and -0.18), diverging from single-cell crack-hotspot coupling.Because the Health Score's Power Performance component is defined directly from Power_Loss, its principal interpretation is as a composite condition and maintenance-prioritisation index rather than as an independent predictor of Power_Loss; it showed substantial rank stability under ±5 percentage-point weight perturbation (Spearman ρ ≥ 0.98, top-20-worst overlap ≥ 85%).These findings offer an empirically evaluated, fleet-scale framework for EL-informed operations-and-maintenance prioritisation in tropical PV installations.
Language: English
Field: Energy
Open Access status: hybrid
OpenAlex ID: W7234847847
FWCI:
Cites: 25
Cited by:
Related to: 10