Research On The Interpolation Calculation Method Of Ground Motions And Structural Responses Under Urban Regional Earthquakes

Research on regional and urban earthquake is greatly significant to urban earthquake prevention and mitigation,as well as post-earthquake damage assessment.Nowadays,measuring seismic responses directly at measuring points using accelerometers is an effective method for monitoring the seismic responses of structures.However,it is costly and inefficient to measure vibration conditions of every building by installing seismometers everywhere under the research scenario of regional and urban earthquakes,which have impact on large regions and a great deal of building.In order to solve this problem,a method for interpolation calculation of seismic responses of target structures is proposed,which is realized by applying machine learning algorithms to process seismic observational networks’ data.The research in this paper mainly focuses on:(1)According to the requirements of seismic time history data under different application scenarios,two approaches to obtain seismic time history data are designed: 1)Downloading and analyzing time history data from ground motion databases in batches.2)Introducing finite element software to calculate and analyze soil-structure models.As to the first approach,author developed a browser plug-in to download and process seismic data in batches.As to the second approach,a finite element model considering soil-structure interaction,geo-stress balance,and viscoelastic artificial boundary is established to directly calculate the ground seismic responses and structural seismic responses.(2)The long short-term memory network is used to excavate the relationship between the seismic responses at several observation stations and those at the target structure,and the observed records at stations could be used to directly calculate the response of the target structure.Considering the discrete distribution of stations and the target structure in space,the phases of time history records at each measuring point are different in time domain.The correlation among records recorded by the stations and the target structure can be established in the time domain by introducing the long short-term memory network equipped with the time pattern attention mechanism,which improves accuracies and efficiencies of calculations in both the time domain and the spatial domain.(3)The transmission path of seismic wave is divided into two parts: propagation in soil and propagation in structures.The two steps are appointed to be solved by Gaussian process regression(GPR)and deconvolution seismic interferometry(DSI),respectively.In order to explore the influences of the original data set,time-frequency transformation method and the dimension of input value of Gaussian process regression on the calculation results,discrete Fourier coefficients and discrete wavelet coefficients are applied into two-dimensional and three-dimensional Gaussian process regression to calculate the ground seismic motions adjacent to the target structures.Besides,the calculation results of Gaussian process regression models with different dimensions were evaluated.To figure out the process of seismic wave propagation in the structures,the threedimensional deconvolution seismic interferometry model was adopted here,by means of which can fully take the interaction mechanism among orthogonal seismic waves into account.(4)The soil-steel frame system is established by using finite element method.The geo-tress balance and viscoelastic artificial boundary are considered.The specified ground motion is input into the finite element model to excite the whole system and simulate responses of the whole system under an earthquake.According to the application conditions of different interpolation calculation methods,the acceleration time histories of specific points are read from the finite element model as the original dataset,and the calculation methods based on LSTM network or GPR-DSI method are used to calculate the seismic responses of the target structures.According to the calculation results,the efficiency and accuracy of the two methods are compared,and the applicability of the two methods in different scenarios is analyzed.