Main Girder Elevation Forecasting And Parameter Identification Of Long Span Concrete Cable-stayed Bridge During Work Progress

Bridges are the basic facilities for connecting roads and crossing rivers,ravines and obstacles.The large span bridge mainly adopts the method of segmental construction.Its construction comes down to people's lives and property safety,to ensure the construction sequence and safety is very important.Construction monitoring is an important work to ensure the quality of construction.In this process,we focus on the bridge alignment and the force of the bridge to meet the requirements of design.Because of construction deviation,external environment and other factors,the actual elevation(alignment)of the bridge and the weight of the concrete main girder may deviate from the theoretical value.If in the course of construction:(1)Using the actual measured data of the finished segment to forecast the bottom elevation of the subsequent section.(2)Using the finite element model and the existing monitoring data to identify the bulk density of the bridge,and then update the parameters of the finite element model.So that the results calculated by model are closer to the ones by actually constructed.This help us to rectify some of deviation in advance,so that the alignment of the bridge is meeting design requirements finally.In this paper,the Fanhe Harbor Bridge of Huizhou(long span prestressed concrete cable-stayed bridge)as the research object,using the nonlinear mapping ability of BP neural network,a nonlinear mapping relationship is established between the measured values of elevation and the influencing factors.Prediction of the elevation of the bottom section of the subsequent segment by the metabolic method.Aiming at the problem of parameter identification,two methods of least squares support vector machine and BP neural network are used to identify the parameters.Firstly,the learning samples of ‘bulk density' to ‘elevation' are obtained by the finite element model,and the identification model of the above two methods is established through the sample.Then in the actual construction monitoring process to reverse the learning process,from the 'measured elevation variables' to' bulk density'.We complete the identification of bulk density finally.The main contents of this paper include the following aspects:1)The finite element model of the bridge is established by software Midas/Civil,and the influence factors are analyzed;2)The main factors affecting the elevation of the main beam are analyzed.A BP neural network model for the elevation and influence factors of main girder segments is established.The stability of the network is tested by the normalized volatility and the normalized normalized volatility.3)We forecast The north tower cross 11# to 23# segment elevation in Final pull operating mode in metabolic way.Through the comparison with the measured value,the influence of parameters such as the number of nodes in the hidden layer,the transfer function and the expected error on the network is analyzed.4)The least squares support vector machine and BP neural network are used to identify the concrete weight of the main girder.Analyze the recognition effect and compare the characteristics of two different methods.