Next-Day Electricity Demand Forecasting Using Regression∗


ÇETİNKAYA M. , ACARMAN T.

2021 International Conference on Artificial Intelligence and Smart Systems, ICAIS 2021, Coimbatore, India, 25 - 27 March 2021, pp.1549-1554 identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/icais50930.2021.9395926
  • City: Coimbatore
  • Country: India
  • Page Numbers: pp.1549-1554
  • Keywords: Electricity Consumption, Hourly Forecasting, Regression Modeling, Seasonality

Abstract

© 2021 IEEE.In this paper, electricity demand is predicted on an hourly basis in Turkey for the next day. Two naiïve forecast approaches used also by grid operators are compared with respect to advanced regression techniques. Considering these modeling approaches, daily and hourly seasonality is highly considered. The naiïve approaches are based on using directly previous electricity consumption values to forecast demand whilst the regression approaches are tuned by Ridge and SARIMAX models. Proposed regression approaches are more successful with respect to naiïve approaches according to yearly test results for electricity demand forecasting. As an additional analysis, the best naiïve approach and the best regression approach is selected based on yearly forecasting success performance and they are tested and evaluated by using marketclearing price information. Hence, the effectiveness of the proposed regression model for portfolio optimization of day ahead market participants is also demonstrated showing the decrease in portfolio deviation compared to the naiïve approach. Because of the hourly changing consumption behavior, 24 different regression models are generated for each hour of a day and the best forecasting performance is sought. The most successful regression method is Ridge regression such as a penalized regression methodology.