Study On Bridge Monitoring Signal Recovery Method Based On LSTM Model

The safety and reliability of bridge structures have always been valued by people.The establishment of structural health monitoring system is the main way to understand the operation status of bridges.However,due to the influence of unfavorable factors such as sensor failure and noise interference,the monitoring data of the bridge health monitoring system often have problems such as data distortion and missing,so that the structural response of the bridge cannot be grasped in time.Therefore,the accuracy and completeness of monitoring data are crucial to accurately understand the health status of bridges.Therefore,it is urgent to develop a method that can accurately restore bridge monitoring data.Aiming at the problems of unclear recovery mechanism of existing bridge monitoring data,insufficient research on hybrid model based on signal decomposition and lack of consideration of multivariate factors,this thesis studies the recovery method of bridge monitoring data based on long short-term memory(LSTM)model from the aspects of statistical analysis,mathematical statistics,signal processing and predictive modeling.The main research contents and conclusions are as follows:(1)The statistical characteristics and correlation of bridge monitoring data are analyzed.Firstly,based on the real bridge monitoring data,the causes of data loss and common processing methods are mined.Secondly,from the perspective of statistical analysis,the test and measurement methods suitable for the correlation analysis of bridge monitoring data are given.Finally,based on the real bridge,according to the correlation test and measurement method,the statistical characteristics of bridge monitoring data and the temporal autocorrelation and spatial-temporal correlation are analyzed,so as to grasp the statistical characteristics of bridge monitoring data,explore the correlation between data,and extract the variable candidate set with high correlation.The analysis results show that there is a time series correlation between the value of the bridge monitoring data at one time and the value at another time.In addition,the monitoring data also have a strong spatial and temporal correlation,especially in the structural symmetry and adjacent positions.(2)A bridge monitoring data recovery mechanism based on time series correlation is proposed.Firstly,the problem of bridge monitoring data recovery based on time series correlation is analyzed,and the univariate and multivariate data recovery strategies are obtained.Secondly,a mask matrix recovery mechanism based on time series correlation is proposed for the lack of univariate bridge monitoring data.Then,based on the temporal correlation of monitoring data and the mask recovery mechanism,a univariate bridge monitoring data recovery model based on LSTM model is constructed and compared with other algorithm prediction models.Finally,the effectiveness and applicability of the proposed method are verified by actual monitoring data under different data missing rates and different missing modes.The research results show that the method is feasible and effective.Compared with other algorithm prediction models,the proposed method has certain advantages.When the data loss rate reaches 60 %,it can still effectively restore the missing data of bridge monitoring.(3)A TVFEMD-LSTM univariate bridge monitoring signal recovery model based on time series correlation is established.Firstly,aiming at the complexity of bridge monitoring data and the excellent filtering characteristics of time varying filter empirical mode decomposition(TVFEMD)and its advantages in dealing with modal aliasing problems,a bridge monitoring data recovery mechanism based on TVFEMD is proposed.Then,based on the temporal correlation of monitoring data,a TVFEMD-LSTM univariate bridge monitoring signal recovery model based on temporal correlation is established,and its recovery performance is compared with that of a single LSTM prediction model.Finally,the effectiveness of the proposed method is verified under different sample lengths and different data missing lengths.The research results show that the method solves the problem of non-linearity and non-stationarity of bridge monitoring data,and makes up for the defect of insufficient prediction accuracy of single LSTM model.The effectiveness of the method in realizing the recovery of bridge monitoring data is verified by an example analysis.(4)A LSTM multivariate bridge monitoring signal recovery method based on spatiotemporal correlation is proposed.Firstly,based on the spatial-temporal correlation of bridge monitoring data and the multivariate data masking mechanism,a LSTM multivariate bridge monitoring signal recovery method based on spatial-temporal correlation is proposed.Secondly,combined with the real bridge monitoring data,the effectiveness of the multivariate data recovery method based on spatio-temporal correlation is analyzed.Finally,based on the relative entropy theory,the reliability of the LSTM multivariate bridge monitoring signal recovery method based on temporal correlation is studied.The results show that this method solves the problem of efficient utilization of multivariate monitoring data,makes up for the defect that univariate data recovery cannot effectively utilize the spatio-temporal correlation between data,effectively improves the recovery efficiency of bridge monitoring missing data,and can provide a useful reference for restoring bridge monitoring missing data.