Research and application of EMD-BiLSTM in bridge data reconstruction

Abstract

Structural monitoring systems are increasingly used in bridge engineering because real environmental and structural response data can be obtained directly. In order to accurately assess bridge conditions and provide basic data for new bridge design, it is important to ensure the quality of the data, and when data are missing, various methods are needed to reconstruct the missing data. In this paper, we propose an EMD-BiLSTM model to reconstruct the missing deflection data by predicting the original data and the decomposed subsequences. The core of this method is to make the data subsequence more correlated by using EMD decomposition, and to obtain the before-and-after correlation of the data subsequence by BiLSTM. The EMD-BiLSTM model can effectively reconstruct the missing bridge deflection data with a root-mean-square error of 0.07759. The subseries of the original data decomposed by EMD improves the prediction accuracy of the BiLSTM model, and the BiLSTM also outperforms other machine learning algorithms to obtain more features of the data.