RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publicación
Artículo científico (JournalArticle)
Deep learning for urban air quality: a traffic-based prediction and alert system for Valencia
[EN] Urban traffic congestion is a critical issue with significant implications for mobility and urban planning. In this
study, we develop a real-time predictive alarm system capable of forecasting whether a street is likely to
experience unusually high traffic within the next 30 min. The system classifies road segments into three alert
levels based on traffic data updated every 10 min, providing timely information that can support decision-making
in traffic management. The prediction model is built using deep learning techniques trained on a whole year of
traffic data in the city of Valencia, and tested with the following year¿s data. We evaluated different neural
network architectures, including long short-term memory (LSTM) networks, an extended LSTM variant
(xLSTM), and Graph Neural Networks (GNNs). Our results show that LSTM provides the best balance between
accuracy and computational efficiency, making it the most suitable model for real-time deployment. In addition to
traffic data, we incorporate meteorological variables such as wind speed, wind direction, and precipitation to
explore their potential impact on traffic dynamics. Although the relationship between traffic and environmental
conditions warrants further study, this work demonstrates the feasibility of using real-time predictions to improve
urban mobility strategies. The proposed system offers a data-driven approach that can be integrated into broader
traffic management frameworks to improve efficiency and responsiveness., The research of MGF is supported by the Margarita Salas postdoctoral fellowship from the Ministerio de Educación MS21-038 and the Generalitat Valenciana ASFAE/2022/020. The research of VS is supported by the Generalitat Valenciana PROMETEO/2021/083 and the Ministerio de Ciencia e Innovacion PID2020-113644GB-I00. The research of JH is supported by the RESECARIN project TED2021-129682B-I00 from the Spanish Ministerio de Ciencia e Innovación. The research of JFU and ELS is supported by the Agència Valenciana de la Innovació (AVI) INNEST/2021/263.