<![CDATA[Accurate daily feeder load forecasting is essential for supporting operation and maintenance strategies in electrical distribution systems, where nonlinear and dynamic load variations may limit the performance of conventional forecasting methods. This study develops a machine learning-based framework for daily load forecasting of six 20 kV distribution feeders (PLR16–PLR21) supplied by Transformer 1 at Palur Substation, Indonesia. We used daily operational load data from January 2020 to April 2026, incorporating customer-growth information as an additional forecasting feature. We used Random Forest as the primary model and Holt–Winters Exponential Smoothing as a benchmark. We evaluated forecasting performance using MAE, RMSE, MAPE, and sMAPE. Random Forest outperformed Holt–Winters, achieving MAE of 292.45 A, RMSE of 417.85 A, and MAPE of 12.98%, compared with 720.30 A, 853.37 A, and 26.79%, respectively. The selected model was further applied to forecast feeder loading through 2028 and identify priority feeders for operation and maintenance planning. The proposed framework provides a practical, data-driven approach to feeder load forecasting and operational decision support in 20 kV distribution systems.]]>
Copyrights © 2026
