Intelligent reservoir porosity prediction method based on seismic multi-attribute fusion

Through seismic multi-attribute fusion and correction of U-net network, the problem of insufficient data fitting ability in traditional reservoir porosity prediction technology is solved, and more efficient and accurate reservoir porosity prediction is achieved, making full use of seismic data non-infrastructure Linear feature information.

CN118938308APending Publication Date: 2024-11-12CHINA PETROLEUM & CHEMICAL CORP +1

Patent Information

Application Number
CN202310515820.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-09
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

Traditional reservoir porosity prediction technology is based on simple elastic parameter fitting, which cannot meet the needs of favorable reservoir fine characterization, and deep learning methods do not fully utilize the nonlinear characteristics and attribute information of seismic data.

Method used

Using a method based on seismic multi-attribute fusion, the nonlinear inversion of the first three parameters and the instantaneous properties of the AVO attribute and the three-parameter wavelet transformation are constructed to construct a corrected U-net network, and the well logging porosity curve is used for training to realize reservoir Intelligent prediction of porosity.

Benefits of technology

The universality and accuracy of reservoir porosity prediction are improved, the reservoir characteristic information in seismic data is fully explored, and the strong nonlinear fitting ability and efficient prediction efficiency are provided.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118938308A_ABST
    Figure CN118938308A_ABST
Patent Text Reader

Abstract

The invention provides an intelligent reservoir porosity prediction method based on seismic multi-attribute fusion. The method comprises the following steps: step 1, carrying out pre-stack three-parameter nonlinear inversion; step 2, extracting AVO attributes; 3, extracting instantaneous amplitude and instantaneous frequency attributes of the seismic data based on three-parameter wavelet transform; 4, the seven attributes extracted through inversion and the logging porosity curve serve as network training data, and a sample set is constructed; and step 5, training and correcting the U-net network by using the training sample to obtain a porosity prediction network model. According to the reservoir porosity intelligent prediction method based on seismic multi-attribute fusion, seismic data are fully utilized, feature information, reflected in the seismic data, of reservoir physical properties is mined from different theoretical angles, and the complex mapping relation between various seismic attributes and reservoir porosity is learned by utilizing the powerful nonlinear fitting capability of the corrected U-net network; therefore, intelligent prediction of the porosity is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of geophysical exploration and interpretation, and in particular to an intelligent reservoir porosity prediction method based on seismic multi-attribute fusion. Background Art

[0002] Reservoir porosity is one of the important parameters for describing reservoir characteristics and a key indicator for evaluating reservoir quality. Accurate porosity prediction based on seismic data is crucial for the fine characterization of favorable reservoirs. Traditional reservoir porosity prediction technology is generally based on the elastic parameters (elastic properties) estimated by seismic inversion technology, and rock physics modeling is performed by fitting the porosity curve of well logging to obtain a relatively simple porosity empirical formula. Practice has shown that the empirical formula has low universality in the work area, and its conversion accuracy often cannot meet the actual production needs of fine characterization of favorable reservoirs. This is mainly due to the lack of nonlinear fitting ability of traditional statistical fitting methods for data.

[0003] In recent years, the new generation of machine learning theory represented by deep learning in the field of artificial intelligence has developed rapidly, and has made breakthroughs in many fields such as image pattern recognition and natural language processing. Deep learning is a method that uses mathematical models (deep neural networks) to model the inherent laws and representation levels of sample data in a certain field to solve similar problems in the field. The key to its success lies in the excellent fitting ability of deep neural networks for nonlinear relationships and high prediction efficiency. At present, deep learning has been preliminarily applied in the fields of seismic data processing and interpretation to save labor and time costs and improve the accuracy of data processing and interpretation.

[0004] The use of deep learning for reservoir parameter prediction has been attempted by some scholars, but most of the training data used is one of the inverted elastic parameters or extracted time-frequency attributes, and the network used is a traditional fully connected network or convolutional network. The shortcomings of these methods are reflected in two aspects: first, the feature types contained in the training sample data are insufficient, and the complete reservoir feature information contained in the seismic data is not fully mined; second, the structure of the network used is relatively simple, and the strong nonlinear fitting ability of deep learning is not fully utilized.

[0005] In the Chinese patent application with application number: CN201910706564.X, a carbonate reservoir prediction method based on prestack multi-attribute and paleo-geomorphology fusion technology is involved. The method includes: 1) For the area where the layer section to be predicted is located, rock physical analysis is performed based on the drilling data to determine the prestack elastic parameters sensitive to the carbonate reservoir; 2) For the layer section to be predicted, the relationship between the longitudinal and transverse wave velocities is determined based on the drilling data, and the relationship between density and longitudinal wave velocity is determined, and then the prestack AVO attribute inversion calculation is performed to obtain the specific values ​​of each prestack elastic parameter sensitive to the carbonate reservoir; 3) For the layer section to be predicted, the current structural morphology and paleo-geomorphology are determined respectively; 4) For the layer section to be predicted, the determined current structural morphology, paleo-geomorphology and the determined prestack AVO attribute inversion are respectively normalized and fused, and the carbonate reservoir is predicted using the fused results.

[0006] In the Chinese patent application with application number: CN202010175162.4, a reservoir oil and gas prediction method and device based on multi-seismic attribute optimization is involved, the method comprising: obtaining first seismic attribute data of the study area, wherein the first seismic attribute data includes: parameter data of multiple seismic attributes; based on the first seismic attribute data, screening second seismic attribute data that meets the preset seismic attribute conditions; inputting the second seismic attribute data into a pre-trained classifier, and outputting the third seismic attribute data after dimensionality reduction, wherein the classifier is a seismic attribute classification model obtained by machine learning training based on the relationship between known drilling data and seismic attributes; based on the third seismic attribute data, reservoir oil and gas prediction is performed in the study area. This invention utilizes the vertical high resolution of logging data and the horizontal high resolution of seismic attribute data, combined with the advantages of multi-seismic attribute reservoir prediction, can improve the accuracy of reservoir oil and gas prediction, and can be applied to oil and gas prediction in tight sandstone reservoirs.

[0007] In the Chinese patent application with application number: CN201911425586.5, a reservoir porosity prediction method based on the Hilbert spectrum and BP neural network is involved. The method first performs a Hilbert-Huang transform on the seismic record, then calculates its Hilbert spectrum, and on this basis calculates the Hilbert instantaneous energy and marginal spectrum to construct the input vector of the BP neural network. The seismic trace records at the drilling / logging location are used as training-test samples, and the expected output of the sample is obtained from the drilling / logging data, and the BP neural network is trained and tested. Finally, the trained and tested BP neural network is used to predict the distribution of porosity with depth at the remaining detection points. Compared with the prior art, the invention has the advantages of high prediction accuracy and detailed reservoir description.

[0008] In the Chinese patent application with application number: CN201910897976.6, a method and device for quantitative prediction of reservoir porosity are involved, which includes: calculating the P-wave impedance parameters and P-wave velocity ratio parameters based on the HM rock physics theoretical model according to the rock matrix physical parameters of the target reservoir section; constructing a near-linear rock physics interpretation template using the P-wave impedance parameters in the double logarithmic domain and the P-wave velocity ratio parameters in the double logarithmic domain; fitting the porosity trend line based on the near-linear rock physics interpretation template to determine the relationship between porosity and the P-wave impedance parameters in the double logarithmic domain and the P-wave velocity ratio parameters in the double logarithmic domain; determining the porosity of the target reservoir section according to the P-wave impedance seismic data body, the P-wave velocity ratio seismic data body and the relationship between the porosity and the P-wave impedance parameters in the double logarithmic domain and the P-wave velocity ratio parameters in the double logarithmic domain. This invention reduces the multi-solution of reservoir porosity prediction and improves the accuracy of reservoir porosity prediction.

[0009] The above existing technologies are significantly different from the present invention and fail to solve the technical problem we want to solve. Therefore, we have invented a new intelligent reservoir porosity prediction method based on seismic multi-attribute fusion. Summary of the invention

[0010] The purpose of the present invention is to provide a reservoir porosity intelligent prediction method based on seismic multi-attribute fusion, which is based on multiple attributes of seismic data and uses deep neural networks to perform reservoir porosity intelligent prediction.

[0011] The purpose of the present invention can be achieved by the following technical measures: a reservoir porosity intelligent prediction method based on seismic multi-attribute fusion, the reservoir porosity intelligent prediction method based on seismic multi-attribute fusion includes:

[0012] Step 1, perform pre-stack three-parameter nonlinear inversion;

[0013] Step 2, extract AVO attributes;

[0014] Step 3, extracting instantaneous amplitude and instantaneous frequency attributes of seismic data based on three-parameter wavelet transform;

[0015] Step 4, using the seven attributes extracted by the above inversion and the logging porosity curve as network training data to construct a sample set;

[0016] Step 5: Use the training samples to train the modified U-net network to obtain the porosity prediction network model.

[0017] The purpose of the present invention can also be achieved by the following technical measures:

[0018] In step 1, under the constraints of well logging curves, seismic horizons, low-frequency background, and rock physics models, the three pre-stack parameters, including P-wave velocity, S-wave velocity, and density, are inverted.

[0019] In step 1, the objective function used in prestack inversion is:

[0020]

[0021] m=[V P1 ,…,V PN ,V S1 ,…V SN ,ρ1,…,ρ N ] T (2)

[0022] Where m is the model parameter, w is the seismic wavelet, θ is the incident angle, r(m,θ) is the reflection coefficient that varies with the incident angle, and d obs (θ) is the amplitude-preserving gather, φ i (m) is the regularization constraint term, λ i is the regularization term coefficient, P is the total number of regularization terms; Vp is the longitudinal wave velocity, Vs is the shear wave velocity, and ρ is the density.

[0023] In step 1, the regularization constraints include low-frequency background constraints, layer information constraints, logging curve constraints, rock physics model constraints between three parameters, and TV regularization; the gradients of each item in the inversion objective function are as follows:

[0024]

[0025]

[0026] The Levenberg–Marquardt optimization algorithm is used to solve the objective function, and the correction value Δm of the model parameters can be calculated according to the following formula:

[0027]

[0028]

[0029] Where K(m) is the Jacobian matrix of the objective function; high-precision pre-stack three parameters are obtained.

[0030] In step 2, based on the amplitude-preserving optimization of the pre-stack angle gathers, the intercept P and gradient G attributes are extracted based on the Shuey two-term reflection coefficient formula.

[0031] In step 2, the Shuey two-term approximation formula is as follows:

[0032] R PP (θ)=P+Gsin2 θ (3)

[0033] Among them, θ is the incident angle, Rpp is the reflection coefficient, G is the gradient attribute, and P is the intercept attribute.

[0034] In step 3, the time domain expression of the three-parameter wavelet is as follows:

[0035]

[0036] Λ=(σ,τ,β) (5)

[0037]

[0038]

[0039]

[0040] Where t is time, σ is the modulation frequency of the analytical wavelet, τ is the parameter that controls the decay speed, and β is the parameter that controls the phase change. Λ represents the set of σ, τ, and β.

[0041] In step 3, the wavelet transform and Hilbert transform of the seismic trace satisfy the following equation:

[0042]

[0043] Where S(t,a) is the three-parameter wavelet transform of signal s(t), H[·] is the Hilbert transform, j is the imaginary unit, and Cg is the integral of the ratio function of the Fourier transform of the wavelet function g(t) and the frequency ω. The above formula can be used to calculate the instantaneous attributes of the seismic signal:

[0044]

[0045]

[0046]

[0047] Among them, e(t), θ(t) and f(t) are the instantaneous amplitude, instantaneous phase and instantaneous frequency of the seismic signal s(t), respectively.

[0048] In step 4, the P-wave velocity, S-wave velocity, density, intercept-gradient attributes and various instantaneous attributes extracted from the inverted seismic traces near each well are extracted as the input data of the network. The porosity curve of each well is appropriately filtered, and the frequency band range is kept consistent with the inversion parameters as the label data of the network.

[0049] In step 5, the loss function of the neural network is:

[0050]

[0051] Among them, d is the training sample data, m is the label data obtained after filtering the logging porosity curve, θ is the parameter to be trained in the neural network, and G(·) represents the neural network.

[0052] In step 5, after the network converges, it is applied to the entire work area to obtain the predicted porosity data volume.

[0053] The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion also includes, before step 1, performing offset processing on the pre-stack seismic data of the target work area and extracting amplitude-preserving angle gathers; and correcting the logging curves so that the correlation coefficient of well-seismic matching reaches a higher level.

[0054] The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion in the present invention relates to a method for intelligent prediction of reservoir porosity based on multiple attributes of seismic data and using a deep neural network. The method uses three parameters (P-wave velocity, S-wave velocity and density) of prestack inversion, extracted AVO attributes (gradient and intercept), and instantaneous attributes (instantaneous frequency, instantaneous amplitude) based on three-parameter wavelet transform as network inputs, uses well logging porosity curves as training labels, and uses a modified U-net network as a network model for training. The present invention adopts the above technical solution, which has the following advantages:

[0055] The three parameters of prestack inversion (P-wave velocity, S-wave velocity and density), the extracted AVO attributes (gradient and intercept), and the instantaneous attributes based on three-parameter wavelet transform (instantaneous frequency, instantaneous amplitude) are used as the input of the network. The seismic data is fully utilized to mine the characteristic information of reservoir physical properties reflected in the seismic data from different theoretical perspectives. The neural network is then used to learn the nonlinear mapping relationship between various attribute parameters and porosity. Compared with the traditional estimation method based on rock physics model, it has the advantages of strong universality (generalization) and high efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flowchart of a specific embodiment of the reservoir porosity intelligent prediction method based on seismic multi-attribute fusion of the present invention;

[0057] Figure 2 is a schematic diagram of a three-parameter profile of prestack inversion in a specific embodiment of the present invention, Figure 1 (a) is the longitudinal wave velocity profile, Figure 1 (b) is the shear wave velocity, Figure 1 (c) is the density profile;

[0058] Figure 3 is a schematic diagram of an AVO attribute profile extracted in a specific embodiment of the present invention, Figure 3 (a) is the AVO gradient attribute profile, Figure 3 (b) is the AVO intercept attribute profile;

[0059] Figure 4 is a schematic diagram of instantaneous attributes extracted in a specific embodiment of the present invention, Figure 4 (a) is the instantaneous amplitude attribute profile, Figure 4 (b) is the instantaneous frequency attribute profile;

[0060] Figure 5 A schematic diagram of a modified U-net neural network structure in a specific embodiment of the present invention;

[0061] Figure 6 It is a schematic diagram of the predicted porosity profile in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0062] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0063] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations and / or combinations thereof.

[0064] The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion of the present invention comprises the following steps: (1) pre-stack three-parameter nonlinear inversion to estimate P-wave velocity, S-wave velocity and density; (2) extracting AVO attributes based on Shuey two-term reflection coefficient formula, including intercept P and gradient G; (3) extracting instantaneous amplitude and instantaneous frequency attributes of seismic data based on three-parameter wavelet transform; (4) using the seven attributes extracted by the above inversion and the logging porosity curve as network training data to construct a sample set; (5) using the training samples to train the modified U-net network to obtain a porosity prediction network model. The method makes full use of seismic data, mines the characteristic information reflected in the reservoir physical properties in the seismic data from different theoretical perspectives, and uses the powerful nonlinear fitting ability of the modified U-net network to learn the complex mapping relationship between multiple seismic attributes and reservoir porosity, thereby realizing the intelligent prediction of porosity.

[0065] The following are several specific embodiments of the present invention.

[0066] Example 1

[0067] In a specific embodiment 1 of the present invention, the reservoir porosity intelligent prediction method based on seismic multi-attribute fusion includes the following steps:

[0068] 1. Prestack elastic parameter inversion. Under the constraints of logging curves, seismic layers, low-frequency background, rock physics model, etc., invert the three prestack parameters (P-wave velocity, S-wave velocity and density). The objective function used in prestack inversion is:

[0069]

[0070] m=[V P1 ,…,V PN ,V S1 ,…V SN ,ρ1,…,ρ N ] T (2)

[0071] Where m is the model parameter, w is the seismic wavelet, θ is the incident angle, r(m,θ) is the reflection coefficient that varies with the incident angle, and d obs (θ) is the amplitude-preserving gather, φ i (m) is the regularization constraint term, λ i is the regularization term coefficient, and P is the total number of regularization terms.

[0072] 2. AVO attribute extraction. Based on the amplitude-preserving optimization of the pre-stack angle gather, the intercept P and gradient G attributes are extracted based on the Shuey two-term reflection coefficient formula. The Shuey two-term approximate formula is as follows:

[0073]

[0074] Among them, θ is the incident angle, Rpp is the reflection coefficient, Vp is the longitudinal wave velocity, G is the gradient attribute, and P is the intercept attribute.

[0075] 3. Instantaneous attribute extraction based on three-parameter wavelet transform. Changes in instantaneous frequency and amplitude of seismic traces can often reflect changes in lithological physical parameters, so advanced time-frequency analysis methods are the key to obtaining effective seismic attributes. The time domain expression of the three-parameter wavelet is as follows:

[0076]

[0077] Λ=(σ,τ,β) (5)

[0078]

[0079]

[0080]

[0081] Among them, t is time, σ is the modulation frequency of the analytical wavelet, τ is the parameter that controls the decay speed, and β is the parameter that controls the phase change. The three parameters have clear physical meanings, and the shape of the wavelet can be flexibly adjusted as needed to perfectly match the waveform of the actual seismic wavelet. Therefore, the instantaneous frequency analysis method based on the three-parameter wavelet transform has good noise resistance and higher time resolution.

[0082] 4. Extract the P-wave velocity, S-wave velocity, density, intercept-gradient attributes and various instantaneous attributes of the inverted seismic traces near each well as the input data of the network, filter the porosity curve of each well appropriately, and keep the frequency band range consistent with the inversion parameters as the label data of the network;

[0083] 5. Train the U-Net neural network model and use the sample data in (4) to train the network. The loss function of the neural network is:

[0084]

[0085] Among them, d is the training sample data, m is the label data obtained after filtering the logging porosity curve, and θ is the parameter to be trained in the neural network. After convergence, the trained network model is applied to the entire work area to predict porosity.

[0086] Example 2

[0087] In a specific embodiment 2 of the present invention, the reservoir porosity intelligent prediction method based on seismic multi-attribute fusion includes:

[0088] First, the prestack seismic data of the target area are offset and the amplitude-preserving angle gathers are extracted. The well logging curves are corrected so that the correlation coefficient of well-seismic matching reaches a high level (>0.85). Then, the prestack three-parameter inversion is performed under the constraints of the well logging curves, layers, rock physics models, etc., to obtain the P-wave velocity, S-wave velocity and density parameters, and extract the AVO attributes (gradient and intercept) of the angle gathers. The angle gathers are then stacked, the stacked data are transformed by three-parameter wavelet transform, and the instantaneous amplitude and instantaneous frequency attributes are extracted based on the transformation results. The above-mentioned 7 inverted / extracted parameters are used as training samples, and the well logging porosity curves are used as labels to train the neural network. After the network converges, the porosity prediction network model is obtained.

[0089] The material basis of the present invention is to obtain high-quality amplitude-preserving gathers, and the porosity curve conversion results of logging data are reliable, and the well locations are distributed as evenly as possible in the work area. The flow chart of the reservoir porosity intelligent prediction method based on seismic multi-attribute fusion of the present invention is as follows: Figure 1 As shown, the specific steps are as follows:

[0090] In step 101, prestack elastic parameters are inverted. The objective function used in prestack inversion is:

[0091]

[0092] m=[V P1 ,…,V PN ,V S1 ,…V SN ,ρ1,…,ρ N ] T (11)

[0093] Where m is the model parameter, w is the seismic wavelet, θ is the incident angle, r(m,θ) is the reflection coefficient that varies with the incident angle, and d obs (θ) is the amplitude-preserving gather, φ i (m) is the regularization constraint term, λ i is the regularization term coefficient, and P is the total number of regularization terms. Regularization constraints include low-frequency background constraints, layer information constraints, logging curve constraints, rock physics model constraints between three parameters, and TV regularization. The gradients of each term in the inversion objective function are as follows:

[0094]

[0095]

[0096] The Levenberg–Marquardt optimization algorithm is used to solve the objective function, and the correction value Δm of the model parameters can be calculated according to the following formula:

[0097]

[0098]

[0099] Where K(m) is the Jacobian matrix of the objective function.

[0100] Obtain high-precision pre-stack three parameters.

[0101] In step 102, AVO attributes are extracted. Based on the amplitude-preserving optimization of the pre-stack angle gather, the intercept P and gradient G attributes are extracted based on the Shuey two-term reflection coefficient formula. The Shuey two-term approximate formula is in the following form:

[0102] R PP (θ)=P+Gsin 2 θ (14)

[0103] Among them, θ is the incident angle, Vp is the longitudinal wave velocity, G is the gradient attribute, and P is the intercept attribute.

[0104] In step 103, instantaneous attributes are extracted based on three-parameter wavelet transform. Changes in instantaneous frequency and amplitude of seismic traces can often reflect changes in lithological physical parameters, so advanced time-frequency analysis methods are the key to obtaining effective seismic attributes. In general, the wavelet transform and Hilbert transform of seismic traces satisfy the following formula:

[0105]

[0106] In the formula, S(t,a) is the three-parameter wavelet transform of the signal s(t), H[·] is the Hilbert transform, j is the imaginary unit, and Cg is the integral of the ratio function of the Fourier transform of the wavelet function g(t) and the frequency ω. The above formula can be used to calculate the instantaneous attributes of the seismic signal:

[0107]

[0108]

[0109]

[0110] Among them, e(t), θ(t) and f(t) are the instantaneous amplitude, instantaneous phase and instantaneous frequency of the seismic signal s(t), respectively.

[0111] In step 104, the P-wave velocity, S-wave velocity, density, intercept-gradient attributes, and instantaneous attributes of the inverted seismic traces near each well are extracted as input data for the network, and the porosity curve of each well is appropriately filtered, with the frequency band range kept consistent with the inversion parameters, as label data for the network.

[0112] In step 105, the network structure is as follows Figure 5 As shown. The loss function of the neural network is:

[0113]

[0114] Among them, d is the training sample data, m is the label data obtained after filtering the logging porosity curve, θ is the parameter to be trained in the neural network, and G(·) represents the neural network. After the network converges, it is applied to the entire work area to obtain the predicted porosity data volume.

[0115] Example 3

[0116] In a specific embodiment 3 of the present invention, an actual seismic survey line is selected to test the method of the present invention. The time range of the seismic data is 1.8s to 2.8s, and the sampling rate is 1ms. The CDP range is 2136 to 2620, with a total of 458 traces. First, the three-parameter inversion of the pre-stack data is performed based on equations (10) to (13). Figure 2The results of prestack three-parameter inversion are shown. Figure 2 (a) to (c) are the inverted P-wave velocity, S-wave velocity and density respectively. Then, the gradient and intercept attributes of the angle gather are extracted based on equation (14): Figure 3 The extracted results are shown. Figure 3 (a) to (b) are the normalized gradient and intercept attribute profiles, respectively. Finally, after the stacked seismic data are transformed by three-parameter wavelet transform, the instantaneous attributes are extracted based on equations (15) to (18). Figure 4 The extracted results are shown. Figure 4 (a) to (b) are the extracted instantaneous amplitude and instantaneous frequency attribute profiles, respectively. The attributes (7 features) inverted / extracted from the seismic traces near the wells of each well are used as training samples. In order to make the network have stronger generalization ability, the size of a single sample is N×M. Among them, N refers to the type of feature, that is, N=7. M is the length of the parameter, that is, using M attributes in the time direction to predict the label value at the center of the time window (that is, at the N / 2 position), as shown in detail. Figure 5 shown. Figure 6 The predicted porosity profile is displayed. The strata in the profile have good lateral continuity and vertical resolution, indicating that the network model has strong noise resistance. After the logging porosity curve at CMP=2354 is appropriately low-pass filtered and embedded in the predicted porosity profile, it can be seen that it has a high consistency with the porosity predicted in the wellside area. The correlation coefficient between the prediction result of the wellside channel and the logging curve reaches above 0.75, which verifies the practicability of the design method of the present invention.

[0117] Finally, it should be noted that the above is only 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 aforementioned embodiments, those skilled in the art can still modify the technical solutions recorded in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0118] Except for the technical features described in the specification, all other technical features are known technologies to those skilled in the art.

Claims

1. An intelligent reservoir porosity prediction method based on seismic multi-attribute fusion is characterized by: The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion includes: Step 1, perform pre-stack three-parameter nonlinear inversion; Step 2, extract AVO attributes; Step 3, extracting instantaneous amplitude and instantaneous frequency attributes of seismic data based on three-parameter wavelet transform; Step 4, using the seven attributes extracted by the above inversion and the logging porosity curve as network training data to construct a sample set; Step 5: Use the training samples to train the modified U-net network to obtain the porosity prediction network model.

2. The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion according to claim 1 is characterized in that: In step 1, under the constraints of well logging curves, seismic horizons, low-frequency background, and rock physics models, the three pre-stack parameters, including P-wave velocity, S-wave velocity, and density, are inverted.

3. The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion according to claim 2 is characterized in that: In step 1, the objective function used in prestack inversion is: m=[V P1 ,…,V PN ,V S1 ,…V SN ,ρ1,…,ρ N ] T (2) Where m is the model parameter, w is the seismic wavelet, θ is the incident angle, r(m,θ) is the reflection coefficient that varies with the incident angle, and d obs (θ) is the amplitude-preserving gather, φ i (m) is the regularization constraint term, λ i is the regularization term coefficient, P is the total number of regularization terms; Vp is the longitudinal wave velocity, Vs is the shear wave velocity, and ρ is the density.

4. The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion according to claim 3 is characterized in that: In step 1, the regularization constraints include low-frequency background constraints, layer information constraints, logging curve constraints, rock physics model constraints between three parameters, and TV regularization; the gradients of each item in the inversion objective function are as follows: The Levenberg–Marquardt optimization algorithm is used to solve the objective function, and the correction value Δm of the model parameters can be calculated according to the following formula: Where K(m) is the Jacobian matrix of the objective function; high-precision pre-stack three parameters are obtained.

5. The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion according to claim 1 is characterized in that: In step 2, based on the amplitude-preserving optimization of the pre-stack angle gathers, the intercept P and gradient G attributes are extracted based on the Shuey two-term reflection coefficient formula.

6. The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion according to claim 5 is characterized in that: In step 2, the Shuey two-term approximation formula is as follows: R PP (θ)=P+Gsin 2 i (3) Among them, θ is the incident angle, Rpp is the reflection coefficient, G is the gradient attribute, and P is the intercept attribute.

7. The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion according to claim 1 is characterized in that: In step 3, the time domain expression of the three-parameter wavelet is as follows: in: Λ=(σ,τ,β) (5) Among them, t is time, σ is the modulation frequency of the analytical wavelet, τ is the parameter that controls the decay speed, β is the parameter that controls the phase change; Λ represents the set of σ, τ, and β.

8. The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion according to claim 7 is characterized in that: In step 3, the wavelet transform and Hilbert transform of the seismic trace satisfy the following equation: Where S(t,a) is the three-parameter wavelet transform of signal s(t), H[·] is the Hilbert transform, j is the imaginary unit, and C g is the integral of the ratio function of the Fourier transform of the wavelet function g(t) and the frequency ω; The above formula can be used to calculate the instantaneous attributes of the seismic signal: Among them, e(t), θ(t) and f(t) are the instantaneous amplitude, instantaneous phase and instantaneous frequency of the seismic signal s(t), respectively.

9. The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion according to claim 1 is characterized in that: In step 4, the P-wave velocity, S-wave velocity, density, intercept-gradient attributes and various instantaneous attributes extracted from the inverted seismic traces near each well are extracted as the input data of the network. The porosity curve of each well is appropriately filtered, and the frequency band range is kept consistent with the inversion parameters as the label data of the network.

10. The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion according to claim 1 is characterized in that: In step 5, the loss function of the neural network is: Among them, d is the training sample data, m is the label data obtained after filtering the logging porosity curve, θ is the parameter to be trained in the neural network, and G(·) represents the neural network.

11. The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion according to claim 10 is characterized in that: In step 5, after the network converges, it is applied to the entire work area to obtain the predicted porosity data volume.

12. The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion according to claim 1 is characterized in that: The reservoir porosity intelligent prediction method based on seismic multi-attribute fusion also includes, before step 1, performing offset processing on the pre-stack seismic data of the target work area and extracting amplitude-preserving angle gathers; and correcting the logging curves so that the correlation coefficient of well-seismic matching reaches a higher level.

Citation Information

Patent Citations

  • Reservoir porosity prediction method based on Hilbert spectrum and BP neural network

    CN110988998A

  • Reservoir porosity quantitative prediction method and device

    CN111381279A

  • Carbonatite reservoir prediction method based on pre-stack multi-attribute and ancient landform fusion technology

    CN112305602A

  • Reservoir oil and gas prediction method and device based on multi-seismic attribute optimization

    CN113393335A

Cited By

  • Deposited reservoir prediction method and system based on three-dimensional seismic data

    CN120335006A