ENERGY CONSUMPTION FORECASTING IN TAIWAN BASED ON ARIMA AND ARTIFICIAL NEURAL NETWORKS MODELS

Abstract

This study compares the prediction performance of Taiwan's energy consumption based on a autoregressive integrated moving average (ARIMA) model and artificial neural networks (ANNs) models for forecasting the short term (I year), the medium term (3 years), the medium-long term (5 years), and the long term (10 years) over the period 1965-2010. On the average of the four time periods, the results indicate that the single-variable ARIMA model shows superior performance than that of ANN1. As to multi-variable-models, the prediction accuracy of different models has advantages in the different time periods. ANN4 model including variables of energy consumption and export shows the most accurate prediction in short term and medium-long term, while ANN6 model including energy consumption, GDP and export has the highest accuracy for medium term prediction. Meanwhile, ANN3 model including energy consumption and population has the best accuracy for the long term prediction. Overall, on the average of the four different time periods of ARIMA model and ANNs models, ANN3 proves the most accurate prediction in comparison to the others. This concludes the contributions of this study.