Deploying Arima and Artificial Neural Networks Models to Predict Energy Consumption in Taiwan

Abstract

With the aim of predicting Taiwan's energy consumption for the short term (1 year), the medium term (3 years), the medium-long term (5 years), and the long term (10 years), this study applies autoregressive integrated moving average (ARIMA) models and artificial neural networks (ANNs) models and the mean absolute percentage error (MAPE) approach is employed to measure prediction accuracy. Based on data extracted from over the period 1965-2010, the results indicate that the single variable ARIMA models illustrate superior performance than that of ANNs1. As to multivariable models, ANNs8 model including variables of energy consumption and exports show the most accurate prediction in short term and medium-long term, while ANNs6 model comprising energy consumption, GDP, and exports has the highest accuracy for medium term prediction. Meanwhile, ANNs5 model consisting of energy consumption and population shows the best accuracy for the long term prediction. Overall, it may conclude that exports and population are two essential variables to predict Taiwan's energy consumption for the short, medium, medium-long, and long term periods. The results support the assumption that parsimonious set of variables incorporated in research models may not sacrifice prediction accuracy. This concludes the contributions of this study.