Modelado de series temporales multivariantes y fusión de datos de alto nivel: aplicación a la mejora de la eficiencia energética

• Autores: Ramón Rueda Delgado
• Directores de la Tesis: María del Carmen Pegalajar Jiménez (dir. tes.) , Manuel Pegalajar Cuéllar (dir. tes.)
• Lectura: En la Universidad de Granada ( España ) en 2020
• Idioma: español
• ISBN: 978-84-1306-730-8
• Tribunal Calificador de la Tesis: María Amparo Vila Miranda (presid.) , Daniel Sánchez Fernández (secret.) , Carlos D. Barranco (voc.) , Luis Jiménez Linares (voc.) , Oscar Germán Duarte Velasco (voc.)
• Enlaces
  • Tesis en acceso abierto en: DIGIBUG
• Resumen

  • Achieving an efficient and sustainable energy consumption in the building sector has become one of the main challenges to be solved in this decade. In the transition towards a complete decarbonization in the use of energy, energy efficiency is positioned as a central tool to identify and avoid unnecessary consumption. Consequently, it is expected to reduce the carbon footprint as well as minimizing the risks of climate change. Thanks to sensor technology advances and the development of Artificial Intelligence techniques, the research community has focused its efforts on the development of intelligent algorithms to automatically extract useful knowledge from data related to energy consumption, enabling the identification of the most relevant factors that help reduce energy consumption.

    This thesis is part of the Horizon Europe program, and focuses on the development of Artificial Intelligence techniques to build a tool for modelling and forecasting energy consumption in buildings. More specifically, we attempt to develop an interpretable technique that helps experts in energy management to make decisions, being useful to reduce energy consumption in distributed buildings, focusing in the particular case of the facilities of the University of Granada. To do so, we used Symbolic Regression as a tool to analyse energy consumption time series with the aim of discovering hidden patterns in it consumption and explain it in terms of an algebraic expression.

    The main drawback of Symbolic Regression is its high search space and then, it is necessary to use optimization algorithms to solve it. Consequently, in this thesis we explored different Soft Computing techniques, as well as linear and non-linear data structures to encode algebraic expressions. In particular, we studied the potential of diverse optimization algorithms such as Genetic Algorithms, Ant Colony Optimization or Multi-Objective Genetic Algorithms, among others, to explore the Symbolic Regression search space. Finally, we conclude that the linear structure of Straight Line Programs together the developed algorithms are able to find an interpretable algebraic expression that allows modelling and predicting the energy consumption of the University of Granada’s buildings with high accuracy, becoming in a useful tool that allows to reduce the energy consumption