New predictive models based on ensemble and deep learning applied to the electric market

Abstract

Accurately predicting future electricity demand ensures a reliable and sustainable energy supply. Electricity planners and policymakers utilize various tools and techniques to forecast the future of electricity, allowing them to assess needs and plan the development of new generation and transmission infrastructure. As electrical energy cannot be stored in large quantities, it is necessary to predict the amount needed. This thesis presents a novel prediction method that integrates multiple deep-learning techniques. Moreover, it has been presented as a compendium of publications comprising three main scientific contributions to international conferences and journals with high impact factors in the Journal Citation Reports.

The first approach is a stacking ensemble that combines three successful models in the field: Long Short-Term Memory (LSTM), Gated Recurrent Unit neural networks (GRU), and Temporal Convolutional Networks (TCN). The results of this study have been published in the Energy Journal. The second proposed approach is a bagging ensemble, which involves dividing the data into coherent subsets and implementing a deep-learning model for each subset. Two scientific articles were published as part of this research. A paper focusing on outlier handling was recently published in the Expert Systems With Applications journal. Another paper was presented at the International Conference on Soft Computing Models in Industrial and Environmental Applications (2021), addressing medium and long-term energy forecasting with hybrid models.

The results of all the proposed methods significantly improve the accuracy of electricity consumption prediction for the economic sector in Algeria. At the aggregate-level granularity, our approaches achieved a mean absolute percentage error (MAPE) of 2.04\%. Regarding individual-level granularity, the provided prediction by the developed models was considered very good by over 87\% of the customers. The contributions are significant on various levels, as they address an important issue in Algeria related to forecasting electricity consumption. Moreover, this work extends to economic, technical, and modeling aspects at various levels, making it a valuable contribution to energy management.