Long-term load forecasting: models based on MARS, ANN and LR methods

Nalcaci G., Özmen A., Weber G. W.

CENTRAL EUROPEAN JOURNAL OF OPERATIONS RESEARCH, vol.27, no.4, pp.1033-1049, 2019 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 27 Issue: 4
  • Publication Date: 2019
  • Doi Number: 10.1007/s10100-018-0531-1
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Page Numbers: pp.1033-1049
  • Keywords: Electricity demand, Time series, MARS, ANN, Linear regression, Load forecasting, Accuracy, Stability, REGRESSION, ROBUSTIFICATION, CMARS
  • TED University Affiliated: No


Electric energy plays an irreplaceable role in nearly every person's life on earth; it has become an important subject in operational research. Day by day, electrical load demand grows rapidly with increasing population and developing technology such as smart grids, electric cars, and renewable energy production. Governments in deregulated economies make investments and operating plans of electric utilities according to mid- and long-term load forecasting results. For governments, load forecasting is a vitally important process including sales, marketing, planning, and manufacturing divisions of every industry. In this paper, we suggest three models based on multivariate adaptive regression splines (MARS), artificial neural network (ANN) and linear regression (LR) methods to model electrical load overall in the Turkish electricity distribution network, and this not only by long-term but also mid- and short-term load forecasting. These models predict the relationship between load demand and several environmental variables: wind, humidity, load-of-day type of the year (holiday, summer, week day, etc.) and temperature data. By comparison of these models, we show that MARS model gives more accurate and stable results than ANN and LR models.