A geophysical identification method for medium-deep undercompacted low-velocity mudstone

Abstract

The present invention provides a geophysical identification method for mid-deep undercompacted low-velocity mudstone. By statistically analyzing data, representative P-wave velocity, S-wave velocity, and density of low-velocity mudstone and gas-bearing sandstone in well logging data are obtained. Combining geological information, a theoretical model of the superimposed relationship between low-velocity mudstone and gas-bearing sandstone is designed, and convolution forward modeling and prestack forward modeling are carried out on it. The post-stack and prestack seismic attributes of the model are calculated respectively. Then, sensitive attributes are optimized from these attributes, and the obtained sensitive attributes are used as learning samples, and the superimposed relationship is identified by probability self-organizing neural network clustering. The present invention adds prestack attributes to the optimization of sensitive attributes, making the fuzzy self-organizing neural network more accurate in identifying the superimposed relationship between gas-bearing sandstone and low-velocity mudstone, effectively distinguishing low-velocity mudstone from gas-bearing sandstone, and providing support for subsequent effective reservoir prediction.

Description

Technical Field

[0001] The present invention provides a geophysical identification method for mid - deep undercompacted low - velocity mudstone, belonging to the technical field of geological exploration. Background Art

[0002] The mid - deep layers of the Yingqiong Basin are characterized by high temperature (ultra - high pressure), and there are undercompacted low - velocity mudstones and complex structures developed, resulting in complex seismic responses and increasing the difficulty of reservoir prediction. The undercompacted low - velocity mudstones formed by rapid deposition under high temperature (ultra - high pressure) have a drastic velocity change in some areas, with a large variation range, which directly affects the seismic reflection characteristics of the top surface of the gas layer. From the petrophysical characteristics of low - velocity mudstones analyzed from known logging data, it can be seen that the longitudinal wave impedance of low - velocity mudstones is basically the same as or even lower than that of gas - bearing sandstones. This leads to the formation of "false bright spots" reflections when low - velocity mudstones contact background mudstones with high impedance on the post - stack section, and "dark spots" reflections when low - velocity mudstones contact gas - bearing sandstones, which seriously affects the identification and prediction of effective reservoirs. Therefore, it is necessary to clarify the distribution of low - velocity mudstones under high temperature and high pressure and the characteristic laws of their different stacking patterns with gas - bearing sandstones, so as to provide support for the subsequent selection of favorable target exploration areas and reservoir prediction in development.

[0003] Identifying lithology and thickness by optimizing sensitive attributes is a commonly used method in the field of lithology identification. By optimizing the sensitive attributes of sand body thickness, an intelligent well - seismic combined sand body prediction method is carried out for various types of sand bodies with different stacking relationships. The attributes used by this technology to identify the stacking relationship of sand bodies are limited to post - stack attributes, and it is always difficult to distinguish gas - bearing sandstones from low - velocity mudstones on the post - stack section. Therefore, this method is not applicable to the identification of the stacking relationship between low - velocity mudstones and gas - bearing sandstones. Summary of the Invention

[0004] The present invention mainly focuses on inventing a geophysical identification method for mid - deep undercompacted low - velocity mudstone. This method optimizes the post - stack and pre - stack sensitive attributes of different stacking relationships between low - velocity mudstones and gas - bearing sandstones, and uses them as inputs for clustering with a fuzzy self - organizing neural network, so as to distinguish different stacking relationships between low - velocity mudstones and gas - bearing sandstones, and further identify the distribution and thickness of low - velocity mudstones.

[0005] A geophysical identification method for mid - deep undercompacted low - velocity mudstone. By statistically analyzing data, representative longitudinal wave velocity, transverse wave velocity, and density of low - velocity mudstones and gas - bearing sandstones in logging data are obtained. Combining geological information, a theoretical model of the stacking relationship between low - velocity mudstones and gas - bearing sandstones is designed, and convolution forward modeling and pre - stack forward modeling are carried out on it. The post - stack and pre - stack seismic attributes of the model are calculated respectively, and then these attributes are optimized for sensitive attributes. The obtained sensitive attributes are used as learning samples, and the stacking relationship is identified by clustering with a probabilistic self - organizing neural network.

[0006] Specifically, it includes the following steps:

[0007] Step 1: Analyze well logging data and seismic data to obtain typical petrophysical parameters and contact relationships of low-velocity mudstone, gas-bearing sandstone, and background mudstone, and establish a theoretical geological model of the superposition relationship between low-velocity mudstone and gas-bearing sandstone;

[0008] Step 2: Conduct convolution forward modeling on the superposition model of low-velocity mudstone and gas-bearing sandstone, and analyze whether the forward modeling results can represent the response characteristics of different contact relationships, such as the "dark spot" reflection phenomenon of the reservoir caused by the overlying low-velocity mudstone on the gas-bearing sandstone and the "false bright spot" reflection phenomenon caused by the contact between low-velocity mudstone and background mudstone. If not, return to Step 1 and modify the parameters of the superposition model.

[0009] Step 3: Conduct prestack forward modeling on the superposition mode of low-velocity mudstone and gas-bearing sandstone, and then calculate the intercept attribute P and gradient attribute G in its prestack attributes, calculate different combinations of P and G, and obtain prestack combined attributes.

[0010] Step 4: Select appropriate time windows to extract various types of seismic attributes, and use the method of principal component analysis (PCA) to evaluate the sensitivity of each seismic attribute between different superposition relationships, so as to optimize the sensitive attributes that can distinguish different superposition relationships.

[0011] The implementation steps of PCA are as follows:

[0012] 1. Standardize the data

[0013]

[0014] where \(x_{ij}\) ij represents the \(j\)-th seismic attribute of the \(i\)-th superposition relationship, \(n\) is the number of superposition relationships, and \(p\) is the number of seismic attributes.

[0015]

[0016] where \(\mu_j\) j is the mean of the \(j\)-th seismic attribute, and \(\sigma_j\) j is its standard deviation.

[0017] 2. Conduct principal component analysis on the standardized matrix \(Z\) and calculate its covariance matrix \(C\):

[0018]

[0019] Perform eigenvalue decomposition on the covariance matrix \(C\) to obtain eigenvalues and eigenvectors. The eigenvectors correspond to the principal component directions, while the eigenvalues represent the variance magnitudes of each principal component. By examining the principal component loadings (components of the eigenvectors), determine which seismic attributes contribute the most to the principal components, and thus optimize the sensitive attributes.

[0020] Step 5: Input the multiple sensitive seismic attributes optimized in Step 4 into the fuzzy self-organizing neural network simultaneously. The network conducts self-organizing learning and competition based on the characteristics of the seismic attributes to check whether the results can distinguish the different superimposed relationships between the low-velocity mudstone and gas-bearing sandstone in the model. If not, it indicates that the sensitive attributes optimized in Step 4 are inappropriate and need to be further optimized. If so, it means that the sensitive attributes optimized in Step 4 are sufficient to distinguish different superimposed relationships and can be used for actual data.

[0021] Step 6: Extract the sensitive attributes optimized in Steps 4 and 5 along a suitable time window in the actual seismic data. After preprocessing, input them into the fuzzy self-organizing neural network to obtain the clustering results. Combine with well logging data to interpret the possible superimposed relationships corresponding to each category, and then judge the distribution and thickness of the regional low-velocity mudstone.

[0022] As a method for identifying low-velocity mudstone by optimizing sensitive attributes, the present invention adds prestack attributes to the optimization of sensitive attributes, making the fuzzy self-organizing neural network more accurate in identifying the superimposed relationship between gas-bearing sandstone and low-velocity mudstone, effectively distinguishing low-velocity mudstone from gas-bearing sandstone, and providing support for subsequent effective reservoir prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is the superimposed model of low-velocity mudstone and gas-bearing sandstone of the present invention;

[0024] Figure 2 The recognition results before and after adding prestack attributes of the present invention;

[0025] Figure 3 It is the topological structure diagram of the fuzzy self-organizing neural network of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0026] The present invention will be described in detail below with reference to the accompanying drawings of the embodiments:

[0027] A geophysical identification method for medium-deep undercompacted low-velocity mudstone includes the following steps:

[0028] Step 1: Analyze well logging data and seismic data to obtain the typical petrophysical parameters and contact relationships of low-velocity mudstone, gas-bearing sandstone, and background mudstone, and further establish a theoretical geological model of the superimposed relationship between low-velocity mudstone and gas-bearing sandstone, such as Figure 1 . There are a total of 8 superimposed patterns: ① gas-bearing sandstone; ② "thick sand - thin mud - thin sand" superimposed pattern; ③ "upper sand - lower mud" superimposed pattern; ④ multi-stage sand-mud superimposed pattern; ⑤ "upper mud - lower sand" superimposed pattern; ⑥ "thick mud - thin sand - thin mud" superimposed pattern; ⑦ low-velocity mudstone; ⑧ background mudstone.

[0029] Step 2: Conduct convolution forward modeling on the superposition model of low-velocity mudstone and gas-bearing sandstone, and analyze whether the forward modeling results can represent the response characteristics of different contact relationships, such as the "dark spot" reflection phenomenon of the reservoir caused by the overlying low-velocity mudstone on the gas-bearing sandstone and the "false bright spot" reflection phenomenon caused by the contact between the low-velocity mudstone and the background mudstone. If not, return to Step 1 and modify the parameters of the superposition model.

[0030] Step 3: Conduct pre-stack forward modeling on the superposition pattern of low-velocity mudstone and gas-bearing sandstone, and then calculate the intercept attribute P and gradient attribute G in its pre-stack attributes, calculate different combinations of P and G, such as combination methods like P*G, P+G, etc., to obtain pre-stack combined attributes.

[0031] Step 4: Select appropriate time windows to extract various types of post-stack seismic attributes, such as amplitude-based, waveform-based, and frequency-based seismic attributes, and pre-stack attributes, such as pre-stack attributes like P, G, P*G, P+G, etc., and use the principal component analysis (PCA) method to evaluate the sensitivity of each seismic attribute between different superposition relationships, so as to optimize and select the sensitive attributes that can distinguish different superposition relationships.

[0032] The implementation steps of PCA are as follows:

[0033] 1. Standardize the data

[0034]

[0035] where \(x_{ij}\) ij represents the \(j\)-th seismic attribute of the \(i\)-th superposition relationship, \(n\) is the number of superposition relationships, and \(p\) is the number of seismic attributes.

[0036]

[0037] where \(\mu_j\) j is the mean of the \(j\)-th seismic attribute, and \(\sigma_j\) j is its standard deviation.

[0038] 2. Conduct principal component analysis on the standardized matrix \(Z\) and calculate its covariance matrix \(C\):

[0039]

[0040] Perform eigenvalue decomposition on the covariance matrix \(C\) to obtain eigenvalues and eigenvectors. The eigenvectors correspond to the principal component directions, and the eigenvalues represent the variance magnitudes of each principal component. By examining the principal component loadings (components of the eigenvectors), it is possible to determine which seismic attributes contribute the most to the principal components, thereby optimizing and selecting the sensitive attributes.

[0041] Step 5: Input the multiple sensitive seismic attributes optimized in Step 4 into the fuzzy self-organizing neural network simultaneously. The network conducts self-organizing learning competition based on the characteristics of the seismic attributes to check whether the results can distinguish the different superimposed relationships between the low-velocity mudstone and gas-bearing sandstone in the model. If not, it indicates that the sensitive attributes optimized in Step 4 are inappropriate and need to be further optimized. If it can distinguish, it means that the sensitive attributes optimized in Step 4 are sufficient to distinguish different superimposed relationships and can be used for actual data. Figure 2 These are the recognition results before and after adding pre-stack attributes. It can be seen that the results without adding pre-stack attributes cannot effectively distinguish the superimposed relationships.

[0042] Step 6: Extract the sensitive attributes optimized in Steps 4 and 5 along the appropriate time window in the actual seismic data. After preprocessing, input them into the fuzzy self-organizing neural network to obtain the clustering results. Combine with well logging data to interpret the possible superimposed relationships corresponding to each category, and then judge the distribution and thickness of the regional low-velocity mudstone.

[0043] The fuzzy self-organizing neural network mentioned in Steps 5 and 6 is a neural network model based on the combination of self-organizing learning mechanism and fuzzy logic. Its topological structure includes an input layer, a fuzzy layer, a membership layer, and an output layer, as Figure 3 .

[0044] Among them, the input layer receives external data; the fuzzy layer fuzzifies the data of the input layer and passes it to the membership layer; each node in the membership layer represents the output of a membership function, that is, the membership degree of the input data to a certain fuzzy set (the degree to which each input data belongs to each fuzzy set). The network measures the classification characteristics of the input attributes through the membership layer; the nodes in the output layer represent different categories. After fuzzy reasoning, the network will give the prediction result of the category to which the input data belongs. The steps of fuzzy self-organizing neural network clustering are as follows:

[0045] 1. Normalize the input seismic attributes. The purpose of normalization is to standardize the input seismic attributes to ensure that the numerical ranges of each attribute are similar, so as to avoid some attributes having too much influence on the results. The formula is as follows:

[0046]

[0047] Among them, A represents the seismic attribute value after normalization; A f represents the original seismic attribute data; A max represents the maximum value of the original seismic attribute data; A min represents the minimum value of the original seismic attribute data.

[0048] 2. Initialize the neural network. Use random numbers to determine the initial weight ω ik (t) of the network clustering center. Among them, 0 < ω ik(t)<1, where i = 1, 2, …, N is the number of input data, and k = 1, 2, …, K represents the number of clustering centers.

[0049] 3. Fuzzify the input seismic attributes. The normalized seismic attributes are fuzzified through the Gaussian function to determine the membership degree of the sample points in the seismic attributes to the clustering centers, that is:

[0050]

[0051] In the formula, i = 1, 2, …, N is the number of input data, j = 1, 2, …, M represents the number of sample data points in each seismic attribute, k = 1, 2, …, K represents the number of clustering centers, σ represents the width of the Gaussian window, and t represents the number of iterations.

[0052] 4. Calculate the membership degree of each sample point in the input seismic attributes to the membership layer neurons, that is:

[0053]

[0054] 5. Adjust the weights of the network, that is

[0055]

[0056] Among them, α represents the learning factor exponent.

[0057] 6. Judge whether the network has reached stability. It is judged by checking the change amount of the neuron weights. If the update amount of the weights is less than a certain threshold, the network is considered to be stable, that is:

[0058] |ω ik (t + 1) - ω ik (t)| < ε

[0059] Where ε is the set threshold. If the change amounts of the weights of all neurons are less than this threshold, the network is considered to converge, otherwise return to step 3 for continued training.

[0060] 7. Output the clustering result. When the network reaches stability, output the clustering result, that is:

[0061]

[0062] The present invention designs a theoretical model of different superimposed relationships between gas-bearing sandstones and low-velocity mudstones based on logging and seismic data, extracts and optimizes the attributes of the post-stack and pre-stack forward responses thereof, identifies different superimposed relationships through fuzzy self-organizing neural network clustering, and further identifies the distribution and thickness of low-velocity mudstones.

Claims

1. A geophysical identification method for medium-deep undercompacted low-velocity mudstone, characterized in that By statistically analyzing the data, representative P-wave velocity, S-wave velocity, and density of low-velocity mudstone and gas-bearing sandstone in well logging data are obtained. Combining with geological information, a theoretical model of the superposition relationship between low-velocity mudstone and gas-bearing sandstone is designed, and convolution forward modeling and prestack forward modeling are carried out on it. The post-stack and prestack seismic attributes of the model are calculated respectively, and then these attributes are optimized for sensitive attributes. The obtained sensitive attributes are used as learning samples, and the superposition relationship is identified by probability self-organizing neural network clustering; It includes the following steps: Step 1: Analyze well logging data and seismic data to obtain typical petrophysical parameters and contact relationships of low-velocity mudstone, gas-bearing sandstone, and background mudstone, and establish a theoretical geological model of the superposition relationship between low-velocity mudstone and gas-bearing sandstone; Step 2: Conduct convolution forward modeling on the superposition model of low-velocity mudstone and gas-bearing sandstone, and analyze whether the forward modeling results can represent the response characteristics of different contact relationships, such as the "dark spot" reflection phenomenon of the reservoir caused by gas-bearing sandstone overlying low-velocity mudstone and the "false bright spot" reflection phenomenon caused by the contact between low-velocity mudstone and background mudstone. If not, return to Step 1 and modify the parameters of the superposition model; Step 3: Conduct prestack forward modeling on the superposition mode of low-velocity mudstone and gas-bearing sandstone, and then calculate the intercept attribute P and gradient attribute G in its prestack attributes, and calculate different combinations of P and G to obtain prestack combined attributes; Step 4: Select appropriate time windows to extract various types of seismic attributes, and use the principal component analysis (PCA) method to evaluate the sensitivity of each seismic attribute between different superposition relationships, so as to optimize the sensitive attributes that can distinguish different superposition relationships; Step 5: Input multiple sensitive seismic attributes optimized in Step 4 into the fuzzy self-organizing neural network at the same time. The network conducts self-organizing learning competition according to the characteristics of the seismic attributes to see whether the results can distinguish different superposition relationships between low-velocity mudstone and gas-bearing sandstone in the model. If not, it means that the sensitive attributes optimized in Step 4 are not appropriate and need to be further optimized; if it can be distinguished, it means that the sensitive attributes optimized in Step 4 are sufficient to distinguish different superposition relationships and can be used for actual data; Step 6: Extract the sensitive attributes optimized in Step 4 and Step 5 along the appropriate time window in the actual seismic data, input them into the fuzzy self-organizing neural network after preprocessing to obtain the clustering results, and combine with well logging data to interpret the possible superposition relationships corresponding to each category, and then judge the distribution and thickness of regional low-velocity mudstone.

2. The geophysical identification method of medium-deep undercompacted low-velocity mudstone according to claim 1, wherein The implementation steps of PCA are as follows: (1) Standardize the data wherein represents the j-th seismic attribute of the i-th stacking relationship, is the number of stacking relationships, is the number of seismic attributes; wherein is the mean value of the j-th seismic attribute, is its standard deviation; (2) Conduct principal component analysis on the standardized matrix Z and calculate its covariance matrix C: Perform eigenvalue decomposition on the covariance matrix C to obtain eigenvalues and eigenvectors; the eigenvectors correspond to the principal component directions, while the eigenvalues represent the variance magnitudes of each principal component; by examining the principal component loadings, i.e., the components of the eigenvectors, determine which seismic attributes contribute the most to the principal components, thereby optimizing and selecting sensitive attributes.

Images