Structural Health Inspection of Bridges Based on Optimal Decomposition and Machine Learning

Accurate forecasting of bridge settlement is important for structural health monitoring and maintenance decision-making, particularly when monitoring responses exhibit nonlinear and non-stationary characteristics. This study proposes a hybrid forecasting framework that integrates variational mode decomposition (VMD), enhanced through whale optimization algorithm (WOA) parameter tuning and hereafter termed enhanced variational mode decomposition (EVMD), with component-specific machine learning models. WOA is employed to adaptively optimize the key VMD parameters, enabling the settlement sequence to be decomposed into intrinsic mode functions (IMFs) with distinct temporal characteristics. For each decomposed component, the most suitable predictor is selected from gated recurrent unit (GRU), long short-term memory (LSTM), backpropagation neural network (BP), AdaBoost, kernel ridge regression (KRR), and support vector regression (SVR) according to its prediction performance. The component-level forecasts are subsequently aggregated to reconstruct the final settlement prediction. The proposed framework was evaluated using bridge settlement monitoring data from Lianhua Bridge in Guizhou Province, China. The results show that the proposed method achieves a coefficient of determination (R2) of 0.963 and outperforms forecasting frameworks based on empirical mode decomposition (EMD), ensemble empirical mode decomposition (EEMD), and conventional VMD, demonstrating improved prediction accuracy and robustness for the monitored settlement sequence. These results indicate that adaptive decomposition combined with component-specific prediction provides an effective approach for short-horizon bridge settlement forecasting.