Studi Prediksi Beban Jangka Pendek Menggunakan Neural Network Berbasis Data Time Series pada PT. Siantar Tara Sejati site Tarakan

Abstract

Electrical loads generally exhibit dynamic and fluctuating patterns that follow daily consumption characteristics. This study provides a new contribution to short-term electrical load forecasting studies based on time series data at one of the independent power plants in Tarakan. The study aims to compare two neural network (NN) approaches, namely NN without time series data processing and NN based on time series data (NN-TS). The data used consisted of historical daily electrical load data obtained from the power plant’s data sourcee. The forecasting process was conducted through a stationarity stage in order to obtain the best model. Through the best NN-TS model, different NN architectures were generated. The NN model produced a 24-10-1 layer architecture, whereas the NN-TS model produced a 24-2-1 layer architecture. The result of short-term electrical load forecasting using the NN method achieved a Mean Absolute Percentage Error (MAPE) value of 6.0846%, while the NN-TS method achieved a MAPE value of 4.33%.