Estimating Solar Power Generation with RF, GB, and SVR Algorithms Based on Meteorological Data and Orientation Angles: Adıyaman Case Study

This study attempts to develop precise machine learning algorithms for estimating solar power generation using meteorological data, including air pollution (PM2.5, PM10) for different orientation angles in the dual-axis solar panel tracking system. Our study focused on maximizing energy generation for multivariable input factors in Adıyaman, rather than optimizing environmental parameters for fixed panels as previously studied. Three machine learning algorithms—Random Forest (RF), Gradient Boosting (GB), and Support Vector Regression (SVR)—were studied. The experimental setup was designed to measure power, temperature, humidity, wind speed/direction, pressure, air pollution data, and to adjust orientation angles (tilt and azimuth angles). Algorithm performances were assessed using R², MSE, and RMSE metrics. RF yielded the best results (R²: 0.83, MSE: 5.32, RMSE: 2.26). Including cumulative air pollution data improved prediction accuracy, implying particulate matter indirectly affects solar radiation and energy output.