Online Prediction and Early Safety Warning for Bridge Bearings Based on Streaming Sparse Variational Gaussian Process Regression

Abstract To address the low prediction accuracy of existing offline temperature-displacement regression (TDR) models caused by material aging and changing service environments, this study proposes an online prediction and early warning framework for bridge bearing displacement using streaming sparse variational Gaussian process regression (SSVGPR). The sparse time-varying relationship between bearing displacement and bridge temperature is identified. A sparse variational Gaussian process regression (SVGPR) model was developed by incorporating inducing points and variational inference. An interval-based incremental updating strategy was proposed within the streaming SVGP-TDR framework to update the hyperparameters and posterior distribution using newly acquired monitoring data. Further, an early warning method based on prediction errors and a model adaptation mechanism triggered by warning information were developed to detect changes in bearing performance and mitigate model contamination caused by abnormal data. The effectiveness of the proposed method is validated using long-term monitoring data from an existing suspension bridge. The results show that the SSVGP-TDR model can accurately predict the probability distribution of bearing displacement induced by bridge temperature. The model updates its parameters to maintain long-term prediction accuracy under sparse variations in bridge temperature and bearing displacement. The proposed early warning method enables the performance degradation identification and model adaptive update of the bearing with different amplitudes.