Date of Award

December 2014

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Engineering

First Advisor

John R. Reisel

Committee Members

Junhong Chen, Chris Yuan, Hamid Seifoddini, Adel Nasiri

Keywords

Artificial Neural Networks, Energy Forecast, Energy Modeling, Linear Regression, Renewable Energy, United States

Abstract

The United States is a country which consumes a vast amount of energy. In order to keep the development of the United States sustainable (diverse and productive over the time) energy planning should be carried out comprehensively and precisely. This dissertation presents a specific mathematical modeling approach towards energy demand modeling of the United States and forecast future energy demand. To generate more detailed and accurate results, this dissertation investigates the energy demand of each sector separately using the analysis of trend for unique set of independent parameters which affect the energy demand in that sector.

In solving a forecast problem with artificial neural networks, the most important part is to choose the independent variables that provide the most precise estimate of the dependent variable. While including too many variables makes the model complicated and increases the calculation time significantly, excluding important independent variables makes integrity of the model questionable and reduces its predictive ability. In this study, correlation coefficient analysis is applied to initially select the independent variables.

In terms of forecasting the energy demand in the residential sector, the MLR and ANN models show two different trends while their performances are at a similar level of accuracy during the test period.

ANN model anticipates a small increase in the energy demand of the transportation sector. Although a small increase has been estimated by the ANN, the United States should keep trying to reduce energy consumption in order to reduce CO2 gas and meet its national and international commitments.

ANN is also applied to forecast the industrial energy demand and perform future projections for the period 2013-2030. Based on model trained with historical data of period 1980-2012, the price of energy significantly affects the amount of energy used in the industrial sector. Hence, ascending price scenario and descending price scenario will result in 7% and 25% increase in the energy demand of this sector, respectively.

Based on model trained with historical data of period 1987-2012, the U.S. trade significantly affects the amount of energy used in the commercial sector. Hence, ascending trade scenario and descending trade scenario will result in 5% and 2% increase in the energy demand of this sector, respectively.

Share

COinS