Random Forest Regression with Lag Features for Power Output Prediction of Automatic Street Lighting System Based on Fresnel Solar Concentrator

Muhammad Vivaldi Vikzi Kohar (Politeknik Negeri Sriwijaya)
Evelina (Politeknik Negeri Sriwijaya)
Muhammad Nawawi (Politeknik Negeri Sriwijaya)

Jurnal Informatika dan Teknik Elektro Terapan
Vol. 14 No. 3 (2026)

The variability of DC Street Lighting (PJU) output power based on Fresnel Solar Concentrator due to weather fluctuations presents a challenge in solar energy management. This study develops a prediction model using Random Forest Regression (RFR) with IoT sensor time-series lag features and BMKG data, utilizing 44 predictor variables from 405 records collected over 27 days. Hyperparameter optimization via 5-fold Grid Search Cross-Validation yielded n_estimators=200 and max_depth=15. Evaluation results achieved R²=0.9850 and MAPE=8.29%, satisfying all success criteria. Panel power (P_Panel) was the most dominant predictor with importance score 0.1197. The model successfully predicted output power for 7 days ahead with an average energy estimate of 250 Wh per day, proving that lag features integration into RFR is effective for proactive PJU energy management.