Natural gas consumption forecast with MARS and CMARS models for residential users

Özmen A., Yilmaz Y., Weber G.

ENERGY ECONOMICS, vol.70, pp.357-381, 2018 (SSCI) identifier identifier

  • Publication Type: Article / Article
  • Volume: 70
  • Publication Date: 2018
  • Doi Number: 10.1016/j.eneco.2018.01.022
  • Journal Name: ENERGY ECONOMICS
  • Journal Indexes: Social Sciences Citation Index (SSCI), Scopus
  • Page Numbers: pp.357-381
  • Keywords: Natural gas consumption, Multivariate Adaptive Regression Splines, Conic Multivariate Adaptive, Regression Splines, Conic Quadratic Programming, Multiple Linear Regression, Neural Network, One-day ahead forecasting, NEURAL-NETWORKS, REGRESSION, PREDICTION, DEMAND, OPTIMIZATION, FINANCE, SPLINE
  • TED University Affiliated: No


Prediction natural gas consumption is indispensable for efficient system operation and required for planning decisions at natural gas Local Distribution Companies (LDCs). Residential users are major consumers that usually demand a significant amount of total gas supplied in distribution systems, especially, in the winter season. Natural gas is primarily used for space heating, and for cooking of food by residential users therefore, they should naturally be non-interruptible. Due to the fact that distribution systems have a limited capacity for the gas supply, proper planning and forecasting in high seasons and during the whole year have become critical and essential. This study is conducted for the responsibility area of Baskentgaz which is the LDC of Ankara. Predictive models MARS (Multivariate Adaptive Regression Splines) and CMARS (Conic Multivariate Adaptive Regression Splines) are formed for one-day ahead consumption of residential users. The models not only permit to compare both approaches, but they also analyze the effect of actual daily minimum and maximum temperatures versus the Heating Degree Day (HDD) equivalent of their average. Using the obtained one-day ahead models with daily data during 2009-2012, the daily consumption of each day in 2013 has been predicted and the results are compared with the existing methods Neural Network (NN) and Linear Regression (LR). The outcomes of this study present MARS and CMARS methods for the natural gas industry as two new competitive approaches. (C) 2018 Elsevier B.V. All rights reserved.