Date of Award

Summer 2002

Document Type

Thesis - Restricted

Degree Name

Master of Science (MS)

Department

Electrical and Computer Engineering

First Advisor

Brown, Ronald

Second Advisor

Povinelli, Richard

Third Advisor

Richie, James

Abstract

The focus of this work is on the development of hourly natural gas forecasting models for the purpose of providing hourly gas demand estimation to the utilities. Some utilities need to know hourly gas demand to economically and efficiently operate their gas distribution networks. Therefore, prediction of accurate natural gas demand can help significant cost savings for these utilities. The contributions of this research are the identification of input factors for the natural gas forecasting models and the development of these forecasting models using two techniques, linear regression and artificial neural network. First, input factors are identified and selected for the 106 (hour O to hour 105) natural gas forecasting models. based on identification methods such as genetic algorithm. Each of these 106 models has a unique set of input factors. For each set of the model input factors, a model for forecasting gas demand is then constructed using least squares and an artificial neural network. Both methods are demonstrated to efficiently identify the linear and nonlinear gas consumption problem.

Share

COinS

Restricted Access Item

Having trouble?