Transverse continuous seismic wave impedance inversion method based on knowledge migration

Abstract

The invention relates to the field of geophysical exploration, in particular to a transverse continuous seismic wave impedance inversion method based on knowledge migration. The method comprises the steps that post-stack reflection seismic data are acquired, and a wave impedance inversion method based on a multi-group variation differential evolution algorithm is used for the first channel of seismic data of the post-stack reflection seismic data; for each subsequent channel of seismic data, the inversion optimal wave impedance model of the previous adjacent channel is used as a useful knowledge vector to be migrated to the evolution of the current seismic channel, the migration of the optimal solution is constrained by the constructed mapping matrix, and a migration knowledge vector is formed; and performing wave impedance inversion of the current seismic trace by using a multi-group variation differential evolution algorithm based on knowledge migration to obtain a transverse continuous wave impedance model. According to the method, a more accurate inversion wave impedance model can be obtained, and the transverse continuity of a result is powerfully ensured; and meanwhile, the convergence speed and the convergence precision of the algorithm are greatly improved, and the method is more suitable for depicting lithologic oil and gas reservoir wave impedance parameters.

Description

Technical Field

[0001] The invention relates to the field of geophysical exploration, and in particular to a lateral continuous seismic wave impedance inversion method based on knowledge migration. Background Art

[0002] Resources buried in underground rock formations, such as oil and natural gas, are important strategic resources related to national security and national economic development. For a long time in the foreseeable future, all walks of life will still be unable to get rid of their high dependence on oil and natural gas resources. At present, the most important oil and gas exploration resources in my country are lithological oil and gas reservoirs, and seismic exploration methods are the main methods in oil and gas exploration. In this method, the distribution of underground oil and gas can be revealed by medium parameters such as the wave impedance of the underground medium. However, my country's lithological oil and gas reservoirs have a thin interlayer structure vertically and strong heterogeneity in the horizontal direction in terms of geological structure. Therefore, how to finely characterize the wave impedance parameters of underground lithological oil and gas reservoirs is a key issue faced in seismic exploration.

[0003] For the wave impedance inversion problem, the commonly used inversion methods are mainly linearized inversion methods based on gradient information and nonlinear inversion methods based on global optimization. For the local optimization inversion method based on gradient information, since the gradient information of the target functional is used, this method is more efficient, but it is easy to fall into the local optimal solution and cannot converge to the optimal parameter model. The nonlinear inversion method based on global optimization (such as particle swarm optimization algorithm, differential evolution algorithm, etc.) solves the problems existing in local optimization, avoids falling into local extreme points, and improves the accuracy of the inversion results. For example, CN107390269B discloses a wave impedance inversion particle swarm optimization algorithm constrained by the statistical characteristics of logging data, which specifically includes: 1) performing phase correction on seismic data to make it a zero-phase record; 2) extracting the wavelet amplitude spectrum and obtaining the seismic wavelet; 3) analyzing and statistically analyzing the velocity information based on the existing logging data in the data volume or profile, and establishing a feasible domain for inverting the reflection coefficient, that is, the search space of the reflection coefficient; 4) inverting the reflection coefficient using the improved particle swarm optimization algorithm proposed by the present invention; 5) calculating the wave impedance based on the inverted reflection coefficient. CN112182481B discloses a seismic waveform inversion method and system based on an improved differential evolution algorithm, the method includes: step 1, calculating and obtaining the calculated seismic data of the area to be explored, measuring and obtaining the actual seismic data of the area to be explored; taking the fitting degree of the calculated seismic data and the actual seismic data as the objective function; step 2, using the improved differential evolution algorithm to optimize the objective function obtained in step 1, obtaining the physical parameters of the underground medium model of the area to be explored, and completing the seismic waveform inversion.

[0004] In practice, since underground media and seismic wavelets usually change continuously, there is an inevitable connection between multiple adjacent seismic traces. However, the global optimization algorithm based on the single-trace inversion strategy ignores these similarities when solving seismic inversion problems, resulting in problems such as lateral discontinuity of inversion results.

[0005] In order to overcome these problems, the model-based wave impedance inversion method usually relies on introducing appropriate constraints to make the inversion more stable and to obtain laterally continuous seismic wave impedance. The most commonly used method is to add regularization terms (such as TV regularization, ATV regularization, etc.) to the objective function of the specific inversion problem. This method relies on prior information and the form of regularization is difficult to determine in many inversion problems. In addition, some scholars use norms and Kalman filters as lateral continuity constraints, but these methods cannot achieve good results in global optimization algorithms, and the accuracy of the inversion results cannot be fully guaranteed. Summary of the invention

[0006] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a lateral continuous seismic wave impedance inversion method (KT-MMDE) based on knowledge transfer. It starts directly from seismic data, utilizes seismic wave impedance inversion and global optimization algorithm, and is inspired by the multi-task optimization problem in the field of evolutionary algorithms. The idea of ​​knowledge transfer and dynamic mapping is introduced into multi-channel wave impedance inversion, and the seismic wave impedance parameters can be accurately inverted and their lateral continuity can be guaranteed without relying on regularization constraints.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] The present invention provides a lateral continuous seismic wave impedance inversion method based on knowledge migration, the method comprising the following steps:

[0009] Obtain post-stack reflection seismic data, and for the first seismic data, use a wave impedance inversion method based on a multi-group mutation differential evolution algorithm to obtain an optimal solution, namely, an optimal inversion wave impedance model;

[0010] For each subsequent seismic data, the inversion optimal wave impedance model of the previous adjacent channel is transferred to the evolution of the current seismic channel as a useful knowledge vector, and the migration of the optimal solution is constrained by the constructed mapping matrix to form a migration knowledge vector;

[0011] The wave impedance inversion of the current seismic trace is performed using a multi-group mutation differential evolution algorithm based on knowledge transfer to obtain a laterally continuous wave impedance model.

[0012] Furthermore, the method specifically comprises the following steps:

[0013] (1) Collecting raw seismic data, and then preprocessing the collected seismic data to obtain post-stack reflection seismic data, denoted as d obs (t,n), where t represents the time variable; n is the number of the common depth point gather;

[0014] (2) Determine the density value, time sampling interval, forward model parameters and seismic wavelet;

[0015] (3) For actual seismic data, the wave impedance parameters are extracted from a real well logging data; the low-frequency wave impedance model I can be obtained by using the Butterworth filter. lf (t,n);

[0016] (4) determining the search space range of the wave impedance parameter I according to the obtained low-frequency wave impedance model, and giving the objective function to be optimized in the wave impedance inversion;

[0017] (5) Perform multi-channel wave impedance inversion. For the first seismic trace in the post-stack reflection seismic data obtained in step (1), use a multi-group mutation differential evolution algorithm to optimize the objective function of step (4). Select the individual with the smallest objective function value in the group after the algorithm iteration is completed, and record it as the optimal solution P of the current trace. The optimal solution is the optimal wave impedance model finally searched for the first seismic trace.

[0018] (6) For each subsequent seismic trace, the inversion of the previous seismic trace and the previous one can be regarded as two adjacent tasks. The optimal solution individual optimized in the previous trace can be transferred to the evolution process of the current trace to guide the direction of algorithm mutation.

[0019] (7) The optimal solution migration in step (6) is constrained by the constructed mapping matrix M, thus forming the final migration knowledge vector

[0020] (8) The migration knowledge vector in step (7) Used in the G+1th iteration of the current path to form a new neighborhood-based mutation strategy;

[0021] (9) After the iteration of step (8), the individual with the smallest objective function value in the multi-group mutation differential evolution algorithm based on knowledge transfer is selected, which is the optimal wave impedance model of the current channel.

[0022] Furthermore, in step (2), the forward model parameters and their seismic wavelets are determined according to the shot offset, effective frequency band range and sampling time of the actual seismic data.

[0023] Furthermore, in step (7), the constructed mapping matrix M changes dynamically during the evolutionary iteration of the current path. Given the parameter G m (1 <G m <Gmax ) is used to control whether the mapping matrix changes.

[0024] Furthermore, for the Gth optimization iteration of the current path, when G≠G m When , the mapping matrix is ​​a unit matrix, that is, M = I; applying it to the previous optimal solution P, the knowledge vector of the migration can be obtained

[0025] Furthermore, for the Gth optimization iteration of the current path, when G = G m When the Gth iteration of the current path is completed, a temporary optimal solution individual can be obtained, which is recorded as vector Q G ; At this time, through the known vectors P and Q G , using the constructed mapping operator Φ M (·), the operator consists of multiple linear regression operations, and the mapping matrix M = Φ M (P,Q G ); Apply it to the previous optimal solution P to obtain the transferred knowledge vector

[0026] Furthermore, in step (8), the migration knowledge vector obtained in step (7) is Used in the evolution process of the current path as the basis vector in the algorithm mutation operation.

[0027] Furthermore, based on the new mutation strategy, the mutation vector of the i-th individual in the G-th generation population is It can be expressed as:

[0028]

[0029] In the formula, represents the current individual population, a∈(0,1) is a constant, F∈[0,2] represents the disturbance factor, and n1 and n2 represent the number of two temporary variant individuals. CCDE (·) and g DE / rand / 1 (·) represent the neighborhood-based mutation operations in the “DE / rand / 1” method and the collaborative mutation differential evolution algorithm (CCDE), respectively. and They represent the mutant individuals generated by DE / rand / 1 and CCDE based on the new mutation strategy respectively.

[0030] Furthermore, and Respectively expressed by formula:

[0031]

[0032]

[0033] In the formula, and and They respectively represent individuals randomly selected from the current population, and satisfy r1≠r2≠i, m1≠m2≠i.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] Compared with the commonly used wave impedance inversion method based on global optimization algorithm, the lateral continuous seismic wave impedance inversion method based on knowledge transfer of the present invention can not only obtain a more accurate inversion wave impedance model, but also largely ensure the lateral continuity of the result. In addition, the convergence speed and convergence accuracy of the algorithm are greatly improved, so the method of the present invention is more suitable for the characterization of wave impedance parameters of lithologic oil and gas reservoirs. The reason why the present invention has the above advantages is that: first, because the global optimization algorithm is used to solve the wave impedance inversion problem, the method can invert the global optimal wave impedance parameters; secondly, the method of the present invention improves the global optimization method on the wave impedance inversion problem, and by adding the mechanism of knowledge transfer and dynamic mapping, the correlation between adjacent seismic traces is strengthened, so that the optimization algorithm evolves and mutates in a more accurate direction, thereby better constructing a seismic wave impedance model with good lateral continuity. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of the process of the present invention;

[0037] Figure 2 The following is a comparison of the wave impedance models obtained by different algorithms in the target area with gentle lateral changes: (a) the real model; (b) the impedance model obtained by MMDE; (c) the impedance model obtained by KT-MMDE (without mapping); (d) the impedance model obtained by KT-MMDE;

[0038] Figure 3 It is the convergence curve of different algorithms in the target area with gentle lateral changes;

[0039] Figure 4 The following is a comparison of the wave impedance models obtained by different algorithms in the target area with dramatic lateral changes: (a) the real model; (b) the impedance model obtained by MMDE; (c) the impedance model obtained by KT-MMDE (without mapping); (d) the impedance model obtained by KT-MMDE;

[0040] Figure 5 It is a graph of the convergence curves of different algorithms in the target area with drastic lateral changes. DETAILED DESCRIPTION

[0041] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0042] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations and / or combinations thereof.

[0043] In order to enable those skilled in the art to more clearly understand the technical solution of the present invention, the technical solution of the present invention will be described in detail below in conjunction with specific embodiments.

[0044] Example 1

[0045] like Figure 1 As shown, the lateral continuous seismic wave impedance inversion method based on knowledge transfer has the following specific steps:

[0046] 1) Collect the original seismic data, then perform conventional preprocessing on the collected seismic data, and obtain the post-stack reflection seismic data through migration imaging, which is recorded as Where t represents the time variable; n is the sequence number of the common depth point gather.

[0047] 2) According to the shot offset, effective frequency band, sampling time, etc. of the actual seismic data, the fixed density value is set to 1.0 g / cc, and the forward model parameters including time sampling interval, seismic wavelet used in forward modeling, number of sampling points, etc. are set.

[0048] 3) For model seismic data, the real wave impedance model is known; for actual seismic data, its wave impedance parameters are extracted from a real well logging data. The low-frequency wave impedance model I can be obtained through the Butterworth filter lf (t, n), the cutoff frequency of the Butterworth filter is set to f0 = 3 Hz in the present invention.

[0049] 4) The search space of wave impedance parameters is obtained by taking the low-frequency wave impedance model obtained in step 3) as the reference value, and adding and subtracting a certain set value on the basis of the reference value as the upper and lower bounds of the search space respectively; the objective function of wave impedance inversion is the absolute sum of the errors between the observed post-stack seismic data and the calculated post-stack seismic data synthesized by the model. For a post-stack seismic data, the objective function is expressed as:

[0050]

[0051] Where, d cal (t) is the forward modeled single-channel post-stack reflection seismic data obtained when the wave impedance parameters are given.

[0052] 5) For the first seismic trace in the seismic data, the present invention uses a multi-group mutation differential evolution algorithm (MMDE) to optimize the objective function defined in formula (3) to solve the optimal wave impedance parameters. The multi-group mutation differential evolution algorithm mainly includes: initialization, mutation, crossover and selection operations. The algorithm is based on a population of NP size to solve the optimization problem, and each individual in the population is a vector with the same dimension as the model parameter. The basic process of the algorithm is briefly described as follows:

[0053] Step 1: Randomly generate NP initial individuals within the search space to form a group;

[0054] Step 2: Perform mutation operation on each individual in the group;

[0055] Step 3: Perform crossover operation on each individual in the group;

[0056] Step 4: Perform selection operations on each individual in the group;

[0057] Step 5: Determine whether the algorithm termination condition (maximum number of iterations G) is met max ), if the optimal individual in the output group is satisfied, it is the solution to the optimization problem; otherwise, jump back to the second step and proceed to the next iteration. The details are as follows:

[0058] Initialization operation: For a D-dimensional optimization problem, if the upper and lower bounds of the search space are x min ={x 1,min ,x 2,min ,...,x D,min} and x max ={x 1,max ,x 2,max ,...,x D,max}, then initialize the jth dimension of the i-th individual in the population It can be expressed as follows:

[0059]

[0060] In the formula, rand i,j (0,1) is a random number uniformly distributed between [0,1]. After initialization, the multi-group mutation differential evolution algorithm will iteratively update and evolve all individuals in the population according to the three basic operations of mutation, crossover and selection until the iteration stop condition is met.

[0061] Mutation operation: This operation generates a mutant individual for the current individual. The mutation strategy in the multi-group mutation differential evolution algorithm combines the mutation strategy in the "DE / rand / 1" method and the collaborative mutation differential evolution algorithm (CCDE). The mutation vector of the i-th individual in the G-th generation population is It can be expressed as:

[0062]

[0063] In the formula, f CCDE (·) and f DE / rand / 1 (·) represent the mutation strategies in the “DE / rand / 1” method and the collaborative mutation differential evolution algorithm (CCDE), respectively. represents the current individual population, a∈(0,1) is a constant, F∈[0,2] represents the disturbance factor, and N1 and N2 represent the number of two temporary mutant individuals respectively.

[0064] Crossover operation: This operation is performed on the current individual and its corresponding variant Exchange "genes" between them to obtain a test individual The crossover operation is defined as:

[0065]

[0066] In the formula, rand i,j [0,1]∈U[0,1], Cr∈[0,1] is the crossover rate, and q∈[1,D] is a randomly selected integer, which ensures that even if Cr=0, there is one-dimensional information in the trial individual from the mutant individual.

[0067] Select an operation: This operation is based on the current individual and probing individuals The objective function value is used to update the current individual. The specific form is as follows:

[0068]

[0069] In the formula, is the updated individual, and Ψ(·) represents the objective function of the optimization problem.

[0070] 6) In multi-channel impedance inversion, in order to consider the correlation between adjacent seismic traces, for each seismic trace after the first one, its inversion with the previous seismic trace can be regarded as two adjacent tasks. The optimal solution individual P of the previous trace is [p1, p2, ..., p D ] T It can be migrated to the evolution process of the current path to guide the direction of optimization algorithm mutation.

[0071] 7) For the previous optimal solution P in step 6), it is often unreasonable to completely migrate it to the inversion evolution of the current trace. It is necessary to constrain the migration vector based on the correlation between the two seismic traces. Therefore, the mapping matrix M is introduced to constrain the previous optimal solution P that needs to be migrated. The constructed mapping matrix M changes dynamically during the evolutionary iteration of the current trace. Given the parameter G m (1 <G m <G max ), which is used to control whether the mapping matrix changes. For the Gth optimization iteration of the current path, the mapping matrix is ​​mainly divided into two cases:

[0072] Case 1: G≠G m

[0073] At this time, the mapping matrix is ​​an identity matrix, that is, M = I. Applying it to the previous optimal solution P, we can get the transferred knowledge vector In this case, it is equivalent to completely migrating the optimal solution of the previous path to the evolutionary inversion process of the current path.

[0074] Case 2: G = G m

[0075] After the Gth iteration of the current path, a temporary optimal solution individual can be obtained, which is recorded as vector At this time, through the known vectors P and Q G , using the constructed mapping operator Φ M (·), we can get the mapping matrix M = Φ M (P,Q G ). Apply it to the previous optimal solution P to obtain the transferred knowledge vector

[0076] The mapping matrix M is defined as a tridiagonal square matrix, that is, there are element values ​​on the main diagonal and the subdiagonals on both sides thereof, and the element values ​​at other positions are all 0. The mapping matrix M is expressed as follows:

[0077]

[0078] In order to ensure the accuracy and effectiveness of the migrated knowledge vector for the current trace inversion task, the migration loss function is defined as follows:

[0079]

[0080] When the migration loss value is the smallest, the optimal mapping matrix M between the two seismic traces can be obtained. * At this time, formula (10) must hold:

[0081] Q G =M *·P (10)

[0082] Therefore, the temporary optimal solution Q obtained at the end of the Gth iteration of the current seismic trace is G It can be expressed as the linear multiplication and addition of matrix elements, that is:

[0083]

[0084] Among them, the multiplication coefficient m 11 ,m 12 ,...,m D,D-1 ,m D,D This is the element value in the required mapping matrix M. Since vectors P and Q G It is known that the element value of each row in the mapping matrix M can be obtained by the linear regression operator, which is expressed as follows:

[0085]

[0086] In the formula, Φ M (·) represents a mapping operator, which consists of D sub-mapping operators Each sub-mapping operator corresponds to a linear regression operation. According to the mapping operator, the mapping matrix M of the constraint migration can be obtained.

[0087] In summary, the knowledge vector transferred between the inversion tasks of two adjacent seismic traces can be expressed as:

[0088]

[0089] 8) The migration knowledge vector obtained in step 7) It is used in the evolution process of the current track inversion as the basis vector in the mutation operation, so as to better guide the mutation direction of the candidate solution individuals. Based on the new mutation strategy, the mutation vector of the i-th individual in the G-th generation population is It can be expressed as:

[0090]

[0091] In the formula, represents the current individual population, a∈(0,1) is a constant, F∈[0,2] represents the disturbance factor, and n1 and n2 represent the number of two temporary variant individuals. CCDE (·) and g DE / rand / 1 (·) represent the neighborhood-based mutation operations in the “DE / rand / 1” method and the collaborative mutation differential evolution algorithm (CCDE), respectively. and They represent the mutant individuals generated by DE / rand / 1 and CCDE based on the new mutation strategy, and are specifically expressed as follows:

[0092]

[0093]

[0094] In the formula, and and They respectively represent individuals randomly selected from the current population, and satisfy r1≠r2≠i, m1≠m2≠i.

[0095] 9) After the iteration in step 8), the individual with the smallest objective function value in the global optimization algorithm based on knowledge transfer is selected, which is the optimal wave impedance model of the current channel.

[0096] Example 2

[0097] The method described in Example 1 is applied to the MarmousiⅡ geological model. After downsampling, the model has a total of 5670 meters in the horizontal direction and 0.8 seconds in the time direction. The model is discretized with parameters of 5 meters in spatial sampling interval and 0.008 seconds in time sampling interval. The discretized wave impedance model has a total of 1134 channels, each with 100 dimensions. Among them, channels 1-300 change slowly in the horizontal direction, and there is a low-speed anomaly around channel 200. The model of channels 400-700 changes dramatically in the horizontal direction, and faults appear. The method of the present invention is used to test the target area of ​​channels 1-300 with gentle lateral changes and the target area of ​​channels 401-700 with dramatic lateral changes.

[0098] Figure 2 This is a comparison chart of the wave impedance models obtained by different algorithms in the target area with gentle lateral changes, which respectively represent the real wave impedance model, the wave impedance model obtained by MMDE inversion, the wave impedance model obtained by KT-MMDE (without mapping), and the wave impedance model obtained by KT-MMDE inversion. Figure 2 As shown in (b), the traditional global optimization algorithm is based on single-channel inversion, so the inverted wave impedance will have obvious lateral discontinuity; Figure 2 The KT-MMDE inversion results using knowledge transfer and dynamic mapping in (d) not only greatly improve the inversion accuracy, but also well ensure the lateral continuity of the wave impedance model. In addition, in order to test the effectiveness and necessity of adding dynamic mapping to the algorithm, Figure 2 (c) shows the inversion results of the KT-MMDE algorithm without mapping, which is improved compared with the results of the traditional method, but there are large errors near the low-velocity anomaly.

[0099] Figure 3It is a convergence curve diagram of different algorithms in the target area with gentle lateral changes. Since the method proposed in the present invention introduces knowledge sharing and dynamic mapping mechanism between multiple channels, the method has greatly improved convergence accuracy and convergence speed.

[0100] Figure 4 It is a comparison diagram of the wave impedance models obtained by different algorithms in the target area with gentle lateral changes, which respectively represent the real wave impedance model, the wave impedance model obtained by MMDE inversion, the wave impedance model obtained by KT-MMDE (without mapping), the wave impedance model obtained by KT-MMDE inversion, and the difference profiles between the latter three and the real impedance. It can be seen that in the target area with drastic lateral changes, the method proposed in the present invention can still obtain wave impedance inversion results with high accuracy and good lateral continuity, while the traditional global optimization algorithm is not applicable.

[0101] Figure 5 This is the convergence curve of different algorithms in the target area with drastic lateral changes. It can be seen that both the convergence accuracy and the convergence speed have been greatly improved.

[0102] The wave impedance inversion method proposed in the present invention can obtain wave impedance inversion results with good lateral continuity and high consistency with the real wave impedance model without relying on regularization constraints, and is suitable for geological areas with different degrees of lateral changes.

[0103] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A method for lateral continuous seismic wave impedance inversion based on knowledge transfer, characterized in that: The method comprises: Obtain post-stack reflection seismic data, and for the first seismic data, use a wave impedance inversion method based on a multi-group mutation differential evolution algorithm to obtain an optimal solution, namely, an optimal inversion wave impedance model; For each subsequent seismic data, the inversion optimal wave impedance model of the previous adjacent channel is transferred to the evolution of the current seismic channel as a useful knowledge vector, and the migration of the optimal solution is constrained by the constructed mapping matrix to form a migration knowledge vector; The wave impedance inversion of the current seismic trace is performed using a multi-group mutation differential evolution algorithm based on knowledge transfer to obtain a laterally continuous wave impedance model.

2. The method for lateral continuous seismic wave impedance inversion based on knowledge transfer according to claim 1, characterized in that: The specific steps include: (1) Collecting raw seismic data, and then preprocessing the collected seismic data to obtain post-stack reflection seismic data, denoted as d obs (t,n), where t represents the time variable; n is the number of the common depth point gather; (2) Determine the density value, time sampling interval, forward model parameters and seismic wavelet; (3) For actual seismic data, the wave impedance parameters are extracted from a real well logging data; the low-frequency wave impedance model I can be obtained by using the Butterworth filter. lf (t,n); (4) determining the search space range of the wave impedance parameter I according to the obtained low-frequency wave impedance model, and giving the objective function to be optimized in the wave impedance inversion; (5) Perform multi-channel wave impedance inversion. For the first seismic trace in the post-stack reflection seismic data obtained in step (1), use a multi-group mutation differential evolution algorithm to optimize the objective function of step (4). Select the individual with the smallest objective function value in the group after the algorithm iteration is completed, and record it as the optimal solution P of the current trace. The optimal solution is the optimal wave impedance model finally searched for the first seismic trace. (6) For each subsequent seismic trace, the inversion of the previous seismic trace and the previous one can be regarded as two adjacent tasks. The optimal solution individual optimized in the previous trace can be transferred to the evolution process of the current trace to guide the direction of algorithm mutation. (7) The optimal solution migration in step (6) is constrained by the constructed mapping matrix M, thus forming the final migration knowledge vector (8) The migration knowledge vector in step (7) Used in the G+1th iteration of the current path to form a new neighborhood-based mutation strategy; (9) After the iteration of step (8), the individual with the smallest objective function value in the multi-group mutation differential evolution algorithm based on knowledge transfer is selected, which is the optimal wave impedance model of the current channel.

3. The lateral continuous seismic wave impedance inversion method based on knowledge transfer according to claim 2 is characterized in that: In step (2), the forward model parameters and their seismic wavelets are determined according to the shot offset, effective frequency band range and sampling time of the actual seismic data.

4. The method for lateral continuous seismic wave impedance inversion based on knowledge transfer according to claim 2, characterized in that: In step (7), the constructed mapping matrix M changes dynamically during the evolutionary iteration of the current path. Given the parameter G m (1 <G m <G max ) is used to control whether the mapping matrix changes.

5. The method for lateral continuous seismic wave impedance inversion based on knowledge transfer according to claim 4 is characterized in that: For the Gth optimization iteration of the current path, when G≠G m When , the mapping matrix is ​​an identity matrix, that is, M = I; Apply it to the previous optimal solution P to obtain the transferred knowledge vector 6. The method for lateral continuous seismic wave impedance inversion based on knowledge transfer according to claim 4, characterized in that: For the Gth optimization iteration of the current path, when G = G m When the Gth iteration of the current path is completed, a temporary optimal solution individual can be obtained, which is recorded as vector Q G ; At this time, through the known vectors P and Q G , using the constructed mapping operator Φ M (·), the operator consists of multiple linear regression operations, and the mapping matrix M = Φ M (P,Q G ); Apply it to the previous optimal solution P to obtain the transferred knowledge vector 7. The method for lateral continuous seismic wave impedance inversion based on knowledge transfer according to claim 2, characterized in that: In step (8), the migration knowledge vector obtained in step (7) is Used in the evolution process of the current path as the basis vector in the algorithm mutation operation.

8. The method for lateral continuous seismic wave impedance inversion based on knowledge transfer according to claim 7, characterized in that: Based on the new mutation strategy, the mutation vector of the i-th individual in the G-th generation population is It can be expressed as: In the formula, represents the current individual population, a∈(0,1) is a constant, F∈[0,2] represents the disturbance factor, and n1 and n2 represent the number of two temporary variant individuals. CCDE (·) and g DE / rand / 1 (·) respectively represent the neighborhood-based mutation operations in the "DE / rand / 1" method and the collaborative mutation differential evolution algorithm (CCDE), and They represent the mutant individuals generated by DE / rand / 1 and CCDE based on the new mutation strategy respectively.

9. The method for lateral continuous seismic wave impedance inversion based on knowledge transfer according to claim 8, characterized in that: and Respectively expressed by formula: In the formula, and and They respectively represent individuals randomly selected from the current population, and satisfy r1≠r2≠i, m1≠m2≠i.

Images