Forecasting future energy production using hybrid artificial neural network and arima model

Abstract

The objective of this research is to obtain an accurate forecasting model for the amount of electricity (in kWh) that is generated from different primary energy sources in the U.S. In this research, Artificial Neural Network (ANN) and hybrid ARIMA and ANN algorithms were developed that can be used for forecasting the amount of energy production in the short, as well as, in the long run. Based on the inferences made from the available data provided by Energy Information Administration from January 2004 to December 2014, two different forecasting models for each primary energy source were constructed. These two models were validated with available data from January 2015 to November 2017, and their performance, as measured by forecasting errors computed, were compared. The results show that ANN algorithm is good for fossil fuels sources such as coal, petroleum, and natural gas. However, ARIMA - ANN hybrid works more accurately for renewable energy sources such as geothermal, hydroelectric, solar, and wind. Finally, the best predictor was selected for each primary energy source which provides valuable information regarding the future electricity generation, and future dominant energy source to generate electricity. This information will hopefully influence energy sector forecasting models and help the government to develop future regulations to shift toward dominant energy sources of the future.