Weekday-Weekend, Day of Week, and Prior Day Effects in Forecasting Daily Natural Gas Demand from Monthly Data

Authors

Maral Fakoor, Marquette University
Ahmadreza Baghaie, Perdue University
George F. Corliss, Marquette UniversityFollow
Ronald H. Brown, Marquette UniversityFollow

Document Type

Conference Proceeding

Language

eng

Publication Date

2018

Publisher

International Institute of Forecasters

Source Publication

International Symposium on Forecasting

Abstract

Local natural gas distribution companies rely on accurate forecasts of daily demand/flow for buying and delivering gas to their customers. Such forecasts are done by devising computational methods that take into account weather data and historical daily flow in regions of interest. However, in some cases, historical measured daily data is not available. In this work, multiparameter linear regression models are built when only monthly/billing-cycle flow data is available for disaggregation and to forecast daily flow. Results show monthly consumption data can be used in conjunction with daily weather data to provide accurate estimates of daily demand. To improve models, adjustments such as Weekday vs. Weekend, Day of Week, and Prior Day weather are incorporated into the models. In comparison to the base linear regression models, these adjustments can decrease the forecast error by up to 20% using the best combination of mentioned adjustments.

Comments

Published version. International Symposium on Forecasting, (2018). Publisher link. © 2018 The Authors. Used with permission.

Recommended Citation

Fakoor, Maral; Baghaie, Ahmadreza; Corliss, George F.; and Brown, Ronald H., "Weekday-Weekend, Day of Week, and Prior Day Effects in Forecasting Daily Natural Gas Demand from Monthly Data" (2018). Electrical and Computer Engineering Faculty Research and Publications. 602.
https://epublications.marquette.edu/electric_fac/602