Conventional ARX and Artificial Neural networks ARX models for prediction of oil consumption in Malaysia

Abstract

This study investigates prediction of oil consumption in Malaysia. Models of oil consumption are developed and validated with respect to training and validation dataset. Available data for Malaysia is annual data from 1982 to 2006 comprises Population, GDP per Capita, and Oil Consumption data. Prediction time target is year 2020 which is commonly used by several energy outlook reports. Two models are developed in this study, conventional Autoregressive Exogenous (ARX) model and Artificial Neural Network ARX (ANN ARX) model. The difference lies on how those models work to find unknown parameters based on training dataset. Conventional model uses Least Square method to calculate the unknown parameter where ANN ARX model uses weight updating strategy to find the unknown parameter. Performance of each model is measured through Root Mean Square Error (RMSE) value. It is shown that ANN ARX model can perform better than conventional ARX especially with small number of training dataset.