Desenvolvimento de instrumentação para aquisição de dados solarimétricos distribuídos espacialmente e com alta resolução temporal

Abstract

A distinctive feature of photovoltaic solar generation is its sensitivity to atmospheric conditions. On cloudy days, module shading significantly reduces energy production, making the source intermittent and stochastic. This variability can lead to undesirable interactions with the electrical grid. Such interactions are better managed when photovoltaic generation is predictable. In this context, this dissertation presents the development, implementation, and validation of an experimental solar monitoring architecture based on photovoltaic modules used as irradiance sensors in low-cost embedded systems. The proposal aims to address the lack of solarimetric data with high temporal resolution (1 Hz) and spatial distribution, especially in tropical regions with high atmospheric variability, such as Brazil. Four prototypes with GPS-synchronized acquisition were developed and installed at different locations on the campus of the Federal University of Minas Gerais (UFMG): School of Engineering (ENG), Extra High Voltage Laboratory (LEAT), Hydraulic Research Center (CPH), and Teaching Activities Center 2 (CAD2), with variations in tilt and azimuth. The system enables continuous irradiance measurements on the module plane and was validated by comparison with meteorological station data from the National Institute of Meteorology (INMET), as well as laboratory testing and statistical analysis using metrics such as RMSE, MAE, BEM, MAPE, nRMSE, and R². Residual analysis at 1-minute intervals was performed to assess the statistical-metrological quality of the estimates. Residuals were defined as r_P (t)=〖Irradiância〗_Protótipo (t)-〖Irradiância〗_Piranômetro (t) e r_c (t)=〖Irradiância〗_Corrigida (t)-〖Irradiância〗_Piranômetro (t), calculated after synchronization and temporal aggregation. Diagnostics included histograms and QQ-plots (normality), ACF and Ljung–Box test (independence), boxplots by range and Bartlett/Breusch–Pagan tests (heteroscedasticity), as well as scatter plots with LOESS and segmented linear regression (residual × irradiance). Results under clear-sky conditions showed that the sensors achieved R² values above 0,95, with average RMSE ranging from 81,6 to 164,9 W/m² and MAPE below 50%, with the CAD2 system standing out. To ensure comparability among measurement points, data were corrected using irradiance decomposition and transposition models, standardizing the series to the reference plane of the INMET meteorological station (A521 – Pampulha). Decomposition separates global horizontal irradiance into direct and diffuse components, while transposition converts these components to the inclined plane of the modules, considering installation geometry. Under partly cloudy and overcast conditions, error levels increased — with nRMSE above 60% and MAPE over 70% in the LEAT system — attributed to the predominance of diffuse radiation and the sensors’ nonlinear response. QQ-plots indicated slight tail deviations; ACFs and Ljung–Box revealed short-term autocorrelation, reduced after correction; boxplots and Bartlett/Breusch–Pagan tests evidenced attenuated physical heteroscedasticity; and segmented regression indicated less pronounced regime shifts. The scientific contribution lies in demonstrating the feasibility of using photovoltaic modules as reliable irradiance sensors, provided that thermal correction protocols, calibration, and statistical validation are applied. The system showed relative uncertainties below 5% even under unstable atmospheric conditions, validating its usefulness for spatial irradiance analysis, support for photovoltaic generation modeling, and very short-term solar forecasting (nowcasting).