Method and device for establishing low-frequency impedance model based on multivariate information control

Abstract

The invention provides a method and a device for establishing a low-frequency impedance model based on multivariate information control. The method comprises the following steps: firstly, carrying out seismic attribute dimension reduction through a local preserving projection algorithm to obtain feature attributes, and calculating a low-frequency impedance model M1 under a depth model framework based on an LSTM network; secondly, well interpolation extrapolation is conducted through inverse proportion weighting, and a well interpolation low-frequency impedance model M2 is obtained; and finally, based on the whale optimization algorithm, obtaining the optimal weight of the combination of the two low-frequency impedance models, and carrying out linear combination to obtain a final low-frequency impedance model Mfinal. Establishment of the low-frequency impedance model is controlled by earthquake, horizon and logging multivariate information, and example application results show that the low-frequency impedance model established by the method has high goodness of fit with the ground, has higher resolution compared with a low-frequency impedance model established by conventional well interpolation, and has extremely high popularization and application value.

Description

Technical Field

[0001] The present invention relates to the field of geophysical exploration technology, and more specifically, to a method and device for establishing a low-frequency impedance model based on multivariate information control. Background Art

[0002] An accurate low-frequency impedance model can effectively improve the accuracy of seismic inversion and thus improve the accuracy of reservoir prediction. However, conventional seismic data often lack low-frequency information, and how to obtain a relatively accurate low-frequency impedance model has become a difficulty in seismic inversion.

[0003] At present, there are several commonly used methods for establishing low-frequency impedance models:

[0004] (1) The low-frequency impedance model establishment method based on well interpolation has the disadvantage that when the well spatial distribution is unreasonable, this method is prone to cause the bull's eye phenomenon:

[0005] (2) The method of establishing a low-frequency impedance model constrained by seismic velocity body. This method is limited by the fact that the velocity body can only compensate for low-frequency information below 2 Hz, and cannot compensate for all the low-frequency information missing in the seismic data:

[0006] (3) Phase-controlled low-frequency impedance model establishment method. The disadvantage of this method is that it is difficult to obtain accurate seismic phases, and seismic phases cannot fully reflect lithology and fluid information, making it difficult to obtain an accurate low-frequency impedance model.

[0007] Since the above methods for establishing low-frequency impedance models all have certain limitations, a method for establishing low-frequency impedance models with strong adaptability is urgently needed. Summary of the invention

[0008] In view of this, the present invention aims to propose a method for establishing a low-frequency impedance model that integrates multivariate information and has strong adaptability.

[0009] According to one aspect of the present invention, a method for establishing a low-frequency impedance model based on multivariate information control is proposed, the method comprising:

[0010] Step 1: pre-process the logging curve to obtain an optimized logging curve, wherein the optimized logging curve includes a low-frequency impedance curve:

[0011] Step 2: Combine the optimized logging curve with seismic data and stratigraphic data to perform well-seismic calibration:

[0012] Step 3, extracting multiple seismic attributes from the seismic data:

[0013] Step 4: Use a local preservation projection algorithm to reduce the dimensions of the multiple seismic attributes to obtain optimized characteristic attributes:

[0014] Step 5: Using the optimized characteristic attributes and the low-frequency impedance curve, obtain a low-frequency impedance model M1 under the deep model framework:

[0015] Step 6: Using the layer data and the low-frequency impedance curve, inverse proportional weighting is used to perform well interpolation and extrapolation to obtain a low-frequency impedance model M2:

[0016] Step 7: Use the whale optimization algorithm to determine the weight coefficients of the low-frequency impedance model M1 and the low-frequency impedance model M2, perform linear combination, and obtain the final low-frequency impedance model M_final.

[0017] In some embodiments, in step 1, preprocessing the well logging curves includes correcting the well logging curves and performing inter-well consistency processing.

[0018] In some embodiments, in step 4, the multiple seismic attributes are dimensionally reduced based on the following formula to obtain an optimized characteristic attribute Y={y 1 ,y 2 ,...,y m}:

[0019] y i =B T x i

[0020] Where X = {x 1 ,x 2 ,...,x n} is the data set of the multiple seismic attributes, B is the projection matrix B, n is the number of the seismic attributes before dimensionality reduction, and m is the number of the characteristic attributes obtained after dimensionality reduction.

[0021] In some embodiments, in step 4, the projection matrix B is determined by minimizing the following objective function:

[0022]

[0023] Among them, W ij represents the weight matrix,

[0024]

[0025] In some embodiments, the optimized characteristic attributes include well bypass characteristic attributes, and step 5 specifically includes:

[0026] The well bypass feature attribute is extracted from the optimized feature attribute, and combined with the low-frequency impedance curve to produce a sample set:

[0027] The nonlinear mapping relationship between the characteristic attributes of the well bypass channel and the low-frequency impedance curve is learned based on the long short-term memory network:

[0028] The nonlinear mapping relationship is extended to the entire target area to obtain the low-frequency impedance properties of all positions in the target area to generate the low-frequency impedance model M1.

[0029] In some implementations, step 7 specifically includes:

[0030] The final low-frequency impedance model M_final is obtained based on the following formula:

[0031] M_final=α1*M1+α2*M2

[0032] Among them, α1 and α2 are weight coefficients determined based on the whale optimization algorithm.

[0033] According to another aspect of the present invention, a low-frequency impedance model establishment device based on multivariate information control is proposed, the device comprising:

[0034] The logging curve preprocessing unit is used to preprocess the logging curve to obtain an optimized logging curve, wherein the optimized logging curve includes a low-frequency impedance curve:

[0035] Well seismic calibration unit is used to combine the optimized logging curve with seismic data and layer data for well seismic calibration:

[0036] A seismic attribute extraction unit, used to extract a plurality of seismic attributes from the seismic data:

[0037] The seismic attribute dimension reduction unit is used to reduce the dimension of the multiple seismic attributes by using a local preservation projection algorithm to obtain optimized characteristic attributes:

[0038] The first low-frequency impedance model acquisition unit is used to acquire a low-frequency impedance model M1 under a deep model framework by using the optimized characteristic attributes and the low-frequency impedance curve:

[0039] The second low-frequency impedance model acquisition unit is used to use the layer data and the low-frequency impedance curve to perform well interpolation and extrapolation by inverse proportional weighting to acquire a low-frequency impedance model M2:

[0040] The final low-frequency impedance model acquisition unit is used to determine the weight coefficients of the low-frequency impedance model M1 and the low-frequency impedance model M2 by using the whale optimization algorithm, perform linear combination, and obtain the final low-frequency impedance model M_final.

[0041] In some embodiments, in the well logging curve preprocessing unit, preprocessing the well logging curve includes correcting the well logging curve and performing inter-well consistency processing.

[0042] In some embodiments, the seismic attribute dimension reduction unit is specifically used to reduce the dimension of the multiple seismic attributes based on the following formula to obtain an optimized characteristic attribute Y={y 1 ,y 2 ,...,y m}:

[0043] y i =B T x i

[0044] Where X = {x 1 ,x 2 ,...,x n} is the data set of the multiple seismic attributes, B is the projection matrix B, n is the number of the seismic attributes before dimensionality reduction, and m is the number of the characteristic attributes obtained after dimensionality reduction.

[0045] In some embodiments, in the seismic attribute dimension reduction unit, the projection matrix B is determined by minimizing the following objective function:

[0046]

[0047] Among them, W ij represents the weight matrix,

[0048]

[0049] In some embodiments, the optimized characteristic attribute includes a well bypass characteristic attribute, and the first low-frequency impedance model acquisition unit is specifically used to:

[0050] The well bypass feature attribute is extracted from the optimized feature attribute, and combined with the low-frequency impedance curve to produce a sample set:

[0051] The nonlinear mapping relationship between the characteristic attributes of the well bypass channel and the low-frequency impedance curve is learned based on the long short-term memory network:

[0052] The nonlinear mapping relationship is extended to the entire target area to obtain the low-frequency impedance properties of all positions in the target area to generate the low-frequency impedance model M1.

[0053] In some implementations, the final low-frequency impedance model acquisition unit is specifically used to:

[0054] The final low-frequency impedance model M_final is obtained based on the following formula:

[0055] M_final=α1*M1+α2*M2

[0056] Among them, α1 and α2 are weight coefficients determined based on the whale optimization algorithm.

[0057] According to another aspect of the present invention, an electronic device is also provided, the electronic device comprising:

[0058] Memory, which stores executable instructions:

[0059] A processor runs the executable instructions in the memory to implement the method for establishing a low-frequency impedance model based on multivariate information control as described above.

[0060] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the method for establishing a low-frequency impedance model based on multivariate information control described above is implemented.

[0061] The beneficial effects of the present invention include at least:

[0062] 1. The present invention integrates seismic, well logging, and stratigraphic multivariate data, and proposes a low-frequency impedance model establishment scheme with strong adaptability through the organic combination of data processing, deep learning, and optimization algorithms:

[0063] 2. The present invention uses the local preservation projection algorithm (LPP) to reduce the dimension of seismic attributes, which can effectively eliminate redundant information in the attributes, refine the characteristic attributes that are sensitive to low-frequency impedance, and provide a good data foundation for subsequent modeling:

[0064] 3. The present invention obtains a low-frequency impedance model under a deep model framework based on a long short-term memory network (LSTM), can learn the nonlinear mapping relationship between feature attributes and low-frequency impedance, and establish a data-driven low-frequency impedance model:

[0065] 4. The present invention adopts the well interpolation method to obtain another low-frequency impedance model, providing a comparison model with physical constraints:

[0066] 5. The present invention uses the whale optimization algorithm to find the optimal combination of two low-frequency impedance models, making full use of the advantages of data-driven models and physical constraint models to obtain a more accurate comprehensive low-frequency impedance model:

[0067] 6. The final low-frequency impedance model obtained according to the present invention has the characteristics of strong adaptability, high quality, good accuracy and high resolution, and can effectively provide support for subsequent seismic inversion and resource assessment work.

[0068] In summary, the technical solution according to the present invention integrates multivariate information, and through the combination of data drive and physical constraints, the final low-frequency impedance model obtained has strong adaptability, high consistency with well data, good resolution, and has achieved very beneficial technical effects.

[0069] The methods and apparatus of the present invention have other features and advantages that will be apparent from or will be described in detail in the accompanying drawings and subsequent detailed descriptions incorporated herein, which together serve to explain the specific principles of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The above and other objects, features and advantages of the present invention will become more apparent through a more detailed description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present invention.

[0071] Figure 1 A flow chart of a method for establishing a low-frequency impedance model based on multivariate information control according to an embodiment of the present invention is shown.

[0072] Figure 2 A typical well calibration map of a target area according to an exemplary embodiment of the present invention is shown.

[0073] Figure 3 A cross-sectional schematic diagram of a well interpolation low-frequency impedance model according to an exemplary embodiment of the present invention is shown.

[0074] Figure 4 FIG. 4 is a schematic cross-sectional view of a final low-frequency impedance model according to an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0075] The preferred embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the preferred embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0076] Example 1

[0077] Figure 1 The flowchart of the method for establishing a low-frequency impedance model based on multivariate information control according to an embodiment of the present invention is shown. As shown in the figure, the example includes steps 1 to 7.

[0078] Step 1: preprocess the well logging curve to obtain an optimized well logging curve, wherein the optimized well logging curve includes a low-frequency impedance curve.

[0079] In some embodiments, preprocessing the well logging curves includes correcting the well logging curves and performing inter-well consistency processing.

[0080] According to this embodiment, preprocessing of the logging curve can achieve the purposes of improving data quality, increasing the signal-to-noise ratio, and improving the consistency of data between wells.

[0081] Step 2: Combine the optimized logging curve with seismic data and stratigraphic data to perform well-seismic calibration.

[0082] Well seismic tie refers to the process of linking well logging data with seismic data. The present invention adopts well seismic tie mainly to integrate information from different data sources into a unified depth coordinate system.

[0083] Specifically, the workflow of well seismic calibration can include the following steps:

[0084] (1) Logging data preprocessing: Correct the logging data, eliminate the effects of each layer, and obtain the true impedance curve of the well section. This step can be completed in step 1 above:

[0085] (2) Seismic data preprocessing: perform seismic profile data correction, improve signal-to-noise ratio, enhance the contrast of formation reflection coefficients, etc.

[0086] (3) Merge well location information: Link the well logging curve with the layer position of the underground stratum to establish the corresponding relationship between well depth and time depth:

[0087] (4) Extracting seismic amplitude: Extract the seismic profile amplitude near the well location to obtain the seismic reflection coefficient:

[0088] (5) Calibration: specific algorithms, such as linear regression, statistical regression, etc., can be used as needed to calibrate the logging impedance and seismic reflection coefficient and establish a mathematical correspondence between the two;

[0089] (6) Verification: Based on the established correspondence, calculate the theoretical seismic profile and compare it with the actual seismic profile to determine the calibration effect;

[0090] (7) Iterative optimization: According to the comparison results, adjust the parameters and repeat the above steps until the calibration effect reaches the desired level.

[0091] According to the present invention, the seismic, stratigraphic and logging data can be used to carry out fine well-seismic calibration, and the information obtained by different measurement methods can be converted into the same domain, laying a good foundation for the subsequent establishment of a low-frequency impedance model.

[0092] Step 3: extract multiple seismic attributes from the seismic data.

[0093] Seismic data can be used to extract various seismic attributes such as amplitude, frequency, and phase.

[0094] Step 4: Use a local preservation projection algorithm to reduce the dimensions of the multiple seismic attributes to obtain optimized characteristic attributes.

[0095] For the various seismic attributes extracted in step 3, seismic attribute dimensionality reduction can be performed based on the local preserving projection algorithm (LPP) to obtain characteristic attributes, remove redundant information in the various seismic attributes, and obtain characteristic attributes that are sensitive to low-frequency impedance, such as well bypass characteristic attributes.

[0096] In some embodiments, the multiple seismic attributes may be dimensionally reduced based on the following formula to obtain an optimized characteristic attribute Y={y 1 ,y 2 ,...,y m}:

[0097] y i =B T x i (1)

[0098] Where X = {x 1 ,x 2 ,...,x n} is the data set of the multiple seismic attributes, B is the projection matrix B, n is the number of the seismic attributes before dimensionality reduction, and m is the number of the characteristic attributes obtained after dimensionality reduction.

[0099] In some examples, the projection matrix B can be determined by minimizing the following objective function:

[0100]

[0101] Among them, W ij represents the weight matrix,

[0102]

[0103] By introducing constraints and combining with equation (1), the problem can be transformed into solving the following generalized eigenvalue problem:

[0104] XL T B=λXDX T B (4)

[0105] Among them, λ is the generalized eigenvalue, L is the generalized eigenvector, D is the degree matrix, and X is the high-dimensional sample matrix, that is, X = {x 1 ,x 2 ,...,xn}, B is the projection matrix.

[0106] When solving, we can first find the generalized eigenvalue λ in formula (4), substitute the found λ into the original problem, solve to get the eigenvector, and finally combine the eigenvectors to get the projection matrix B.

[0107] Step 5: Using the optimized characteristic attributes and the low-frequency impedance curve, obtain a low-frequency impedance model M1 under a deep model framework.

[0108] In some embodiments, the optimized characteristic attributes include well bypass characteristic attributes, and step 5 specifically includes:

[0109] The well bypass feature attribute is extracted from the optimized feature attribute, and combined with the low-frequency impedance curve to produce a sample set:

[0110] The nonlinear mapping relationship between the characteristic attributes of the well bypass channel and the low-frequency impedance curve is learned based on a long short-term memory network (LSTM):

[0111] The nonlinear mapping relationship is extended to the entire target area to obtain the low-frequency impedance properties of all positions in the target area to generate the low-frequency impedance model M1.

[0112] In this embodiment, the well bypass characteristic attribute is a characteristic attribute sensitive to low-frequency impedance obtained after dimension reduction using a local preservation projection algorithm.

[0113] The long short-term memory network (LSTM) used in this embodiment is a variant of the recurrent neural network (RNN), which can effectively solve the problem that the RNN may miss long-term memory.

[0114] In this embodiment, a nonlinear mapping relationship between feature attributes and low-frequency impedance is obtained based on LSTM, and then a low-frequency impedance model M1 under a deep model framework is obtained.

[0115] In this embodiment, after the LSTM model learns the nonlinear mapping relationship between the characteristic attributes of the well bypass channel and the low-frequency impedance, this mapping relationship is extended to the entire target area, that is, the trained LSTM model is used to predict the entire area to obtain the low-frequency impedance attributes of all locations in the target area, thereby generating a low-frequency impedance model M1 for the entire area.

[0116] Step 6: Using the layer data and the low-frequency impedance curve, inverse proportional weighting is used to perform well interpolation and extrapolation to obtain a low-frequency impedance model M2.

[0117] In some embodiments, inverse proportional weighting may be performed by the following steps:

[0118] (1) Calculate the distance between each target interpolation point and the adjacent sampling points:

[0119] (2) Assign a weight to each adjacent sampling point, and the weight is inversely proportional to the above distance:

[0120] (3) Multiply the low-frequency impedance value of each sampling point by the corresponding weight to obtain the contribution of the sampling point to the target difference point:

[0121] (4) The contributions of all adjacent logging sampling points are accumulated and summed to obtain the final low-frequency impedance value of the target interpolation point.

[0122] The present invention adopts inverse proportional weighted interpolation, and the near-well point data contributes more, which conforms to the principle of distance attenuation and is conducive to obtaining a high-quality low-frequency impedance model.

[0123] According to the present invention, local interpolation can be performed first in the area with wells, and then the interpolation result can be extrapolated to the entire area without wells, so as to realize the low-frequency impedance prediction of the entire target area.

[0124] Step 7: Use the whale optimization algorithm to obtain the weight coefficients of the low-frequency impedance model M1 and the low-frequency impedance model M2, perform linear combination, and obtain the final low-frequency impedance model M_final.

[0125] In some embodiments,

[0126] The final low-frequency impedance model M_final is obtained based on the following formula:

[0127] M_final=α1*M1+α2*M2 (5)

[0128] Among them, α1 and α2 are weight coefficients determined based on the whale optimization algorithm.

[0129] The whale optimization algorithm is an optimization algorithm that imitates the whale hunting process. Each whale individual corresponds to a solution, and the whale individual position is continuously updated to search for the solution of the problem until the termination condition is met.

[0130] Whale Optimization Algorithm is a bionic optimization algorithm that simulates the behavior of whales in finding food. The algorithm corresponds one whale to one solution, and searches for the solution of the problem by continuously updating the position of the whale until the termination condition is met. The algorithm is based on the following three whale movement modes:

[0131] (1) Encirclement and hunting

[0132] Simulate the whale's spiral movement around the food. In the algorithm, by updating the position of individual whales, this spiral movement is simulated to approach the optimal solution:

[0133] (2) Searching for food

[0134] Simulate whales hunting through bubble nets. In the algorithm, randomly simulating whale position mutations is equivalent to searching for new solutions through bubble nets:

[0135] (3) Search

[0136] Simulate whales randomly searching for food. In the algorithm, random perturbations are added to individuals to increase population diversity and avoid falling into local optimality:

[0137] The present invention determines the weight coefficients of the low-frequency impedance model M1 and the low-frequency impedance model M2 by adopting the whale optimization algorithm, so that the solution process has the advantages of strong global search capability and fast convergence speed, and can quickly find the global optimal solution, further improving the quality of the final low-frequency impedance model M_final.

[0138] Compared with the existing technology, this embodiment has at least the following innovative features:

[0139] (1) This embodiment is based on the local preservation projection algorithm, which can effectively eliminate redundant information in seismic attributes and provide a solid data foundation for the subsequent establishment of a low-frequency impedance model;

[0140] (2) This embodiment obtains the nonlinear mapping relationship between the characteristic attributes and the low-frequency impedance curve based on the LSTM network according to the time series characteristics of the seismic signal;

[0141] (3) This embodiment can effectively obtain the optimal weight coefficient based on the whale optimization algorithm, and then optimally combine the low-frequency impedance model under the deep model framework with the well interpolation low-frequency impedance model to obtain the final low-frequency impedance model;

[0142] (4) This embodiment establishes a multi-information-controlled low-frequency impedance model establishment method flow, and fully integrates seismic, layer and well logging multi-information in the low-frequency impedance model establishment process.

[0143] Example 2

[0144] According to an embodiment of the present invention, a low-frequency impedance model building device based on multivariate information control is provided.

[0145] The device comprises:

[0146] The logging curve preprocessing unit is used to preprocess the logging curve to obtain an optimized logging curve, wherein the optimized logging curve includes a low-frequency impedance curve:

[0147] Well seismic calibration unit is used to combine the optimized logging curve with seismic data and layer data for well seismic calibration:

[0148] A seismic attribute extraction unit, used to extract a plurality of seismic attributes from the seismic data:

[0149] The seismic attribute dimension reduction unit is used to reduce the dimension of the multiple seismic attributes by using a local preservation projection algorithm to obtain optimized characteristic attributes:

[0150] The first low-frequency impedance model acquisition unit is used to acquire a low-frequency impedance model M1 under a deep model framework by using the optimized characteristic attributes and the low-frequency impedance curve:

[0151] The second low-frequency impedance model acquisition unit is used to use the layer data and the low-frequency impedance curve to perform well interpolation and extrapolation by inverse proportional weighting to acquire a low-frequency impedance model M2:

[0152] The final low-frequency impedance model acquisition unit is used to determine the weight coefficients of the low-frequency impedance model M1 and the low-frequency impedance model M2 by using the whale optimization algorithm, perform linear combination, and obtain the final low-frequency impedance model M_final.

[0153] In some embodiments, in the well logging curve preprocessing unit, preprocessing the well logging curve includes correcting the well logging curve and performing inter-well consistency processing.

[0154] In some embodiments, the seismic attribute dimension reduction unit is specifically used to reduce the dimension of the multiple seismic attributes based on the following formula to obtain an optimized characteristic attribute Y={y 1 ,y 2 ,...,y m}:

[0155] y i =B T x i (1)

[0156] Where X = {x 1 ,x 2 ,...,x n} is the data set of the multiple seismic attributes, B is the projection matrix B, n is the number of the seismic attributes before dimensionality reduction, and m is the number of the characteristic attributes obtained after dimensionality reduction.

[0157] In some embodiments, in the seismic attribute dimension reduction unit, the projection matrix B is determined by minimizing the following objective function:

[0158]

[0159] Among them, W ij represents the weight matrix,

[0160]

[0161] By introducing constraints and combining with equation (1), the problem can be transformed into solving the following generalized eigenvalue problem:

[0162] XL T B=λXDX T B (4)

[0163] Among them, λ is the generalized eigenvalue, L is the generalized eigenvector, D is the degree matrix, and X is the high-dimensional sample matrix, that is, X = {x 1 ,x 2 ,...,x n}, B is the projection matrix.

[0164] When solving, we can first find the generalized eigenvalue λ in formula (4), substitute the found λ into the original problem, solve to get the eigenvector, and finally combine the eigenvectors to get the projection matrix B.

[0165] In some embodiments, the optimized characteristic attribute includes a well bypass characteristic attribute, and the first low-frequency impedance model acquisition unit is specifically used to:

[0166] The well bypass characteristic attribute is extracted from the optimized characteristic attribute, and combined with the low-frequency impedance curve to produce a sample set:

[0167] The nonlinear mapping relationship between the characteristic attributes of the well bypass channel and the low-frequency impedance curve is learned based on the long short-term memory network:

[0168] The nonlinear mapping relationship is extended to the entire target area to obtain the low-frequency impedance properties of all positions in the target area to generate the low-frequency impedance model M1.

[0169] In this embodiment, the well bypass characteristic attribute is a characteristic attribute sensitive to low-frequency impedance obtained after dimension reduction using a local preservation projection algorithm.

[0170] The long short-term memory network (LSTM) used in this embodiment is a variant of the recurrent neural network (RNN), which can effectively solve the problem that the RNN may miss long-term memory.

[0171] In this embodiment, a nonlinear mapping relationship between feature attributes and low-frequency impedance is obtained based on LSTM, and then a low-frequency impedance model M1 under a deep model framework is obtained.

[0172] In this embodiment, after the LSTM model learns the nonlinear mapping relationship between the characteristic attributes of the well bypass channel and the low-frequency impedance, this mapping relationship is extended to the entire target area, that is, the trained LSTM model is used to predict the entire area to obtain the low-frequency impedance attributes of all locations in the target area, thereby generating a low-frequency impedance model M1 for the entire area.

[0173] In some implementations, in the second low-frequency impedance model acquisition unit, inverse proportional weighting may be performed by the following steps:

[0174] (1) Calculate the distance between each target interpolation point and the adjacent sampling points:

[0175] (2) Assign a weight to each adjacent sampling point, and the weight is inversely proportional to the above distance:

[0176] (3) Multiply the low-frequency impedance value of each sampling point by the corresponding weight to obtain the contribution of the sampling point to the target difference point:

[0177] (4) The contributions of all adjacent logging sampling points are accumulated and summed to obtain the final low-frequency impedance value of the target interpolation point.

[0178] The present invention adopts inverse proportional weighted interpolation, and the near-well point data contributes more, which conforms to the principle of distance attenuation and is conducive to obtaining a high-quality low-frequency impedance model.

[0179] According to the present invention, local interpolation can be performed first in the area with wells, and then the interpolation result can be extrapolated to the entire area without wells, so as to realize the low-frequency impedance prediction of the entire target area.

[0180] In some implementations, the final low-frequency impedance model acquisition unit is specifically used to:

[0181] The final low-frequency impedance model M_final is obtained based on the following formula:

[0182] M_final=α1*M1+α2*M2

[0183] Among them, α1 and α2 are weight coefficients determined based on the whale optimization algorithm.

[0184] The whale optimization algorithm is an optimization algorithm that imitates the whale hunting process. Each whale individual corresponds to a solution, and the whale individual position is continuously updated to search for the solution of the problem until the termination condition is met.

[0185] Whale Optimization Algorithm is a bionic optimization algorithm that simulates the behavior of whales in finding food. The algorithm corresponds one whale to one solution, and searches for the solution of the problem by continuously updating the position of the whale until the termination condition is met. The algorithm is based on the following three whale movement modes:

[0186] (1) Encirclement and hunting

[0187] Simulate the whale's spiral movement around the food. In the algorithm, by updating the position of individual whales, this spiral movement is simulated to approach the optimal solution:

[0188] (2) Searching for food

[0189] Simulate whales hunting through bubble nets. In the algorithm, randomly simulating whale position mutations is equivalent to searching for new solutions through bubble nets:

[0190] (3) Search

[0191] Simulate whales randomly searching for food. In the algorithm, random perturbations are added to individuals to increase population diversity and avoid falling into local optimality:

[0192] The present invention determines the weight coefficients of the low-frequency impedance model M1 and the low-frequency impedance model M2 by adopting the whale optimization algorithm, so that the solution process has the advantages of strong global search capability and fast convergence speed, and can quickly find the global optimal solution, further improving the quality of the final low-frequency impedance model M_final.

[0193] For other detailed descriptions and advantages of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0194] Example 3

[0195] According to another aspect of the present invention, an electronic device is provided. The electronic device comprises:

[0196] Memory, which stores executable instructions:

[0197] A processor runs the executable instructions in the memory to implement the low-frequency impedance model establishment method based on multivariate information control according to the present invention.

[0198] The method comprises the following steps:

[0199] Step 1: pre-process the logging curve to obtain an optimized logging curve, wherein the optimized logging curve includes a low-frequency impedance curve:

[0200] Step 2: Combine the optimized logging curve with seismic data and stratigraphic data to perform well-seismic calibration:

[0201] Step 3, extracting multiple seismic attributes from the seismic data:

[0202] Step 4: Use a local preservation projection algorithm to reduce the dimensions of the multiple seismic attributes to obtain optimized characteristic attributes:

[0203] Step 5: Using the optimized characteristic attributes and the low-frequency impedance curve, obtain a low-frequency impedance model M1 under the deep model framework:

[0204] Step 6: Using the layer data and the low-frequency impedance curve, inverse proportional weighting is used to perform well interpolation and extrapolation to obtain a low-frequency impedance model M2:

[0205] Step 7: Use the whale optimization algorithm to determine the weight coefficients of the low-frequency impedance model M1 and the low-frequency impedance model M2, perform linear combination, and obtain the final low-frequency impedance model M_final.

[0206] In some embodiments, in step 1, preprocessing the well logging curves includes correcting the well logging curves and performing inter-well consistency processing.

[0207] In some embodiments, in step 4, the multiple seismic attributes are dimensionally reduced based on the following formula to obtain an optimized characteristic attribute Y={y 1 ,y 2 ,...,y m}:

[0208] y i =B T x i

[0209] Where X = {x 1 ,x 2 ,...,x n} is the data set of the multiple seismic attributes, B is the projection matrix B, n is the number of the seismic attributes before dimensionality reduction, and m is the number of the characteristic attributes obtained after dimensionality reduction.

[0210] In some embodiments, in step 4, the projection matrix B is determined by minimizing the following objective function:

[0211]

[0212] Among them, W ij represents the weight matrix,

[0213]

[0214] In some embodiments, the optimized characteristic attributes include well bypass characteristic attributes, and step 5 specifically includes:

[0215] The well bypass characteristic attribute is extracted from the optimized characteristic attribute, and combined with the low-frequency impedance curve to produce a sample set:

[0216] The nonlinear mapping relationship between the characteristic attributes of the well bypass channel and the low-frequency impedance curve is learned based on the long short-term memory network:

[0217] The nonlinear mapping relationship is extended to the entire target area to obtain the low-frequency impedance properties of all positions in the target area to generate the low-frequency impedance model M1.

[0218] In some implementations, step 7 specifically includes:

[0219] The final low-frequency impedance model M_final is obtained based on the following formula:

[0220] M_final=α1*M1+α2*M2

[0221] Among them, α1 and α2 are weight coefficients determined based on the whale optimization algorithm.

[0222] Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc.

[0223] The processor may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In one embodiment of the present invention, the processor is used to run the computer-readable instructions stored in the memory.

[0224] The technical solution according to this embodiment integrates multivariate information, and through the combination of data drive and physical constraints, the final low-frequency impedance model obtained has strong adaptability, high consistency with well data, good resolution, and has achieved very beneficial technical effects.

[0225] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0226] Example 4

[0227] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the method for establishing a low-frequency impedance model based on multivariate information control according to the present invention is implemented.

[0228] The method comprises the following steps:

[0229] Step 1: pre-process the logging curve to obtain an optimized logging curve, wherein the optimized logging curve includes a low-frequency impedance curve:

[0230] Step 2: Combine the optimized logging curve with seismic data and stratigraphic data to perform well-seismic calibration:

[0231] Step 3, extracting multiple seismic attributes from the seismic data:

[0232] Step 4: Use a local preservation projection algorithm to reduce the dimensions of the multiple seismic attributes to obtain optimized characteristic attributes:

[0233] Step 5: Using the optimized characteristic attributes and the low-frequency impedance curve, obtain a low-frequency impedance model M1 under the deep model framework:

[0234] Step 6: Using the layer data and the low-frequency impedance curve, inverse proportional weighting is used to perform well interpolation and extrapolation to obtain a low-frequency impedance model M2:

[0235] Step 7: Use the whale optimization algorithm to determine the weight coefficients of the low-frequency impedance model M1 and the low-frequency impedance model M2, perform linear combination, and obtain the final low-frequency impedance model M_final.

[0236] In some embodiments, in step 1, preprocessing the well logging curves includes correcting the well logging curves and performing inter-well consistency processing.

[0237] In some embodiments, in step 4, the multiple seismic attributes are dimensionally reduced based on the following formula to obtain an optimized characteristic attribute Y={y 1 ,y 2 ,...,y m}:

[0238] y i =B T x i

[0239] Where X = {x 1 ,x 2 ,...,x n} is the data set of the multiple seismic attributes, B is the projection matrix B, n is the number of the seismic attributes before dimensionality reduction, and m is the number of the characteristic attributes obtained after dimensionality reduction.

[0240] In some embodiments, in step 4, the projection matrix B is determined by minimizing the following objective function:

[0241]

[0242] Among them, W ij represents the weight matrix,

[0243]

[0244] In some embodiments, the optimized characteristic attributes include well bypass characteristic attributes, and step 5 specifically includes:

[0245] The well bypass feature attribute is extracted from the optimized feature attribute, and combined with the low-frequency impedance curve to produce a sample set:

[0246] The nonlinear mapping relationship between the characteristic attributes of the well bypass channel and the low-frequency impedance curve is learned based on the long short-term memory network:

[0247] The nonlinear mapping relationship is extended to the entire target area to obtain the low-frequency impedance properties of all positions in the target area to generate the low-frequency impedance model M1.

[0248] In some implementations, step 7 specifically includes:

[0249] The final low-frequency impedance model M_final is obtained based on the following formula:

[0250] M_final=α1*M1+α2*M2

[0251] Among them, α1 and α2 are weight coefficients determined based on the whale optimization algorithm.

[0252] The computer-readable storage medium according to the embodiment of the present invention stores non-transitory computer-readable instructions, and when the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the above-mentioned methods of the embodiments of the present invention are executed.

[0253] The above-mentioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or mobile hard disk), media with built-in rewritable non-volatile memory (e.g., memory card) and media with built-in ROM (e.g., ROM box).

[0254] Those skilled in the art should be able to understand that in order to solve the technical problem of how to obtain a good user experience, the present embodiment may also include well-known structures such as a communication bus and an interface, and these well-known structures should also be included in the protection scope of the present invention.

[0255] The technical solution according to this embodiment integrates multivariate information, and through the combination of data drive and physical constraints, the final low-frequency impedance model obtained has strong adaptability, high consistency with well data, good resolution, and has achieved very beneficial technical effects.

[0256] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0257] Example 5

[0258] Figure 2 , Figure 3 and Figure 4 An application example according to an exemplary embodiment of the present invention is shown.

[0259] Figure 2 A typical well calibration map of the target area is shown.

[0260] Figure 3 FIG. 4 shows a cross-sectional schematic diagram of a well interpolation low-frequency impedance model according to an exemplary embodiment of the present invention. Figure 3 It can be seen that in the blind well area, the trend of the low-frequency impedance model M2 is less consistent with the trend of the low-frequency impedance on the well.

[0261] Figure 4 FIG. 4 shows a cross-sectional schematic diagram of a final low-frequency impedance model according to an exemplary embodiment of the present invention. Figure 4 It can be seen that in the blind well area, the consistency between the low-frequency impedance model trend and the low-frequency impedance trend on the well is improved to a certain extent compared with the low-frequency impedance model M2 obtained by conventional well interpolation. At the same time, the final low-frequency impedance model M_final integrated with seismic data has a significantly improved resolution compared with the low-frequency impedance model M2 obtained by conventional well interpolation.

[0262] For other detailed descriptions of this exemplary embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0263] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terms used herein are selected to best explain the principles of the embodiments, practical applications, or technical improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for establishing a low-frequency impedance model based on multivariate information control, It is characterized in that The method comprises: Step 1: pre-process the logging curve to obtain an optimized logging curve, wherein the optimized logging curve includes a low-frequency impedance curve: Step 2: Combine the optimized logging curve with seismic data and stratigraphic data to perform well-seismic calibration: Step 3, extracting multiple seismic attributes from the seismic data: Step 4: Use a local preservation projection algorithm to reduce the dimensions of the multiple seismic attributes to obtain optimized characteristic attributes: Step 5: Using the optimized characteristic attributes and the low-frequency impedance curve, obtain a low-frequency impedance model M1 under the deep model framework: Step 6: Using the layer data and the low-frequency impedance curve, inverse proportional weighting is used to perform well interpolation and extrapolation to obtain a low-frequency impedance model M2: Step 7: Use the whale optimization algorithm to determine the weight coefficients of the low-frequency impedance model M1 and the low-frequency impedance model M2, perform linear combination, and obtain the final low-frequency impedance model M_final.

2. The method according to claim 1, It is characterized in that, In step 1, preprocessing the well logging curve includes correcting the well logging curve and performing inter-well consistency processing.

3. The method according to claim 1, It is characterized in that In step 4, the multiple seismic attributes are dimensionally reduced based on the following formula to obtain an optimized characteristic attribute Y={y 1 ,y 2 ,...,y m }: y i =B T x i Where X = {x 1 ,x 2 ,...,x n } is the data set of the multiple seismic attributes, B is the projection matrix B, n is the number of the seismic attributes before dimensionality reduction, and m is the number of the characteristic attributes obtained after dimensionality reduction.

4. The method according to claim 3, It is characterized in that In step 4, the projection matrix B is determined by minimizing the following objective function: Among them, W ij represents the weight matrix, 5. The method according to claim 1, It is characterized in that The optimized characteristic attributes include the well bypass characteristic attributes, and the step 5 specifically includes: The well bypass feature attribute is extracted from the optimized feature attribute, and combined with the low-frequency impedance curve to produce a sample set: The nonlinear mapping relationship between the characteristic attributes of the well bypass channel and the low-frequency impedance curve is learned based on the long short-term memory network: The nonlinear mapping relationship is extended to the entire target area to obtain the low-frequency impedance properties of all positions in the target area to generate the low-frequency impedance model M1.

6. The method according to claim 1, It is characterized in that The step 7 specifically includes: The final low-frequency impedance model M_final is obtained based on the following formula: M_final=α1*M1+α2*M2 Among them, α1 and α2 are weight coefficients determined based on the whale optimization algorithm.

7. A low-frequency impedance model building device based on multivariate information control, It is characterized in that The device comprises: The logging curve preprocessing unit is used to preprocess the logging curve to obtain an optimized logging curve, wherein the optimized logging curve includes a low-frequency impedance curve: Well seismic calibration unit is used to combine the optimized logging curve with seismic data and layer data for well seismic calibration: A seismic attribute extraction unit, used to extract a plurality of seismic attributes from the seismic data: The seismic attribute dimension reduction unit is used to reduce the dimension of the multiple seismic attributes by using a local preservation projection algorithm to obtain optimized characteristic attributes: The first low-frequency impedance model acquisition unit is used to acquire a low-frequency impedance model M1 under a deep model framework by using the optimized characteristic attributes and the low-frequency impedance curve: The second low-frequency impedance model acquisition unit is used to use the layer data and the low-frequency impedance curve to perform well interpolation and extrapolation by inverse proportional weighting to acquire a low-frequency impedance model M2: The final low-frequency impedance model acquisition unit is used to determine the weight coefficients of the low-frequency impedance model M1 and the low-frequency impedance model M2 by using the whale optimization algorithm, perform linear combination, and obtain the final low-frequency impedance model M_final.

8. The device according to claim 7, It is characterized in that The earthquake attribute dimension reduction unit is specifically used for: The multiple seismic attributes are dimensionally reduced based on the following formula to obtain an optimized characteristic attribute Y={y 1 ,y 2 ,...,y m }: y i =B T x i Where X = {x 1 ,x 2 ,...,x n } is the data set of the multiple seismic attributes, B is the projection matrix B, n is the number of the seismic attributes before dimensionality reduction, and m is the number of the characteristic attributes obtained after dimensionality reduction.

9. The device according to claim 7, It is characterized in that The optimized characteristic attributes include well bypass characteristic attributes, and the first low-frequency impedance model acquisition unit is specifically used for: The well bypass feature attribute is extracted from the optimized feature attribute, and combined with the low-frequency impedance curve to produce a sample set: The nonlinear mapping relationship between the characteristic attributes of the well bypass channel and the low-frequency impedance curve is learned based on the long short-term memory network: The nonlinear mapping relationship is extended to the entire target area to obtain the low-frequency impedance properties of all positions in the target area to generate the low-frequency impedance model M1.

10. An electronic device, It is characterized in that The electronic device comprises: Memory, which stores executable instructions: A processor, wherein the processor runs the executable instructions in the memory to implement the method according to any one of claims 1 to 7.

Images