Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Graph Neural Network–Based Spatiotemporal Structural Response Modeling in Buildings
https://orcid.org/0000-0003-1492-9673
Modeling structural responses is vital in building structural health monitoring. This study proposed the graph network–based structure simulator (GNSS), a method employing graph neural networks, for spatiotemporal structural response modeling in buildings. GNSS considered both the spatial positions and connections of structural components and the temporal correlations of time-series structural data. The entire 6-story building was represented as a graph, with nodes representing mass and edges representing columns and beams. These nodes and edges captured time-series data about structural information, responses, and ground motion. GNSS included three components: encoder, processor, and decoder. Four GNSS model variations were explored (GNSS-NE, GNSS-N2E, GNSS-NUEU, and GNSS-Full), each investigating different feature integrations and graph network architectures. To assess GNSS’s predictive performance for structural responses (displacement and acceleration) under varying test conditions, three case studies were conducted: One-Step, Rollout, and Rollout&Calibration. Among the four model variations, GNSS-NE demonstrated superior performance in predicting both displacement and acceleration across all three case studies, except for displacement prediction in the Rollout scenario. Overall, GNSS models performed best in the One-Step case study, followed by Rollout&Calibration, with the lowest performance observed in the Rollout case study. These results highlight the significant potential of GNSS for extensive application in structural response modeling by effectively integrating spatial and temporal information.
Graph Neural Network–Based Spatiotemporal Structural Response Modeling in Buildings
https://orcid.org/0000-0003-1492-9673
Modeling structural responses is vital in building structural health monitoring. This study proposed the graph network–based structure simulator (GNSS), a method employing graph neural networks, for spatiotemporal structural response modeling in buildings. GNSS considered both the spatial positions and connections of structural components and the temporal correlations of time-series structural data. The entire 6-story building was represented as a graph, with nodes representing mass and edges representing columns and beams. These nodes and edges captured time-series data about structural information, responses, and ground motion. GNSS included three components: encoder, processor, and decoder. Four GNSS model variations were explored (GNSS-NE, GNSS-N2E, GNSS-NUEU, and GNSS-Full), each investigating different feature integrations and graph network architectures. To assess GNSS’s predictive performance for structural responses (displacement and acceleration) under varying test conditions, three case studies were conducted: One-Step, Rollout, and Rollout&Calibration. Among the four model variations, GNSS-NE demonstrated superior performance in predicting both displacement and acceleration across all three case studies, except for displacement prediction in the Rollout scenario. Overall, GNSS models performed best in the One-Step case study, followed by Rollout&Calibration, with the lowest performance observed in the Rollout case study. These results highlight the significant potential of GNSS for extensive application in structural response modeling by effectively integrating spatial and temporal information.
Graph Neural Network–Based Spatiotemporal Structural Response Modeling in Buildings
https://orcid.org/0000-0003-1492-9673
Modeling structural responses is vital in building structural health monitoring. This study proposed the graph network–based structure simulator (GNSS), a method employing graph neural networks, for spatiotemporal structural response modeling in buildings. GNSS considered both the spatial positions and connections of structural components and the temporal correlations of time-series structural data. The entire 6-story building was represented as a graph, with nodes representing mass and edges representing columns and beams. These nodes and edges captured time-series data about structural information, responses, and ground motion. GNSS included three components: encoder, processor, and decoder. Four GNSS model variations were explored (GNSS-NE, GNSS-N2E, GNSS-NUEU, and GNSS-Full), each investigating different feature integrations and graph network architectures. To assess GNSS’s predictive performance for structural responses (displacement and acceleration) under varying test conditions, three case studies were conducted: One-Step, Rollout, and Rollout&Calibration. Among the four model variations, GNSS-NE demonstrated superior performance in predicting both displacement and acceleration across all three case studies, except for displacement prediction in the Rollout scenario. Overall, GNSS models performed best in the One-Step case study, followed by Rollout&Calibration, with the lowest performance observed in the Rollout case study. These results highlight the significant potential of GNSS for extensive application in structural response modeling by effectively integrating spatial and temporal information.
Graph Neural Network–Based Spatiotemporal Structural Response Modeling in Buildings
J. Comput. Civ. Eng.
Liu, Fangyu (Autor:in) / Xu, Yongjia (Autor:in) / Li, Junlin (Autor:in) / Wang, Linbing (Autor:in)
01.03.2025
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
Multi-site solar irradiance prediction based on hybrid spatiotemporal graph neural network
| American Institute of Physics | 2024
| American Institute of Physics | 2022
| DOAJ | 2022