A supervised learning inversion method and system for Rayleigh wave dispersion curve

Abstract

The present invention relates to a supervised learning inversion method and system for Rayleigh wave dispersion curve, and belongs to the technical field of seismic shear wave velocity inversion. First, an inversion equation is constructed based on a dispersion equation, and iterative optimization is performed using the inversion equation to obtain the first seismic shear wave velocity structure of each station. Then, the first seismic shear wave velocity structure of all stations is used as input, and multiple groups of velocity structure sets are determined using a sequential Gaussian simulation method, and a training velocity set and a training curve set are further constructed to train the inversion model to obtain a trained inversion model. Finally, all first Rayleigh wave dispersion curves are used as input, and the second Rayleigh wave dispersion curve of each inter-station to be estimated point is determined using a Kriging interpolation method, and the second seismic shear wave velocity structure of each inter-station to be estimated point is obtained using a trained inversion model. The present invention can simultaneously obtain the seismic shear wave velocity structure of stations and stations, and can greatly improve the lateral resolution.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic shear wave velocity inversion, and in particular to a supervised learning inversion method and system for Rayleigh wave dispersion curves with enhanced lateral resolution. Background Art

[0002] At present, the seismic shear wave velocity structure is mostly obtained by inverting the Rayleigh wave dispersion curve. However, this method can only obtain the seismic shear wave velocity structure of a station, but cannot obtain the seismic shear wave velocity structure between stations. In engineering applications, there is a problem of low lateral resolution of the inversion results.

[0003] Based on this, an inversion technology that can improve the lateral resolution is urgently needed. Summary of the invention

[0004] The purpose of the present invention is to provide a supervised learning inversion method and system for Rayleigh wave dispersion curve, which can simultaneously obtain the seismic shear wave velocity structure of stations and between stations, and greatly improve the lateral resolution.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A supervised learning inversion method for Rayleigh wave dispersion curve, the inversion method comprising:

[0007] For each station, a first initial seismic shear wave velocity structure and a first initial Rayleigh wave dispersion curve of the station are obtained; an inversion equation is constructed based on a dispersion equation; the first initial seismic shear wave velocity structure and the first initial Rayleigh wave dispersion curve are used as inputs, and the inversion equation is used for iterative optimization to obtain the first seismic shear wave velocity structure of the station; the Rayleigh wave dispersion curve is a curve showing the change of Rayleigh wave phase velocity with frequency;

[0008] Taking the first seismic shear wave velocity structure of all the stations as input, a plurality of velocity structure sets are determined by sequential Gaussian simulation method; the velocity structure sets include the second initial seismic shear wave velocity structure of each inter-station to-be-estimated point; the inter-station to-be-estimated point is a grid node between two adjacent stations;

[0009] For each of the velocity structure sets, the velocity structure set and all the first seismic shear wave velocity structures form a training velocity set;

[0010] For each of the training velocity sets, taking the first seismic shear wave velocity structure as input, using the dispersion equation to calculate a first Rayleigh wave dispersion curve, taking the second initial seismic shear wave velocity structure as input, using the dispersion equation to calculate a second initial Rayleigh wave dispersion curve, all of the first Rayleigh wave dispersion curves and all of the second initial Rayleigh wave dispersion curves constitute a training curve set;

[0011] Using the training curve set as input and the training velocity set as output, the inversion model is trained to obtain a trained inversion model;

[0012] Taking all the first Rayleigh wave dispersion curves as input, the second Rayleigh wave dispersion curve of each of the inter-station points to be estimated is determined using the Kriging interpolation method; taking all the first Rayleigh wave dispersion curves and all the second Rayleigh wave dispersion curves as input, the second seismic shear wave velocity structure of each of the inter-station points to be estimated is obtained using the trained inversion model.

[0013] A supervised learning inversion system for Rayleigh wave dispersion curves, the inversion system comprising:

[0014] The station velocity inversion module is used for obtaining, for each station, a first initial seismic shear wave velocity structure and a first initial Rayleigh wave dispersion curve of the station; constructing an inversion equation based on a dispersion equation; taking the first initial seismic shear wave velocity structure and the first initial Rayleigh wave dispersion curve as input, performing iterative optimization using the inversion equation to obtain the first seismic shear wave velocity structure of the station; the Rayleigh wave dispersion curve is a curve of the change of Rayleigh wave phase velocity with frequency;

[0015] The inter-station initial velocity determination module is used to determine multiple groups of velocity structure sets using the first seismic shear wave velocity structure of all the stations as input and using the sequential Gaussian simulation method; the velocity structure set includes the second initial seismic shear wave velocity structure of each inter-station to-be-estimated point; the inter-station to-be-estimated point is a grid node between two adjacent stations;

[0016] A training module is used for, for each of the velocity structure sets, using the velocity structure set and all the first seismic shear wave velocity structures to form a training velocity set; for each of the training velocity sets, using the first seismic shear wave velocity structure as input, using the dispersion equation to calculate a first Rayleigh wave dispersion curve, using the second initial seismic shear wave velocity structure as input, using the dispersion equation to calculate a second initial Rayleigh wave dispersion curve, all the first Rayleigh wave dispersion curves and all the second initial Rayleigh wave dispersion curves to form a training curve set; using the training curve set as input and the training velocity set as output, training an inversion model to obtain a trained inversion model;

[0017] The inter-station velocity inversion module is used to use all the first Rayleigh wave dispersion curves as input and use the Kriging interpolation method to determine the second Rayleigh wave dispersion curve of each inter-station point to be estimated; use all the first Rayleigh wave dispersion curves and all the second Rayleigh wave dispersion curves as input and use the trained inversion model to obtain the second seismic shear wave velocity structure of each inter-station point to be estimated.

[0018] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0019] The present invention is used to provide a supervised learning inversion method and system for Rayleigh wave dispersion curves. First, the first initial seismic shear wave velocity structure and the first initial Rayleigh wave dispersion curve of the station are used as inputs, and the inversion equation is used for iterative optimization to obtain the first seismic shear wave velocity structure of each station. Then, the first seismic shear wave velocity structure of all stations is used as input, and multiple groups of velocity structure sets are determined by sequential Gaussian simulation method, and training velocity sets and training curve sets are further constructed to train the inversion model to obtain a trained inversion model. Finally, all the first Rayleigh wave dispersion curves are used as inputs, and the second Rayleigh wave dispersion curve of each inter-station to be estimated point is determined by Kriging interpolation method, and the second seismic shear wave velocity structure of each inter-station to be estimated point is obtained by using the trained inversion model. The present invention can simultaneously obtain the seismic shear wave velocity structure of the station and the inter-station, and can greatly improve the lateral resolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 A method flow chart of the inversion method provided in Example 1 of the present invention;

[0022] Figure 2 This is a system block diagram of the inversion system provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] The purpose of the present invention is to provide a supervised learning inversion method and system for Rayleigh wave dispersion curve, which can simultaneously obtain the seismic shear wave velocity structure of stations and between stations, greatly improve the lateral resolution, and obtain the seismic shear wave velocity structure with higher lateral resolution.

[0025] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] Embodiment 1:

[0027] This embodiment is used to provide a supervised learning inversion method for Rayleigh wave dispersion curve, such as Figure 1 As shown, the inversion method includes:

[0028] S1: For each station, obtain the first initial seismic shear wave velocity structure and the first initial Rayleigh wave dispersion curve of the station; construct an inversion equation based on the dispersion equation; use the first initial seismic shear wave velocity structure and the first initial Rayleigh wave dispersion curve as input, use the inversion equation to perform iterative optimization, and obtain the first seismic shear wave velocity structure of the station; the Rayleigh wave dispersion curve is a curve of the change of Rayleigh wave phase velocity with frequency;

[0029] The station in this embodiment is a seismic station, which refers to a basic seismic observation institution having at least one seismic observation site, observation facilities and implementation management functions.

[0030] In S1, obtaining the first initial seismic shear wave velocity structure of the station may include: first obtaining the rock type (or lithology) of each layer of the station, and then for each layer, determining the initial velocity of the layer according to the rock type of the layer, specifically, the rock type corresponds to the velocity range, such as the shear wave velocity range of sedimentary basalt is 3050-4500m / s, randomly selecting a velocity from the velocity range as the initial velocity of the layer, and the initial velocities of all layers constitute the first initial seismic shear wave velocity structure, so as to roughly estimate the first initial seismic shear wave velocity structure of the station according to the rock type of each layer of the station, and the first initial seismic shear wave velocity structure includes the initial velocity of each layer. Since the rock type and the seismic shear wave velocity structure only roughly correspond, it is still necessary to determine the accurate first seismic shear wave velocity structure of the station through subsequent inversion, and the first initial seismic shear wave velocity structure will be used as an initial model for subsequent inversion.

[0031] In S1, obtaining the first initial Rayleigh wave dispersion curve of the station may include: collecting and acquiring time domain seismic waveform data of the station; extracting Rayleigh wave components of the time domain seismic waveform data, generally, the signals on both sides of the seismic trace except for obvious reflection signals are regarded as Rayleigh waves, so the Rayleigh wave components can be obtained by directly cutting this part of the signal, and the Rayleigh wave components can be understood as the curve of the change of Rayleigh wave phase velocity with time; using the cross-correlation method to process the Rayleigh wave components, and estimate the first initial Rayleigh wave dispersion curve, and the Rayleigh wave dispersion curve is the curve of the change of Rayleigh wave phase velocity with frequency.

[0032] Before constructing the inversion equation based on the dispersion equation, this embodiment first determines the form of the inversion solution of the Rayleigh wave dispersion curve.

[0033] The dispersion equation generally includes the frequency f j , Rayleigh wave phase velocity C Rj , seismic shear wave velocity V s , seismic longitudinal wave velocity V p , density ρ and thickness d, which are in the following form:

[0034] F(f j ,C Rj ,V s ,V p ,ρ,d)=0 (j=1,2,…,m);

[0035] Where m represents the number of frequencies; C Rj is a real number, V s ,V p ,ρ are N-dimensional vectors, d is an N-1-dimensional vector, and N is the number of layers. In the subsequent inversion process, when the Rayleigh wave dispersion curve is known, the parameter V in the dispersion equation can be solved s ,V p ,ρ and d. Usually, the Rayleigh wave phase velocity is only sensitive to the seismic shear wave velocity, and is not sensitive to the seismic longitudinal wave velocity, density and thickness. Therefore, the seismic longitudinal wave velocity, density and thickness can all be used as known quantities. Furthermore, the dispersion equation can be regarded as j The Rayleigh wave phase velocity and seismic shear wave velocity V are described below: s An implicit function of m dispersion equations is solved to obtain the Rayleigh wave dispersion curve corresponding to a certain seismic shear wave velocity structure. Therefore, the solution of the inversion of the Rayleigh wave dispersion curve is the seismic shear wave velocity structure. Based on this, this embodiment uses the roughly estimated first initial seismic shear wave velocity structure as the initial model for inversion, performs inversion processing on the initial model, and finally obtains a more accurate first seismic shear wave velocity structure. The initial model is equivalent to the starting model of the seismic shear wave velocity structure, and more accurate results can be obtained through inversion.

[0036] This embodiment can collect relevant geological data of the work area to determine the seismic longitudinal wave velocity, density and thickness, so as to collect prior geological information in advance. Among them, the thickness of each layer is determined as d = (d1, d2, ..., d N-1 ), N represents the number of layers, and the last layer is assumed to be a uniform half-space with infinite thickness.

[0037] In this embodiment, constructing the inversion equation based on the dispersion equation may include:

[0038] (1) Perform Taylor expansion on the dispersion equation to obtain the first function;

[0039] Assume that the seismic shear wave velocity structure of the underground medium at the seismic station location is V s =(v s1 ,v s2 ,…,v sN ) T , the Rayleigh wave phase velocity is C R =(C R1 ,C R2 ,…,C Rm ) T , the Rayleigh wave phase velocity can be obtained according to the Rayleigh wave dispersion curve. Taylor expansion of the dispersion equation is performed, and the first function obtained is:

[0040] JΔV s =ΔC R ;

[0041] Where J is the Jacobian matrix; ΔV s Indicates a small disturbance in the velocity structure of the first earthquake shear wave. In the subsequent iterative optimization process of inversion, this small disturbance represents the difference in the velocity structure of the first earthquake shear wave between two adjacent iterations; ΔC R It represents a small perturbation of the Rayleigh wave phase velocity. In the subsequent iterative optimization process of inversion, this small perturbation represents the difference between the Rayleigh wave phase velocities in two adjacent iterations.

[0042] (2) constructing a first objective function based on the first function;

[0043] In this embodiment, the first objective function is as follows:

[0044]

[0045] Where Φ is the first objective function; J is the Jacobian matrix; ΔV s is the difference between the first seismic shear wave velocity structure in two consecutive iterations; ΔC R is the difference in the Rayleigh wave phase velocity between two adjacent iterations; W is a positive definite diagonal weight matrix used to balance the residual effects of different frequencies. W can be a known matrix given artificially or a covariance matrix of data errors. W = LT L, where L is a diagonal matrix and λ is the regularization parameter.

[0046] (3) Minimize the first objective function to obtain the inversion equation.

[0047] The first objective function is used to evaluate the inversion results and solve the drop. The first seismic shear wave velocity structure obtained by inversion should make the objective function value as small as possible. Minimize Φ to obtain the inversion equation. Minimization is to instruct the first objective function to minimize ΔV s The process of finding the minimum value of a function by taking the partial derivative and setting it to 0. The inverse equation is used to describe the amount of descent obtained by solving the descent direction of the objective function, as follows:

[0048] ΔV s =V(Λ 2 +λI) -1 ΛU T d;

[0049] Among them, A=LJ is subjected to singular value decomposition to obtain A=UΛV T , U, Λ, V are the left eigenvector matrix, diagonal matrix and right eigenvector matrix respectively; I is the unit matrix; d = LΔC R represents the weighted phase velocity vector.

[0050] Based on the above inversion equation, the following iterative format can be constructed:

[0051] (ΔV s ) n+1 =V(Λ 2 +λI) -1 ΛU T (d) n ;

[0052] (V s ) n+1 =(V s ) n +(ΔV s ) n+1 ;

[0053] Where (ΔV s ) n+1 represents the increment of seismic shear wave velocity structure in the n+1th iteration; (d) n represents the weighted phase velocity vector in the nth iteration; (V s ) n+1 and (V s ) n They represent the seismic shear wave velocity structures of the n+1th and nth iterations respectively.

[0054] This embodiment uses the first initial seismic shear wave velocity structure and the first initial Rayleigh wave dispersion curve as input, and uses the above inversion equation to perform continuous iterative optimization until (ΔV s ) n+1 When the modulus is less than a certain threshold, such as 0.001, the loop is exited to obtain the first earthquake shear wave velocity structure of the seismic station.

[0055] Since the first objective function contains a regularization parameter λ, this embodiment can use the L curve method to quantify the relationship between the regularization parameter and the first objective function, by setting the value range of the regularization parameter λ to [100, 0.01], and determining its value by the L curve method, and then outputting the inversion result. Specifically, a plurality of values ​​are randomly selected within the value range of the regularization parameter λ; for each value, the above-mentioned inversion process is performed once to obtain the seismic shear wave velocity structure, and then the seismic shear wave velocity structure is brought into the first objective function to obtain the first function value (i.e., the fitting difference before the plus sign) and the second function value (i.e., the roughness value after the plus sign) of the first objective function, thereby obtaining the fitting difference and roughness value corresponding to each value; with the fitting difference as the ordinate, the roughness value as the abscissa, and the fitting difference and roughness value corresponding to all values ​​as data points, L curve fitting is performed, and the regularization parameter at the inflection point is used as the optimal regularization parameter. The seismic shear wave velocity structure corresponding to the optimal regularization parameter is the first seismic shear wave velocity structure of the station output by S1.

[0056] Through S1, the inversion process of the seismic shear wave velocity structure of the underground medium at the station location can be completed, and the first seismic shear wave velocity structure of each station can be obtained.

[0057] This embodiment uses the sequential Gaussian simulation method to estimate the seismic shear wave velocity structure of the underground medium between two seismic stations, and obtains supervised learning training samples with spatial correlation, randomness and diversity, so as to use a deep neural network to estimate the seismic shear wave velocity structure between the stations with higher resolution, see subsequent S2-S4.

[0058] S2: Taking the first seismic shear wave velocity structure of all the stations as input, a plurality of velocity structure sets are determined by sequential Gaussian simulation method; the velocity structure sets include the second initial seismic shear wave velocity structure of each inter-station to-be-estimated point; the inter-station to-be-estimated point is a grid node between two adjacent stations;

[0059] This embodiment discretizes the underground medium in the working area to obtain multiple grid nodes. The working area includes each station in S1, and the seismic shear wave velocity structure of the grid node corresponding to the station is calculated by S1. The grid nodes between stations are the inter-station points to be estimated, and their seismic shear wave velocity structure is calculated by S2-S4. The positions of the inter-station points to be estimated can be planned in advance. On the basis of the known positions of each station, some grid nodes are evenly selected between the stations as the inter-station points to be estimated, and the inter-station points to be estimated are evenly distributed.

[0060] S2 is a random simulation of the seismic shear wave velocity structure between stations. The input is the first seismic shear wave velocity structure of each station obtained by inversion in S1, and the output is a set of M groups of seismic shear wave velocity structures between stations (i.e., M groups of velocity structure sets, each group of velocity structure sets includes a second initial seismic shear wave velocity structure of each inter-station point to be estimated). The determination of the M value is empirical and subjective to a certain extent. It is generally not less than the number of depth domain velocity parameters of the medium below each station location. The number of depth domain velocity parameters is the number of station layers N, in order to enhance the diversity of random simulation data. For example, if the number of depth domain velocity parameters is 10, the M value should be greater than or equal to 10, that is, the number of groups of velocity structure sets is greater than or equal to 10.

[0061] In this embodiment, the first seismic shear wave velocity structure of each station can be used to estimate the second initial seismic shear wave velocity structure of the inter-station estimated point at any position between stations within the depth range of the working area by using the sequential Gaussian simulation method in geostatistics, thereby generating M sets of velocity structure sets. Specifically, in S2, taking the first seismic shear wave velocity structure of all stations as input, using the sequential Gaussian simulation method to determine multiple sets of velocity structure sets may include:

[0062] (1) Randomly define the estimation order of multiple groups of all inter-station estimated points to obtain multiple random simulation orders;

[0063] (2) For each random simulation order, the first seismic shear wave velocity structure of all stations is used as input, and the second initial seismic shear wave velocity structure of each inter-station point to be estimated is determined by the sequential Gaussian simulation method to obtain the velocity structure set corresponding to the random simulation order; the velocity structure sets corresponding to all random simulation orders are composed of multiple groups of velocity structure sets.

[0064] S3: for each of the velocity structure sets, the velocity structure set and all the first seismic shear wave velocity structures form a training velocity set; for each of the training velocity sets, the first seismic shear wave velocity structure is used as input, and the dispersion equation is used to calculate a first Rayleigh wave dispersion curve; the second initial seismic shear wave velocity structure is used as input, and the dispersion equation is used to calculate a second initial Rayleigh wave dispersion curve; all the first Rayleigh wave dispersion curves and all the second initial Rayleigh wave dispersion curves form a training curve set; the training curve set is used as input and the training velocity set is used as output to train the inversion model to obtain a trained inversion model;

[0065] This embodiment can obtain M groups of training velocity sets, each group of training velocity sets includes a velocity structure set and the first seismic shear wave velocity structure of each station, and then each group of training velocity sets is used as input, and the training curve set corresponding to the group of training velocity sets can be obtained by using the dispersion equation, and each group of training curve sets is used as sample data, and the training velocity set corresponding to the group of training curve sets is used as label data to construct a data set.

[0066] Among them, taking the training velocity set as input and using the dispersion equation to obtain the training curve set corresponding to the training velocity set may include: taking each first seismic shear wave velocity structure in the training velocity set as input, and using the dispersion equation to obtain the first Rayleigh wave dispersion curve of each first seismic shear wave velocity structure; taking each second initial seismic shear wave velocity structure in the training velocity set as input, and using the dispersion equation to obtain the second initial Rayleigh wave dispersion curve of each second initial seismic shear wave velocity structure; all the first Rayleigh wave dispersion curves and all the second initial Rayleigh wave dispersion curves constitute the training curve set corresponding to the training velocity set.

[0067] The inversion model of this embodiment is a fully connected deep neural network. The number of layers of the established fully connected deep neural network is m (generally greater than or equal to 5), and the number of neurons in each layer is N. p ,p=1,2,…,m, (generally no less than 10 per layer), and its activation function is the ReLU function.

[0068] In this embodiment, the inversion model is trained by the cross-validation method, and the training process of the inversion model may include: dividing the above-established data set into a training set, a test set and a validation set according to a ratio of 6:2:2. The cross-validation method is used to perform different divisions on the training set and the test set, and the training is performed using the divided training set. After the training is completed, the test is performed using the divided test set. According to the error of the test set, the weight is trained using the stochastic gradient descent method to adjust the weight parameters, and a training process is completed. The generalization ability of the inversion model is evaluated by the validation set, and its evaluation standard has a certain degree of experience. It is generally believed that if the error of the validation set is slightly higher than the error of the test set (that is, the error of the validation set and the error of the test set are less than the preset value), the training result is acceptable, and the training is terminated. Otherwise, the next training is performed until the requirements are met or the maximum number of iterations is reached, thereby obtaining a trained inversion model.

[0069] S4: Using all the first Rayleigh wave dispersion curves as input, determine the second Rayleigh wave dispersion curve of each of the inter-station points to be estimated using the Kriging interpolation method; using all the first Rayleigh wave dispersion curves and all the second Rayleigh wave dispersion curves as input, obtain the second seismic shear wave velocity structure of each of the inter-station points to be estimated using the trained inversion model.

[0070] The estimation process of the second seismic shear wave velocity structure of the inter-station to-be-estimated point may include: inputting the first Rayleigh wave dispersion curve obtained through the first seismic shear wave velocity structure of each seismic station, taking this as known data, using the Kriging interpolation method in geological statistics to estimate the second Rayleigh wave dispersion curve of the inter-station to-be-estimated point A at any position between the stations, and using the second Rayleigh wave dispersion curve of the inter-station to-be-estimated point A as the input variable of the trained inversion model, and then outputting the second seismic shear wave velocity structure of the inter-station to-be-estimated point A. By repeating this process, the seismic shear wave velocity structure of the underground medium at any position in the working area can be obtained, thereby obtaining the seismic shear wave velocity structure of the underground medium in the entire working area.

[0071] This embodiment is different from the traditional technology of determining the shear wave velocity between seismic stations through linear interpolation or neural network methods. The random simulation method adopted in S2 can increase the spatial correlation, randomness and diversity of supervised learning training samples under the premise of fully considering the spatial statistical correlation. The construction and training process of the inversion model is the guarantee for automatically obtaining the seismic shear wave velocity structure. In the inversion process of the trained inversion model, the application of the Kriging interpolation method ensures the spatial correlation between the Rayleigh wave dispersion curve between stations and the Rayleigh wave dispersion curve of the known station. The inversion result has the characteristics of spatial correlation, randomness and diversity of the training samples, thereby reflecting a higher resolution and obtaining a seismic shear wave velocity structure with a higher lateral resolution.

[0072] Embodiment 2:

[0073] This embodiment is used to provide a supervised learning inversion system for Rayleigh wave dispersion curves. Figure 2 As shown, the inversion system comprises:

[0074] The station velocity inversion module M1 is used for obtaining the first initial seismic shear wave velocity structure and the first initial Rayleigh wave dispersion curve of each station; constructing an inversion equation based on the dispersion equation; taking the first initial seismic shear wave velocity structure and the first initial Rayleigh wave dispersion curve as input, using the inversion equation to perform iterative optimization to obtain the first seismic shear wave velocity structure of the station; the Rayleigh wave dispersion curve is a curve of the change of Rayleigh wave phase velocity with frequency;

[0075] The inter-station initial velocity determination module M2 is used to determine multiple groups of velocity structure sets using the first seismic shear wave velocity structure of all the stations as input, and using the sequential Gaussian simulation method; the velocity structure set includes the second initial seismic shear wave velocity structure of each inter-station to be estimated point; the inter-station to be estimated point is a grid node between two adjacent stations;

[0076] The training module M3 is used for, for each of the velocity structure sets, using the velocity structure set and all the first seismic shear wave velocity structures to form a training velocity set; for each of the training velocity sets, using the first seismic shear wave velocity structure as input, using the dispersion equation to calculate a first Rayleigh wave dispersion curve, using the second initial seismic shear wave velocity structure as input, using the dispersion equation to calculate a second initial Rayleigh wave dispersion curve, all the first Rayleigh wave dispersion curves and all the second initial Rayleigh wave dispersion curves to form a training curve set; using the training curve set as input and the training velocity set as output, training the inversion model to obtain a trained inversion model;

[0077] The inter-station velocity inversion module M4 is used to use all the first Rayleigh wave dispersion curves as input and use the Kriging interpolation method to determine the second Rayleigh wave dispersion curve of each inter-station point to be estimated; use all the first Rayleigh wave dispersion curves and all the second Rayleigh wave dispersion curves as input and use the trained inversion model to obtain the second seismic shear wave velocity structure of each inter-station point to be estimated.

[0078] Each embodiment in this specification focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0079] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A supervised learning inversion method for Rayleigh wave dispersion curve, characterized in that: The inversion method comprises: For each station, a first initial seismic shear wave velocity structure and a first initial Rayleigh wave dispersion curve of the station are obtained; an inversion equation is constructed based on a dispersion equation; the first initial seismic shear wave velocity structure and the first initial Rayleigh wave dispersion curve are used as inputs, and the inversion equation is used for iterative optimization to obtain the first seismic shear wave velocity structure of the station; the Rayleigh wave dispersion curve is a curve showing the change of Rayleigh wave phase velocity with frequency; Taking the first seismic shear wave velocity structure of all the stations as input, a plurality of velocity structure sets are determined by sequential Gaussian simulation method; the velocity structure sets include the second initial seismic shear wave velocity structure of each inter-station to-be-estimated point; the inter-station to-be-estimated point is a grid node between two adjacent stations; For each of the velocity structure sets, the velocity structure set and all the first seismic shear wave velocity structures form a training velocity set; For each of the training velocity sets, taking the first seismic shear wave velocity structure as input, using the dispersion equation to calculate a first Rayleigh wave dispersion curve, taking the second initial seismic shear wave velocity structure as input, using the dispersion equation to calculate a second initial Rayleigh wave dispersion curve, all of the first Rayleigh wave dispersion curves and all of the second initial Rayleigh wave dispersion curves constitute a training curve set; Using the training curve set as input and the training velocity set as output, the inversion model is trained to obtain a trained inversion model; Taking all the first Rayleigh wave dispersion curves as input, the second Rayleigh wave dispersion curve of each of the inter-station points to be estimated is determined using the Kriging interpolation method; taking all the first Rayleigh wave dispersion curves and all the second Rayleigh wave dispersion curves as input, the second seismic shear wave velocity structure of each of the inter-station points to be estimated is obtained using the trained inversion model.

2. The inversion method according to claim 1, characterized in that: Acquiring the first initial seismic shear wave velocity structure of the station specifically includes: Obtain the rock type of each layer of the station; For each of the layers, the initial velocity of the layer is determined according to the rock type of the layer; the initial velocities of all the layers constitute a first initial seismic shear wave velocity structure.

3. The inversion method according to claim 1, characterized in that: Obtaining the first initial Rayleigh wave dispersion curve of the station specifically includes: Acquiring time-domain seismic waveform data of the station; extracting Rayleigh wave components of the time-domain seismic waveform data; The Rayleigh wave component is processed by using a cross-correlation method to obtain a first initial Rayleigh wave dispersion curve.

4. The inversion method according to claim 1, characterized in that: The inversion equation constructed based on the dispersion equation specifically includes: Taylor expansion is performed on the dispersion equation to obtain the first function; A first objective function is constructed based on the first function; the first objective function is Where Φ is the first objective function; J is the Jacobian matrix; ΔV s is the difference between the first seismic shear wave velocity structure in two consecutive iterations; ΔC R is the difference between the Rayleigh wave phase velocities in two consecutive iterations; W is the positive definite diagonal weight matrix; λ is the regularization parameter; The first objective function is minimized to obtain an inversion equation.

5. The inversion method according to claim 4, characterized in that: The regularization parameter is calculated using the L-curve method.

6. The inversion method according to claim 1, characterized in that: The method of using the first seismic shear wave velocity structure of all the stations as input and using the sequential Gaussian simulation method to determine multiple groups of velocity structure sets specifically includes: Randomly define the estimation order of multiple groups of all inter-station estimated points to obtain multiple random simulation orders; For each of the random simulation orders, the first seismic shear wave velocity structure of all the stations is taken as input, and the second initial seismic shear wave velocity structure of each inter-station point to be estimated is determined by the sequential Gaussian simulation method to obtain the velocity structure set corresponding to the random simulation order; the velocity structure sets corresponding to all the random simulation orders form multiple groups of velocity structure sets.

7. The inversion method according to claim 1, characterized in that: The number of groups of the velocity structure set is greater than or equal to the number of layers of the station.

8. The inversion method according to claim 1, characterized in that: The inversion model is a fully connected deep neural network.

9. The inversion method according to claim 1, characterized in that: The training of the inversion model specifically includes: training the inversion model by using a cross-validation method.

10. A supervised learning inversion system for Rayleigh wave dispersion curves, characterized in that: The inversion system comprises: The station velocity inversion module is used for obtaining, for each station, a first initial seismic shear wave velocity structure and a first initial Rayleigh wave dispersion curve of the station; constructing an inversion equation based on a dispersion equation; taking the first initial seismic shear wave velocity structure and the first initial Rayleigh wave dispersion curve as input, performing iterative optimization using the inversion equation to obtain the first seismic shear wave velocity structure of the station; the Rayleigh wave dispersion curve is a curve of the change of Rayleigh wave phase velocity with frequency; The inter-station initial velocity determination module is used to determine multiple groups of velocity structure sets using the first seismic shear wave velocity structure of all the stations as input and using the sequential Gaussian simulation method; the velocity structure set includes the second initial seismic shear wave velocity structure of each inter-station to-be-estimated point; the inter-station to-be-estimated point is a grid node between two adjacent stations; A training module is used for, for each of the velocity structure sets, using the velocity structure set and all the first seismic shear wave velocity structures to form a training velocity set; for each of the training velocity sets, using the first seismic shear wave velocity structure as input, using the dispersion equation to calculate a first Rayleigh wave dispersion curve, using the second initial seismic shear wave velocity structure as input, using the dispersion equation to calculate a second initial Rayleigh wave dispersion curve, all the first Rayleigh wave dispersion curves and all the second initial Rayleigh wave dispersion curves to form a training curve set; using the training curve set as input and the training velocity set as output, training an inversion model to obtain a trained inversion model; The inter-station velocity inversion module is used to use all the first Rayleigh wave dispersion curves as input and use the Kriging interpolation method to determine the second Rayleigh wave dispersion curve of each inter-station point to be estimated; use all the first Rayleigh wave dispersion curves and all the second Rayleigh wave dispersion curves as input and use the trained inversion model to obtain the second seismic shear wave velocity structure of each inter-station point to be estimated.

Images