Prediction of monitoring data and state assessment of heritage stone-timber structures based on spatio-temporal graph neural networks

Abstract

This study aims to improve the accuracy of monitoring data prediction and structural state assessment for preserving cultural heritage stone-timber structures. Significant spatial correlation and temporal dependence are revealed in the monitoring data. A novel spatio-temporal graph neural network (GLSTNN) is therefore proposed to predict monitoring data and assess the structural state of heritage building. The GLSTNN combines Graph Attention Networks and Long Short-Term Memory Networks to model the spatio-temporal characteristics of monitoring data. This study constructs irregular topological graphs based on physical proximity and data similarity to represent the spatial relationships between sensors. Additionally, residual connections are introduced in GLSTNN to enhance the model’s ability to represent non-linear data. To validate the proposed approach, a case study is conducted on a heritage stone-timber structure located in Tibet, China, where sensors monitor structural strain, tilt angles, and environmental temperature and humidity. Data from 18 monitored parameters are used for validation. Experimental results show that incorporating temperature and humidity effects improves predictive performance by strengthening the spatio-temporal relationships among sensors and better capturing environmental impacts. The GLSTNN model outperforms existing techniques in structural state prediction and anomaly detection, providing accurate identification of anomalies and prediction of key parameters from the measured data. The proposed approach has a high potential to offer a promising solution for complex heritage stone-timber structures.