Elevation Prediction Method Of Rigid Frame Bridge Deck Based On Bayesian Network

Elevation control is an important part of the construction of prestressed concrete continuous rigid frame bridges.In this paper,a series of studies are carried out on the elevation control of prestressed concrete continuous rigid frame bridges,aiming to propose a preferred method of elevation control.The main work and conclusions are as follows:(1)Sensitivity analysis of factors influencing elevation.According to the statistical characteristics of the seven elevation influencing factors,such as concrete bulk density and elastic modulus,prestressed tensile force,pipeline friction coefficient,pipeline deviation coefficient,hanging basket load,concrete shrinkage creep relative humidity,etc.,independent or not according to each factor In both cases,sensitivity analysis is performed with sensitivity coefficient and partial correlation coefficient as indicators.The analysis found that: 1.The sensitivity of each factor is in the same order in different situations,indicating that the factors are independent of each other;2.The sensitivity of the three factors of concrete bulk density,concrete elastic modulus,and prestressed tensile force is the largest and is the impact The key variable of structural elevation.(2)Kalman filter estimation of factors affecting elevation.Based on the measured data of the field elevation,a nonlinear Kalman filter local iterative algorithm is used to estimate the three key factors of concrete bulk density,concrete elastic modulus,and prestressed tensile force,and the initial parameter estimation of the nonlinear Kalman filter local iterative algorithm is studied.Effect of results and iteration convergence rate.The results show that the initial parameter variance,innerand outer loop weighting coefficients,measurement variance,and system variance do not have much impact on the Kalman filter estimation results,but only affect its iterative convergence speed.Inversely proportional to the convergence rate,the variance of the system has little effect on the convergence rate,and the effect of the initial parameter variance on the convergence rate depends on the specific situation.(3)Deflection prediction based on Bayesian network.The Bayesian network topology is constructed according to the causal relationship between the influencing factors of the elevation and the deflection deformation error.The Latin hypercube sampling method is used to sample the initial probability distribution data of each influencing factor to obtain complete data samples and train the Bayesian network.The learning ability of Bayesian network is used to predict the deflection and deformation of the bridge structure of each beam section,and the prediction error is analyzed.The research shows that in the short cantilever stage of the rigid frame bridge,due to insufficient learning samples,the prediction effect is poor.The maximum deviation between the predicted value and the measured value is 7mm.As the construction progresses,the learning sample increases,and the prediction effect becomes better and better.The predicted value The maximum difference from the actual value does not exceed 3mm.The above results indicate that the deflection of Bayesian network can be used to predict the deflection of cantilever construction.(4)Comparison of deflection prediction methods.Combining with the previous research results,a method for predicting the elevation of rigid-structure bridge deck based on Kalman filtering method and Bayesian network is proposed.The actual engineering application shows that the prediction result of this method is less than 2mm.Compared with the conventional Kalman filter estimation method(prediction error is 5mm),the accuracy is higher,and it can be used to guide the elevation control of the rigid frame bridge cantilever construction process.