Seismic attribute fusion method based on thin plate spline base function

Abstract

The application provides a seismic attribute fusion method based on a thin-plate spline base function, comprising the following steps: step 1, calculating an attribute value set X of all seismic attributes participating in fusion; step 2, defining a thin-plate spline base function; step 3, establishing an initial network; step 4, calculating the distance of sample data to a cluster center; step 5, calculating the nearest neighbor cluster of the kth sample data x k ; step 6, forming an activation function of the thin-plate spline base function; step 7, establishing a network learning error judgment function; step 8, correcting a training weight coefficient, and completing seismic attribute fusion calculation according to the trained weight coefficient w i . The seismic attribute fusion method based on the thin-plate spline base function is more consistent with geological characteristics, meets the requirement of nonlinear seismic attribute fusion on a mathematical model, solves the nonlinear relationship fusion processing of seismic attribute data, and can enhance the accuracy of seismic attributes in reservoir prediction.

Description

Technical Field

[0001] This invention relates to the field of oilfield development technology, and in particular to a seismic attribute fusion method based on thin plate spline basis functions. Background Technology

[0002] Seismic attribute fusion technology is an attribute analysis method that has emerged in recent years. Due to the complexity of subsurface geological conditions, the application of any single seismic attribute cannot accurately interpret seismic events. As more and more seismic attributes are proposed, the optimization of seismic attributes and the comprehensive interpretation of multiple attributes have become particularly important. Multi-attribute fusion technology combines multiple attributes representing different characteristics of reservoirs through mathematical operations to obtain the optimal result. Seismic attribute fusion technology not only considers the impact of different attributes on reservoir prediction but also eliminates redundant geological information. It improves the accuracy of reservoir prediction while retaining as much effective information as possible. Commonly used attribute fusion methods today include cluster analysis, multi-source linear regression fusion technology, kernel principal component analysis, and color-based fusion technology.

[0003] Chinese patent application CN202010459269.1 discloses a seismic attribute fusion method and apparatus. The method includes: acquiring a sample set of a target reservoir region; extracting multi-scale seismic attribute sets for each target reservoir sample block in the sample set; inputting the multi-scale seismic attribute sets of each target reservoir sample block into a trained adaptive function neural network to identify the fused seismic attributes of the target reservoir region sample set; and updating the network parameters by iteratively deleting the weight coefficients of the weight matrix between the input layer and the hidden unit layer of the adaptive function neural network. This invention continuously updates the trained adaptive function neural network by iteratively deleting the weight coefficients of the weight matrix, maximizing the information contained in the fused seismic attributes identified by the trained adaptive function neural network, highlighting the geological characteristics of the target reservoir region, and improving the reliability and accuracy of seismic interpretation.

[0004] Chinese patent application CN201410427217.0 discloses a method and apparatus for multi-dimensional seismic attribute fusion. The method includes: extracting multiple seismic attribute information along geological interpretation stratigraphic data; dividing each seismic attribute information into a predetermined number of regional block point sets; selecting a seismic attribute information to characterize the geological target features of the common regional block point sets of all the seismic attribute information using an algorithm; and integrating the selected seismic attribute plane information point sets of all the seismic attribute information into a single seismic attribute plane information point set to obtain the fused seismic attribute information. The seismic attribute information fused by this scheme can effectively reflect the different geological characteristics of the entire geological body.

[0005] Chinese patent application CN201810912109.0 discloses a seismic attribute fusion method, apparatus, and storage medium, belonging to the field of seismic data interpretation technology. The method includes: for each location point of a target reservoir in a target oil and gas block, acquiring the attribute value corresponding to each of multiple seismic attributes at that location point; determining the fusion weight of each of the multiple seismic attributes, whereby the fusion weight characterizes the importance of the corresponding seismic attribute in the seismic attribute fusion process; and fusing the multiple attribute values ​​at each location point according to the fusion weight of each seismic attribute to obtain the seismic attribute fusion result of the target reservoir. This invention solves the problem of multiple solutions for the distribution results of oil and gas reservoirs in related technologies by setting different fusion weights for each of the multiple seismic attributes and then fusing the multiple seismic attributes according to the fusion weights of each seismic attribute.

[0006] The existing technologies described above are significantly different from the present invention and have failed to solve the technical problem we want to address. Therefore, we have invented a new seismic attribute fusion method based on thin plate spline basis functions. Summary of the Invention

[0007] The purpose of this invention is to provide a seismic attribute fusion method based on thin-plate spline basis functions that can enhance the accuracy of seismic attributes in reservoir prediction.

[0008] The objective of this invention can be achieved through the following technical measures: a seismic attribute fusion method based on thin-plate spline basis functions, which includes:

[0009] Step 1: Calculate the set of attribute values ​​X for all seismic attributes involved in the fusion;

[0010] Step 2, define the basis functions for the thin plate spline;

[0011] Step 3, establish the initial network;

[0012] Step 4: Calculate the distance from the sample data to the cluster center;

[0013] Step 5, calculate the k-th sample data x k Nearest neighbor clustering;

[0014] Step 6: Form the activation function of the thin-plate spline basis function;

[0015] Step 7: Establish the network learning error judgment function;

[0016] Step 8: Adjust the training weight coefficients based on the trained weight coefficients w. i Complete the earthquake attribute fusion calculation.

[0017] The objective of this invention can also be achieved through the following technical measures:

[0018] In step 1, based on the well coordinates, in a certain seismic attribute data s i Extract the attribute values ​​of all wells and form a set denoted as x. n Using the same method, the x-values ​​of all seismic attributes involved in the fusion were calculated. n This forms a data set X.

[0019] In step 1, based on the attribute data x in set X n The number of nodes is used to determine the number of input layer nodes n, and the above set X is imported as the input layer;

[0020] X = {x1, x2, ..., x} n},x i ∈R n Formula 1

[0021] Where each x n They are all composed of several pairs of seismic attribute data (x 1 ,y 1 )composition.

[0022] In step 2, the width r of the thin plate spline basis function is determined, a vector A(l) is defined to store the sum of the output vectors belonging to each class, and a counter B(l) is defined to count the number of samples belonging to each class, where l is the number of classes. The thin plate spline basis function is as follows:

[0023] f i (x)=||xc i 2 ln(||xc i ||) i=1,2,…,m Equation 2

[0024] In the formula: x is an n-dimensional input vector, i.e., x in step 1. n c i is the center value of the thin-plate spline basis function of the i-th attribute, a vector with the same dimension as x; m is the number of sensing units, i.e., the number of hidden layer nodes; ||xc i || is the vector xc i The norm of x, which represents the relationship between x and c. i The distance between them; f in Equation 2 i (x) in c i There is a unique maximum value at a certain point, as ||xc i As || increases, f i (x) decays rapidly to zero; for a given input x∈R n Only a small portion near the center of x is activated.

[0025] In step 3, arbitrarily select a seismic attribute data x from set X. n From x n The first data pair (x 1 ,y 1 (Starting at x) 1 Establish a cluster center on the x-axis, and let c1 = x. 1 A(1)=y 1 , B(1)=1; Establish the initial network. At this time, there is only one hidden unit. The center of the hidden unit is c1. The weight vector from the hidden unit to the output layer is w1=A(1) / B(1).

[0026] In step 4, calculate x n The second sample data pair (x 2 ,y 2 Find x. 2 Distance to cluster center c1 |x 2 -c1|.

[0027] In step 4, if |x 2 -c1|≤r, where r is the width of the thin plate spline basis function defined in step 2, then c1 is x 2 The nearest neighbor clustering, and let A(1) = y 1 +y 2 , B(1)=2, w1=A(1) / B(1).

[0028] In step 4, if |x 2 -c1|>r, then x 2 As a new cluster center, let c2 = x 2 A(2)=y 2 B(2) = 1; Add another hidden unit to the network established in step 3, and the weight vector from the hidden unit to the output layer is w2 = A(2) / B(2).

[0029] In step 5, assume that the calculation is performed up to the kth sample data pair (x k ,y k When k = 3, 4, ..., there are M cluster centers, meaning the network established above has M hidden units with center points c1, c2, ..., c3. M Then calculate the distances |x| to each of the M cluster centers. k -c i |, i = 1, 2, ..., M, if |x k -c j | is the minimum distance among these distances, i.e., c j For x k The nearest neighbor clustering.

[0030] In step 5, if |x k -c j |>r, then x k As a new cluster center, and making c M+1 =x k M = M + 1, A(M) = y k B(M) = 1; and keep the values ​​of A(i) and B(i) unchanged, i = 1, 2, ..., M-1, and add the Mth hidden unit to the established network.

[0031] In step 5, if |x k -c j If |≤r, calculate as follows: A(j)=A(j)+y k B(j) = B(j) + 1; when i ≠ j, i = 1, 2, ..., M, and keep the values ​​of A(i) and B(i) unchanged, the weight vector from the hidden unit to the output layer is w i =A(i) / B(i), i=1,2,…,M.

[0032] In step 6, the network structure established according to the above rules and the initial weight coefficients w i The activation function that forms the basis functions of the thin-plate spline is given by the following formula:

[0033]

[0034] The size of the radius r determines the complexity of the dynamic adaptive radial basis function network; the smaller r is, the more clusters are obtained, and the greater the computational cost. An appropriate r is found through experiments and error information.

[0035] In step 7, the error judgment function is established. Let the input sample data be {x}. i The expression {i = 1, 2, ..., N} and the output sample data {y} are given. i {i = 1, 2, ..., N}, where N represents the number of wells, establish a network learning error judgment function:

[0036]

[0037] Where q is the iteration number, E q Let f(x) be the error between the actual value and the expected value in the qth iteration. i ) is the expected value, which is calculated by the activation function.

[0038] In step 8, the training weight coefficients are adjusted. Partial derivatives of Equation 4 are calculated to establish Equation 5. Iterative calculations are then performed, continuously adjusting the weight coefficients w. i :

[0039]

[0040] Where α1 is the learning rate, i.e., a given constant, E is the error in Equation 5, and w i For the corrected weighting coefficients, w' i These are the weighting coefficients before correction.

[0041] In step 8, based on the trained weight coefficients w i The following formula is used to complete the seismic attribute fusion calculation:

[0042] S = w1·s1 + w2·s2 + ... + w i ·s i Formula 6

[0043] Where S represents the fused seismic attribute, w i The weights s are the weights after the above training. i This is earthquake attribute data.

[0044] This invention relates to a seismic attribute fusion method based on thin-plate spline basis functions. This method is applicable to areas where seismic attributes do not clearly reflect geological features and exhibit strong ambiguity. It addresses situations where nonlinear relationships exist between seismic attributes, making it difficult to predict seismic data based on any single attribute. By optimizing the radial basis function based on thin-plate spline basis functions, a new activation function is formed, resulting in higher accuracy in training the weight coefficients. This eliminates redundant information between seismic attributes and improves the accuracy of seismic attribute prediction. This invention combines the slow-changing characteristics of thin-plate spline basis functions with the strong ability of neural networks to handle nonlinear relationships, constructing a new activation function. This allows for the training of more accurate weight coefficients, obtaining seismic attribute fusion data that better reflects geological significance, achieving the goal of eliminating redundant information and improving the accuracy of seismic interpretation. Attached Figure Description

[0045] Figure 1 This is an energy half-time property diagram in a specific embodiment of the present invention;

[0046] Figure 2 This is a diagram showing the instantaneous frequency attribute in a specific embodiment of the present invention;

[0047] Figure 3 This is a maximum amplitude attribute diagram in a specific embodiment of the present invention;

[0048] Figure 4 This is a fused sand body thickness attribute map (without well display) in a specific embodiment of the present invention;

[0049] Figure 5 This is a fusion-based sand body thickness attribute map (with well display) in a specific embodiment of the present invention;

[0050] Figure 6 This is a flowchart of a specific embodiment of the seismic attribute fusion method based on thin plate spline basis functions of the present invention. Detailed Implementation

[0051] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0052] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, and / or combinations thereof.

[0053] This invention leverages the advantages of thin-plate spline basis functions to construct a novel neural network activation function, thereby training more accurate weight coefficients and forming a completely new seismic attribute fusion method. This overcomes the inherent limitations of existing linear data processing methods. Addressing the nonlinear relationships between seismic attributes, this invention establishes an activation function suitable for geological characteristics and proposes a nearest-neighbor clustering method, a thin-plate spline basis function model, and a weight coefficient correction model. Its advantages lie in combining the slow-changing characteristics of thin-plate spline functions with the strong ability of neural networks to handle nonlinear relationships. This results in a slow-changing characteristic not found in Gaussian functions, better reflecting geological characteristics and meeting the mathematical model requirements for nonlinear seismic attribute fusion. It solves the problem of nonlinear relationship fusion processing of seismic attribute data, enhancing the accuracy of seismic attributes in reservoir prediction.

[0054] The following are several specific embodiments of the application of the present invention.

[0055] Example 1

[0056] In a specific embodiment 1 of the present invention, such as Figure 6 As shown, Figure 6 This is a flowchart of the seismic attribute fusion method based on thin plate spline basis functions according to the present invention.

[0057] Step 101: Calculate the set of attribute values ​​X for all earthquake attributes involved in the fusion.

[0058] Step 102: Define the basis functions for the thin plate spline.

[0059] Step 103: Establish the initial network.

[0060] Step 104, calculate x nThe second sample data pair (x 2 ,y 2 Find x. 2 Distance to cluster center c1 |x 2 -c1|.

[0061] Step 105, calculate the k-th sample data x k The nearest neighbor clustering.

[0062] Step 106: Form the activation function of the thin plate spline basis function.

[0063] Step 107: Establish the network learning error judgment function.

[0064] Step 108: Adjust the training weight coefficients.

[0065] Step 109, based on the trained weight coefficients w i Complete the earthquake attribute fusion calculation.

[0066] Example 2

[0067] In a specific embodiment 2 of the present invention, the seismic attribute fusion method based on thin plate spline basis functions includes the following steps:

[0068] Step 1: Based on the well coordinates, in a certain seismic attribute data s i Extract the attribute values ​​of all wells and form a set denoted as x. n The same method was used to calculate the x-values ​​of all seismic attributes involved in the fusion. n This forms a data set X. Based on the attribute data x in set X... n The number of nodes is used to determine the number of input layer nodes n, and the above set X is imported as the input layer.

[0069] Step 2: Set the width r (a given constant) of the thin-plate spline basis function, define a vector A(l) to store the sum of the output vectors belonging to each class, and define a counter B(l) to count the number of samples belonging to each class, where l is the number of classes. The thin-plate spline basis function is as follows:

[0070] f i (x)=||xc i || 2 ln(||xc i ||)i=1,2,…,m Equation 2

[0071] In the formula: x is an n-dimensional input vector (x in step 1) n );c i is the center value of the thin-plate spline basis function of the i-th attribute, a vector with the same dimension as x; m is the number of sensing units (i.e., the number of hidden layer nodes); ||xci || is the vector xc i The norm of x, which represents the relationship between x and c. i The distance between them; f in Equation 2 i (x) in c i There is a unique maximum value at a certain point, as ||xc i As || increases, f i (x) decays rapidly to zero. For a given input x∈R n Only a small portion near the center of x is activated.

[0072] Step 3: Randomly select a seismic attribute data x from set X. n From x n The first data pair (x 1 ,y 1 (Starting at x) 1 Establish a cluster center on the x-axis, and let c1 = x. 1 A(1)=y 1 B(1) = 1. Establish the initial network. At this time, there is only one hidden unit. The center of the hidden unit is c1. The weight vector from the hidden unit to the output layer is w1 = A(1) / B(1).

[0073] Step 4: Calculate x n The second sample data pair (x 2 ,y 2 Find x. 2 Distance to cluster center c1 |x 2 -c1|.

[0074] Step 5: Assume that the calculation is performed up to the kth sample data pair (x k ,y k When k = 3, 4, ..., there are M cluster centers (the network established above has M hidden units), and their center points are c1, c2, ..., c3. M Then calculate the distances |x| to each of the M cluster centers. k -c i |, i = 1, 2, ..., M, if |x k -c j | is the minimum distance among these distances, i.e., c j For x k The nearest neighbor clustering.

[0075] Step 6: Establish the network structure and initial weight coefficients w based on the above rules. i The activation function that forms the basis functions of the thin-plate spline is given by the following formula:

[0076]

[0077] The size of the radius r determines the complexity of the dynamic adaptive radial basis function network. The smaller r is, the more clusters are obtained, and the greater the computational cost. Usually, an appropriate r can be found through experiments and error information.

[0078] Step 7: Establish the error judgment function. Let the input sample data be {x}. i The expression {i = 1, 2, ..., N} and the output sample data {y} are given. i Let i = 1, 2, ..., N, where N represents the number of wells. Establish a network learning error judgment function.

[0079]

[0080] Where q is the iteration number, E q Let f(x) be the error between the actual value and the expected value in the qth iteration. i ) is the expected value, which is calculated by the activation function (Equation 3).

[0081] Step 8: Adjust the training weight coefficients. Take the partial derivative of Equation 4 to establish Equation 5. Iterate the calculation and continuously adjust the weight coefficients w. i .

[0082]

[0083] Where α1 is the learning rate (a given constant), E is the error in Equation 5, and w i For the corrected weighting coefficients, w' i These are the weighting coefficients before correction.

[0084] Step 9: Based on the trained weight coefficients w i The following formula is used to complete the earthquake attribute fusion calculation.

[0085] S = w1·s1 + w2·s2 + ... + w i ·s i Formula 6

[0086] Where S represents the fused seismic attribute, w i The weights s are the weights after the above training. i For seismic attribute data (with step 1s) i same).

[0087] Example 3

[0088] In a specific embodiment 3 of the present invention, the seismic attribute fusion method based on thin plate spline basis functions of the present invention includes the following steps:

[0089] Step 1: Based on the well coordinates, in a certain seismic attribute data s i Extract the attribute values ​​of all wells. Figures 1 to 3 Let x be the set of the given set. n The same method was used to calculate the x-values ​​of all seismic attributes involved in the fusion. n This forms a data set X. Based on the attribute data x in set X... n The number of nodes is used to determine the number of input layer nodes n, and the above set X is imported as the input layer.

[0090] X = {x1, x2, ..., x} n},x i ∈R n Formula 1

[0091] Where each x n They are all composed of several pairs of seismic attribute data (x 1 ,y 1 )composition.

[0092] Step 2: Set the width r (a given constant) of the thin-plate spline basis function, define a vector A(l) to store the sum of the output vectors belonging to each class, and define a counter B(l) to count the number of samples belonging to each class, where l is the number of classes. The thin-plate spline basis function is as follows:

[0093] f i (x)=||xc i || 2 ln(||xc i (i = 1, 2, ..., m) Equation 2

[0094] In the formula: x is an n-dimensional input vector (x in step 1) n );c i is the center value of the thin-plate spline basis function of the i-th attribute, a vector with the same dimension as x; m is the number of sensing units (i.e., the number of hidden layer nodes); ||xc i || is the vector xc i The norm of x, which represents the relationship between x and c. i The distance between them; f in Equation 2 i (x) in c i There is a unique maximum value at a certain point, as ||xc i As || increases, f i (x) decays rapidly to zero. For a given input x∈R n Only a small portion near the center of x is activated.

[0095] Step 3: Randomly select a seismic attribute data x from set X. n From x n The first data pair (x 1 ,y 1 (Starting at x) 1Establish a cluster center on the x-axis, and let c1 = x. 1 A(1)=y 1 B(1) = 1. Establish the initial network. At this time, there is only one hidden unit. The center of the hidden unit is c1. The weight vector from the hidden unit to the output layer is w1 = A(1) / B(1).

[0096] Step 4: Calculate x n The second sample data pair (x 2 ,y 2 Find x. 2 Distance to cluster center c1 |x 2 -c1| is calculated as follows:

[0097] Where, if |x 2 -c1|≤r (where r is the width of the thin plate spline basis function defined in step 2), then c1 is x 2 The nearest neighbor clustering, and let A(1) = y 1 +y 2 , B(1)=2, w1=A(1) / B(1);

[0098] If |x 2 -c1|>r, then x 2 As a new cluster center, let c2 = x 2 A(2)=y 2 B(2) = 1. Add another hidden unit to the network established in step 3. The weight vector from this hidden unit to the output layer is w2 = A(2) / B(2).

[0099] Step 5: Assume that the calculation is performed up to the kth sample data pair (x k ,y k When k = 3, 4, ..., there are M cluster centers (the network established above has M hidden units), and their center points are c1, c2, ..., c3. M Then calculate the distances |x| to each of the M cluster centers. k -c i |, i = 1, 2, ..., M, if |x k -c j | is the minimum distance among these distances, i.e., c j For x k The nearest neighbor clustering.

[0100] Where, if |x k -c j |>r, then x k As a new cluster center, and making c M+1 =x k M = M + 1, A(M) = y kB(M) = 1. Keep the values ​​of A(i) and B(i) unchanged, i = 1, 2, ..., M-1, and add the Mth hidden unit to the established network.

[0101] If |x k -c j If |≤r, calculate as follows: A(j)=A(j)+y k B(j) = B(j) + 1. When i ≠ j, i = 1, 2, ..., M, and the values ​​of A(i) and B(i) remain unchanged, the weight vector from the hidden unit to the output layer is w. i =A(i) / B(i), i=1,2,…,M.

[0102] Step 6: Establish the network structure and initial weight coefficients w based on the above rules. i The activation function that forms the basis functions of the thin-plate spline is given by the following formula:

[0103]

[0104] The size of the radius r determines the complexity of the dynamic adaptive radial basis function network. The smaller r is, the more clusters are obtained, and the greater the computational cost. Usually, an appropriate r can be found through experiments and error information.

[0105] Step 7: Establish the error judgment function. Let the input sample data be {x}. i The expression {i = 1, 2, ..., N} and the output sample data {y} are given. i Let i = 1, 2, ..., N, where N represents the number of wells. Establish a network learning error judgment function.

[0106]

[0107] Where q is the iteration number, E q Let f(x) be the error between the actual value and the expected value in the qth iteration. i ) is the expected value, which is calculated by the activation function (Equation 3).

[0108] Step 8: Adjust the training weight coefficients. Take the partial derivative of Equation 4 to establish Equation 5. Iterate the calculation and continuously adjust the weight coefficients w. i .

[0109]

[0110] Where α1 is the learning rate (a given constant), E is the error in Equation 5, and w i For the corrected weighting coefficients, w' i These are the weighting coefficients before correction.

[0111] Step 9: Based on the trained weight coefficients w iThe following formula is used to complete the earthquake attribute fusion calculation.

[0112] S = w1·s1 + w2·s2 + ... + w i ·s i Formula 6

[0113] Where S represents the fused seismic attribute ( Figure 4 , Figure 5 ), w i The weights s are the weights after the above training. i For seismic attribute data (with step 1s) i same).

[0114] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0115] Except for the technical features described in the specification, all other technologies are known to those skilled in the art.

Claims

1. A seismic attribute fusion method based on thin plate spline basis functions, characterized in that, The seismic attribute fusion method based on thin-plate spline basis functions includes: Step 1: Calculate the set of attribute values ​​for all seismic attributes involved in the fusion. ; Step 2, define the basis functions for the thin plate spline; Step 3, establish the initial network; Step 4: Calculate the distance from the sample data to the cluster center; Step 5, calculate the first... Sample data Nearest neighbor clustering; Step 6: Form the activation function for the thin-plate spline basis function; Step 7: Establish the network learning error judgment function; Step 8: Adjust the training weight coefficients based on the trained weight coefficients. Complete the earthquake attribute fusion calculation; In step 2, the width of the thin plate spline basis function is defined as... Define a vector A counter is defined to store the sum of the output vectors belonging to each class. Used to count the number of samples belonging to each category. Let be the number of categories, where the thin-plate spline basis functions are as follows: Formula 2; In the formula: yes 3D input vector ; It is the first The central values ​​of the thin-plate spline basis functions of each attribute, and Vectors with the same dimension; It is the number of sensing units, i.e., the number of hidden layer nodes; It is a vector The norm of , which represents and The distance between them; in Equation 2 exist There is a unique maximum value at that point, and with The increase, It decays rapidly to zero for a given input. Only a small portion are close The center is activated; In step 3, from the set of attribute values Select any earthquake attribute data ,from The first data pair ( ) beginning, in Establish a cluster center on the top, and let , ; Establish the initial network, which at this point has only one hidden unit, and the center of this hidden unit is The weight vector from the hidden unit to the output layer is ; In step 4, calculate The second sample data pair ( Find out arrive The distance to this cluster center ; In step 4, if ,but for The nearest neighbor cluster, and let , , ; In step 4, if Then As a new cluster center, and making , , Add another hidden unit to the network established in step 3. The weight vector from this hidden unit to the output layer is... .

2. The seismic attribute fusion method based on thin plate spline basis functions according to claim 1, characterized in that, In step 1, based on the well coordinates, in a certain seismic attribute data Extract the attribute values ​​of all wells and form a set denoted as . The same method was used to calculate all seismic attributes involved in the fusion. , forming a set of attribute values .

3. The seismic attribute fusion method based on thin plate spline basis functions according to claim 2, characterized in that, In step 1, based on the set of attribute values Attribute data The number of nodes is used to determine the number of input layer nodes n, and the above attribute values ​​are set together. Imported as an input layer; Formula 1; Each of them They are all composed of several pairs of seismic attribute data ( )composition.

4. The seismic attribute fusion method based on thin plate spline basis functions according to claim 1, characterized in that, In step 5, assuming the calculation reaches the [number]th ... Each sample data pair ( When ), there exists The network established by each cluster center contains There are 1 hidden unit, whose center points are respectively , Then calculate the results separately. Distance between cluster centers | |, If | | is the minimum distance among these distances, then for The nearest neighbor clustering.

5. The seismic attribute fusion method based on thin plate spline basis functions according to claim 4, characterized in that, In step 5, if Then As a new cluster center, and making , , , And maintain The value remains unchanged And add the third one to the established network. Hidden unit.

6. The seismic attribute fusion method based on thin plate spline basis functions according to claim 5, characterized in that, In step 5, if The following calculations are performed: ;when hour, and maintain The value remains unchanged, and the weight vector from the hidden unit to the output layer is... , .

7. The seismic attribute fusion method based on thin plate spline basis functions according to claim 6, characterized in that, In step 6, based on the established network structure and the initial weight coefficients... The activation function that forms the basis functions of the thin-plate spline is given by the following formula: Formula 3; The size of the variable determines the complexity of the dynamic adaptive radial basis network; The smaller the value, the more clusters are obtained, and the greater the computational cost. An appropriate value needs to be found through experiments and error information. .

8. The seismic attribute fusion method based on thin plate spline basis functions according to claim 1, characterized in that, In step 7, the error judgment function is established, assuming the input sample data is... and output sample data , To represent the number of wells, establish a function to judge the network learning error: Equation 4; in, For the number of iterations, For the first The error between the true value and the expected value The expected value is calculated using the activation function.

9. The seismic attribute fusion method based on thin plate spline basis functions according to claim 8, characterized in that, In step 8, the training weight coefficients are adjusted by taking the partial derivative of Equation 4, establishing Equation 5, and iteratively calculating and continuously adjusting the weight coefficients. : Formula 5; in, The learning rate is a given constant. The error in Equation 5, These are the corrected weighting coefficients. These are the weighting coefficients before correction.

10. The seismic attribute fusion method based on thin plate spline basis functions according to claim 9, characterized in that, In step 8, based on the trained weight coefficients The following formula is used to complete the seismic attribute fusion calculation: Formula 6; in, For the merged earthquake attributes, These are the weight coefficients after the above training. This is earthquake attribute data.

Images