Gestión de la Eficiencia Energética Mediante Técnicas de Minería de Datos

Abstract

The energy efficiency arises as one of the areas of the greatest government interest in current times. On the other hand, the increase of computational capabilities for information processing and its storage, along with the wide availability of sensing nets have led to a massive increase in data production in the Energy field. Consequently, the challenge lies in treating and processing such information so as to obtain consumption profiles, in addition, to use that information to support the decision-making process. In this context, Artificial Intelligence emerges as an adequate tool for solving this problem. As a result, specific applications for efficient energy management are recently being exploited. Thus, this thesis comprises the creation of predictive models to estimate the energy consumption of buildings and the optimization of those models, together with the implementation of a software prototype for visual energy monitoring and knowledge representation. To achieve this aim, we propose to use recurrent neural networks capable of modelling data with the particularity that have the time dependencies; such as, non-linear autoregressive neural networks (NAR), its version with external inputs to enhance accuracy (NARX), the Elman recurrent neural networks and the well-known LSTM. The results proved that the combination of these models with genetic algorithms as optimization method provides an excellent improvement in term of accuracy to predict energy consumption in buildings.