Parameter Sensitivity Analysis Of Seismic Vulnerability Of RC Frame Structure Based On Neural Network

The seismic vulnerability assessment of existing regional buildings can provide an important scientific basis for urban earthquake prevention and disaster reduction.However,the parameter investigation of regional buildings can be effectively reduced and if the structural characteristic parameters related to structural seismic vulnerability closely are found by the means of sensitivity analysis,which significantly helps to improve the efficiency of regional seismic vulnerability assessment.The deep neural network algorithm,which has a strong data learning ability,can be used to realize the nonlinear mapping between the characteristic parameters of RC frame structures and structural seismic damage,and provide a more accurate mathematical relationship model for the sensitivity analysis of seismic vulnerability of RC frame structures.In this paper,the sensitivity analysis method based on neural network was applied to the parameter sensitivity analysis of seismic vulnerability of RC frame structures in order to obtain scientific and accurate influence ranking of structural characteristic parameters with the seismic damage assessment method based on component deformation.The main research work and results are as follows:(1)7 structural characteristic parameters,including fortification intensity,site category,height of standard story,number of structural stories,span parallel to the loading direction of seismic wave,number of spans parallel to the loading direction of seismic wave and number of spans perpendicular to the loading direction of seismic wave,were set as variables,and 2592 typical RC frame structure models were designed by YJK software correspondingly.Then the dynamic incremental time history analysis was carried out by using Open SEES software after 7 seismic waves were selected.In order to evaluate the seismic vulnerability of frame structure,four indexes,such as the maximum story shear force,the maximum story drift ratio,the residual story drift ratio and the maximum horizontal displacement are selected as the structure-level seismic damage index,while the proportion of beam and column component s in 7 performance states of each story were taken as the component-level seismic damage index.Finally,62208 groups of seismic damage index data are extracted from the results of elastoplastic analysis.(2)Based on the deep forward neural network algorithm,the seismic damage index prediction models at structure-level and component-level were designed and trained respectively,and the structural damage results could be predicted by inputting PGA and structural characteristic parameters.Among them,the average prediction accuracy of structure-level seismic damage index on the test set is 86.87%,while the accuracy of component-level seismic damage index on the test set is slightly lower.For the proportion of beams whose performance state is in 1,2 and 3,the average prediction accuracy is 74.07%,and that of columns is 82.91% respectively.However,the fitting effect of the component performance state distribution on each story is good,which is enough to support the judgment of the specific damage of the structure.(3)The MIV sensitivity analysis method based on neural network was used to explore the influence of different characteristic parameters on the seismic damage index of structure-level and component-level respectively.For the seismic damage index of structure-level,the comprehensive influence order of characteristic parameters is PGA> Fortification Intensity> Height of Standard Story> Number of Structural Stories> Site Category> Span Parallel to the Loading Direction of Seismic Wave > Number of Spans Parallel to the Loading Direction of Seismic Wave > Number of Spans Perpendicular to the Loading Direction of Seismic Wave,while for component-level,it is PGA> Site Category> Fortification Intensity> Height of Standard Story >Number of Structure Stories> Span Parallel to the Loading Direction of Seismic Wave > Number of Spans Parallel to the Loading Direction of Seismic Wave > Number of Spans Perpendicular to the Loading Direction of Seismic Wave.