Seasonal Autoregressive Integrated Moving Average Ensemble via Artificial Neural Networks for Electrical Load Forecasting Mega Silfiani(a,b,*), Surya Puspita Sari (a)
a) Department of Statistics, Institut Teknologi Kalimantan, Jalan Soekarno Hatta KM.15, Balikpapan, Indonesia
b)2Department of Economics, Gdansk University
Armii Krajowej Street, 114-116, Sopot, Poland
*) megasilfiani[at]lecturer.itk.ac.id
Abstract
The purpose of this research is to explore the performance of forecasting electrical load utilizing artificial neural network for combining the seasonal autoregressive integrated moving average (SARIMA) ensemble. A SARIMA ensemble member is created by modelling different autoregressive orders or moving averages. Meanwhile, the SARIMA ensemble is combined using artificial neural network. The datasets contain monthly electrical load for four categories, i.e., households, businesses, industries, and the public, from January 2016 to December 2020. The results show that in both categories and forecast horizons, SARIMA ensemble-based artificial neural networks outperformed all models. Further research should look into different techniques for creating and combining ensemble members.
