Study On The Processing Method Of Bridge Health Monitoring Information Based On Time Series Analysis

The bridge health monitoring system is now been extensively applied due to the mixture of the increasingly prominent of the in-service bridge safety state problem and the rapid development on the technology of bridge monitoring, and how to make a quick and accurate estimation of a bridage safety state using the real-time bridge health monitoring information is of great concern. In this paper, on the basis of time series analysis, we successively analysis the processing method of data preprocessing, similarity query and anomaly detection based on bridge health monitoring multivariate information, realizing the processing method of bridge health monitoring information by contructing the anomaly detective model based on time series analysis.A method of outliers detection based on cluster is proposed, which can identify all outliers in each single variable time series effectively through segmenting the univariate time series by using local standard deviationas as a measurement, making the elements in a same class having high similarity while the elements in neighboring class having comparatively low similarity, then dig out the class whose elements are not more than the corresponding threshold value as the outliers. Then choosing the method based on Nearest Neighbour which is fast, accurate and suitable for dealing with the massive data to form the missing value by finding the most possibly similar elements and calculating the average value of these elements. Then extract the characteristics of Masangxi bridge health monitoring dataset by using the PCA to set up the CMTS, verifying the feasibility of using the PCA to compress a bridge health monitoring dataset.To remedy the deficiency of original K-means, a modified K-means algorithm based on meshing and the trilateral theorem of triangle is proposed. The modified K-means can meshes the sample space into correspondingly dimensional blocks due to the distribution of the sample itself to evaluate the k and initial position of center points, additionally, by introducing the trilateral theorem of triangle, reduces the iterations and computation complexity greatly. By contrast the processing accuracy and efficiency of this two algorithm when dealing with same data set called PAMAP2 Physical Activity Monitoring in UCI, verifying the high accuracy and efficiency of modified K-means.A KNN algorithm based on B+-tree structure is set up, and by using the CMTS of Masangxi bridge health monitoring dataset, verifying its high accuracy and efficiency.By analysis the anomaly detection of the multivariate time series and the multivariate bridge health monitoring dataset by using the LOF as a measure, the anomaly detective model of the multivariate bridge health monitoring dataset is set up. Then integrate all the methods and algorithms above, set up a information analysis system of bridge health monitoring information on Matlab platform based on the anomaly detection model, and import the Niupeng bridge real-time health monitoring dataset into this system to verify this anomaly detection model, the conclusion is equates with the periodic inspection report of Niupeng Bridge, verifying the feasibility of the anomaly detective model of bridge health monitoring information based on time series analysis.