Forecasting Electric Substation’s Load Curve Using GMDH-type Neural Network

Abstract

Electric load forecasting plays an important role in the planning and operation of the power system for high productivity in any province in the region. A short-term electrical energy forecast for Sibalom, Antique, Philippines Substation was carried out using a GMDH-type neural network as its core algorithm, and the result was compared to that of regression analysis. The GMDH-type neural network was used to train and test weekly energy consumed in the substation from July 2020 to August 2020. The neural network was trained using a quadratic neural function. Root mean square error (RMSE) and mean absolute percentage error (MAPE) were used as performance indices to test the accuracy of the forecast. The neural network model gave a root mean square error (RMSE) of 3.57177%, a mean absolute percentage error (MAPE) of 2.35499%, and a correlation (R) value of 0.983591 while the regression analysis method gave a standard error of 99.39524 and a correlation (R) value of 0.907673. Results obtained show the efficacy of the GMDH-type neural network model in forecasting over the regression analysis method.