Bike sharing demand forecasting using Recurrent Neural Networks

García Martínez, Máximo (2021). Bike sharing demand forecasting using Recurrent Neural Networks. Proyecto Fin de Carrera / Trabajo Fin de Grado, E.T.S. de Ingenieros Informáticos (UPM), Madrid, España.

Description

Title: Bike sharing demand forecasting using Recurrent Neural Networks
Author/s:
  • García Martínez, Máximo
Contributor/s:
  • García Dopico, Antonio
Item Type: Final Project
Degree: Grado en Ingeniería Informática
Date: January 2021
Subjects:
Freetext Keywords: Redes neuronales, Redes neuronales recurrentes, Predicción, Series de tiempo, Alquiler de bicicletas, Neural network, Recurrent neural network, Forecasting, Time series, Bike sharing
Faculty: E.T.S. de Ingenieros Informáticos (UPM)
Department: Arquitectura y Tecnología de Sistemas Informáticos
Creative Commons Licenses: Recognition - No derivative works - Non commercial

Full text

[img]
Preview
PDF - Requires a PDF viewer, such as GSview, Xpdf or Adobe Acrobat Reader
Download (6MB) | Preview

Abstract

Las redes neuronales recurrentes (RNN en inglés) son un tipo de redes neuronales que utilizan conexiones de retroalimentación. Existen diferentes aplicaciones para este tipo de redes neuronales como por ejemplo el estudio de la demanda de un activo durante un tiempo. Este proyecto se llevó a cabo para estudiarlas teóricamente y desarrollar modelos para predecir la demanda de bicicletas en la ciudad de Chicago basados en el enfoque de las RNN y comparar distintas arquitecturas. Los modelos empleados han sido redes neuronales Feedforward, Simple Recurrent Neural Network, LSTM y arquitecturas auto-regresivas. Los modelos predicen en base a una ventana deslizante de intervalos de 1 hora. La salida de los modelos es un vector con valores con la predicción de cada estación de la ciudad para un determinado intervalo. La fecha y la hora es la única entrada que la red neuronal y con ella se han obtenido resultados sobresalientes. Los resultados obtenidos con las redes LSTM y auto-regresivas han obtenido resultados que compiten con proyectos de última generación, a pesar de que se han encontrado varios puntos de fallo y posibles mejoras.---ABSTRACT---Recurrent neural networks (RNN) are a type of neural network that uses feedback connections. There are different applications for this type of neural network, such as studying the demand for an asset over time. This project was carried out to study them theoretically and develop models to predict the demand for bicycles in the city of Chicago based on the RNN approach and to compare different architectures. The models used were Feedforward Neural Networks, Simple Recurrent Neural Network, LSTM and auto-regressive architectures. The models predict on the basis of a sliding window of 1 hour intervals. The output of the models is a vector with values with the prediction of each station in the city for a given interval. The date and time is the only input that the neural network and with it outstanding results have been calculated. The results obtained with the LSTM and autoregressive networks have obtained results that compete with state-of-the-art projects, despite the fact that several points of failure and possible improvements have been found.

More information

Item ID: 66259
DC Identifier: https://oa.upm.es/66259/
OAI Identifier: oai:oa.upm.es:66259
Deposited by: Biblioteca Facultad de Informatica
Deposited on: 04 Mar 2021 11:20
Last Modified: 04 Mar 2021 11:20
  • Logo InvestigaM (UPM)
  • Logo GEOUP4
  • Logo Open Access
  • Open Access
  • Logo Sherpa/Romeo
    Check whether the anglo-saxon journal in which you have published an article allows you to also publish it under open access.
  • Logo Dulcinea
    Check whether the spanish journal in which you have published an article allows you to also publish it under open access.
  • Logo de Recolecta
  • Logo del Observatorio I+D+i UPM
  • Logo de OpenCourseWare UPM