GMDH-type neural network-based monthly electricity demand forecasting of Turkey

Abstract

In this study, it was aimed to develop an accurate forecasting model for the monthly electricity demand of Turkey in the medium-term. For this purpose, the Group Method of Data Handling (GMDH)-type Neural Network (NN) approach was used to structure a nonlinear time-series based forecasting model. A large dataset containing monthly electricity demand was considered for the period of 2003-2018. The developed model was tested in the period of 2019/01-2019/11 in order to determine the generalization ability of the model. The test results showed that the developed model was very close to actual values. The obtained test performances were 2.10 % for mean absolute percentage error (MAPE), 2.36 % for root mean square percentage error (RMSPE) and 0.869 for coefficient of determination (R2). In addition, results of the proposed GMDH-type NN model were compared with the forecasting results of a literature study. The comparison revealed that GMDH-type NN was a better approach for forecasting the monthly electricity demand of Turkey. Finally, the developed model was utilized to forecast monthly electricity demand in the period of 2019/12-2020/12.