The Study On Full Waveform Inversion Based On Low-frequency Seismic Wavefield Reconstruction

As the focus of exploration for oil and gas resources is gradually shifting to the complex structural and deep areas,the requirements for imaging quality of underground structures are also increasing.High-precision velocity construction is the key to improving the quality of seismic data migration imaging.At present,the velocity construction methods using seismic data mainly include velocity analysis,tomography,and full waveform inversion.Among them,the first two methods only use the travel time information of the seismic wavefield,which theoretically limits the accuracy and resolution of their results.Full waveform inversion is currently the velocity construction method with the highest accuracy in seismic exploration field.It uses the norm of the residuals between synthetic data and observed data as objective function,and can comprehensively utilize the kinematic and dynamic characteristics of the seismic wavefield.Although full waveform inversion is currently the most promising method to realize high-precision velocity construction,many problems still remain to be solved.Among them,the most famous and also the most urgent problem is the initial model dependency problem of full waveform inversion,or the cycle-skipping problem caused by poor initial velocity models.The root of this problem is that full waveform inversion uses Born approximation to transform the strong nonlinear inversion problem into a local optimization problem.Although this facilitates the calculation,the inversion is easy to fall into local minima when the initial model is far away from the true model.In this paper,three methods and strategies are proposed to overcome the problems of initial model dependency and cycle-skipping of full waveform inversion.The first one is to propose a joint full waveform inversion method for multi-source seismic data,applying the passive source seismic data to the inversion process of conventional active source seismic data,and to make full use of the low frequency advantages of passive source seismic data.The second one is to propose a multi-scale full-waveform inversion method based on sparse constraints,using sparse constraints and sparse blind constraints to reconstruct low-frequency seismic data,thereby constructing a more accurate large-scale velocity model.The third one is to propose a low frequency reconstruction method and envelope-waveform inversion method based on the envelope operator.Using the ultra-low frequency information of the envelope wavefield,a large-scale strong perturbation velocity model is constructed.The main methods proposed in this paper and the main results achieved can be summarized as follows:(1)From the perspective of imaging,the advantages of the combined application of active and passive source seismic data are comprehensively studied.A passive source seismic data reconstruction method based on the multitaper-spectral method is proposed,which can suppress the reconstruction noise while maintaining the relative amplitude characteristics of the waveform.According to the noise characteristics of passive source reconstruction records,a multi-domain iterative denoising algorithm is proposed.In the time domain,joint interpolation of active and passive seismic data and joint data imaging methods are proposed.The complementary effects of active and passive seismic data under different acquisition conditions are studied.The results demonstrate that a small amount of active seismic data can restrain and stimulate the passive source imaging in the active area where the passive source is relatively active.In the case of sparse acquisition of the active seismic data,the reconstructed passive seismic data are used to interpolate the shot gathers,which can enhance the continuity of the imaging results and provide more detailed structural information.Using the low-frequency advantages of passive seismic data,an energy-matching low-frequency compensation method is proposed to reconstruct the missing low-frequency active seismic data,and the deep imaging quality is significantly improved.(2)Joint acoustic and elastic wave full waveform inversion methods using active and passive seismic data are proposed.In order to solve the problems of weak signal identification,source location uncertainty and source wavelet unknown for passive seismic data,a passive seismic full waveform inversion method based on seismic interferometry and source-independent algorithm is proposed.The effects of passive sources number and recording time on the inversion results are discussed separately.It is found that the larger the number of sources and the longer the recording time,the more favorable the construction of the macro-velocity model is.It is proposed to use the low-frequency information of the passive seismic data to construct the initial velocity model for the full waveform inversion of active seismic data,and the corresponding inversion strategies are given.The above inversion methods and strategies are extended to elastic media,and a full waveform inversion method for elastic multi-component passive seismic data is proposed.The study found that the passive virtual source data that corresponding to different source and receiver components contribute differently to P-and S-wave velocity inversion.The final series inversion strategy using elastic active and passive seismic data can not only obtain accurate large-scale background information of P-and S-wave velocity,but also obtain accurate small-scale details of P-and S-wave velocity.(3)Multiscale full waveform inversion methods for acoustic and elastic waves based on sparse constraints are proposed.The sparse constraint inversion can extract the broadband reflection impulse response of seismic data,thereby enhancing the low-frequency components of the seismic data.The wavelet convolution reconstruction method can not only extract the frequency band information to be used,but also ensure the wavelet determinism of the reconstructed data.The sparse blind constrained low-frequency reconstruction algorithm combines sparse constraints with Tikhonov regularization constraints to form the objective function,and iteratively solves the reflected impulse response and source wavelet,and can finally obtain the low-frequency reconstructed data while correcting the source wavelet error.Using sparsely constrained low-frequency reconstruction data as observed data in sequence,multiscale full waveform inversion can be performed.Combined with multi-source hybrid coding strategy,efficient multi-scale full waveform inversion can be achieved.The above methods were extended to elastic media,and a multi-scale elastic wave full waveform inversion method based on sparse constraint is proposed.In order to suppress the crosstalk caused by multi-source aliasing,an elastic wave full waveform inversion method based on the anisotropic total variation constraint is proposed.(4)In order to use the ultra-low frequency information of the envelope wavefield more effectively,envelope-waveform inversion based on hybrid scale separation and source-independent direct envelope-waveform inversion are proposed.Based on the modulation signal model,the shortcomings of the conventional envelope demodulation effect are analyzed,and the demodulation on the basis of the improved linear scale separation is proposed,namely the hybrid scale separation method.The method can further purify the long wavelength response of the media contained in the envelope data,and has the advantage of suppressing strong low-frequency noise.The envelope-waveform inversion based on hybrid scale separation can still obtain good inversion results under strong low-frequency noise.Conventional full waveform inversion and conventional envelope inversion are based on the waveform Fréchet derivative,which is difficult to invert large-scale and strong perturbation media.Direct envelope inversion overcomes this shortcoming,but it is sensitive to source wavelet errors.This paper proposed a source-independent direct envelope-waveform inversion method.It can not only obtain large-scale strong perturbation velocity structures,but also can construct relatively accurate small-scale weak perturbation velocity structures,and can also overcome the effects of source-wavelet errors on the final inversion results.