Depth domain multi-channel collaborative seismic inversion method and system based on structure constraint

Abstract

The invention discloses a construction-constrained depth-domain multi-channel collaborative seismic inversion method and system, and the method comprises the steps: firstly constructing a depth-domain wavelet matrix, generating a synthetic record consistent with an actual seismic feature in combination with an impedance model, and constructing a pre-training data set; gaussian noise is superposed on the synthetic data, and layer dislocation is simulated; then, a plane wave decomposition method is adopted to extract construction inclination angle information from the enhanced data, and a transverse space constraint relation is established; embedding and constructing an inclination angle constraint term in the space-time neural network, and training a multi-channel collaborative inversion model through synthetic data; extracting inclination angle data of a real seismic record, fusing logging and horizon interpretation results, and establishing a depth domain low-frequency initial impedance model; and finally, actually measured seismic data, construction constraint information, an initial model and well data are combined, network parameters are adjusted through transfer learning, and finally high-precision and high-stability impedance inversion is achieved. The problems of insufficient result stability, low resolution, error accumulation effect and the like in the existing seismic inversion method are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic inversion, and particularly to a method and system for multi-channel collaborative seismic inversion in the depth domain with structural constraints. Background Art

[0002] Seismic exploration generates seismic waves through artificial seismic sources and uses geophones to record the propagation characteristics of the wave field in the subsurface medium, and then inversely calculates the physical properties of the subsurface geological structure. As the core link connecting seismic data and geological interpretation, seismic impedance (defined as the product of rock density and P-wave velocity) can comprehensively reflect key parameters such as reservoir porosity, fluid saturation, and lithology, providing a quantitative basis for hydrocarbon reservoir identification and mineral resource assessment. With the continuous innovation of seismic exploration technology, reservoir prediction methods have developed from early empirical analysis to a high-precision technical system integrating multiple disciplines. The current exploration focus is on complex and subtle targets, such as thin interbedded reservoirs with thickness less than the seismic resolution limit and microstructural traps blocked by faults. Such targets pose higher requirements for the signal-to-noise ratio of seismic data and the resolution of inversion algorithms.

[0003] In oil and gas exploration under complex geological conditions, traditional seismic inversion methods have long relied on a single-channel independent processing mode, which is characterized by high computational efficiency and simple algorithm implementation. These methods are usually based on the assumption of horizontally layered media and use a convolutional model with a fixed wavelet shape to solve each seismic trace one by one. However, this simplified processing mode reveals significant limitations when dealing with complex structures: insufficient lateral resolution leads to blurred boundaries of thin interbeds, and the problem of single-channel independent solution being sensitive to noise further reduces the reliability of reservoir parameter prediction. Especially when there are lateral velocity mutations or anisotropic characteristics in the subsurface medium, traditional methods are difficult to meet the current requirements of oil and gas exploration for fine characterization of thin reservoirs and complex structure imaging due to ignoring the spatial correlation of seismic data and the geometric constraints of geological structures, and there is an urgent need to break through the existing technical bottlenecks through more advanced inversion strategies. In addition, traditional seismic inversion methods are mainly carried out in the time domain, and the core technical bottleneck lies in the insufficient ability to represent thin reservoirs. This limitation stems from the irreversible loss of high-frequency components during the conversion of logging data from depth to time, resulting in a significant decrease in the vertical identification accuracy of thin layers (such as formations with thickness less than the conventional seismic resolution). Based on the above limitations, seismic inversion technology has undergone an important technical iteration from the time domain to the depth domain in the field of oil and gas reservoir characterization. In recent years, with the breakthrough of high-performance computing and algorithm theory, prestack depth migration technology has been applied in engineering and has become a standard link in seismic data processing. However, there are still many challenges at the methodological level in carrying out seismic attribute inversion and reservoir parameter prediction based on the results of depth migration. Essentially, this difficulty stems from the fundamental differences in the seismic response mechanism in the depth domain: when the subsurface medium velocity changes continuously with depth, the shape of the seismic wavelet will exhibit spatial non-stationarity characteristics, and the forward simulation framework based on the convolution of a fixed wavelet and reflection coefficient in traditional time-domain inversion is no longer applicable in the depth domain, resulting in the difficulty of directly migrating mature time-domain inversion algorithms to the depth domain. Currently, the processing of depth-domain seismic data in the oil and gas exploration field generally adopts an expedient strategy, which essentially realizes a technical transition through "time-depth conversion": first, the depth-domain migration imaging data is converted to the time domain, the mature time-domain inversion technology is used to complete reservoir parameter prediction, and finally the results are inversely converted back to the depth domain for geological interpretation. This roundabout path does not touch the core mechanism of depth-domain inversion, and due to multiple domain conversions, errors are gradually amplified during the conversion process, resulting in information distortion of geological details such as thin interbeds during repeated resampling, significantly reducing the characterization accuracy of complex reservoirs.

[0004] In summary, the current seismic impedance inversion methods mainly have the following problems: 1. They rely on a single-channel independent processing mode, ignoring the spatial correlation of seismic data and geological structure constraints, resulting in blurred boundaries of thin interbeds, distorted fault zone structures, and being sensitive to noise, making it difficult to depict the spatial distribution of complex reservoirs; 2. In the complex "depth-time-depth" domain conversion process, high-frequency components are irreversibly lost, significantly reducing the vertical identification accuracy of thin layers and restricting the fine characterization of subtle reservoirs; 3. The seismic wavelets in the depth domain show spatial non-stationarity with velocity changes, and the fixed wavelet assumption in the time domain does not match the actual dynamic wavelet characteristics in the depth domain, leading to amplitude response errors and reservoir parameter prediction deviations; 4. High-resolution algorithms such as multi-channel collaboration and anisotropy correction have high computational complexity, and the fusion ability of multi-scale data (seismic, logging, and structural information) is weak, making it difficult to perform high-precision prediction of complex reservoirs. Summary of the Invention

[0005] The purpose of the present invention is to provide a tectonic constraint-based multi-channel collaborative seismic inversion method and system in the depth domain. It adopts a deep learning-driven multi-channel joint inversion method, establishes inter-channel spatial correlation by introducing a tectonic dip constraint term, and uses a dynamic window alignment mechanism to achieve stratigraphic continuity inversion modeling. A global mean square error constraint term based on dip orientation is designed to calculate the formation offset through tectonic dip data and dynamically adjust the spatial alignment relationship of adjacent channel window data, synchronously optimizing the longitudinal data matching degree and the lateral geological structure continuity in the loss function, and realizing the fusion of dip constraint and inversion process based on the spatio-temporal neural network (STNN) framework. The network parameters and geological structure features are updated synchronously through backpropagation, enhancing the reservoir characterization ability under complex tectonic backgrounds while maintaining high resolution, and providing a highly reliable geophysical solution for the fine description of oil and gas reservoirs under thin interbeds and complex tectonic backgrounds.

[0006] To achieve the above object, the present invention provides the following technical solutions: On the one hand, the present invention provides a tectonic constraint-based multi-channel collaborative seismic inversion method in the depth domain, including the following steps: Including the following steps: S1. Use the interpolation method to construct a depth-domain wavelet matrix, and synthesize depth-domain seismic records using impedance model data based on the non-steady-state convolution model, providing a benchmark data set for subsequent network training to ensure the consistency between the synthesized data and real depth-domain seismic data; S2. Superimpose Gaussian noise on the seismic records to simulate random noise and acquisition interference in actual seismic data, and displace the upper and lower horizons of the noisy seismic records to enhance the robustness of the model to complex structures; S3. Use the plane wave decomposition method (PWD) to extract the tectonic dip data of the noisy seismic records, providing highly reliable lateral constraint information for subsequent seismic inversion; S4, using a spatiotemporal neural network (STNN) and establishing spatial correlation constraints on the inversion results through a loss function, and using noisy seismic records and extracted structural dip data for network pre-training; S5, extracting structural dip data of real seismic records by the method described in S3, and establishing a low-frequency initial impedance parameter model in the depth domain by using the seismic data layer interpretation data and logging data in the work area; S6. Use real seismic records and their corresponding structural dip data and initial impedance data to adjust the pre-trained network parameters to obtain the final impedance inversion results.

[0007] In some embodiments, in S1, the non-steady-state convolution model is: ; The depth domain seismic record is: ; in, , , ; In the formula, is the seismic wavelet that varies with depth; is the displacement of the seismic wavelet in the depth direction; for depth; is the reflection coefficient corresponding to the seismic wavelet in the depth direction; For depth domain seismic records: is the non-stationary deep variable wavelet matrix; is the depth domain reflection coefficient; for; The depth is The seismic wavelet at is the depth domain reflection coefficient corresponding to the depth z.

[0008] In some embodiments, S2 includes the following steps: S21, randomly generate Gaussian noise intensity according to uniform distribution; S22, dividing the noisy seismic data into a plurality of non-overlapping blocks with every 5 traces as a group; S23. Set random offsets for non-overlapping blocks in the tectonic intense range (where the absolute value of the structural dip angle in seismic data is greater than 30°), and fill in the missing values ​​after the offset using the cubic spline interpolation method.

[0009] In some embodiments, the calculation expression for superimposing Gaussian noise on seismic records is: ; In the formula, is noisy seismic data; is the number of deep sampling points; is the number of seismic traces; is the normalization result of the synthetic depth-domain seismic record data; is the noise intensity, ; is the standard Gaussian distribution.

[0010] In some embodiments, S3 includes the following steps: Obtain the local slope through the local plane wave differential equation. When the local slope is constant, obtain the plane wave field; Obtain the structural dip equation through the filtering operator; Obtain the slope increment in each iteration of the structural dip equation, which is used to update the initial slope. Subsequently, solve iteratively until the solution converges, and finally obtain the structural dip field.

[0011] In some embodiments, the structural dip equation is: ; In the formula, is the increment of the slope in the th iteration; is the seismic record data; is the th iteration of the filtering operator; is the first-order partial derivative of where is the initial slope value,

[0012] In some embodiments, S4 includes the following steps: Adopt the Huber loss constraint to predict the matching degree between the impedance and the label data. The auxiliary task dynamically calculates the waveform similarity of adjacent seismic traces under the structural guidance through the structural dip constraint term; Construct the structural dip constraint term and, based on the sliding window mechanism, calculate the formation offset through the dip data, dynamically align the adjacent trace seismic data, and quantify the spatial continuity error of the structural direction; Use the synthetic depth-domain seismic record data, the initial impedance model obtained by filtering the impedance model data (the open-source Overthrust classic nappe impedance model on the network), and the extracted structural dip data to perform network pre-training to obtain the prediction result of the model impedance data.

[0013] In some embodiments, the loss function is: ; where, ; ; In the formula, is the loss function in neural network training; is the label constraint term in the loss function; is the structural dip angle constraint term in the loss function; is the set of network parameters to be trained; 、 are the weights of different losses contributing to the training process; is the label data corresponding to the seismic data; is the network to be trained; is the seismic record input to the network; is the initial impedance parameter model input to the network; , is the robustness threshold for controlling the Huber loss; is the number of sample points; is the number of depth sampling points; is the number of seismic traces; is the relative offset index within the sliding window; is the local offset correlation range; is the impedance value of the th trace obtained during the network training process; is the offset of each sample point; ; is the structural dip angle value of each sampling point extracted from the seismic data; is the rounding function.

[0014] On the other hand, a structural constraint-based multi-channel collaborative seismic inversion system in the depth domain of the present invention uses the above method and includes the following modules: Synthesis module: Synthesize the depth-domain seismic record by constructing a depth-domain wavelet matrix and based on the non-steady-state convolution model; Noise addition module: Used to add Gaussian noise to the seismic record and perform dislocation of the upper and lower horizons on the noisy seismic record; Structural dip angle module: Extract the structural dip angle data of the noisy seismic record by using the plane wave decomposition method; Pre-training module: Use a spatio-temporal neural network to establish a spatial correlation constraint for the inversion result through a loss function, and perform network pre-training with the noisy seismic record and the extracted structural dip angle data; Impedance parameter module: Establish a low-frequency initial impedance parameter model in the depth domain by extracting the structural dip angle data of the real seismic record and using the layer interpretation data of the seismic data in the work area and well logging data; Acquisition module: Adjust the pre-trained network parameters by using the real seismic record and its corresponding structural dip angle data and initial impedance data to obtain the final impedance inversion result.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. Through the joint inversion of the constructed dip angle constraint term and the dynamic window alignment mechanism, the present invention synchronously optimizes the longitudinal data matching and the lateral geological structure continuity, effectively suppressing the divergence problem of the solution space of the traditional inversion method and significantly reducing the multi-solution property of the inversion result.

[0016] 2. By combining the fault processing strategy of synthesizing seismic records with the non-steady wavelet matrix and extracting dip angles with PWD, the present invention improves the lateral continuity of the inversion result in complex structural areas such as faults and pinch-outs, and enhances the adaptability of the model to complex geological conditions.

[0017] 3. Based on the two-stage strategy of pre-training with synthetic data and fine-tuning with real data, the present invention reduces the dependence on large-scale labeled data, accelerates network convergence, and at the same time ensures a high degree of consistency between the inversion result and well data.

[0018] 4. The present invention completes wavelet construction, seismic record synthesis, and inversion modeling in the depth domain, avoiding the accumulation of time-depth conversion errors and improving the accuracy of the inversion result in the depth domain. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a schematic diagram of the overall process of the present invention; Figure 2 is a schematic diagram of the depth-domain seismic record synthesized based on the single-channel model data according to Embodiment 1 of the present invention using the non-steady convolution model; Figure 3 is a schematic diagram of the synthetic noisy post-stack depth-domain seismic data (a), the initial impedance parameter model (b), the extracted structural dip angle data (c), and the predicted result (d) of the output model impedance data for pre-training input according to Embodiment 1 of the present invention; Figure 4 is a schematic diagram of the real post-stack depth-domain seismic data (a), the initial impedance parameter model (b), and the extracted structural dip angle data (c) input according to Embodiment 1 of the present invention; Figure 5 is a schematic diagram of the impedance prediction result obtained by the depth-domain multi-channel collaborative intelligent seismic inversion method with structural constraints according to Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0021] Embodiment 1: Please refer to Figures 1-5 a multi-channel collaborative seismic inversion method for structural constraints in the depth domain. First, based on the non-steady convolution theory, a wavelet matrix in the depth domain is constructed, and combined with the impedance model to generate synthetic records consistent with the actual seismic characteristics, and a high-precision pre-training dataset is constructed. Secondly, Gaussian noise is added to the synthetic data and layer dislocation is simulated to enhance the robustness of the model to actual noise interference and structural distortion; the plane wave decomposition method (PWD) is used to extract the structural dip information from the enhanced data and establish the lateral spatial constraint relationship; then, the structural dip constraint term is embedded in the spatio-temporal neural network (STNN), and the multi-channel collaborative inversion model is trained through synthetic data to strengthen the geological correlation between channels; the PWD dip data based on the real seismic records is extracted, and combined with the logging and horizon interpretation results to establish a low-frequency initial impedance model in the depth domain; finally, by jointly using the measured seismic data, structural constraint information, initial model and well data, the network parameters are fine-tuned through transfer learning to finally achieve high-precision and high-stability impedance inversion.

[0022] Specifically, it includes the following steps: S1. Interpolate the original time-domain seismic wavelet to obtain a wavelet matrix in the depth domain with uniformly depth-sampled points, and obtain the convolution model of the depth-domain seismic record through the non-steady convolution model: convolve the wavelet matrix in the depth domain with the depth-domain reflection coefficient sequence obtained from the model impedance parameter data to synthesize the depth-domain seismic record.

[0023] In the depth domain, due to the influence of the formation velocity change on the seismic wavelet, the depth-domain seismic data can be regarded as a seismic profile in which the seismic wavelet changes with depth, and the convolution model of the depth-domain seismic record can be expressed in the form of non-steady convolution: ; For the convenience of constructing the inversion objective function, it can be written in matrix form: ; where , , ; Among them, the wavelet matrix in the depth domain is obtained by processing the original time-domain wavelet using the interpolation method. Each seismic wavelet in the wavelet matrix in the depth domain is different, reflecting the characteristics of the depth-domain seismic wavelet changing with depth. As shown in Figure 2 (a) is a one-dimensional impedance parameter model constructed by experience, (b) is the reflection coefficient corresponding to the model shown in (a), and (c) is a one-dimensional depth-domain seismic record synthesized using the non-steady convolution model. It can be seen that its waveform is not symmetric at the impedance interface, showing non-steadiness.

[0024] S2. Add Gaussian noise (SNR = 1 - 5 dB) to the seismic data synthesized in S1 to simulate random noise and system interference in field acquisition. Secondly, perform horizon dislocation processing on the noisy record to simulate the formation dislocation caused by faults (the dislocation distance is 1 - 7 sampling points).

[0025] Add Gaussian noise to each trace of the synthesized seismic record in the depth domain through the following formula: ; where the noise intensity is randomly generated according to a uniform distribution to obtain noisy seismic data with a signal-to-noise ratio of 1 - 5 dB; divide the noisy seismic data into several non-overlapping blocks with every 5 traces as a group, set a random offset for the blocks in the area with intense structure (the absolute value of the structural dip angle of the seismic data is greater than 30°), and fill the missing values after offset through cubic spline interpolation. The two-dimensional simulated seismic record obtained after the above artificial interference is as shown in Figure 3 (a).

[0026] S3. Use the plane wave decomposition method (PWD) to extract the synthesized seismic dip angle data: Based on the local plane wave assumption, optimize the dip angle estimation through multi-scale window scanning, perform lateral smoothing constraint by combining multi-trace data, solve the local slope through non-linear inversion, and finally convert it into high-precision dip angle data.

[0027] Obtain the local slope through the local plane wave differential equation: ; The local plane wave differential equation is: ; In the formula, is the plane wave field; is the local slope.

[0028] When the local slope is constant, the plane wave field expression can be obtained: ; In the formula, is an arbitrary waveform; is the slope value; is the distance; is the time.

[0029] In order to estimate the local slope through the PWD filter, by introducing the filtering operator (representing the convolution of the data and the two-dimensional filter), the structural dip angle solution equation can finally be obtained:

[0030] In each iteration of constructing the dip angle equation, the slope increment is obtained to update the initial slope, and then the loop solution is carried out until the solution converges. According to the above method, the dip angle field corresponding to the two-dimensional simulated seismic record is extracted as Figure 3 (c) as shown.

[0031] S4. Utilize the spatio-temporal neural network (STNN) and introduce a dip angle constraint mechanism. Through the loss function, establish the spatial correlation constraint of the inversion result. The loss function is designed in a multi-task fusion mode: the main task uses the Huber loss to constrain the matching degree between the predicted impedance and the label data, and the auxiliary task dynamically calculates the similarity of the inversion impedance data of adjacent seismic traces under the structural guidance through the dip angle constraint term. Among them, the structural constraint module is based on the sliding window mechanism. Through the dip angle data, calculate the formation offset, dynamically align the seismic data of adjacent traces, and quantify the spatial continuity error in the structural direction. Then, use the data obtained in S2 and S3 as the input data for network pre-training.

[0032] Among them, the network training uses the Adam optimizer (initial learning rate 0.01, momentum parameter 0.9 / 0.99), and the total number of training cycles is 1500 rounds.

[0033] The loss function is in a multi-task fusion mode: the main task uses the Huber loss (weight 0.975) to constrain the matching degree between the predicted impedance and the label data, and the auxiliary task dynamically calculates the waveform similarity of adjacent seismic traces under the structural guidance through the dip angle constraint term (weight 0.025). The calculation expression is: ; Among them, ; ; Among them, the dip angle constraint term is based on the sliding window mechanism (window size 9), and calculates the formation offset through the dip angle data: ; Dynamically align the seismic data of adjacent traces, and quantify the spatial continuity error in the structural direction.

[0034] Then, use the above synthetic seismic record data, the initial impedance model obtained by filtering the impedance model data, and the extracted dip angle data, as shown in Figure 3 (a), (b), (c) for network pre-training to obtain the prediction result of the model impedance data, as shown in Figure 3 (d), and save the network parameters. In particular, the seismic records participating in the label loss calculation in the pre-training stage are 3 traces (the 100th, 600th, and 2500th traces).

[0035] S4. Extract the measured seismic data according to the method described in S3 to construct dip angle data, and fuse the seismic horizon interpretation results of the work area with well logging data to construct a low-frequency initial impedance model in the depth domain.

[0036] Extract the structural dip angle data (as shown in Figure 4 (a)) of the true seismic record (as shown in Figure 4 (c)). Using seismic structural interpretation data (such as spatial structural features of fault systems, stratigraphic contact relationships, etc.), combined with regional sedimentary pattern analysis, including the distribution law of sedimentary facies belts, stratigraphic stacking patterns, and lateral lithofacies change trends, construct a geological structure framework model, which characterizes elements such as the geometric shape of the strata (such as stratum thickness, attitude), division of tectonic units, and spatial contact relationships. Under the constraint of this tectonic framework, the wave impedance parameters obtained from well logging data are extrapolated through a spatially interpolated algorithm controlled by tectonic trends, and the discrete well point data are extended to the positions of each seismic interpretation line. Finally, an initial wave impedance model that matches the geological structure and sedimentary characteristics is established (as shown in Figure 4 (b)). The establishment of the impedance parameter model is based on the spatial interpolation method: two-dimensional plane modeling of seismic interpretation horizons through discrete data interpolation to form a geological horizon profile to depict the spatial shape of sedimentary bodies, and then, with the horizon profile as the constraint condition, perform geologically guided lateral interpolation on well logging impedance data, expand the point-like well logging information into a continuous distribution along the two-dimensional plane, and extrapolate it to the area without well control. Finally, a two-dimensional initial impedance parameter model covering the entire work area is generated.

[0037] S6. Adopt a transfer learning strategy to achieve multi-channel seismic data collaborative inversion: load the pre-trained network weights, freeze the parameters of the feature extraction layer to retain the general features of seismic data; based on the true seismic record, dip angle data, and initial impedance model in the depth domain, jointly optimize the network parameters through a dynamic weighting mechanism, where the dip angle data constrains the spatial prediction direction, and well data corrects the low-frequency trend error; finally, through iterative training, make the inversion result simultaneously satisfy seismic waveform matching, tectonic guidance continuity, and well logging data hard constraints to obtain a high-precision impedance inversion result.

[0038] Specifically, according to the measured seismic record, the dip angle data, the initial impedance parameter model, and the measured well impedance data (channels 80, 127, 233) obtained by the above method, fix the parameters of the feature extraction layer (including two convolutional modules and a GRU time series layer) in the transfer learning framework, and use the true seismic data to update the weight matrix parameters of the neural network linear layer. When the loss of the validation set does not decrease for 3 consecutive training epochs, terminate the training. The output of the network is the impedance parameter model in the depth domain obtained by inversion (as shown in Figure 5 ).

[0039] Example 2 A depth-domain multi-channel collaborative seismic inversion system with structural constraints, using the above method, includes the following modules: Synthesis module: Synthesize depth-domain seismic records by constructing a depth-domain wavelet matrix and based on the non-steady-state convolution model; Noise addition module: Used to add Gaussian noise to the seismic records and perform dislocation of the upper and lower horizons on the noisy seismic records; Structural dip module: Extract the structural dip data of the noisy seismic records by using the plane wave decomposition method; Pre-training module: Use a spatio-temporal neural network to establish spatial correlation constraints for the inversion results through a loss function, and perform network pre-training with the noisy seismic records and the extracted structural dip data; Impedance parameter module: Establish a low-frequency initial impedance parameter model in the depth domain by extracting the structural dip data of the real seismic records and using the horizon interpretation data and logging data of the seismic data in the work area; Obtaining module: Adjust the pre-trained network parameters by using the real seismic records and their corresponding structural dip data and initial impedance data to obtain the final impedance inversion result.

[0040] A depth-domain multi-channel collaborative seismic inversion system with structural constraints of the present invention can be installed in a computer device. The computer device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a depth-domain multi-channel collaborative seismic inversion program with structural constraints. Among them, the memory includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (such as: SD or DX memory, etc.), magnetic memory, magnetic disk, optical disk, etc. The processor is the control core of the electronic device, connecting various components of the entire computer device through various interfaces and lines, and by running or executing the programs or modules stored in the memory, and calling the data stored in the memory, to perform various functions of the computer device and process data.

[0041] The module of the present invention refers to a series of computer program segments that can be executed by the processor of a computer device and can complete fixed functions, and are stored in the memory of the computer device.

[0042] Those skilled in the art will easily think of other implementation schemes of the present application after considering the specification and practicing the content disclosed herein. The present application aims to cover any variations, uses, or adaptive changes of the present application, and these variations, uses, or adaptive changes follow the general principles of the present application and include the common general knowledge or conventional technical means in the technical field not disclosed in the present application.

Claims

1. A structurally constrained depth domain multi-channel collaborative seismic inversion method, characterized in that: The following steps are involved: S1. Construct a wavelet matrix in the depth domain and synthesize the seismic records in the depth domain based on the non-steady-state convolution model; S2, superimpose Gaussian noise on the seismic records, and perform upper and lower layer displacement on the noisy seismic records; S3, extracting structural dip data from noisy seismic records using plane wave decomposition method; S4. In the spatiotemporal neural network, the spatial correlation constraints of the inversion results are established through the loss function, and the network is pre-trained using noisy seismic records and extracted structural dip data; S5. Extract structural dip data from real seismic records, and use the seismic data layer interpretation data and well logging data in the work area to establish a low-frequency initial impedance parameter model in the depth domain; S6. Use real seismic records and their corresponding structural dip data and initial impedance data to adjust the pre-trained network parameters to obtain the final impedance inversion results.

2. The method for structurally constrained depth domain multi-channel collaborative seismic inversion according to claim 1, characterized in that: In S1, the non-steady-state convolution model is: ； The depth domain seismic record is: ； in, ， ， ； In the formula, is the seismic wavelet that varies with depth; is the displacement of the seismic wavelet in the depth direction; for depth; is the reflection coefficient corresponding to the seismic wavelet in the depth direction; For depth domain seismic records: is the non-stationary deep variable wavelet matrix; is the depth domain reflection coefficient; for; The depth is The seismic wavelet at is the depth domain reflection coefficient corresponding to the depth z.

3. The method for structurally constrained depth domain multi-channel coordinated seismic inversion according to claim 1, characterized in that: S2 includes the following steps: S21, randomly generate Gaussian noise intensity according to uniform distribution; S22, dividing the noisy seismic data into a plurality of non-overlapping blocks with every 5 traces as a group; S23. For non-overlapping blocks of seismic data with an absolute structural inclination angle greater than 30°, a random offset is set, and missing values ​​after the offset are filled by cubic spline interpolation.

4. The method for structurally constrained depth domain multi-channel coordinated seismic inversion according to claim 3, characterized in that: The calculation expression for superimposing Gaussian noise on seismic records is: ； In the formula, is noisy seismic data; is the number of depth sampling points; is the seismic trace number; It is the normalization result of synthetic depth domain seismic record data; is the noise intensity, ; is a standard Gaussian distribution.

5. The method for structurally constrained depth domain multi-channel coordinated seismic inversion according to claim 1, characterized in that: S3 includes the following steps: The local slope is obtained through the local plane wave differential equation. When the local slope is constant, the plane wave field is obtained. The solution equation of the structural inclination angle is obtained through the filtering operator; In each iteration of the structural dip angle solution equation, the slope increment is obtained to update the initial slope, and then the solution is cyclically solved until the solution converges, and finally the structural dip angle field is obtained.

6. The method for structurally constrained depth domain multi-channel coordinated seismic inversion according to claim 5, characterized in that: The structural inclination solution equation is: ； In the formula, For the The increment of the slope in iterations; Recording data for earthquakes; For the The filter operator in the iteration; for The first-order partial derivative of is the initial slope value, is the number of iterations.

7. The method for structurally constrained depth domain multi-channel coordinated seismic inversion according to claim 1, characterized in that: S4 includes the following steps: The Huber loss constraint is used to predict the matching degree between impedance and label data. The auxiliary task dynamically calculates the waveform similarity of adjacent seismic traces under structural guidance through the structural dip constraint term. The dip constraint term is constructed and based on the sliding window mechanism, the formation offset is calculated through the dip data, the adjacent seismic data are dynamically aligned, and the spatial continuity error of the structural direction is quantified; The prediction results of model impedance data are obtained by network pre-training using synthetic depth domain seismic record data, the initial impedance model obtained by filtering the impedance model data, and the extracted structural dip data.

8. The method for structurally constrained depth domain multi-channel coordinated seismic inversion according to claim 7, characterized in that: The loss function is: ； in, ； ； In the formula, is the loss function in neural network training; is the label constraint term in the loss function; is the construction angle constraint term in the loss function; is the set of network parameters to be trained; , The weights that different losses contribute to the training process; is the label data corresponding to the earthquake data; The network to be trained; earthquake records as input to the network; Initial impedance parameter model for network input; , is the robustness threshold for controlling Huber loss; is the number of sample points; is the number of depth sampling points; is the seismic trace number; is the relative offset index within the sliding window; is the local offset related range; is the first Channel impedance value; is the offset for each sample point; ； Constructing dip values ​​for each sampling point extracted from seismic data; is the rounding function.

9. A structurally constrained depth domain multi-channel coordinated seismic inversion system, using a structurally constrained depth domain multi-channel coordinated seismic inversion method according to claims 1 to 8, characterized in that: include Synthesis module: constructs a wavelet matrix in the depth domain and synthesizes the depth domain seismic records based on the non-steady-state convolution model; Noise adding module: used to superimpose Gaussian noise on seismic records and perform upper and lower layer displacement on noisy seismic records; Structural dip module: extracts structural dip data from noisy seismic records by using plane wave decomposition method; Pre-training module: Using the spatiotemporal neural network, the spatial correlation constraints of the inversion results are established through the loss function, and the network is pre-trained using noisy seismic records and extracted structural dip data; Impedance parameter module: By extracting structural dip data from real seismic records and using the seismic data layer interpretation data and logging data in the work area, a low-frequency initial impedance parameter model in the depth domain is established; Acquisition module: Use real seismic records and their corresponding structural dip data and initial impedance data to adjust the pre-trained network parameters and obtain the final impedance inversion results.

Images