Research On High Precision Seismic Inversion Algorithm Based On Fully Convolutional Network

With the continuous development of high-speed railways construction in China,it has great significance that how to effectively use the high-speed railways as a new repeatable and broad band discrete spectrum artificial seismic source to realize high-precision detection of surface structure and physical properties near high-speed railways viaduct.The high-speed railways seismic waves excited by the high-speed trains have obvious scale effects.That is,the response is related to the microstructure characteristics of the medium.High-speed railways seismic waves can sense changes at the microstructure scale.Therefore,it is feasible to use the inversion algorithm to construct a microstructure scale parameter model from the high-speed railway seismic records.Based on the previous research results of the research group combining theory and actual experimental data,deep learning algorithm is used to reverse the characteristic length scale parameter of medium from synthetic seismic data.The characteristic length scale parameter of medium represents microstructure scale characteristics of medium.The validity of the inversion of characteristic length scale parameter of medium is verified.It lays a foundation for the subsequent real-time monitoring of the surface geological state of the high-speed railways foundation.In this paper,based on the previous research results of numerical modelling and response analysis of high-speed railways seismic waves of the research group,a high precision seismic inversion strategy based on fully convolutional network(FCN)is proposed to invert the characteristic length scale parameter of medium.It mainly includes the following parts:(1)Firstly,the computational efficiency and search accuracy are improved by optimizing the conventional quantum particle swarm optimization(QPSO)algorithm,and an optimized finite difference(FD)operator is proposed based on the random-enhanced QPSO algorithm to suppress the interference of numerical dispersion(the error due to the substitution of the differential operator by the difference operator)to seismic data and achieve high precision numerical modelling.The optimized FD operator is used to realize the numerical modelling of elastic waves.The numerical modelling results indicate that the optimized FD operator based on the random-enhanced QPSO algorithm can effectively reduce numerical dispersion and achieve more accurate numerical modelling of wave field.The high precision numerical modelling method can better express the effect of discontinuous microstructure on seismic waves.It is also one of the necessary means to realize the inversion of the characteristic length scale parameter of medium.(2)Secondly,based on the asymmetric elastic wave equations derived by the research group,the model of the characteristic length scale parameter of medium is introduced into the numerical modelling,with the application of FCN and the optimized FD operator based on the random-enhanced QPSO algorithm,an inversion strategy based on FCN is proposed.The strategy includes:(1)Using the asymmetric elastic wave equations to generate synthetic seismic records that include scale effects.(2)Establishing the nonlinear mapping relationship between synthetic seismic records and the characteristic length scale parameter of medium.The model of the characteristic length scale parameter of medium,which represents microstructure scale characteristics of medium,is extracted from the synthetic seismic record containing the scale effect.(3)Appling to different models to verify the effectiveness of the inversion method of the characteristic length scale parameter of medium.(3)Finally,the characteristic length scale parameter of the medium is inverted and the result is analyzed.The research contents include the construction of velocity model,density model and characteristic length scale parameter of medium model,the generation of synthetic seismic data,the preprocessing of synthetic seismic data,and inversion results analysis on layered homogeneous model and multilayer salt-model.By inverting different types of models of the characteristic length scale parameter of medium,the inversion effect of the characteristic length scale parameter of medium based on FCN is analyzed.The effectiveness of the high-precision inversion method for the characteristic length scale parameter of medium based on FCN is evaluated.To sum up,the theory and experimental results show that the proposed FD operator based on the random-enhanced QPSO algorithm can effectively suppress numerical dispersion and better express the scale effect of seismic waves when applied to numerical modelling.At the same time,the proposed seismic inversion strategy based on FCN can directly invert characteristic length scale parameter of medium through the row seismic data.This indicates that the influence of microstructure scale characteristics of medium on seismic waves is not negligible.The high-precision seismic inversion algorithm based on FCN relies on the research of high-speed railways seismology.Not only the complexity of high-speed trains as a special moving source should be considered,but also the complexity of the medium interaction in the propagation process of high-speed railways seismic waves should be considered.It will be helpful for the real-time monitoring of the surface geological state of the high-speed railways foundation.