Long-term energy demand in Malaysia as a function of energy supply: A comparative analysis of Non-Linear Autoregressive Exogenous Neural Networks and Multiple Non-Linear Regression Models

Abstract

Energy modeling and forecasting are key to providing insightful energy-related policy decisions that could propel positive economic growth and sustainable development. In this study, the Non-Linear Autoregressive Exogenous Neural Networks (NARX) and Multiple Non-Linear Regression (MNLR) models were employed for modeling the final energy demand per capita in Malaysia. The non-linear relationship between the time-series data such as GDP per capita, population, crude oil supply, natural gas supply, coal and coke supply, hydropower, electricity demand per capita, and the final energy demand per capita were modeled using five NARX and MNLR models refer to as NARX-1, NARX-3, NARX-5, MNLR-1, and MNLR-2. The effect of varying the network time delay and the hidden neurons on the NARX model performance were investigated. The best model configuration was obtained using 1 network delay and 17 hidden neurons. Both the NARX-1 and MNLR-1 models accurately learn the time series data for the prediction of the final energy consumption per capita. With R2 of 0.999 and 0.996 obtained for the NARX-1 and MLNR-1, respectively, the predicted final energy demand per capita was in close agreement with the actual values. Using the NARX-1 and MNLR-1 models, the final energy demand per capita was projected to be 2.3 toe/person and 2.1toe/person, respectively by 2050. The level of importance analysis of the NARX-1 models and the parameter estimates of the MNLR-1 model revealed that an increase in population and natural gas supply had a significant influence on the predicted final energy demand per capita.

Keywords

• Non-linear autoregressive exogenous
• Artificial neural networks
• Long-term energy demand
• Multiple non-linear regression
• Predictive energy modeling