Técnicas de machine learning para el tratamiento de series temporales de big data en el ámbito energético

Abstract

Currently, one of the greatest challenges in the energy sector is the development of production and distribution systems that enable the use of clean, efficient, and sustainable energy. Advances in sensors and storage systems have provided a wealth of data that facilitates modeling energy consumption. Machine Learning models, particularly Artificial Neural Networks, have become the primary tool for accurately modeling energy consumption due to their high precision. However, training and optimizing the hyperparameters of these models can be computationally expensive. This high complexity can result in significant economic and environmental costs when deploying new models or retraining them, situations that frequently arise due to the dynamic nature of energy consumption. To address these challenges, this thesis focuses on studying dimensionality reduction techniques and parallel implementations using GPUs. By doing so, we aim to explore various methods that allow us to efficiently train and optimize Machine Learning models, advancing toward methodologies that enable a cleaner, more efficient, and sustainable energy future.