This study aims to explore a machine learning based bridge safety monitoring system to improve the accuracy and efficiency of bridge safety monitoring. Through the ConvLSTM (Convolutional Long Short-Term Memory) model algorithm, this system can predict and evaluate the structural health status of bridges in real time. The research results indicate that compared with traditional methods, this system has significant improvements in accuracy and response time, and has important practical application value for bridge safety management and maintenance. Four experiments in the experimental stage evaluated the performance of the high-precision safety monitoring system for bridges based on the ConvLSTM model. In the prediction accuracy test experiment, the AUC (Area Under the Curve) value of the ConvLSTM model was 0.95, the AUC value of the logistic regression model was 0.85, the AUC value of the Support Vector Machine (SVM) was 0.78, and the AUC value of the Random Forest (RF) algorithm was 0.90. In the real-time data processing capability evaluation experiment, the average response time of the ConvLSTM model was 0.005 s. In extreme climate assessment experiments, ConvLSTM can maintain a 95% accuracy even under extreme climate conditions. In the final long-term stability and reliability experiment, the average accuracy of the ConvLSTM model remained at 94% within one year. In the above data conclusions, the ConvLSTM model is applicable in bridge safety monitoring, with high accuracy and superior response speed, especially in the face of environmental challenges and long-term operational needs.