Bridge Health Monitoring Data Analysis Based On Machine Learning Support Vector Machine Model

In view of the many difficulties in the data analysis of bridge health monitoring in the era of big data,and the various algorithms in machine learning field are coming to be more mature,this thesis implements machine learning theory as the foundamental theory to study the key technologies of data preprocessing and damage identification on bridge structure.This paper applied support vector machine model(SVM)and K nearest neighbor algorithm(KNN)in the process of bridge vibration acceleration data.Its feasibility and validity are verified through the application in golden gate bridge in the United States.The main contents of this thesis are shown as follows:First of all,the current status of the bridge health monitoring system and the basic theories of machine learning are researched and studied significantly.Due to limitations and inefficiency of the current model of bridge health monitoring systems that implement "implementing sensors to collect data with FEA" in the process of mass monitoring data,it is stated that how to apply SVM and KNN algorithms into analyzing the data generated from the bridge health monitoring system.The accuracy and efficiency of results are significantly improved.Espacially,the original problem,Lagrange function expression,and the optimal solution of support vector machine were derived by mathematical formulas,which laid a theoretical foundation for the subsequent implementation of algorithms.Considering the heavy noise,inconsistent dimensions,and mass data of the structure output response information,this paper adopts a series of data preprocessing measures including repeated value processing,missing value filling,and data nondimensionalizing.Particularly,in the processing of missing values,this article innovatively uses the random forest algorithm to complete the filling of missing values,so as to fill the original data set without changing its probability distribution curve.This enriches the data samples for the subsequent model learning process.In addition,due to the limited dimensions of the feature matrix of the data set,this article uses PCA to perform noise reduction processing on the data,which not only avoilds of noise,but also retains the original dimension space of the feature matrix,therefore retaining the effective information as much as possible.Then,K-means algorithm(KMeans)is used to perform cluster analysis on the data after deleting the noise and the observation on silhouette coefficient of different K values is performed.Combined the results generated from cluster analysis and the statistical indicators,the sample data is given a new label "Level" which indicates the levels of structural damage.They are used as the learning target for subsequent machine learning models.Finally,this paper relies on the real-time monitoring data of the Golden Gate Bridge in the United States to verify the implementation of the KNN and SVM model.The outcomes generated from these models have shown that SVM model has a higher accuracy of 93.2%,outperforming the KNN model.Considering the SVM’s higher processing efficiency will become more outstanding as the volume of data increases,the SVM model is eventually recommended as a reference model for future bridge structure helath monitoring data analysis.