Predicting Electricity Demand in Low-Voltage Distribution Networks Using an Optimized Support Vector Regression

Accurate electricity demand forecasting is crucial for maintaining grid stability and optimizing energy resources. Traditional methods often fail to capture power load demand’s complex and variable nature. Hence, artificial intelligence techniques, such as Support Vector Regression, have emerged as promising tools for improving forecasting accuracy. Even though SVR has been used for electricity demand forecasting, most of the research has concentrated on input variables such as temperature and weather. This study aims to explore the potential of SVR in predicting electricity demand based on other characteristics such as land size, floor area ratio, build-up area, load density, diversity factor, and group diversity factor. An SVR model was built with data collected from industrial consumers in Malaysia Vision Valley. Hyperparameter tuning, including the selection of the kernel function and regularization parameters, was employed to enhance the model’s accuracy. The SVR model accurately predicted electricity demand, achieving high R² values, low MAE and MAPE across different training and testing scenarios. The best performance was recorded at a 70/30 training/testing split, yielding a mean absolute error (MAE) of 0.0059, a mean absolute percentage error (MAPE) of 0.82%, and an R-squared value of 99.97%. These findings highlight the effectiveness of SVR in capturing complex relationships between input variables and electricity demand, which can aid utilities in effective planning and resource allocation, ultimately enhancing grid reliability and efficiency, reducing operational costs, and making informed decisions regarding energy resources.