Power Performance Assessment in a Grid-Connected PV System in the Brazilian Northeast
Abstract
This paper presents a data-driven framework for performance assessment and anomaly detection in a 0.815 kWp grid-connected photovoltaic (PV) system located in Recife, Brazil. A simplified linear model was implemented to estimate AC power output as a function of plane-of-array irradiance and module temperature, incorporating a calibrated global loss factor (η = 0.90) representative of real operating conditions. The model achieved strong agreement with measured data (R² = 0.95, rRMSE = 12.8%), confirming its suitability for low-cost monitoring applications. Two complementary anomaly detection methods were applied: an envelope-based approach using modeled versus measured power binning, and a residual-based approach using boxplot-derived statistical thresholds. Both techniques successfully identified known low-generation events and controlled shading experiments, validating their reliability under real field conditions. The proposed methodology offers a simple, interpretable, and computationally efficient tool for operational PV monitoring and can be extended to real-time fault detection in small and medium-scale distributed systems.
Keywords
PV performace, PV modeling, anomaly detection, power performance assessment