Peramalan Beban Jangka Panjang pada Gardu Induk Bangil dengan Metode Generalized Regression Neural Network

Given the rising energy demands, the existing electrical infrastructure, notably distribution transformers 3 and 4 at the Bangil Substation, faces the risk of overload. Accurate load forecasting is imperative to inform timely interventions like transformer replacement. This study aims to forecast the load for Transformers 3 and 4 at the Bangil Substation using 2 difference methods, comparing Feed Forward Backpropagation Neural Network (FFBNN) and Generalized Regression Neural Network (GRNN). This research also evaluates potential transformer overloads based on forecasted peak loads. This research employed a STL Decomposition to decompose monthly peak load data in each transformer into trend, seasonal and residual components and developing forecasting model for each transformer trend component data. Simultaneously, separate forecasting models were developed for the Gross Regional Domestic Product (GRDP) and the Industrial Sector of GRDP. The forecasted trend components from the transformer data were combined with the GRDP and Industrial Sector of GRDP forecasts using an approximation model. This approach aimed to approximate the monthly peak load more accurately, incorporating both energy demand trends and economic indicators. The forecasting models' accuracy was gauged using Mean Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE). The analysis indicates that Transformer 3 is projected to reach overload by August 2038, with a forecasted peak load of 1407.7465 A. Conversely, Transformer 4 is expected to experience overload by February 2028, with a peak load of 1269.2173 A. FFBNN exhibited superior accuracy for Transformer 3, recording a MAPE of 10.522% and MAE of 74.204. In contrast, GRNN displayed better performance for Transformer 4, achieving a MAPE of 6.051% and MAE of 46.557. Timely interventions, such as transformer replacement, are essential to mitigate potential overloads. The research underscores the importance of employing tailored forecasting approaches, emphasizing the peak load transformer data with economic indicators for more precise load approximations