Forecasting the dynamic response of in-service bridges is essential for real-time structural condition assessment and early abnormal detection. Current research efforts mainly focus on statistical learning for near-term forecasting (e.g., one-step ahead) of structural response, such as displacements and strains. However, there is a growing need for data-driven methods capable of long-range (e.g., multi-step ahead) forecasting. To this end, we present a novel sparse matrix learning method for long-term time series prediction in the field of structural health monitoring (SHM). This approach integrates Fourier transform and sparse matrix multiplication techniques to learn spatiotemporal features of SHM time series data. More precisely, the Fourier transform module captures global patterns, while the sparse matrix multiplication module focuses on local spatial features. The performance of the proposed method is evaluated by using a real-world dataset featuring multi-year temperature and strain time series from a concrete bridge. Specifically, we assess the robustness of our model against noise interference. Remarkably, the proposed sparse matrix learning method outperforms baseline models, demonstrating its superior capability in long-range strain forecasting. We anticipate that the proposed sparse matrix learning method will significantly advance data-informed safety assessment of critical civil infrastructure. The data and source code is available at https://github.com/DUAT-Maker/FT-SMNet.

中文翻译：

---

FT-SMNet：用于结构健康监测时间序列数据预测的傅里叶变换稀疏矩阵网络


预测在役桥梁的动态响应对于实时结构状况评估和早期异常检测至关重要。目前的研究工作主要集中在统计学习上，用于结构响应的近期预测（例如，提前一步），例如位移和应变。然而，对能够进行长期（例如，提前多步）预测的数据驱动方法的需求越来越大。为此，我们在结构健康监测 （SHM） 领域提出了一种新的稀疏矩阵学习方法，用于长期时间序列预测。这种方法集成了傅里叶变换和稀疏矩阵乘法技术，以学习 SHM 时间序列数据的时空特征。更准确地说，傅里叶变换模块捕获全局模式，而稀疏矩阵乘法模块专注于局部空间特征。通过使用具有混凝土桥梁多年温度和应变时间序列的真实数据集来评估所提出的方法的性能。具体来说，我们评估了模型对噪声干扰的鲁棒性。值得注意的是，所提出的稀疏矩阵学习方法优于基线模型，展示了其在远程应变预测方面的卓越能力。我们预计，拟议的稀疏矩阵学习方法将显着推进关键民用基础设施的数据知情安全评估。数据和源代码可在 https://github.com/DUAT-Maker/FT-SMNet 获取。