Short-Term Load Forecasting using Artificial Neural Network in Indonesia

Abstract

Short-term Load Forecast (STLF) is a load forecasting that is very important to study because it determines the operating pattern of the electrical system. Forecasting errors, both positive and negative, result in considerable losses because operating costs increase and ultimately lead to waste. STLF research in Indonesia, especially the State Electricity Company (PLN Sulselrabar), has yet to be widely used. Methods mainly used are manual and conventional methods because they are considered adequate. In addition, Indonesia's geographical conditions are extensive and diverse, and the electricity system is complex. As a result, the factors affecting each country's electricity demand are different, so unique forecasting methods are needed. Artificial Neural Network (ANN) is one of the Artificial Intelligent (AI) methods widely used for STLF because it can model complex and non-linear relationships from networks. This paper aims to build an STLF forecasting model that is suitable for Indonesia's geographical conditions using several ANN models tested. Based on several ANN forecasting models, the test results obtained the best model is Model-6 with ANN architecture (9-20-1). This model has one hidden layer, 20 neurons in the hidden layer, a sigmoid logistic activation function (binary sigmoid), and a linear function. Forecasting performance values obtained mean squared error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) of 430.48 MW2, 15.07 MW, and 2.81%, respectively.