<![CDATA[Accurate solar irradiance prediction is fundamental for planning and operating solar photovoltaic (PV) power systems. This study compares the performance of four machine learning algorithms — Random Forest (RF), Support Vector Regression (SVR), XGBoost, and Artificial Neural Network (ANN) — in predicting daily Global Horizontal Irradiance (GHI) in Malang, East Java. The dataset was obtained from NASA POWER spanning 10 years (2014–2023), comprising 3,646 daily records with 11 input features including meteorological parameters, temporal features, and autoregressive features. Data splitting was performed chronologically (70% training, 15% validation, 15% testing). Results show that XGBoost achieved the best performance with R² = 0.6797, RMSE = 0.5212 kWh/m²/day, and MAPE = 8.35%. Seasonal analysis reveals all models perform better during the dry season (R² = 0.74; MAPE = 6.63%) compared to the wet season (R² = 0.54; MAPE = 11.06%). A 5 kWp PV system in Malang is estimated to produce 7,626 kWh/year.]]>
Copyrights © 2026
