Predicting Propane Demand Generation with Autoregressive Artificial Neural Networks

Abstract

Propane is a major crude oil byproduct generated by oil refineries. Its applications range from home heating to varied industrial and commercial purposes. Due to the nonlinear nature of the demand generation process, predicting this demand with traditional econometric approaches leads to inaccurate results. In this paper, we thus propose an alternative Autoregressive Neural Network (ARNN) based approach. We also employed the Autoregressive Integrated Moving Average (ARIMA) model to benchmark the performance of ARNN. The results show a 55% reduction in Mean Squared Error when ARNN is used over ARIMA. This improvement bears significant consequences for planning and decision-making by refineries.