Electricity Demand Forecasting using A Supervised Machine Learning Random Forest Algorithm

Abstract

The design and development of national power grids are critical to delivering a reliable electricity supply in Malaysia, where accurate load forecasting in low-voltage distribution networks is crucial in ensuring grid efficiency and stability. Traditional forecasting methods often lack of precision and adaptability required for modern energy management systems, particularly in dynamic environments such as the oil and gas industry. This study leverages supervised machine learning, specifically Random Forest, to enhance load-prediction accuracy in low-voltage networks, with a focus on Gas and Condensate Receiving facilities in the oil and gas sector in Kerteh, Terengganu, Malaysia. Based on analyses of variables such as equipment rating (kW), coincident factor, equipment efficiency (%), and equipment load duty, a robust forecasting model that demonstrates significant improvements over conventional approaches is developed. The result is reported as a best value of 80:20% training and testing split (scenario 4) after optimization with MAE of 1.01, MAPE of 0.49%, RMSE of 1.43, and R² of 99.07%. This show that the proposed method not only achieves higher forecasting accuracy but will also improves operational efficiency, reduces energy waste and enhances grid reliability. These findings highlight the transformative potential of machine learning for data-driven in low voltage distribution network, facilitating the integration of renewable energy sources and supporting Malaysia’s grid modernization efforts.