Predicting Power Deviation in the Turkish Power Market Based on Adaptive Factor Impacts


Güven D., Ozozen A., Kayakutlu G., Kayalıca M. Ö.

8th IFIP WG 12.6 International Workshop on Artificial Intelligence for Knowledge Management, AI4KM 2021 held in conjunction with International Joint Conference on Artificial Intelligence, IJCAI 2020, Virtual, Online, 7 - 08 January 2021, vol.614, pp.213-234 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Volume: 614
  • Doi Number: 10.1007/978-3-030-80847-1_14
  • City: Virtual, Online
  • Page Numbers: pp.213-234
  • Keywords: Energy market balancing, Auxiliary power market modeling, Turkish power market, Adaptive prediction, Genetic algorithm and recursive neural networks, RECURRENT NEURAL-NETWORKS, BALANCING MARKET, ELECTRICITY, PRICE, MODEL
  • Istanbul Technical University Affiliated: Yes

Abstract

© 2021, IFIP International Federation for Information Processing.Energy market models are generally focused on energy balancing using the optimum energy mix. In countries where the energy markets are not fully liberalised, the State Regulators reflect any cost of being off-balance on the utility companies and this affects the consumers as well. The right short term prediction of the market trends is beneficial both to optimise the physical energy flow and commercial revenue balance for suppliers and utility companies. This study is aimed to predict the sign trends in the power market by selecting the influencing factors adaptive to the conditions of the day ahead, 10 h, 5 h, 2 h and 1 h before the electricity balance is active. There are numerous factors consisting of weather conditions, resource costs, operation costs, renewable energy conditions, regulations, etc. with a considerable impact on the predictions. The contribution of this paper is to choose the factors with the highest impacts using the Genetic Algorithm (GA) with Akaike Information Criteria (AIC), which are then used as input of a Recursive Neural Network (RNN) model for forecasting the deviation trends. The proposed hybrid method does not only reduce the prediction errors but also avoid dependency on expert knowledge. Hence this paper will allow both the market regulator and the suppliers to take precautions based on a confident prediction.