| TAZ-TFG-2023-3703 |
Redes neuronales recurrentes para la predicción de series temporales en sistemas dinámicos complejos y sus aplicaciones
Fredes Cáceres, Arturo
Gutiérrez Rodrigo, Sergio (dir.) ; Arbiol Herrera, César (dir.)
Elipe Sánchez, Antonio (ponente)
Universidad de Zaragoza,
CIEN,
2023
Departamento de Física Aplicada, Área de Física Aplicada
Graduado en Matemáticas
Resumen: Artificial intelligence and machine learning techniques are currently a very hot topic, and recent
advances in the field have raised hopes, doubts and fear in society. These techniques are useful
for tackling several types of problems like classification, regression, generation of content or
forecasting. The focus of this work was set on forecasting time series, which plays an important
role in different areas such as business, weather or economics providing insights into future trends
or outcomes.
The objective was to a study how recurrent neural networks perform when forecasting future
steps of time series, by working with different sets of data of increasing complexity. A proof of
concept was made working with one-dimensional data of the sum of two waves and some noise.
Different types of neurons and architectures were used to compare performance in different
scenarios. Next, the study was continued using data from a 3D chaotic system (the Lorenz
Attractor), and finally it was attempted to apply these techniques to real business data.
advances in the field have raised hopes, doubts and fear in society. These techniques are useful
for tackling several types of problems like classification, regression, generation of content or
forecasting. The focus of this work was set on forecasting time series, which plays an important
role in different areas such as business, weather or economics providing insights into future trends
or outcomes.
The objective was to a study how recurrent neural networks perform when forecasting future
steps of time series, by working with different sets of data of increasing complexity. A proof of
concept was made working with one-dimensional data of the sum of two waves and some noise.
Different types of neurons and architectures were used to compare performance in different
scenarios. Next, the study was continued using data from a 3D chaotic system (the Lorenz
Attractor), and finally it was attempted to apply these techniques to real business data.
Tipo de Trabajo Académico: Trabajo Fin de Grado
Enlace permanente:
El registro pertenece a las siguientes colecciones:
Trabajos académicos > Trabajos Académicos por Centro > Facultad de Ciencias
Trabajos académicos > Trabajos fin de grado