Research on the Completion Method for Missing Data in Structural Monitoring System Based on CNN-BiLSTM

Structural monitoring systems often have missing or abnormal data of key monitoring indicators under the influence of multiple categories, which seriously affects the effective assessment of the structural condition and cannot provide a more reliable basis for maintenance and management decisions. In this paper, we built a monitoring data missing completion method for based on the combination of convolutional neural network and bidirectional long-short term neural network (CNN-BiLSTM). The Root Mean Squared Error (RMSE), Determination Coefficient (R2) and Mean Absolute Error (MAE) were adopted to evaluate the validity and accuracy of the supposed method in comparison tests. Comparing with BP, CNN-LSTM and CNN-LSTM-Attention methods for bridge deflection data prediction and completion, the CNN-BiLSTM method has higher prediction accuracy and smaller error for monitoring data completion, showing certain superiority. The K-fold cross-validation results show that the CNN-BiLSTM method does not fall into an overfitting state under different data subsets, and has good stability. The errors of CNN-BiLSTM method under different signal-to-noise ratios are smaller than other methods, and it still has relatively good prediction performance under low signal-to-noise ratio.