A Novel Bridge Deflection Missing Data Repair Model Based on Two-Stage Modal Decomposition and Deep Learning

Abstract

The bridge structural health monitoring (SHM) system will inevitably experience missing data. To ensure the integrity and practicability of the bridge SHM system, it is essential to repair the missing data. The existing data recovery methods mainly use the spatial correlation with other monitoring data but cannot adequately capture the time dependence of the raw monitoring data. This paper uses historical monitoring data to predict future data and complete the task of repairing missing data. A hybrid prediction model based on the gated recurrent unit (GRU) neural network, complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), and variational mode decomposition (VMD) is proposed. By decomposing the raw monitoring data, the input of the GRU model is optimized, resulting in improved accuracy of prediction and enabling the model to operate independently from other sensors. The accuracy of the method is verified based on the SHM data of a cable-stayed bridge. The prediction results of the proposed model are stable and reliable, with a prediction accuracy reaching 95%, indicating that the CEEMDAN-VMD-GRU model is suitable for repairing missing deflection data in bridge SHM systems.