Bridge Stability Prediction Using Bayesian-Optimized ConvLSTM

Abstract Bridge structures are affected by wind loads, traffic loads, and long-term environmental changes. Therefore, stability prediction is crucial for achieving effective bridge health monitoring. This study proposes a convolutional long short-term memory network model based on Bayesian optimization improvement. This model can capture spatiotemporal features in monitoring data. The study collects vibration and wind speed data from cable-stayed bridges. Then, the study uses denoising, interpolation, and normalization techniques for preprocessing. To further optimize feature extraction, the model introduces a grouping space spectrum pyramid pooling structure. Meanwhile, to achieve adaptive tuning of hyperparameters, Bayesian optimization algorithm is adopted. Both experimental and ablation studies show that the proposed model outperforms traditional methods in terms of prediction accuracy and generalization ability. The combination of grouping space spectrum pyramid pooling and Bayesian optimization significantly advances the model's ability to learn spatiotemporal features. This helps to obtain more robust and reliable prediction results. This model can serve as a practical tool for real-time evaluation of bridge stability. Then, the model can provide support for intelligent decision-making in long-term structural health monitoring.