The integration of DER such as PV and V2G systems, presents a novel and significant challenge to centralized power systems like Jamali grid. This study aims to model DER penetration in the Jamali grid, to evaluate and select the optimal machine learning algorithm for predicting its impact on peak demand and to analyze the effects of a 1-5% DER penetration level on the system.
Employing a quantitative approach, this research compares five machine learning models; XGBoost, LightGBM, CatBoost, AdaBoost and Gradient Boosting. The models were analyzed using a five-year historical dataset (2020 - 2024) that included Jamali power system data, socio-economic indicators, ambient and solar irradiance data and metrics related to DER penetration. The model with the highest performance, evaluated using MAE, MAPE, MSE, MdAPE, RMSE and R² metrics, was subsequently used to forecast peak demand from January 2025 to December 2030 and to simulate DER penetration scenarios.
The modeling analysis results indicate that Gradient Boosting emerged as the most accurate and stable model following a hyperparameter tuning process, achieving a R2 of 0,9961, a MAE of 3,65% and a MdAPE of 2,47%. Utilizing this Gradient Boosting forecasting model, the maximum feasible DER penetration within the power system for the 2025 to 2030 period is projected for the Jakarta-Banten and Bali subsystems, with penetration levels reaching 3,86% and 5,63% of peak demand, respectively.
These penetration levels served as the basis for a simulation of a contingency event involving the simultaneous disconnection of all DER units. The simulation revealed a significant frequency decline, particularly in the Jakarta-Banten system, where the frequency dropped to 48,33 Hz. This value is below the normal operational threshold stipulated by the Grid Code 2020, indicating a high degree of system vulnerability to disturbances.