Abstract: The bridge deflection periodic changes under the temperature action.Exploring its time-varying characteristics can reflect the service performance of the main girder’s boundary constraints,such as bearings and expansion joints.Establishing a correlation model between temperature and deflection is crucial for detecting abnormal changes in bridge deflection.The existing research ignores the hysteresis effect between temperature and deflection,which leads to low modeling accuracy and affects the accurate detection of bridge deflection anomalies.Therefore,a method for monitoring bridge deflection anomalies that adaptively addresses the hysteresis effect is proposed.Firstly,the accurate extraction of the thermal-induced bridge deflection is realized based on the wavelet decomposition method,and the main temperature variables affecting the bridge deflection are screened.Secondly,a temperature-deflection correlation model is established using the gated recurrent unit (GRU) neural network,which adjusts for the hysteresis effect between variables through reset and update gates.This model enables accurate prediction of thermal-induced deflection.Then,an anomaly detection indicator is proposed that reflects deflection changes caused by the deterioration of service performance in the main girder constraint components.Finally,the method is validated with health monitoring data from a real bridge.Results show that this method can effectively identify thermal-induced deflection anomalies,providing a basis for online monitoring and diagnosing performance degradation in bridge constraint components.