<![CDATA[This study developed and evaluated a machine learning–based Digital Twin proof of concept for a physical solar panel with a nominal capacity of 100 W. Environmental and electrical data were collected using sensors connected to an ESP32 and transmitted to a server for power prediction using Random Forest Regression. Exhaustive Scenario Search was used to compare panel angles of 25°, 35°, and 45°, while the dashboard presented measured power, predicted power, and angle recommendations. The analysis used 9,720 measurement records divided chronologically into training, validation, and testing datasets. Physical-panel validation was performed by comparing the predicted values with the measured power. For the 1,944 testing records, the model achieved an MAE of 2.769 W, an RMSE of 4.145 W, and an R² of 0.9446. The angle scenario evaluation produced 2,492.97 Wh of energy, which was 3.06% higher than the 2,418.91 Wh obtained with a fixed angle of 35°. These findings support the implementation of a Digital Twin proof of concept for panel-condition monitoring, power prediction, and angle recommendation under the tested conditions.]]>
Copyrights © 2026
