Skip to main content

RED DE INVESTIGACION EN TRANSPORTE Y MOVILIDAD SOSTENIBLE, SEGURA E INTELIGENTE

RED2024-154236-E

•
Funding agency name Agencia Estatal de Investigación
Funding agency acronym AEI
Program Programa Estatal de Transferencia y Colaboración
Subprogram Subprograma Estatal para impulsar la Colaboración en y entre los Sectores Público y Privado
Call Redes Investigación /Redes Excelencia
Call year 2024
State Plan 2024-2027 Plan Estatal de Investigación Científica y Técnica y de Innovación
Beneficiary centre UNIVERSITAT POLITECNICA DE CATALUNYA

Resultados relacionados

Resultados totales (Incluyendo duplicados): 1
Encontrada(s) 1 página(s)

Recercat. Dipósit de la Recerca de Catalunya
Publication Artículo científico (JournalArticle)

GeoLSTM: A road geometry-aware neural network model for spatio-temporal highway traffic speed and intensity forecasting

  • Franco Panadès, Gerard
  • Llatje i Hierro, Òscar
  • Pagès Giralt, Laia
  • Jofre Roca, Lluís
Accurate prediction of dynamic states in complex physical systems is critical for intelligent decision-support and control. This paper presents a novel spatio-temporal deep learning framework tailored to expert systems operating in mobility and transportation environments. Using a large-scale, multi-source dataset collected over two years from a 100-kilometer highway segment, the proposed model integrates static road geometry features—specifically curvature and slope—into a Long Short-Term Memory (LSTM) architecture. We introduce GeoLSTM, a geometry-aware neural network model that enhances mid-term traffic forecasting accuracy across multiple metrics, including vehicle speed, total intensity and heavy-vehicle (HV) intensity, addressing the common omission of static road geometry by embedding curvature and slope. Our experiments demonstrate that incorporating static spatial features significantly reduces prediction error (up to 26% for speed and 16% for heavy-vehicle intensity) and improves model stability, outperforming both traditional baselines and transformer-based alternatives. We further quantify computational costs to assess real-time deployment feasibility. These results establish a robust methodology for integrating contextual spatial data into deep learning-based expert systems, with potential generalization across other domains requiring predictive modelling in structured, data-rich environments., The first author would like to personally acknowledge the Catalan Traffic Authority and the Industrial Doctorates Plan of the Department of Research and Universities of the Generalitat de Catalunya for supporting with funding the research project which frames this research article, developed in the framework of his PhD thesis. Additional support was provided by the European Institute of Technology (EIT) for funding mobility to attend conferences through the Doctoral Training Network (DTN) organization. Furthermore, the research has also been developed in the context of the following projects: • i-MovE (Interoperabilidad para una Movilidad Eficiente, sEgura y Equitativa) project, funded by TSI-100121-2024-25 and by the EuropeanUnion NextGenerationEU/PRTR; • METROPOLIS project, PLEC2021-007609, funded by MCIN/AEI/ 10.13039/501100011033 and by the EuropeanUnion NextGenerationEU/PRTR; •And the REDES project, aid RED2024-154236-E funded by MICIU/ AEI/10.13039/501100011033., Peer Reviewed, Postprint (published version)