Font Size: a A A

Reserch On Dynamic Prediction Methods Of Bridge Extreme Stresses Based On Monitoring Data Assimilation

Posted on:2020-05-26Degree:MasterType:Thesis
Country:ChinaCandidate:G QuFull Text:PDF
GTID:2392330596487380Subject:Engineering, Construction and Civil Engineering
Abstract/Summary:PDF Full Text Request
During the long-term service periods,bridge health monitoring system has produced a large amount of monitoring data,and how to dynamically predict bridge dynamic responses with these data,which can help guide important decisions about bridge preventive maintenance and provide the theoretical foundations for further bridge dynamic reliability assessment,has become one of the key scientific problems in the field of bridge health monitoring(BHM).In this paper,the maximum value of monitoring stress data at the given time interval is defined as the monitoring extreme stress.Considering the coupling of the monitoring information including randomness,time-varying characteristic,periodicity and non-stationarity,the data assimilation methods including the decoupling prediction methods,Bayesian dynamic models and improved particle filters,are utilized for dynamically predicting bridge extreme stresses.The specific contents include:(1)Two kinds of bridge extreme stress prediction methods based on the fusion of decoupling monitored extreme stresses and Bayesian dynamic model are proposed: 1)the monitored data is decoupled through the moving average method,and the corresponding Bayesian dynamic models are built with decoupled data,then the dynamic prediction of the bridge extreme stresses can be achieved with the combination of the decoupled extreme stress prediction data;2)with the empirical mode decomposition(EMD)method,the intrinsic mode functions(IMFs)extracted from monitored information are utilized to build the corresponding Bayesian dynamic models,further,the bridge extreme stress prediction can be made with the fusion method;finally,the two kinds of methods are compared with each other based on the monitoring data of the actual bridges.(2)Two kinds of Bayesian dynamic nonlinear prediction models for bridge non-stationary extreme stresses are proposed: 1)for the nonlinear monitored data with long periodicity,the Fourier dynamic nonlinear model(FDNM),established based on monitored extreme stress data,is converted into the Fourier dynamic linear model(FDLM)through the Taylor series expansion technique,further,the periodic bridge extreme stresses can be dynamically predicted with Bayes method;2)for the general nonlinear extreme stress data,to reduce the prediction errors caused by the state equations relying on the historical monitoring information,local polynomial theory and monitored extreme stress time series analysis method are introduced and combined to build the Bayesian dynamic local trend model(BDLTM),further,the adaptive and dynamic prediction for non-stationary extreme stresses can be achieved.(3)Two kinds of improved particle filter algorithms are proposed for dynamic prediction of bridge extreme stresses: 1)for the nonlinear and non-Gaussian monitored information,the EM algorithm and the K-MEANS algorithm are combined and embedded into the Gaussian mixture filter for estimating the probability distribution of the target state in a high-precision,further,based on monitored stress data,the improved Gaussian mixture particle filter algorithm(IGMPF)is utilized for dynamically predicting bridge extreme stresses;2)based on the nonlinear and Gaussian monitored information,at the importance sampling stage,the Bayesian dynamic filtering algorithm is utilized to generate the proposed distribution which can solve the particle degradation problem of traditional particle filter,and considering the possible environmental mutation,the Bayesian discount factor is used for increasing the robustness of the particle filter,finally,based on the improved particle filter(IPF)approach,the dynamic prediction of bridge extreme stresses is achieved.
Keywords/Search Tags:monitored extreme stresses, coupling, dynamic prediction, Fourier dynamic linear model, Bayesian dynamic local trend model, improved particle filter
PDF Full Text Request
Related items