An economic approach of assessing the performance of ANN – based models in predicting energy comsumption: a study case on Romania

The increasing demand for energy is part of the challenges facing the transformation of the energy sector today. The transition to new ecologically sustainable energy sources is a priority of the European Union and therefore of Romania, a member state with diverse energy sources. A prediction of energy demand and possible peaks would be very useful in the future energy landscape, both for domestic and industrial consumers. In this paper, we compare the use of two artificial neural network architectures for building predictive models, namely the Long-Short Term Memory architecture and the Gated Recurrent Unit one. The analysis includes the comparison between the best performing models in terms of the optimization algorithm and the weight distribution method used. The purpose of this work is to assess their performance in predicting the national energy consumption of Romania by using publicly available data for training and testing the models.