Multi-parameter lithology prediction method and device based on deep feed-forward neural network, electronic equipment and storage medium

Abstract

The invention discloses a multi-parameter lithology prediction method and device based on a deep feedforward neural network, electronic equipment and a storage medium. The method comprises the following steps: acquiring a high-resolution longitudinal wave impedance body, a gamma data body, a shale content data body and a sand content data body based on frequency-broadening post-stack seismic data of a research area; obtaining a high-resolution sandstone probability body and a high-resolution mudstone probability body based on the high-resolution longitudinal wave impedance body; constructing a training sample set based on the sandstone probability body, the mudstone probability body, the gamma data body, the shale content data body, the sand content data body and the sand mudstone lithologic logging interpretation data; training the lithology prediction model based on the deep feedforward neural network by using the training sample set; and carrying out lithology prediction based on the trained lithology prediction model. According to the method, the lithology prediction model based on the deep feedforward neural network is trained to obtain a nonlinear relation model between multiple parameters and lithology, and high-resolution prediction of the lithology is realized.

Description

Technical Field

[0001] This invention belongs to the field of geophysical exploration technology, and more specifically, relates to a multi-parameter lithology prediction method, electronic equipment, storage medium and device based on a deep feedforward neural network. Background Technology

[0002] Tight sandstone reservoirs are characterized by distinct stratification and thin reservoir thickness. With the continuous development of oil exploration, impedance spectroscopy inversion has become an important method for identifying thin layers and classifying lithology in reservoir prediction. However, seismic data is band-limited, limiting its ability to resolve thin tight sandstone reservoirs. Relying solely on impedance spectroscopy for lithology classification has limitations. It is necessary to combine this with high-resolution information from well logging curves to identify thin layers and classify lithology.

[0003] In recent years, deep learning methods have developed rapidly, providing new approaches to solving data-driven seismic inversion problems. They can directly learn the mapping function between seismic data and well logging target parameters end-to-end from the training dataset, and have achieved good results in some applications. However, data-driven deep learning methods are highly dependent on the training sample set, requiring the pre-collection of large amounts of observational data and high-resolution data pairs to train the neural network.

[0004] The information disclosed in the background section of this invention is intended only to enhance the understanding of the general background of this invention, and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to propose a multi-parameter lithology prediction method, electronic device, storage medium and apparatus based on a deep feedforward neural network to achieve high-resolution lithology prediction.

[0006] To achieve the above objectives, this invention proposes a multi-parameter lithology prediction method, electronic device, storage medium, and apparatus based on a deep feedforward neural network.

[0007] According to a first aspect of the present invention, a multi-parameter lithology prediction method based on a deep feedforward neural network is proposed, comprising:

[0008] A comprehensive seismic wavelet and well logging interpolation impedance model was obtained based on the upsampling post-stack seismic data of the study area.

[0009] Based on the well logging interpolation impedance model and the integrated seismic wavelet, the post-stack impedance inversion of the frequency-spreading post-stack seismic data is performed to obtain a high-resolution P-wave impedance volume.

[0010] Bayesian probability classification was performed on the high-resolution longitudinal wave impedance volume to obtain high-resolution sandstone probability volumes and mudstone probability volumes.

[0011] Based on the aforementioned frequency-stacked seismic data, a waveform indication simulation method was used to obtain high-resolution gamma data volumes, clay content data volumes, and sand content data volumes.

[0012] A training sample set is constructed based on the sandstone probability volume, mudstone probability volume, gamma data volume, mud content data volume, sand content data volume, and sandstone and mudstone lithology well logging interpretation data.

[0013] A lithology prediction model based on a deep feedforward neural network is constructed, and the lithology prediction model is trained using the training sample set.

[0014] Lithology prediction is carried out based on the trained lithology prediction model.

[0015] Optionally, based on the superimposed seismic data, broadband Rick wavelet is used to perform well-seismic calibration on each well in the study area to determine the dominant frequency of the wavelet for each well, obtain the wavelet corresponding to each well, and obtain the comprehensive seismic wavelet of the study area based on the wavelet corresponding to each well.

[0016] Optionally, based on the superimposed post-stack seismic data, stratigraphic interpretation is performed to construct a seismic stratigraphic framework, and kriging interpolation is performed on the well logging data along the seismic stratigraphic framework to obtain a well logging interpolation impedance model.

[0017] Optionally, the calculation expression for the Kriging interpolation is:

[0018]

[0019] Among them, z * (x0) represents the impedance value at the interpolation point, λ i The weights for the Kriging interpolation at the i-th wellbore, z(x) i ) represents the known impedance value at the i-th wellbore.

[0020] Optionally, the construction of the training sample set based on the sandstone probability volume, mudstone probability volume, gamma data volume, mud content data volume, sand content data volume, and sandstone-mudstone lithology well logging interpretation data includes:

[0021] The sandstone probability volume, mudstone probability volume, gamma data volume, mud content data volume, and sand content data volume are normalized and used as input data for the training sample set.

[0022] The well logging interpretation data of the sandstone and mudstone lithology serves as the label data for the training sample set;

[0023] The training sample set is constructed based on the input data and the label data.

[0024] Optionally, the network structure of the lithology prediction model includes:

[0025] Input layer h (0) Output layer h (L) And L-1 hidden layers;

[0026] Neurons in adjacent layers are fully connected to each other, while neurons in the same hidden layer and between adjacent hidden layers are not connected to each other.

[0027] Optionally, the lithology prediction model is:

[0028] y = f(x, θ);

[0029] Where x is the input data and θ is the neural network learning parameters for each layer.

[0030] According to a second aspect of the present invention, a multi-parameter lithology prediction device based on a deep feedforward neural network is proposed, comprising:

[0031] The acquisition module is used to acquire the integrated seismic wavelet and well logging interpolation impedance model of the study area based on the post-stacked frequency-spreading seismic data;

[0032] The inversion module is used to perform post-stack impedance inversion on the extended frequency stacked seismic data based on the well logging interpolation impedance model and the integrated seismic wavelet to obtain a high-resolution P-wave impedance volume.

[0033] The classification module is used to perform Bayesian probability classification on the high-resolution longitudinal wave impedance volume to obtain high-resolution sandstone probability volumes and mudstone probability volumes.

[0034] The waveform indication simulation module is used to obtain high-resolution gamma data volume, clay content data volume and sand content data volume based on the frequency-stacked seismic data using the waveform indication simulation method.

[0035] The construction module is used to construct a training sample set based on the sandstone probability volume, mudstone probability volume, gamma data volume, mud content data volume, sand content data volume and sandstone-mudstone lithology well logging interpretation data.

[0036] The construction and training module is used to construct a lithology prediction model based on a deep feedforward neural network and to train the lithology prediction model using the training sample set.

[0037] The lithology prediction module is used to perform lithology prediction based on the trained lithology prediction model.

[0038] According to a third aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0039] At least one processor; and,

[0040] A memory communicatively connected to the at least one processor; wherein,

[0041] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the multi-parameter lithology prediction method based on a deep feedforward neural network as described in any of the first aspects.

[0042] According to a fourth aspect of the invention, a non-transitory computer-readable storage medium is provided, which stores computer instructions for causing a computer to execute the multi-parameter lithology prediction method based on a deep feedforward neural network as described in any of the first aspects of the invention.

[0043] The beneficial effects of this invention are as follows: This invention obtains a comprehensive seismic wavelet and well logging interpolation impedance model from the extended frequency post-stack seismic data of the study area, performs post-stack impedance inversion on the extended frequency post-stack seismic data, obtains a high-resolution P-wave impedance volume, performs Bayesian probability classification on the high-resolution P-wave impedance volume, obtains high-resolution sandstone probability volume and mudstone probability volume, and obtains high-resolution gamma data volume, mud content data volume and sand content data volume by using waveform indication simulation method based on the extended frequency post-stack seismic data; uses multi-parameter data of sandstone probability volume, mudstone probability volume, gamma data volume, mud content data volume, and sand content data volume and well logging interpretation data of sandstone and mudstone lithology to construct a training sample set, trains the lithology prediction model based on depth feedforward neural network to obtain a nonlinear relationship model between multi-parameters and lithology, and realizes high-resolution prediction of lithology.

[0044] The system of the present invention has other features and advantages that will be apparent from or will be set forth in detail in the accompanying drawings and following detailed description, which together serve to explain the particular principles of the invention. Attached Figure Description

[0045] The above and other objects, features and advantages of the present invention will become more apparent from the accompanying drawings, in which like reference numerals generally denote like parts.

[0046] Figure 1 A flowchart illustrating the steps of a multi-parameter lithology prediction method based on a deep feedforward neural network according to the present invention is shown.

[0047] Figure 2 A flowchart illustrating the steps of a multi-parameter lithology prediction method based on a deep feedforward neural network according to Embodiment 2 of the present invention is shown.

[0048] Figure 3 A schematic diagram of a time-domain overlay seismic profile according to Embodiment 2 of the present invention is shown.

[0049] Figure 4 A schematic diagram of a high-resolution longitudinal wave impedance profile according to Embodiment 2 of the present invention is shown.

[0050] Figure 5 A schematic diagram of a high-resolution sandstone probability volume according to Embodiment 2 of the present invention is shown.

[0051] Figure 6 A schematic diagram of a high-resolution gamma data volume according to Embodiment 2 of the present invention is shown.

[0052] Figure 7 A schematic diagram of a lithological prediction profile according to Embodiment 2 of the present invention is shown. Detailed Implementation

[0053] The invention will now be described in more detail with reference to the accompanying drawings. While preferred embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0054] like Figure 1 As shown, a multi-parameter lithology prediction method based on a deep feedforward neural network according to the present invention includes:

[0055] A comprehensive seismic wavelet and well logging interpolation impedance model was obtained based on the upsampling post-stack seismic data of the study area.

[0056] Based on the well logging interpolation impedance model and the integrated seismic wavelet, post-stack impedance inversion is performed on the frequency-spreading post-stack seismic data to obtain a high-resolution P-wave impedance volume.

[0057] Bayesian probability classification was performed on the high-resolution longitudinal wave impedance volume to obtain high-resolution sandstone probability volume and mudstone probability volume.

[0058] High-resolution gamma data volumes, clay content data volumes, and sand content data volumes were obtained using waveform indication simulation based on post-stacked frequency-spreading seismic data.

[0059] A training sample set was constructed based on sandstone probability volume, mudstone probability volume, gamma data volume, mud content data volume, sand content data volume, and sandstone and mudstone lithology well logging interpretation data.

[0060] A lithology prediction model based on a deep feedforward neural network was constructed, and the lithology prediction model was trained using a training sample set.

[0061] Lithology prediction is carried out based on the trained lithology prediction model.

[0062] Specifically, based on the upscaling post-stack seismic data, broadband Ricker wavelets were used to perform well-seismic calibration on each well in the study area to determine the dominant wavelet frequency of each well and obtain the corresponding wavelet for each well. Based on the wavelets corresponding to each well, a comprehensive seismic wavelet for the study area was obtained. Stratigraphic interpretation was performed based on the upscaling post-stack seismic data to construct a seismic stratigraphic framework. Kriging interpolation was performed on the well logging data along the seismic stratigraphic framework to obtain a well logging interpolation impedance model. Based on the well logging interpolation impedance model and the comprehensive seismic wavelet, post-stack impedance inversion was performed on the upscaling post-stack seismic data to obtain a high-resolution P-wave impedance volume. For example, constrained sparse pulse inversion, generalized linear inversion, geostatistical inversion, and genetic algorithm inversion methods were used for post-stack impedance inversion. Bayesian probability classification was performed on the high-resolution P-wave impedance volume. Based on the high-resolution P-wave impedance volume and well logging lithology interpretation data, the conditional probability density function for each lithology was estimated using a Gaussian kernel function to obtain high-resolution sandstone and mudstone probability volumes. Based on the upscaling post-stack seismic data, waveform indication simulation methods were used to obtain high-resolution gamma data volumes and mudstone data volumes. Taking gamma-ray logs as an example, the content data volumes and sand content data volumes are obtained by comparing each simulated seismic trace with all well-side seismic traces. Wells in the same sedimentary environment have similar seismic waveforms and logging curves. The top n wells with the highest similarity to the simulated seismic trace waveforms (n is the number of effective samples) are selected. Based on the similarity of the seismic waveforms, the seismic waveforms are grouped to establish a sample set of seismic waveforms and logging curves indicating different sedimentary environments in the study area. Multi-scale filtering of the logging curves in the sample set is performed in the wavelet domain to obtain the high-frequency, mid-frequency, and low-frequency parts of the logging curves. The mid-low frequency part of the logging curves is selected to establish a matched filter. The high-frequency part of the logging curves has randomness, but the mid-low frequency part has high similarity, and the frequency band of the mid-low frequency part with high similarity greatly exceeds the effective frequency band of the seismic logs. In this way, a matched filter is established to establish the intrinsic relationship between the gamma-ray logging curves and the seismic waveforms. The simulated seismic trace waveforms are processed using the matched filter to obtain the corresponding gamma-ray data volume. Similarly, the mud content and sand content volumes can be obtained. Generally, sandstone is characterized by low gamma ray content, low argillaceous content, and high sand content. Lithology can be classified using the cross-sectional analysis of P-wave impedance and gamma ray content in well logging curves. Therefore, multi-parameter data related to lithology classification, such as P-wave impedance volume, gamma ray volume, argillaceous content volume, sandy content volume, sandstone probability volume, and mudstone probability volume, are normalized and used as input data. Well logging interpretation results for sandstone and mudstone lithology are used as label data to construct a training sample set for a lithology prediction neural network. The normalization formula for the multi-parameter data is: Where s and s norm These represent the data before and after multi-parameter normalization, s and s respectively. max and s minThese represent the maximum and minimum values ​​of the data before multi-parameter normalization. Normalization can improve the computational speed of the convolutional kernel and activation function of the deep feedforward neural network. The relationship between multi-parameters and lithology is non-linear, and deep feedforward neural networks can effectively construct this non-linear relationship. Therefore, a lithology prediction model based on a deep feedforward neural network is constructed and trained using a training sample set to obtain a non-linear relationship model between multi-parameters and lithology, thereby achieving lithology prediction.

[0063] In one example, broadband Rick wavelet is used to perform well-seismic calibration of each well in the study area based on the overlay seismic data, determine the dominant frequency of the wavelet for each well, obtain the wavelet corresponding to each well, and obtain the comprehensive seismic wavelet of the study area based on the wavelet corresponding to each well.

[0064] In one example, stratigraphic interpretation is performed based on post-stacked topology seismic data to construct a seismic stratigraphic framework. Kriging interpolation is then performed on well logging data along the seismic stratigraphic framework to obtain a well logging interpolation impedance model.

[0065] In one example, the expression for calculating Kriging interpolation is:

[0066]

[0067] Among them, z * (x0) represents the impedance value at the interpolation point, λ i The weights for the Kriging interpolation at the i-th wellbore, z(x) i ) represents the known impedance value at the i-th wellbore.

[0068] In one example, the training sample set constructed based on sandstone probability volumes, mudstone probability volumes, gamma data volumes, mudstone content data volumes, sand content data volumes, and sandstone-mudstone lithology well logging interpretation data includes:

[0069] The sandstone probability volume, mudstone probability volume, gamma data volume, mud content data volume, and sand content data volume were normalized and used as input data for the training sample set.

[0070] Well logging interpretation data of sandstone and mudstone lithology were used as label data for the training sample set;

[0071] A training sample set is constructed based on the input data and the label data.

[0072] In one example, the network structure of the lithology prediction model includes:

[0073] Input layer h (0) Output layer h (L) And L-1 hidden layers;

[0074] Neurons in adjacent layers are fully connected to each other, while neurons in the same hidden layer and between adjacent hidden layers are not connected to each other.

[0075] In one example, the lithology prediction model is:

[0076] y = f(x, θ);

[0077] Where x is the input data and θ is the neural network learning parameters for each layer.

[0078] Specifically, for the input data x(z)∈R in the training sample set Set(z) 2 Given the output data y(z)∈R, the output of its hidden layer is:

[0079]

[0080] The number of network layers, the number of neurons in each layer, and the activation function are as follows:

[0081]

[0082] The input layer has 6 neurons (n0 = 6). These 6 neurons are respectively input to the longitudinal wave impedance volume, gamma volume, mud content volume, sand content volume, sandstone probability volume, and mudstone probability volume. The input layer activation function... The ReLU function is chosen, and the Sigmoid function is chosen as the activation function for both the output and hidden layers. The number of neurons in the output layer is n. L =1, the output value is 0 (mudstone) or 1 (sandstone);

[0083] The parameters to be learned in the deep feedforward neural network are:

[0084]

[0085] The relationship between the input and output of a deep feedforward neural network is as follows:

[0086]

[0087] Based on the above formula, the lithology prediction model based on a deep feedforward neural network can be obtained as follows:

[0088] y = f(x, θ).

[0089] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the invention. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present invention can be combined with each other.

[0090] Example 1

[0091] This embodiment provides a multi-parameter lithology prediction method based on a deep feedforward neural network, including:

[0092] A comprehensive seismic wavelet and well logging interpolation impedance model was obtained based on the upsampling post-stack seismic data of the study area.

[0093] Post-stack impedance inversion was performed on the overlay seismic data based on the well logging interpolation impedance model and the integrated seismic wavelet to obtain a high-resolution P-wave impedance volume. Well-seismic calibration of each well in the study area was performed using the broadband Ricker wavelet based on the overlay seismic data to determine the dominant wavelet frequency of each well, obtaining the corresponding wavelet for each well. The integrated seismic wavelet for the study area was then obtained based on the wavelets corresponding to each well. Stratigraphic interpretation was performed on the overlay seismic data to construct a seismic stratigraphic framework. Kriging interpolation was performed on the well logging data along the seismic stratigraphic framework to obtain the well logging interpolation impedance model. The calculation expression for Kriging interpolation is as follows: Among them, z * (x0) represents the impedance value at the interpolation point, λ i The weights for the Kriging interpolation at the i-th wellbore, z(x) i ) represents the known impedance value at the i-th wellbore.

[0094] Bayesian probability classification was performed on the high-resolution P-wave impedance volume. Based on the high-resolution P-wave impedance volume and well logging lithology interpretation data, the conditional probability density function of each lithology was estimated using the Gaussian kernel function to obtain the high-resolution sandstone probability volume and mudstone probability volume.

[0095] High-resolution gamma-ray, shale content, and sand content data volumes were obtained using waveform indication simulation based on post-stacked frequency-spreading seismic data. Taking gamma-ray volumes as an example, each seismic trace to be simulated was compared with all well-side seismic traces. Wells in the same sedimentary environment exhibited similar seismic waveforms and logging curves. The top n wells with the highest waveform similarity to the seismic traces to be simulated (n being the effective sample number) were selected. Seismic waveforms were grouped according to their similarity to establish a sample set of seismic waveforms and logging curves indicating different sedimentary environments in the study area. The logging curves in the sample set were then analyzed. Multi-scale filtering in the wavelet domain is performed to obtain the high-frequency, mid-frequency, and low-frequency components of the well logging curve. The mid- and low-frequency components of the well logging curve are selected to establish a matched filter. The high-frequency component of the well logging curve has randomness, but the mid- and low-frequency components have high similarity. Moreover, the frequency band of the mid- and low-frequency components with high similarity greatly exceeds the effective frequency band of the seismic waveform. In this way, a matched filter is established to establish the intrinsic relationship between the gamma logging curve and the seismic waveform. The waveform of the seismic trace to be simulated is processed using the matched filter to obtain the corresponding gamma data volume. Similarly, the mud content and sand content volumes can be obtained.

[0096] A training sample set is constructed based on sandstone probability volumes, mudstone probability volumes, gamma data volumes, argillaceous content data volumes, sandy content data volumes, and well logging interpretation data of sandstone and mudstone lithology. Generally, sandstone is characterized by low gamma, low argillaceous content, and high sandy content. Lithology can be classified using the cross-sectional analysis of P-wave impedance and gamma data from well logging curves. Therefore, the multi-parameter data related to lithology classification, such as P-wave impedance volumes, gamma data, argillaceous content data, sandy content data, sandstone probability volumes, and mudstone probability volumes, are normalized and used as input data. Well logging interpretation results of sandstone and mudstone lithology are used as label data to construct the training sample set for a lithology prediction neural network. The normalization formula for the multi-parameter data is: Where s and s norm These represent the data before and after multi-parameter normalization, s and s respectively. max and s min These are the maximum and minimum values ​​of the data before multi-parameter normalization, respectively. Normalization can improve the computation speed of convolution kernels and activation functions in deep feedforward neural networks.

[0097] A lithology prediction model based on a deep feedforward neural network was constructed and trained using a training sample set. Lithology prediction was then performed based on the trained model. The deep feedforward neural network is a fully connected neural network, and its network structure includes an input layer h. (0) Output layer h (L) There are L-1 hidden layers. Neurons in adjacent layers are fully connected, while neurons within the same hidden layer and between adjacent hidden layers are not connected. For the input data x(z)∈R in the training sample set Set(z)... 2 Given the output data y(z)∈R, the output of its hidden layer is:

[0098]

[0099] The number of network layers, the number of neurons in each layer, and the activation function are as follows:

[0100]

[0101] The input layer has 6 neurons (n0 = 6). These 6 neurons receive inputs for the longitudinal wave impedance volume, gamma data volume, mud content volume, sand content volume, sandstone probability volume, and mudstone probability volume, respectively. The input layer activation function is... The ReLU function is chosen, and the Sigmoid function is chosen as the activation function for both the output and hidden layers. The number of neurons in the output layer is n. L =1, the output value is 0 (mudstone) or 1 (sandstone);

[0102] The parameters to be learned in the deep feedforward neural network are:

[0103]

[0104] The relationship between the input and output of a deep feedforward neural network is as follows:

[0105]

[0106] Based on the above formula, the lithology prediction model based on a deep feedforward neural network can be obtained as follows:

[0107] y = f(x, θ).

[0108] The relationship between multiple parameters and lithology is nonlinear, and deep feedforward neural networks can effectively construct this nonlinear relationship. Therefore, a lithology prediction model based on a deep feedforward neural network is constructed and trained using a training sample set to obtain a nonlinear relationship model between multiple parameters and lithology, thereby achieving lithology prediction.

[0109] Example 2

[0110] like Figure 2 As shown, this embodiment provides a multi-parameter lithology prediction method based on a deep feedforward neural network, including:

[0111] (1) Based on the overlay seismic data, well-seismic calibration of each well in the study area was performed using broadband Ricker wavelet to determine the dominant wavelet frequency of each well and obtain a comprehensive broadband wavelet applicable to the study area. The Ricker wavelet is a good approximation of the impulse function, with a narrow main lobe, small side lobe amplitude, and simple waveform. Under the same main lobe width, the broadband Ricker wavelet has a higher actual resolution. Based on the overlay seismic data, well-seismic calibration of each well in the study area was performed to determine the dominant wavelet frequency of each well, obtain the corresponding wavelet for each well, and obtain a comprehensive broadband wavelet applicable to the study area. Figure 3 The post-stacked seismic profile with a time domain range of 0.6s-1.3s and a duration of 0.7s is shown in this embodiment. The tight sandstone reservoir is located in the range of H32 to H36 and exhibits thin-layered seismic response characteristics.

[0112] (2) The well logging data is interpolated along the seismic horizon using the Kriging method to obtain the well logging interpolation impedance model. First, the actual post-stack seismic data is used for horizon interpretation, and a stratigraphic framework is constructed. Then, the well logging data is interpolated along the seismic horizon using the Kriging method to obtain the well logging interpolation impedance model. There are n wells in total, and the impedance value at the i-th wellbore is known to be z(x i Set the weights λ for Kriging interpolation. i Then the impedance z at the interpolation point * (x0) is: The target layers in this embodiment are mainly H32, H33, H34, H35, and H36. A stratigraphic framework is constructed based on these five layers, and then the Kriging method is used to interpolate the well logging data along the seismic layers to obtain the well logging interpolation impedance model.

[0113] (3) The post-stack impedance inversion method is used to obtain the P-wave impedance volume of the target layer. Commonly used post-stack impedance inversion methods include constrained sparse pulse inversion, generalized linear inversion, geostatistical inversion, and genetic algorithm inversion. This embodiment uses the constrained sparse pulse inversion method to invert the post-stack seismic data and obtain a high-resolution P-wave impedance volume, such as... Figure 4 As shown.

[0114] The objective function for wave impedance inversion is:

[0115] F = L p (R)+λL q (SD)+α -1 L1(ΔZ);

[0116]

[0117] Where D is the original seismic record, R is the reflection coefficient, i is the sampling point number, W is the composite seismic wavelet extracted in step (1); S represents the full-wavefield synthetic seismic record, calculated by the convolution of R and W; λ represents the equilibrium factor, α represents the trend soft constraint relative uncertainty factor; p and q are L-mode factors, Z is the full-wavefield absolute impedance, Z_mod is the well logging interpolation impedance model; F is the objective function for full-wavefield seismic data inversion, L p Let L be the L-norm of distance p. q L1 represents the L-norm at distance q, L2 represents the L-norm at distance 1, and ΔZ is the difference between the absolute impedance of the full-wave field and the low-frequency impedance of the layer velocity. i (lower) represents the minimum value of the absolute impedance constraint range of the full-wave field at the i-th sample point, Z i Z represents the absolute impedance of the full-wave field at the i-th sample point. i (upper) represents the maximum value of the absolute impedance constraint range of the full-wave field at the i-th sample point.

[0118] (4) Using the Bayesian probabilistic classification method, based on the high-resolution P-wave impedance volume and well logging lithology interpretation data, the conditional probability density function of each lithology is estimated using the Gaussian kernel function to obtain high-resolution sandstone and mudstone probability data volumes. The sandstone probability volume is shown below. Figure 5 As shown.

[0119] (5) Based on post-stack seismic data, a waveform indication simulation method was used to obtain high-resolution gamma, argillaceous content, and sand content data volumes. Each seismic trace to be simulated was compared with all well-side traces. Wells in the same sedimentary environment exhibited similar seismic waveforms and logging curve characteristics. The top 5 wells with the highest waveform similarity to the seismic traces to be simulated were selected. Seismic waveforms were grouped according to their similarity to establish a sample set of seismic waveforms and logging curves indicating different sedimentary environments in the study area. Multi-scale wavelet filtering was performed on the logging curves in the sample set, and a matched filter was established by selecting the mid-to-low frequency portions of the logging curves. The waveforms of the seismic traces to be simulated were processed using the matched filter to obtain the high-resolution gamma data volume, such as... Figure 6 As shown. Similarly, the mud content and sand content can be obtained.

[0120] (6) Normalize the multi-parameter data such as P-wave impedance volume, gamma body, argillaceous content volume, sandy content volume, sandstone probability volume, and mudstone probability volume and use them as input data. Use the well logging interpretation results of sandstone and mudstone lithology as label data to construct the training sample set of the lithology prediction neural network. Generally, sandstone is characterized by low gamma, low argillaceous content, and high sandy content. The cross-sectional analysis of P-wave impedance and gamma of the well logging curve can classify lithology. Therefore, normalize the multi-parameter data related to lithology classification, such as P-wave impedance volume, gamma body, argillaceous content volume, sandy content volume, sandstone probability volume, and mudstone probability volume, and use them as input data. Use the well logging interpretation results of sandstone and mudstone lithology as label data to construct the training sample set of the lithology prediction neural network. The normalization formula for the multi-parameter data is: Where s and s norm These represent the data before and after multi-parameter normalization, s and s respectively. max and s min These are the maximum and minimum values ​​of the data before multi-parameter normalization, respectively. Normalization can improve the computation speed of convolution kernels and activation functions in deep feedforward neural networks.

[0121] (7) The relationship between multiple parameters and lithology is nonlinear, and deep feedforward neural networks can effectively construct this nonlinear relationship. Therefore, a lithology prediction model based on a deep feedforward neural network is constructed and trained using a training sample set to obtain a nonlinear relationship model between multiple parameters and lithology, thereby realizing the lithology prediction function. The deep feedforward neural network is a fully connected neural network, and its network structure includes an input layer h... (0) Output layer h (L) There are L-1 hidden layers. Neurons in adjacent layers are fully connected, while neurons within the same hidden layer and between adjacent hidden layers are not connected. For the input data x(z)∈R in the training sample set Set(z)... 2 Given the output data y(z)∈R, the output of its hidden layer is:

[0122]

[0123] The number of network layers, the number of neurons in each layer, and the activation function are as follows:

[0124]

[0125] The input layer has 6 neurons (n0 = 6). These 6 neurons are respectively input to the longitudinal wave impedance volume, gamma volume, mud content volume, sand content volume, sandstone probability volume, and mudstone probability volume. The input layer activation function... The ReLU function is chosen, and the Sigmoid function is chosen as the activation function for both the output and hidden layers. The number of neurons in the output layer is n. L =1, the output value is 0 (mudstone) or 1 (sandstone);

[0126] The parameters to be learned in the deep feedforward neural network are:

[0127]

[0128] The relationship between the input and output of a deep feedforward neural network is as follows:

[0129]

[0130] Based on the above formula, the lithology prediction model based on a deep feedforward neural network can be obtained as follows:

[0131] y = f(x, θ).

[0132] (8) Based on the trained network model, lithology prediction is performed on actual seismic data. Based on the trained network model, high-resolution impedance inversion, high-resolution gamma ray, mudstone, and sandstone content simulation, and sandstone probability and mudstone probability classification are performed on actual post-stack seismic data to obtain a multi-parameter data volume. The multi-parameter data volume is normalized and input into the previously trained neural network model to obtain the corresponding lithology results, such as… Figure 7 As shown. Comparison Figure 4 and Figure 7 It can be seen that the multi-parameter lithology prediction method based on deep neural networks proposed in this invention has a higher prediction resolution than wave impedance inversion, and can better distinguish tight sandstone thin reservoirs in detail, which is more consistent with actual drilling, and provides strong data support for the geophysical exploration of oil and gas in tight sandstone thin reservoirs in the study area.

[0133] Example 3

[0134] This embodiment provides a multi-parameter lithology prediction device based on a deep feedforward neural network, including:

[0135] The acquisition module is used to acquire the integrated seismic wavelet and well logging interpolation impedance model of the study area based on the post-stacked frequency-spreading seismic data;

[0136] The inversion module is used to perform post-stack impedance inversion on the overlay seismic data based on the well logging interpolation impedance model and the integrated seismic wavelet to obtain a high-resolution P-wave impedance volume.

[0137] The classification module is used to perform Bayesian probability classification on high-resolution longitudinal wave impedance volumes to obtain high-resolution sandstone probability volumes and mudstone probability volumes.

[0138] The waveform indication simulation module is used to obtain high-resolution gamma data volume, clay content data volume, and sand content data volume based on the waveform indication simulation method of the post-stacked frequency-spreading seismic data.

[0139] The module is used to construct a training sample set based on sandstone probability volume, mudstone probability volume, gamma data volume, mud content data volume, sand content data volume and sandstone-mudstone lithology well logging interpretation data.

[0140] The construction and training module is used to build a lithology prediction model based on a deep feedforward neural network and train the lithology prediction model using a training sample set.

[0141] The lithology prediction module is used to perform lithology predictions based on the trained lithology prediction model.

[0142] Example 4

[0143] This disclosure also provides an electronic device, which includes:

[0144] At least one processor; and,

[0145] A memory communicatively connected to the at least one processor; wherein,

[0146] The memory stores instructions that can be executed by the at least one processor, which enables the at least one processor to perform the multi-parameter lithology prediction method based on a deep feedforward neural network in Embodiment 1.

[0147] An electronic device according to embodiments of the present disclosure includes a memory and a processor. The memory is used to store non-transitory computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0148] The processor may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In one embodiment of this disclosure, the processor is used to execute computer-readable instructions stored in the memory.

[0149] Those skilled in the art will understand that, in order to solve the technical problem of how to achieve a good user experience, this embodiment may also include well-known structures such as communication buses and interfaces, and these well-known structures should also be included within the protection scope of this disclosure.

[0150] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0151] Example 5

[0152] This disclosure provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the multi-parameter lithology prediction method based on a deep feedforward neural network in Embodiment 1.

[0153] A computer-readable storage medium according to embodiments of the present disclosure stores non-transitory computer-readable instructions. When these non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the methods described in the foregoing embodiments of the present disclosure are performed.

[0154] The aforementioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or portable hard drive), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).

[0155] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A method of multi-parameter lithology prediction based on deep feedforward neural networks, characterized in that, The method comprises the following steps: obtaining a comprehensive seismic wave and a logging interpolation impedance model based on the extended frequency post-stack seismic data of a study area; performing post-stack impedance inversion on the extended frequency post-stack seismic data based on the logging interpolation impedance model and the comprehensive seismic wave to obtain a high-resolution P-wave impedance volume; performing Bayesian probability classification on the high-resolution P-wave impedance volume to obtain a high-resolution sandstone probability volume and a shale probability volume; obtaining a high-resolution gamma data volume, a shale content data volume and a sandstone content data volume by using waveform indication simulation based on the extended frequency post-stack seismic data; constructing a training sample set based on the sandstone probability volume, the shale probability volume, the gamma data volume, the shale content data volume, the sandstone content data volume and sand-shale lithology logging interpretation data; constructing a lithology prediction model based on a deep feedforward neural network and training the lithology prediction model by using the training sample set; performing lithology prediction based on the trained lithology prediction model.

2. The deep feedforward neural network-based multi-parameter lithology prediction method of claim 1, wherein, performing well-seismic calibration on each well in the study area by using a wideband Ricker wavelet based on the extended frequency post-stack seismic data, determining the wavelet main frequency of each well, obtaining a wavelet corresponding to each well, and obtaining a comprehensive seismic wave of the study area based on the wavelet corresponding to each well.

3. The deep feedforward neural network-based multi-parameter lithology prediction method of claim 1, wherein, performing horizon interpretation based on the extended frequency post-stack seismic data to build a seismic stratigraphic framework, and performing Kriging interpolation on the logging data along the seismic stratigraphic framework to obtain a logging interpolation impedance model.

4. The deep feedforward neural network-based multi-parameter lithology prediction method of claim 3, wherein, The calculation expression of the Kriging interpolation is as follows: wherein z * (x0) is the impedance value of the point to be interpolated, λ i is the weight of the Kriging interpolation at the i-th borehole, z(x i ) is the known impedance value at the i-th borehole.

5. The deep feedforward neural network-based multi-parameter lithology prediction method of claim 1, wherein, The construction of the training sample set based on the sandstone probability volume, the shale probability volume, the gamma data volume, the shale content data volume, the sandstone content data volume and sand-shale lithology logging interpretation data comprises the following steps: performing normalization processing on the sandstone probability volume, the shale probability volume, the gamma data volume, the shale content data volume and the sandstone content data volume as input data of the training sample set; the sand-shale lithology logging interpretation data is used as label data of the training sample set; the training sample set is constructed based on the input data and the label data.

6. The deep feedforward neural network-based multi-parameter lithology prediction method of claim 1, wherein, The network structure of the lithology prediction model comprises the following steps: input layer h (0) output layer h (L) and L-1 hidden layers; neurons between adjacent layers are fully connected with each other, while neurons in the same hidden layer and neurons separated by hidden layers are not connected with each other.

7. The deep feedforward neural network-based multi-parameter lithology prediction method of claim 1, wherein, The lithology prediction model is as follows: y = f(x, θ); wherein x is input data and θ is a neural network learning parameter of each layer.

8. An electronic device, comprising: The electronic device comprises: at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the multi-parameter lithology prediction method based on a deep feedforward neural network according to any one of claims 1-7.

9. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium stores computer instructions for causing a computer to perform the multi-parameter lithology prediction method based on a deep feedforward neural network according to any one of claims 1-7.

10. A multi-parameter lithology prediction apparatus based on deep feedforward neural networks, characterized in that, The method comprises the following steps: obtaining a comprehensive seismic wave and a logging interpolation impedance model based on the extended frequency post-stack seismic data of a study area; an inversion module configured to perform post-stack impedance inversion on the frequency-extended post-stack seismic data based on the well-interpolated impedance model and the synthetic seismic wavelet to obtain a high-resolution P-wave impedance volume; a classification module configured to perform Bayesian probability classification on the high-resolution P-wave impedance volume to obtain a high-resolution sandstone probability volume and a shale probability volume; a waveform-indication simulation module configured to obtain a high-resolution gamma data volume, a shale content data volume and a sand content data volume by using a waveform-indication simulation method based on the frequency-extended post-stack seismic data; a construction module configured to construct a training sample set based on the sandstone probability volume, the shale probability volume, the gamma data volume, the shale content data volume, the sand content data volume and sand-shale lithology well-logging interpretation data; a construction and training module configured to construct a lithology prediction model based on a deep feedforward neural network and train the lithology prediction model by using the training sample set; a lithology prediction module configured to perform lithology prediction based on the trained lithology prediction model.

Images