Data reconstruction for tunnel structural health monitoring: An updated KNN model with gray relational analysis

Abstract

Structural health monitoring (SHM) system is an important way to evaluate the tunnel structural performance. In practice, the missing data is inevitably induced in the SHM dataset, which may cause deviation or even misleading results of the data analysis, and hence, need to be accurately reconstructed. This study proposes a new KNN data reconstruction method based on a gray correlation measure (GRA-KNN). Compared to the traditional KNN, the GRA-KNN can measure the structural similarity of data better, and the pre-filling can make full use of the information of known data to estimate missing data. By comparing with the other five machine learning methods (i.e., Ridge Regression, Support Vector Regression, Multilayer Perceptron, RandomForest, and XGBoost), the imputation performance of this method is examined using two real-time SHM datasets from Nanjing Yangtze River tunnel and Hong Kong-Zhuhai-Macao Bridge Undersea Tunnel. Results show that the GRA-KNN method possesses a better performance and stronger robustness under a wide range of missing ratios from 10% to 90%, showing great potential in SHM data reconstruction. In particular, when the missing ratio is larger than 60%, the imputation error of the proposed method is the smallest compared to that of the other machine learning methods.