Probabilistic prediction and early warning for bridge bearing displacement using sparse variational Gaussian process regression

Abstract

Investigating the relationship between temperature variations and bridge bearing displacement is crucial for ensuring structural integrity and safety. However, the current temperature-displacement regression (TDR) model fails to account for inherent uncertainties in monitoring data and model errors. This paper proposes a probabilistic prediction and early warning framework for displacement of bridge bearing using the sparse variational Gaussian process regression (SVGPR) model. The time-varying relationships between temperature and bearing displacement at different time scales are analyzed. The SVGP-TDR model is constructed based on the fully independent training condition (FITC), and the induced points and hyperparameters are optimized simultaneously by combining variational learning and gradient descent method. An early warning method for bearing performance is proposed based on the model estimation error and Shewhart control chart theory, along with the implementation procedure provided. The effectiveness of the proposed method is verified using long-term monitoring data from an existing suspension bridge. The results show that the SVGP-TDR model can predict probability distribution of bearing displacement caused by temperature. Moreover, it can not only consider the uncertainty in the monitoring data, but also quantify the model error and prediction uncertainty. The proposed early warning method performs satisfactorily in assessing the service performance of bridge bearing.