Please use this identifier to cite or link to this item: http://hdl.handle.net/10609/149579
Title: Análisis del uso de las principales fuentes de energía en España: análisis causal y predicción mediante técnicas de Business Intelligence
Author: Juez Martel, Pedro
Abstract: The current public energy policies, as well as the Agenda 2030, highlight the importance of promoting the use of clean energy at home. However, the choice of energy source for households depends on variables that are often unknown. It is important for public policies and energy companies to know this variables and predict the consumer behaviour to take correct decisions.Objectives: The objective of the work is twofold:  Apply Artificial Intelligence and multivariate analysis techniques to model and predict the energy source a household will use.  Apply these same techniques to identify the explanatory variables of the use of each energy source This can be use later for the public policies and the business strategy of energy companies. Methodology: To study the causality and the prediction of the different energy sources we will use: 1. Multivariate analysis techniques such as contingency tables or the study of standardized residuals to detect relationships between variables. 2. Use of classification models like the logit or probit model. 3 Use of Artificial Intelligence techniques belonging to the field of machine learning, such as decision trees and random forest and Artificial Intelligence techniques belonging to the field of deep learning, such as neural networks. Conclusions:  At an explanatory level, we find that the most polluting energy sources are more strongly correlated with rural households, cold temperatures in the Autonomous Community, and elementary education.  Gas is more closely associated with Autonomous Communities with cold temperatures, new and urban housing, higher education, and women.  Among the three models, the multilayer perceptron achieves the highest accuracy, reaching up to 70%. Both the random forest and the perceptron models achieve higher accuracy than the multinomial probit model.  Thanks to these AI techniques, we have achieved high prediction accuracy and also conducted a causal analysis of the main energy sources, performing an analysis for each of them. Of all, electricity achieves the lowest level of precision, while gas is predicted more accurately and its causal relationships are easier to explain.  Electricity is the least predictably modeled energy source. It is particularly linked to Autonomous Communities with high temperatures and single-member households. It is the most challenging energy source to predict.
Keywords: deep learning
machine learning
Document type: info:eu-repo/semantics/bachelorThesis
Issue Date: 13-Jan-2024
Publication license: http://creativecommons.org/licenses/by-nc-nd/3.0/es/  
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
pjuezTFG0124memoria.pdfMemoria del TFG906,44 kBAdobe PDFThumbnail
View/Open
Share:
Export:
View statistics

This item is licensed under aCreative Commons License Creative Commons