Comparing Various Combined Techniques at Seasonal Autoregressive Integrated Moving Average (SARIMA) for Electrical Load Forecasting

The objective of this study is to investigate the accuracy of forecasting electrical consumption using various combined techniques at the seasonal autoregressive integrated moving average (SARIMA) ensemble. A SARIMA ensemble is created by modeling various SARIMA models and combining techniques such as arithmetic averaging, Bates-Granger weight, Akaike weight, and Principal Component Regression (PCR) weight. Results indicate that SARIMA’s ensemble-based PCR outperformed all the remaining models in both categories and forecast horizons. In general, ensemble models produced by Bates–Granger perform better than ensemble models produced by Akaike weight and averaging. Meanwhile, an ensemble model based on Akaike weight and averaging have equal performance. In addition, the public electrical load forecast has the best performance in forecast horizon and various models. Generally, household and electrical industry loads have equal performance three months ahead and twelve months ahead of the forecast horizon. A further investigation ought to explore the various construction and combination techniques implemented by ensemble members.