Publicação

Predicting Bike Stations Occupation using Time-Series Models: The Case Study of Lisbon

Ver documento

Detalhes bibliográficos
Resumo:Urban bike-sharing systems have become a vital component of sustainable mobility strategies in modern cities, helping reduce traffic congestion and environmental impact while improving transportation accessibility. In Lisbon, the GIRA bike-sharing network plays a significant role in meeting short-distance travel needs, yet ensuring consistent bike availability across its many stations remains a complex challenge due to spatio-temporal demand fluctuations influenced by external factors like weather and air quality. Although various forecasting methods have been proposed to tackle this issue, many rely on limited data types or struggle with inconsistent data quality, particularly due to missing values in environmental datasets. To better understand and address this problem, we aimed to train various robust forecasting models capable of predicting hourly bike availability at GIRA stations while assessing the impact of different imputation strategies on prediction performance. We employed deep learning architectures, including Long Short-Term Memory (LSTM) and Transformer models, and trained them using historical bike usage data enriched with meteorological and air quality variables. This thesis presents a comprehensive performance evaluation of these models under multiple imputation approaches, analyzing errors across time periods, stations, and input conditions. Our findings reveal that the Transformer architecture consistently achieved the best performance. Grouped statistical imputation produced more reliable forecasts on non-imputed data, although biases emerged in predicting imputed test rows. Furthermore, temporal patterns in error distributions highlight specific hours of the day with heightened predictive difficulty, typically during morning and evening peak times. In conclusion, deep learning models—especially Transformers—offer promising accuracy in forecasting bike availability in a real-world context, provided that imputation strategies are carefully considered. These results support the integration of predictive analytics into bike-sharing operations to enhance decision-making for redistribution, planning, and service quality optimization.
Autores principais:Moreira, Guilherme Valler
Assunto:Bike-sharing Time series forecasting Transformer Data imputation Urban mobility Demand prediction SDG 11 - Sustainable cities and communities SDG 13 - Climate action
Ano:2025
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso aberto
Instituição associada:Universidade Nova de Lisboa
Idioma:inglês
Origem:Repositório Institucional da UNL
_version_ 1865920611261349888
author Moreira, Guilherme Valler
author_facet Moreira, Guilherme Valler
Moreira, Guilherme Valler
author_role author
contributor_name_str_mv Jardim, João Bruno Morais de Sousa
Oliveira, Rita Madalena Cardoso da Silva
RUN
country_str PT
creators_json_str [{\"Person.name\":\"Moreira, Guilherme Valler\"}]
datacite.contributors.contributor.contributorName.fl_str_mv Jardim, João Bruno Morais de Sousa
Oliveira, Rita Madalena Cardoso da Silva
RUN
datacite.creators.creator.creatorName.fl_str_mv Moreira, Guilherme Valler
datacite.date.Accepted.fl_str_mv 2025-10-28T00:00:00Z
datacite.date.available.fl_str_mv 2025-11-10T12:23:35Z
datacite.date.embargoed.fl_str_mv 2025-11-10T12:23:35Z
datacite.rights.fl_str_mv http://purl.org/coar/access_right/c_abf2
datacite.subjects.subject.fl_str_mv Bike-sharing
Time series forecasting
Transformer
Data imputation
Urban mobility
Demand prediction
SDG 11 - Sustainable cities and communities
SDG 13 - Climate action
datacite.titles.title.fl_str_mv Predicting Bike Stations Occupation using Time-Series Models: The Case Study of Lisbon
dc.contributor.none.fl_str_mv Jardim, João Bruno Morais de Sousa
Oliveira, Rita Madalena Cardoso da Silva
RUN
dc.creator.none.fl_str_mv Moreira, Guilherme Valler
dc.date.Accepted.fl_str_mv 2025-10-28T00:00:00Z
dc.date.available.fl_str_mv 2025-11-10T12:23:35Z
dc.date.embargoed.fl_str_mv 2025-11-10T12:23:35Z
dc.format.none.fl_str_mv application/pdf
dc.identifier.none.fl_str_mv http://hdl.handle.net/10362/190399
dc.language.none.fl_str_mv eng
dc.rights.cclincense.fl_str_mv http://creativecommons.org/licenses/by/4.0/
dc.rights.none.fl_str_mv http://purl.org/coar/access_right/c_abf2
dc.subject.none.fl_str_mv Bike-sharing
Time series forecasting
Transformer
Data imputation
Urban mobility
Demand prediction
SDG 11 - Sustainable cities and communities
SDG 13 - Climate action
dc.title.fl_str_mv Predicting Bike Stations Occupation using Time-Series Models: The Case Study of Lisbon
dc.type.none.fl_str_mv http://purl.org/coar/resource_type/c_bdcc
description Urban bike-sharing systems have become a vital component of sustainable mobility strategies in modern cities, helping reduce traffic congestion and environmental impact while improving transportation accessibility. In Lisbon, the GIRA bike-sharing network plays a significant role in meeting short-distance travel needs, yet ensuring consistent bike availability across its many stations remains a complex challenge due to spatio-temporal demand fluctuations influenced by external factors like weather and air quality. Although various forecasting methods have been proposed to tackle this issue, many rely on limited data types or struggle with inconsistent data quality, particularly due to missing values in environmental datasets. To better understand and address this problem, we aimed to train various robust forecasting models capable of predicting hourly bike availability at GIRA stations while assessing the impact of different imputation strategies on prediction performance. We employed deep learning architectures, including Long Short-Term Memory (LSTM) and Transformer models, and trained them using historical bike usage data enriched with meteorological and air quality variables. This thesis presents a comprehensive performance evaluation of these models under multiple imputation approaches, analyzing errors across time periods, stations, and input conditions. Our findings reveal that the Transformer architecture consistently achieved the best performance. Grouped statistical imputation produced more reliable forecasts on non-imputed data, although biases emerged in predicting imputed test rows. Furthermore, temporal patterns in error distributions highlight specific hours of the day with heightened predictive difficulty, typically during morning and evening peak times. In conclusion, deep learning models—especially Transformers—offer promising accuracy in forecasting bike availability in a real-world context, provided that imputation strategies are carefully considered. These results support the integration of predictive analytics into bike-sharing operations to enhance decision-making for redistribution, planning, and service quality optimization.
dirty 0
eu_rights_str_mv openAccess
format masterThesis
fulltext.url.fl_str_mv https://run.unl.pt/bitstreams/fe9afd6a-47a2-4889-80bd-059cc12a1a7b/download
id run_4349e8d73a2ced754b5bf605e72f03e9
identifier.url.fl_str_mv http://hdl.handle.net/10362/190399
instacron_str unl
institution Universidade Nova de Lisboa
instname_str Universidade Nova de Lisboa
language eng
network_acronym_str run
network_name_str Repositório Institucional da UNL
oai_identifier_str oai:run.unl.pt:10362/190399
organization_str_mv urn:organizationAcronym:unl
person_str_mv Moreira, Guilherme Valler
publishDate 2025
reponame_str Repositório Institucional da UNL
repository_id_str urn:repositoryAcronym:run
service_str_mv urn:repositoryAcronym:run
spelling engpt_PTUrban bike-sharing systems have become a vital component of sustainable mobility strategies in modern cities, helping reduce traffic congestion and environmental impact while improving transportation accessibility. In Lisbon, the GIRA bike-sharing network plays a significant role in meeting short-distance travel needs, yet ensuring consistent bike availability across its many stations remains a complex challenge due to spatio-temporal demand fluctuations influenced by external factors like weather and air quality. Although various forecasting methods have been proposed to tackle this issue, many rely on limited data types or struggle with inconsistent data quality, particularly due to missing values in environmental datasets. To better understand and address this problem, we aimed to train various robust forecasting models capable of predicting hourly bike availability at GIRA stations while assessing the impact of different imputation strategies on prediction performance. We employed deep learning architectures, including Long Short-Term Memory (LSTM) and Transformer models, and trained them using historical bike usage data enriched with meteorological and air quality variables. This thesis presents a comprehensive performance evaluation of these models under multiple imputation approaches, analyzing errors across time periods, stations, and input conditions. Our findings reveal that the Transformer architecture consistently achieved the best performance. Grouped statistical imputation produced more reliable forecasts on non-imputed data, although biases emerged in predicting imputed test rows. Furthermore, temporal patterns in error distributions highlight specific hours of the day with heightened predictive difficulty, typically during morning and evening peak times. In conclusion, deep learning models—especially Transformers—offer promising accuracy in forecasting bike availability in a real-world context, provided that imputation strategies are carefully considered. These results support the integration of predictive analytics into bike-sharing operations to enhance decision-making for redistribution, planning, and service quality optimization.application/pdfpt_PTPredicting Bike Stations Occupation using Time-Series Models: The Case Study of LisbonMoreira, Guilherme VallerJardim, João Bruno Morais de SousaOliveira, Rita Madalena Cardoso da SilvaHostingInstitutionOrganizationalRUNe-mailmailto:run@unl.ptrun@unl.ptURNurn:tid:2040719332025-11-10T12:23:35Z2025-10-282025-10-28T00:00:00ZHandlehttp://hdl.handle.net/10362/190399http://purl.org/coar/access_right/c_abf2open accessBike-sharingTime series forecastingTransformerData imputationUrban mobilityDemand predictionSDG 11 - Sustainable cities and communitiesSDG 13 - Climate action2966213 bytesliteraturehttp://purl.org/coar/resource_type/c_bdccmaster thesis2025-10-28http://creativecommons.org/licenses/by/4.0/http://purl.org/coar/access_right/c_abf2application/pdffulltexthttps://run.unl.pt/bitstreams/fe9afd6a-47a2-4889-80bd-059cc12a1a7b/download
spellingShingle Predicting Bike Stations Occupation using Time-Series Models: The Case Study of Lisbon
Predicting Bike Stations Occupation using Time-Series Models: The Case Study of Lisbon
Moreira, Guilherme Valler
Bike-sharing
Time series forecasting
Transformer
Data imputation
Urban mobility
Demand prediction
SDG 11 - Sustainable cities and communities
SDG 13 - Climate action
Moreira, Guilherme Valler
Bike-sharing
Time series forecasting
Transformer
Data imputation
Urban mobility
Demand prediction
SDG 11 - Sustainable cities and communities
SDG 13 - Climate action
status NEW
subject.fl_str_mv Bike-sharing
Time series forecasting
Transformer
Data imputation
Urban mobility
Demand prediction
SDG 11 - Sustainable cities and communities
SDG 13 - Climate action
title Predicting Bike Stations Occupation using Time-Series Models: The Case Study of Lisbon
title_full Predicting Bike Stations Occupation using Time-Series Models: The Case Study of Lisbon
title_fullStr Predicting Bike Stations Occupation using Time-Series Models: The Case Study of Lisbon
Predicting Bike Stations Occupation using Time-Series Models: The Case Study of Lisbon
title_full_unstemmed Predicting Bike Stations Occupation using Time-Series Models: The Case Study of Lisbon
Predicting Bike Stations Occupation using Time-Series Models: The Case Study of Lisbon
title_short Predicting Bike Stations Occupation using Time-Series Models: The Case Study of Lisbon
title_sort Predicting Bike Stations Occupation using Time-Series Models: The Case Study of Lisbon
topic Bike-sharing
Time series forecasting
Transformer
Data imputation
Urban mobility
Demand prediction
SDG 11 - Sustainable cities and communities
SDG 13 - Climate action
topic_facet Bike-sharing
Time series forecasting
Transformer
Data imputation
Urban mobility
Demand prediction
SDG 11 - Sustainable cities and communities
SDG 13 - Climate action
url http://hdl.handle.net/10362/190399
visible 1