A method, system, device, and medium for reservoir porosity prediction
By constructing a closed-loop porosity prediction network and combining well logging interpretation and reflectivity forward modeling, the problem of lack of lateral distribution and direct relationship in reservoir porosity evaluation was solved, achieving high-precision porosity prediction and improving the reliability and accuracy of the prediction results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PETROCHINA CO LTD
- Filing Date
- 2023-12-18
- Publication Date
- 2026-06-26
AI Technical Summary
In existing technologies, reservoir porosity evaluation based on well logging interpretation results cannot describe the lateral distribution of the reservoir, and there is no direct analytical relationship between seismic data and porosity, making it difficult to guarantee the reliability and accuracy of the prediction results.
A closed-loop porosity prediction network based on deep learning was constructed, which includes a first temporal convolutional network and a second temporal convolutional network. By training and optimizing the network model, a functional mapping relationship between pre-stack seismic data and porosity was established. Subsurface physical property information was obtained by combining well logging interpretation and reflectivity forward modeling to predict porosity.
It improves the reliability and accuracy of porosity prediction, avoids multiple assumptions in rock physics modeling and the accumulation of pre-stack inversion errors, and achieves high-precision reservoir porosity evaluation.
Smart Images
Figure CN120178328B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of petroleum exploration technology, and in particular to a method, system, equipment and medium for predicting reservoir porosity. Background Technology
[0002] With the continuous development of computer technology, significant research results and application progress have been made in seismic data processing and inversion based on deep learning. Among related technologies, while evaluating reservoir porosity based on well logging interpretation results can achieve extremely high vertical resolution, it cannot describe the lateral distribution of the reservoir. Pre-stack seismic data describes the spatial characteristics of the reservoir, and further, methods for predicting porosity based on the empirical relationship between porosity obtained from well logging interpretation and pre-stack seismic data have been proposed. However, from a rock physics perspective, there is no direct analytical relationship between seismic data and porosity. Porosity cannot be directly evaluated based on empirical formulas or other rock physics methods. It often requires combining knowledge from well logging, geology, and geophysics, and relying on extensive experimental analysis, data interpretation, and data processing to obtain predicted reservoir porosity results. This places high demands on the knowledge reserves of interpreters and computational costs, and the prediction results vary from person to person and are difficult to control, reducing the reliability and accuracy of the prediction results.
[0003] Therefore, there is an urgent need for an effective method for predicting reservoir porosity to solve the above problems. Summary of the Invention
[0004] In view of this, the present invention proposes a method, system, device and medium for predicting reservoir porosity.
[0005] To achieve the above objectives, one aspect of the present invention provides a method for predicting reservoir porosity, specifically including the following steps:
[0006] Pre-stack seismic data of a preset number of well point sample points are obtained to obtain pre-stack seismic gathers, and the porosity parameters corresponding to each well point sample point are obtained as tag porosity parameters.
[0007] A closed-loop porosity prediction network is constructed, comprising a first temporal convolutional network and a second temporal convolutional network. The pre-stack seismic gathers are input into the first temporal convolutional network to train and obtain predicted porosity. The predicted porosity is then input into the second temporal convolutional network to train and obtain synthetic seismic data.
[0008] The predicted porosity is compared with the corresponding tag porosity parameter to obtain a first error value, and the synthetic seismic data is compared with the corresponding pre-stack seismic gather to obtain a second error value. Based on the first error value and the second error value, the closed-loop porosity prediction network is optimized to determine the functional mapping relationship between the pre-stack seismic data and the predicted porosity.
[0009] Collect actual seismic data, and predict the target porosity of the actual seismic data based on the function mapping relationship and the closed-loop porosity prediction network.
[0010] In some embodiments, the step of optimizing the closed-loop porosity prediction network based on the first error value and the second error value to determine the functional mapping relationship between the pre-stack seismic data and the predicted porosity includes:
[0011] Based on the first error value and the second error value, a loss value is obtained, and it is determined whether the loss value meets a preset condition.
[0012] In response to the loss value not meeting the preset condition, the training parameters are adjusted, and the first temporal convolutional network and the second temporal convolutional network are retrained based on the pre-stack seismic gather, the labeled porosity parameter, and the adjusted training parameters to optimize the closed-loop porosity prediction network, and the step of determining whether the loss value meets the preset condition is returned.
[0013] In response to the loss value satisfying the preset condition, the mapping relationship obtained by training the first temporal convolutional network is used as a functional mapping relationship between the pre-stack seismic data and the predicted porosity.
[0014] In some implementations, the step of obtaining the loss value based on the first error value and the second error value includes:
[0015] The loss value is calculated based on the loss function, the first error value, and the second error value. The loss function is expressed as follows:
[0016]
[0017] In the formula, This represents the label porosity parameter. This indicates the predicted porosity. Then, represents the first error value, and s represents the pre-stack seismic data. This refers to the synthetic seismic data. This represents the second error value, where ε is a hyperparameter. The loss value is given.
[0018] In some implementations, the step of inputting the pre-stack seismic gathers into the first temporal convolutional network to train and obtain predicted porosity includes:
[0019] The pre-stack seismic gathers are normalized, and the normalized pre-stack seismic gathers are divided into several one-dimensional vector data of a preset size.
[0020] All the corresponding one-dimensional vector data are input into the first temporal convolutional network to train and obtain the predicted porosity.
[0021] In some embodiments, the steps of obtaining pre-stack seismic data of a preset number of wellpoint sample points to obtain pre-stack seismic gathers, and obtaining the porosity parameters corresponding to the wellpoint sample points as tag porosity parameters, include:
[0022] Acquire historical seismic data and historical well logging data for a preset number of well point sample points;
[0023] Prestack seismic data of each well point sample point is obtained based on the historical seismic data and forward modeling using the reflectivity method to obtain prestack seismic gathers.
[0024] The historical logging data are interpreted to obtain the porosity parameters of each well point sample point, and the porosity parameters are used as tag porosity parameters.
[0025] In some embodiments, the step of acquiring actual seismic data and predicting the target porosity of the actual seismic data based on the function mapping relationship and the closed-loop porosity prediction network includes:
[0026] The actual seismic data is analyzed in relation to the strata absorption between the actual seismic data and the shot-receiver distance. The actual seismic data is then compensated based on the strata absorption. The compensated actual seismic data is then used as the target seismic data for predicting the target porosity.
[0027] The target actual seismic data is input into the closed-loop porosity prediction network that has determined the function mapping relationship to obtain the target porosity.
[0028] In some implementations, the training parameters include hyperparameters and network structure parameters of the closed-loop porosity prediction network.
[0029] In another aspect of the present invention, a reservoir porosity prediction system is provided, characterized in that it includes:
[0030] The acquisition unit is used to acquire pre-stack seismic data of a preset number of well point sample points to obtain pre-stack seismic gathers, and to acquire the porosity parameters corresponding to each well point sample point as tag porosity parameters.
[0031] The construction unit is used to construct a closed-loop porosity prediction network including a first temporal convolutional network and a second temporal convolutional network. The pre-stack seismic gathers are input into the first temporal convolutional network to train and obtain predicted porosity, and the predicted porosity is input into the second temporal convolutional network to train and obtain synthetic seismic data.
[0032] An optimization unit is configured to compare the predicted porosity with the corresponding tag porosity parameter to obtain a first error value, compare the synthetic seismic data with the corresponding pre-stack seismic gather to obtain a second error value, and optimize the closed-loop porosity prediction network based on the first error value and the second error value to determine the functional mapping relationship between the pre-stack seismic data and the predicted porosity.
[0033] The prediction unit is used to collect actual seismic data and predict the target porosity of the actual seismic data based on the function mapping relationship and the closed-loop porosity prediction network.
[0034] In another aspect of the present invention, a computer device is provided, comprising: at least one processor; and a memory storing a computer program executable on the processor, the computer program performing the steps of the method described above when executed by the processor.
[0035] In another aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method steps.
[0036] The present invention has at least the following beneficial technical effects:
[0037] (1) Through the reservoir porosity prediction method of the present invention, a closed-loop porosity prediction network containing two time-domain convolutional network structures is designed. The output results of the first time-domain convolutional network, the output results of the second time-domain convolutional network, porosity parameters, and pre-stack seismic data are combined with a loss function to achieve dual constraints on the closed-loop porosity prediction network, thereby ensuring that the establishment of the function mapping relationship in the first time-domain convolutional network has high accuracy. The optimal closed-loop porosity prediction network that meets the preset conditions is used for porosity prediction and evaluation, ensuring the reliability of the porosity prediction results.
[0038] (2) Porosity parameters and pre-stack seismic data were obtained by well logging interpretation and forward modeling using the reflectivity method, respectively, providing multi-scale and multi-range subsurface physical property information for training the closed-loop porosity prediction network, improving the comprehensiveness of the closed-loop porosity prediction network and ensuring the accuracy of porosity prediction results.
[0039] (3) Basic processing work such as absorption compensation related to shot-receiver distance was carried out on the collected seismic data, which effectively improved the quality of the input data of the closed-loop porosity prediction network, improved the accuracy of porosity prediction results and reduced the difficulty of training the closed-loop porosity prediction network. Attached Figure Description
[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort.
[0041] Figure 1 A block diagram of an embodiment of the reservoir porosity prediction method provided by the present invention;
[0042] Figure 2 This is a schematic diagram of an embodiment of the closed-loop porosity prediction network model provided by the present invention.
[0043] Figure 3 This is a schematic diagram of an embodiment of the porosity prediction result on the training set provided by the present invention;
[0044] Figure 4 This is a schematic diagram of an embodiment of the porosity prediction results on a validation set provided by the present invention.
[0045] Figure 5 This is a schematic diagram of an embodiment of the training error curve of the closed-loop porosity prediction network prediction framework provided by the present invention.
[0046] Figure 6 A schematic diagram illustrating an embodiment of the relationship between seismic amplitude and incident angle under different porosities provided by the present invention;
[0047] Figure 7 A schematic diagram of an embodiment for interpreting porosity parameters from well logging data provided by the present invention;
[0048] Figure 8 This is a schematic diagram of an embodiment of pre-stack seismic data obtained by forward modeling using the reflectivity method provided by the present invention.
[0049] Figure 9 A schematic diagram of an embodiment of pre-stack seismic data before processing;
[0050] Figure 10 This is a schematic diagram of an embodiment of pre-stack seismic data after strata absorption compensation related to shot-receiver distance provided by the present invention.
[0051] Figure 11 A cross-sectional schematic diagram of an embodiment of the porosity prediction results of actual seismic data provided by the present invention;
[0052] Figure 12 A cross-sectional schematic diagram of an embodiment of the porosity prediction results of actual seismic data provided by the present invention;
[0053] Figure 13 A three-dimensional schematic diagram of an embodiment of the porosity prediction result of actual seismic data provided by the present invention;
[0054] Figure 14 A schematic diagram of an embodiment of the reservoir porosity prediction system provided by the present invention;
[0055] Figure 15 A schematic diagram of the structure of an embodiment of the computer device provided by the present invention;
[0056] Figure 16 This is a schematic diagram of an embodiment of the computer-readable storage medium provided by the present invention. Detailed Implementation
[0057] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to specific examples and the accompanying drawings.
[0058] It should be noted that all uses of "first" and "second" in the embodiments of the present invention are for the purpose of distinguishing two entities or parameters with the same name but different names. It is clear that "first" and "second" are only for the convenience of expression and should not be construed as limiting the embodiments of the present invention. Subsequent embodiments will not explain this in detail.
[0059] Reservoir physical parameters are crucial for reservoir description and play a vital role in comprehensive reservoir evaluation. Traditional physical parameter prediction typically involves two parts: rock physics modeling and reservoir parameter inversion. However, rock physics modeling relies on complex assumptions and is difficult to solve. Furthermore, due to the highly nonlinear relationship between seismic elastic parameters and reservoir physical parameters, traditional physical parameter evaluation methods can lead to the accumulation and amplification of errors. Therefore, a direct physical parameter inversion method based on seismic data has been developed. Seismic data and reservoir physical parameter information are closely linked through rock physics and the Zoeppritz equation, and their relationship is complex. Various geophysical inversion methods need to be comprehensively applied to obtain reservoir physical parameter information from pre-stack seismic data, thereby providing reliable evaluation indicators for reservoir spatial distribution, physical properties, and hydrocarbon potential. Given the high-dimensional nonlinearity of the relationship between seismic data and reservoir physical parameters, deep learning is suitable for mining and establishing corresponding mapping relationships. This data-driven inversion method avoids the difficulties of rock physics modeling and the complexity of solving inversion problems, making it an efficient and low-cost feature extraction method. Convolutional Neural Networks (CNNs), due to the unique nature of convolution operations and their similarity to geophysical convolution theory, have been widely used in seismic data processing and inversion. The CNN network structure is highly adaptable and easily designed and modified for specific problems, enabling it to fully extract and integrate reservoir structure and physical property information carried in seismic data, providing a feasible technical means for the intelligent and efficient prediction of reservoir physical parameters.
[0060] This invention, based on the TCN (Temporal Convolutional Network) architecture, designs a closed-loop porosity prediction framework comprising two TCN structures. Combining this framework, a method for predicting reservoir porosity directly based on seismic data is proposed. By leveraging the correlation between seismic amplitude variations and porosity, a complex nonlinear relationship for porosity inversion is extracted and established from seismic data. The inversion process is constrained by seismic data, thereby fully utilizing the subsurface physical property information at different scales and ranges provided by well logging data interpretation and seismic data. This provides essential fundamental understanding and reliable basis for reservoir characterization and prediction.
[0061] Based on the above objectives, a first aspect of the present invention provides an embodiment of a reservoir porosity prediction method. For example... Figure 1 As shown, it includes the following steps:
[0062] Step S100: Obtain pre-stack seismic data of a preset number of well point sample points to obtain pre-stack seismic gathers, and obtain the porosity parameters corresponding to each well point sample point as label porosity parameters.
[0063] Step S200: Construct a closed-loop porosity prediction network including a first time-domain convolutional network and a second time-domain convolutional network; input pre-stack seismic gathers into the first time-domain convolutional network to train and obtain predicted porosity; input predicted porosity into the second time-domain convolutional network to train and obtain synthetic seismic data.
[0064] Step S300: Compare the predicted porosity with the corresponding labeled porosity parameters to obtain a first error value, compare the synthesized seismic data with the corresponding pre-stack seismic gathers to obtain a second error value, and optimize the closed-loop porosity prediction network based on the first and second error values to determine the functional mapping relationship between the pre-stack seismic data and the predicted porosity.
[0065] Step S400: Collect actual seismic data and predict the target porosity of the actual seismic data based on the function mapping relationship and the closed-loop porosity prediction network.
[0066] In some implementations, the TCN network structure acts as a feature extractor, directly extracting reservoir physical parameters from seismic data. This is based on the principle of the difference in seismic reflection AVO (Amplitude versus Offset) under different porosities. Seismic amplitude information at different incident angles and porosity obtained from well logging interpretation are used as inputs to the TCN network. Based on the TCN network architecture, the physical property information carried in the seismic data is mined, and porosity prediction results are provided. The TCN network is a variation and supplement to a one-dimensional convolutional neural network, enabling it to better handle temporal data. It uses one-dimensional causal convolution and extended convolution as standard convolutional layers, and encapsulates every two standard convolutional layers with an identity mapping into a residual module. Through the superposition of multiple residual modules, a deep network is formed, and fully convolutional layers replace fully connected layers in the network structure. Based on the aforementioned network structure, the TCN network possesses both the ability to extract high-dimensional features and the ability to memorize historical information. This effectively avoids the gradient problem encountered during network training. Compared to traditional recurrent neural networks, it offers advantages such as lower training memory, more flexible design, flexible receptive field size, and parallel processing capability. Furthermore, the TCN network incorporates regularization and other processing techniques to facilitate effective training of deep networks.
[0067] In some implementations, porosity parameters are interpreted from well logging data and used as labels for training a closed-loop porosity prediction network. These labels, along with pre-stack seismic data reflecting porosity changes obtained through forward modeling using the reflectivity method, are used as input data for the closed-loop porosity prediction network. In the closed-loop porosity prediction network, a first TCN network is used to extract the physical property information carried in the pre-stack seismic data and to constrain the closed-loop porosity prediction network by constructing an error term using the porosity prediction results and labels. A second TCN network is used to map the porosity prediction results to synthetic seismic data and to construct a consistency error term using the synthetic seismic data and the input pre-stack seismic data. The second TCN network constrains the reliability of porosity inversion in the first TCN network. It is understood that the second TCN network introduces a forward modeling module of the TCN network structure, mapping the porosity prediction results to seismic data through the network and using this module, along with the input pre-stack seismic data, to construct an error term to constrain and control the porosity inversion process, ensuring the reliability of the porosity prediction results and providing valuable evidence for reservoir evaluation.
[0068] The reservoir porosity prediction method of this invention designs a closed-loop porosity prediction network containing two time-domain convolutional network structures. The output results of the first time-domain convolutional network, the output results of the second time-domain convolutional network, porosity parameters, and pre-stack seismic data are combined with a loss function to achieve dual constraints on the closed-loop porosity prediction network. This ensures that the establishment of the function mapping relationship in the first time-domain convolutional network has high accuracy. The optimal closed-loop porosity prediction network that meets the preset conditions is used for porosity prediction and evaluation, ensuring the reliability of the porosity prediction results, avoiding multiple assumptions in rock physics modeling, and avoiding the amplification and accumulation of pre-stack inversion errors.
[0069] In some implementations, the step of optimizing the closed-loop porosity prediction network based on a first error value and a second error value to determine the functional mapping relationship between pre-stack seismic data and predicted porosity includes: obtaining a loss value based on the first error value and the second error value, and determining whether the loss value meets a preset condition; in response to the loss value not meeting the preset condition, adjusting the training parameters, retraining the first temporal convolutional network and the second temporal convolutional network based on the pre-stack seismic gathers, labeled porosity parameters, and the adjusted training parameters to optimize the closed-loop porosity prediction network, and returning to the step of determining whether the loss value meets the preset condition; in response to the loss value meeting the preset condition, using the mapping relationship obtained by training the first temporal convolutional network as the functional mapping relationship between pre-stack seismic data and predicted porosity.
[0070] In some implementations, the step of obtaining the loss value based on the first error value and the second error value includes: calculating the loss value based on a loss function, the first error value, and the second error value, wherein the loss function is expressed as: In the formula, Indicates the label porosity parameter. Indicates predicted porosity. This represents the first error value, where s represents the pre-stack seismic data. This indicates synthetic seismic data. This represents the second error value, where ε is a hyperparameter. This is the loss value.
[0071] In some implementations, the training parameters include hyperparameters and network structure parameters of the closed-loop porosity prediction network.
[0072] In some implementations, the first error value guides the closed-loop porosity prediction network in establishing a mapping relationship between pre-stack seismic data and porosity. The second error value improves the reliability of porosity prediction and further constrains the porosity prediction process. Hyperparameters control the weight of the second error value, i.e., control the weight of the seismic data constraint term. Preset conditions are set according to actual application conditions. When the loss value meets the preset conditions, it indicates that the closed-loop porosity prediction network model has small errors, high accuracy, and stable error and accuracy changes during training. This means the training process is stable and efficient, without overfitting or other problems. The training parameters at this point can be saved for subsequent applications in complex scenarios such as transfer learning.
[0073] In one example, Figure 2 This is a schematic diagram of an embodiment of the closed-loop porosity prediction network model provided by the present invention. Based on the loss values from several training iterations, the closed-loop porosity prediction network model contains five hidden layers, using the ReLU (Linear Rectification function) as the activation function, and a fully connected layer for outputting the porosity prediction result. Through network hyperparameter experiments, the acquired pre-stack seismic data and porosity parameters are divided into training and validation sets, respectively used as training data for the closed-loop porosity prediction network and for evaluating the effectiveness and reliability of the closed-loop porosity prediction network. When the hyperparameter ε in the loss function is 0.1, the training effect of the closed-loop porosity prediction network is better. Figure 3 This is a schematic diagram of an embodiment of the porosity prediction result on the training set provided by the present invention. Figure 4 This is a schematic diagram of an embodiment of the porosity prediction results on the validation set provided by the present invention.
[0074] In some implementations, the network structure parameters of the closed-loop porosity prediction network include the learning rate of the TCN network, the number of samples in the training set and the dataset, and the number of network iterations.
[0075] In one example, the learning rate of both network structures is set to 0.001, and 512 samples are used as a batch for network training. The training is iterated 500 times. The RMSprop (root mean square propagation) algorithm is used as the optimization algorithm to extract the temporal correlation of the data. A loss function is constructed to examine the changes in error value and accuracy of the closed-loop porosity prediction network during training and to optimize the closed-loop porosity prediction network model accordingly. Figure 5 This is a schematic diagram of an embodiment of the training error curve of the closed-loop porosity prediction network prediction framework provided by the present invention.
[0076] In some implementations, the step of inputting pre-stack seismic gathers into a first temporal convolutional network to train and obtain predicted porosity includes: normalizing the pre-stack seismic gathers and dividing the normalized pre-stack seismic gathers into a number of one-dimensional vector data of a preset size; and inputting all the corresponding one-dimensional vector data into the first temporal convolutional network to train and obtain predicted porosity.
[0077] In some implementations, for the porosity prediction problem, since the network needs to process seismic reflection amplitudes at different incident angles under the same porosity simultaneously and needs to consider the relationship of amplitude changes, it is necessary to treat porosity as sequential data with data correlation and treat this sequential data as a one-dimensional object. The correlation between data is processed by causal convolution, and the receptive field of convolution is expanded by dilated convolution to capture longer dependencies. Figure 6 This is a schematic diagram of an embodiment of the relationship between seismic reflection amplitude and incident angle under different porosities provided by the present invention.
[0078] In one example, the seismic waveforms of the pre-stack seismic gathers are normalized to a range of -1 to 1. The pre-stack seismic gathers are then divided into one-dimensional vectors of a preset size of 1*31 as input data for the TCN network. The 31 represents the seismic reflection amplitudes corresponding to incident angles from 0° to 30°. The preset size can be set according to the actual application scenario, and no specific limitation is made here.
[0079] In some implementations, the steps of acquiring pre-stack seismic data of a preset number of wellpoint sample points to obtain pre-stack seismic gathers and acquiring the porosity parameters corresponding to the wellpoint sample points as tag porosity parameters include: acquiring historical seismic data and historical logging data of a preset number of wellpoint sample points; acquiring pre-stack seismic data of each wellpoint sample point based on each historical seismic data and forward modeling using the reflectivity method to obtain pre-stack seismic gathers; performing logging interpretation on each historical logging data to obtain the porosity parameters of each wellpoint sample point, and using the porosity parameters as tag porosity parameters.
[0080] In some implementations, complete logging data is crucial for ensuring the accuracy of pre-stack property estimation and prediction. Conventional logging data typically does not include shear wave logging curves, while shear wave velocity plays a vital role in tasks such as lithology identification and reservoir characterization, and is also one of the fundamental data for pre-stack inversion. Therefore, under the premise of optimizing the limited logging data, shear wave logging curve reconstruction is carried out to ensure the integrity of the logging data. Based on this, logging interpretation is performed to obtain porosity parameters at well locations. Figure 7 This is a schematic diagram of an embodiment of interpreting porosity parameters from well logging data provided by the present invention.
[0081] Forward modeling based on the Zoeppritz equation assumes a single interface and does not consider various propagation effects of the wavefield. This leads to discrepancies between the modeled and actual wavefields and results in missing inversion information. This affects the method's adaptability and transferability in practical data processing and applications, and increases the difficulty for the network to extract reservoir property-related information from seismic records. Therefore, this invention employs reflectivity-based forward modeling to synthesize pre-stack seismic records, obtaining a full-wavefield seismic record encompassing various propagation effects as the foundational data for information extraction by the closed-loop porosity prediction network.
[0082] In one example, the reflectivity method for forward modeling synthesizes pre-stack seismic gathers, which can effectively simulate wavefield records of various modes and is widely used in forward modeling of layered media. The basic principle of the reflectivity method for pre-stack waveform forward modeling is as follows:
[0083] First, the equations of motion and constitutive equations are obtained in the frequency domain.
[0084] Equations of motion:
[0085] Constitutive equation: τ=C:σ;
[0086] In the above formula, τ is stress, σ is strain tensor, C is the elastic coefficient of the generalized Hooke's law at frequency, ρ represents the medium density, ω is angular frequency, u represents displacement vector, and f is the body force term. This represents the gradient operator.
[0087] Under the assumption of a layered medium, parameters such as velocity and density are functions of depth. The frequency domain relationships described above can be transformed into the following ordinary differential equations through simple mathematical transformations: In the formula, b is the stress-displacement vector, F is the body force term, and A is the system matrix of the equation. It is the derivative with respect to the z-direction, where i is the imaginary unit.
[0088] The solutions to ordinary differential equations are usually analytical in mathematics, which makes the reflectivity method one of the commonly used methods for forward modeling of layered media. The solutions to the above ordinary differential equations can be solved by the propagation matrix method, defining the propagation matrix P: b(z1) = P(z1,z2)b(z2).
[0089] The propagation matrix is used to link displacement vectors across different layers, but this method becomes unstable with exponential growth. To overcome this instability, the Kennett recursive algorithm was developed, which uses the eigenvalue and eigenvector matrices of the system matrix to define the reflection and transmission coefficient matrices of the uplink and downlink waves.
[0090]
[0091] Where D is the eigenvector matrix of the system matrix, and M U M D N U N D Its block matrix, M U The operator M represents the transformation from the upward wave vector to the upward wave displacement vector. D Downward wave vector to downward wave displacement vector conversion operator, N U The operator N represents the transformation of the upward wave vector into the upward wave stress vector. D This is an operator that transforms a downwave vector into a downwave stress vector.
[0092] By combining the eigenvector matrix D and the propagation matrix, the wave propagation matrix can be recursively obtained. For a uniform layer, the wave propagation matrix Q can be expressed as:
[0093]
[0094] In the formula, E u =diag(exp[iω(z-z0)q u ]), E d =diag(exp[iω(z-z0)q d ]), This represents the vertical slowness matrix.
[0095] Furthermore, the reflection and transmission matrices of different layers are calculated:
[0096] R D =Q UD (Q DD ) -1 ,
[0097] R U =-(Q DD ) -1 Q DU,
[0098] T D =(Q DD ) -1 ,
[0099] T U =Q UU -Q UD (Q DD ) -1 Q DU ;
[0100] Among them, Q UU Q UD Q DD and Q DU R is a block matrix of the uniform layer propagation matrix. D R U T D and T U These represent the uplink and downlink wave forms of the reflection and transmission coefficients, respectively.
[0101] Based on the above reflection and transmission matrix, the displacement solution in the frequency-ray parameter (ω,p) domain can be further expressed as:
[0102]
[0103] In the formula, u(z) r ) represents the displacement solution expression below the detector, ∑ d (z s () represents the portion radiating downwards from the earthquake source. This represents the wave propagation response received at the receiver from below. The superscripts in the reflection / transmission coefficient matrix denote different layers: F for free surface, S for source, R for receiver, and N for bottom interface. The subscripts in the reflection / transmission coefficient matrix denote different traveling waves: U for upward-traveling wave and D for downward-traveling wave. For example, This is the reflection coefficient matrix that propagates downwards between the detector point and the bottom interface. This is the transmission coefficient matrix that propagates downwards between the seismic source and the receiver.
[0104] The reflectivity method is based on the plane wave assumption; therefore, to obtain a record representing a point source earthquake, spherical wave synthesis of the plane wave is necessary. After obtaining the displacement solution in the (ω,p) domain, the following transformation yields the seismic record in the time-space (t,x) domain:
[0105]
[0106] In the formula, u(ω,p) is the displacement solution in the (ω,p) domain, J0 is the zero-order Bessel function, and x is the offset distance.
[0107] In layered media, the displacement solution of the entire wave field can be obtained analytically using the reflectivity method, and then the seismic record in the (t,x) domain can be obtained using the transformation formula of spherical wave synthesis. Figure 8 This is a schematic diagram of an embodiment of pre-stack seismic data obtained by forward modeling using the reflectivity method provided by the present invention.
[0108] In summary, the reflectivity method forward modeling includes the following three main steps:
[0109] (1) Calculate the reflection and transmission coefficient matrix of different layers;
[0110] (2) Calculate the displacement solution of the full wave field in the frequency-ray parameter domain;
[0111] (3) Calculate earthquake records in the time-space domain.
[0112] The reservoir porosity prediction method of this invention obtains porosity parameters and pre-stack seismic data through well logging interpretation and reflectivity forward modeling, respectively. This provides multi-scale and multi-range subsurface physical property information for training the closed-loop porosity prediction network, improving the comprehensiveness of the closed-loop porosity prediction network and ensuring the accuracy of porosity prediction results.
[0113] In some implementations, the steps of acquiring actual seismic data and predicting the target porosity of the actual seismic data based on function mapping relationships and closed-loop porosity prediction networks include: analyzing the formation absorption of the actual seismic data in relation to the shot-receiver distance; compensating the actual seismic data according to the formation absorption; using the compensated actual seismic data as the target seismic data for predicting the target porosity; and inputting the target seismic data into a closed-loop porosity prediction network with determined function mapping relationships to obtain the target porosity.
[0114] In some implementations, during the practical application testing phase of the closed-loop porosity prediction network, it is necessary to perform gather optimization processing such as absorption compensation related to the shot-receiver distance on the actual seismic data (i.e., actual seismic data) in advance to restore the true amplitude changes and correct the distortion of the reflection amplitude in the shot-receiver distance direction, so as to provide a high-quality data foundation for pre-stack inversion in the closed-loop porosity prediction network.
[0115] The formation absorption is decomposed into the shot-receiver distance direction and the depth direction. The equivalent absorption factor in the shot-receiver distance direction is:
[0116]
[0117] Among them, v rms Let Δt be the mean square root velocity. 0,i For the vertical two-way propagation travel of the i-th layer, t 0,n Let α be the self-excitation and self-reception time of seismic waves in a layered medium.i v i Let be the absorption factor and velocity of the i-th layer, respectively.
[0118] Formation absorption reduces vertical seismic resolution and alters AVO reflection characteristics, leading to interpretation traps in AVO analysis and decreasing lateral consistency of wavegroup relationships among different depth gathers in seismic data. This source-receiver offset-related absorption primarily affects the AVO characteristics of seismic reflections, and this lateral amplitude variation is a crucial basis for porosity prediction and evaluation based on seismic data. Therefore, lateral compensation and processing of seismic data are necessary before practical application of the network.
[0119] Its specific compensation equation is expressed as follows:
[0120] x=(G T G+λI) -1 G T y;
[0121] Where x represents the seismic record before the formation absorption compensation related to the shot-receiver distance, y represents the seismic record after the formation absorption compensation related to the shot-receiver distance, and G is the integral kernel matrix composed of the formation absorption responses related to the shot-receiver distance at different times. Figure 9 This is a schematic diagram of an embodiment of pre-stack seismic data before processing. Figure 10 This is a schematic diagram of an embodiment of pre-stack seismic data after shot-receiver distance-dependent stratigraphic absorption compensation provided by the present invention. After shot-receiver distance-dependent stratigraphic absorption compensation, the lateral variation characteristics of the seismic amplitude are well recovered, providing high-quality basic data for subsequent inversion work and high-quality, reliable data for the subsequent utilization of seismic waveform information.
[0122] The reservoir porosity prediction method of this invention performs basic processing on the acquired seismic data, such as absorption compensation related to shot-receiver distance, which effectively improves the quality of input data to the closed-loop porosity prediction network, enhances the accuracy of porosity prediction results, and reduces the difficulty of training the closed-loop porosity prediction network.
[0123] In one example, the following describes the processing procedure of the reservoir porosity prediction method:
[0124] Step 1: Perform well logging interpretation on the well logging data to obtain the porosity parameters at the well points as label data for network training;
[0125] Step 2: Conduct seismic forward modeling based on the reflectivity method to obtain pre-stack seismic data reflecting porosity changes;
[0126] Step 3: Construct a closed-loop porosity prediction network model based on the TCN network, divide the pre-stack seismic data and porosity parameters into training and validation sets, and perform optimization work such as network training and training parameter adjustment.
[0127] Step 4: Perform gather optimization processing on the actual seismic data, such as absorption compensation related to the shot-receiver distance;
[0128] Step 5: Based on the actual seismic data after absorption compensation processing, conduct porosity prediction work to further examine the practical application potential and applicability of the closed-loop porosity prediction network.
[0129] Figure 11 This is a cross-sectional schematic diagram of an embodiment of the porosity prediction results of actual seismic data provided by the present invention. Figure 12 This is a cross-sectional schematic diagram of an embodiment of the porosity prediction results of actual seismic data provided by the present invention. Figure 13 This is a three-dimensional schematic diagram of an embodiment of the porosity prediction results from actual seismic data provided by the present invention. Figure 11 As shown, judging from the degree of agreement between the inversion profile and the well, this invention provides a reliable basis for reservoir evaluation and is a practical and efficient method for porosity prediction.
[0130] Based on the same inventive concept, according to another aspect of the present invention, such as Figure 14 As shown, a reservoir porosity prediction system is also provided, which includes:
[0131] The acquisition unit 110 is used to acquire pre-stack seismic data of a preset number of well point sample points to obtain pre-stack seismic gathers, and to acquire the porosity parameters corresponding to each well point sample point as label porosity parameters.
[0132] The construction unit 120 is used to construct a closed-loop porosity prediction network including a first time-domain convolutional network and a second time-domain convolutional network. The pre-stack seismic gathers are input to the first time-domain convolutional network to train and obtain predicted porosity, and the predicted porosity is input to the second time-domain convolutional network to train and obtain synthetic seismic data.
[0133] The optimization unit 130 is used to compare the predicted porosity with the corresponding labeled porosity parameter to obtain a first error value, compare the synthetic seismic data with the corresponding pre-stack seismic gather to obtain a second error value, and optimize the closed-loop porosity prediction network based on the first error value and the second error value to determine the functional mapping relationship between the pre-stack seismic data and the predicted porosity.
[0134] Prediction unit 140 is used to acquire actual seismic data and predict the target porosity of the actual seismic data based on function mapping relationships and closed-loop porosity prediction networks.
[0135] The reservoir porosity prediction system of this invention designs a closed-loop porosity prediction network comprising two temporal convolutional network structures. The outputs of the first and second temporal convolutional networks, porosity parameters, and pre-stack seismic data are combined with a loss function to implement dual constraints on the closed-loop porosity prediction network. This ensures high accuracy in establishing the function mapping relationship in the first temporal convolutional network. The optimal closed-loop porosity prediction network that meets the preset conditions is then used for porosity prediction and evaluation, guaranteeing the reliability of the porosity prediction results.
[0136] Based on the same inventive concept, according to another aspect of the present invention, such as Figure 15 As shown, an embodiment of the present invention also provides a computer device 30, which includes a processor 310 and a memory 320. The memory 320 stores a computer program 321 that can be run on the processor. When the processor 310 executes the program, it performs the steps of the method described above.
[0137] Based on the same inventive concept, according to another aspect of the present invention, such as Figure 16 As shown, embodiments of the present invention also provide a computer-readable storage medium 40, which stores a computer program 410 that, when executed by a processor, performs the methods described above.
[0138] Finally, it should be noted that those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The storage medium for the program can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc. The above computer program embodiments can achieve the same or similar effects as any of the corresponding foregoing method embodiments.
[0139] Those skilled in the art will also understand that the various exemplary logic blocks, modules, circuits, and algorithm steps described in conjunction with the disclosure herein can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability between hardware and software, the functionality of various illustrative components, blocks, modules, circuits, and steps has been generally described. Whether this functionality is implemented as software or as hardware depends on the specific application and the design constraints imposed on the system as a whole. Those skilled in the art can implement the functionality in various ways for each specific application, but such implementation decisions should not be construed as departing from the scope of the embodiments disclosed herein.
[0140] The above are exemplary embodiments disclosed in this invention. However, it should be noted that various changes and modifications can be made without departing from the scope of the embodiments of this invention as defined by the claims. The functions, steps, and / or actions of the methods according to the disclosed embodiments described herein do not need to be performed in any particular order. The sequence numbers of the disclosed embodiments of this invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. Furthermore, although the elements disclosed in the embodiments of this invention may be described or claimed individually, they may be understood as multiple unless explicitly limited to a singular number.
[0141] It should be understood that, as used herein, the singular form “a” is intended to include the plural form as well, unless the context clearly supports an exception. It should also be understood that, as used herein, “and / or” refers to any and all possible combinations of one or more of the associated listed items.
[0142] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples. Within the framework of the invention, technical features of the above embodiments or different embodiments can be combined, and many other variations of different aspects of the invention exist, which are not provided in the details for the sake of brevity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the protection scope of the invention.
Claims
1. A method for predicting reservoir porosity, characterized in that, include: Pre-stack seismic data from a predetermined number of wellpoint sample points are acquired to obtain pre-stack seismic gathers. Porosity parameters corresponding to each wellpoint sample point are obtained as label porosity parameters. The pre-stack seismic data is acquired based on historical seismic data of each wellpoint sample point and reflectivity forward modeling. The reflectivity forward modeling includes: calculating the reflection and transmission coefficient matrices of different layers; calculating the displacement solution of the full wavefield in the frequency-ray parameter domain; and calculating the seismic record in the time-space domain. The displacement solution in the frequency-ray parameter domain is expressed as: ； In the formula, The expression for the displacement below the detector is given. This is the portion that radiates downwards from the earthquake source. The reflection and transmission coefficient matrix represents the wave propagation response received by the receiver from below. The superscripts in the matrix denote different layers: F represents the free surface, S represents the source, R represents the receiver, and N represents the bottom interface. The subscripts in the matrix denote different traveling waves: U represents an upward-traveling wave, and D represents a downward-traveling wave. This is the reflection coefficient matrix that propagates downwards between the detector point and the bottom interface. This is the transmission coefficient matrix propagating downwards between the source and the receiver. A closed-loop porosity prediction network is constructed, comprising a first temporal convolutional network and a second temporal convolutional network. The pre-stack seismic gathers are input to the first temporal convolutional network to train and obtain predicted porosity. The predicted porosity is input to the second temporal convolutional network to train and obtain synthetic seismic data. One-dimensional causal convolution and extended convolution are used as standard convolutional layers, and every two standard convolutional layers are encapsulated with an identity mapping into a residual module. The first temporal convolutional network and the second temporal convolutional network are formed by stacking multiple residual modules. The predicted porosity is compared with the corresponding labeled porosity parameter to obtain a first error value, and the synthetic seismic data is compared with the corresponding pre-stack seismic gather to obtain a second error value. Based on the first error value and the second error value, the closed-loop porosity prediction network is optimized to determine the functional mapping relationship between the pre-stack seismic data and the predicted porosity. Acquiring actual seismic data and predicting the target porosity of the actual seismic data based on the function mapping relationship and the closed-loop porosity prediction network includes: analyzing the formation absorption of the actual seismic data related to the shot-receiver distance; compensating the actual seismic data according to the formation absorption; using the compensated actual seismic data as the target seismic data for predicting the target porosity; inputting the target seismic data into the closed-loop porosity prediction network with the function mapping relationship determined to obtain the target porosity; and decomposing the formation absorption into the shot-receiver distance direction and the depth direction, wherein the equivalent absorption factor in the shot-receiver distance direction is: ； in, For mean square root and velocity, For the vertical two-way propagation travel of the i-th layer, The self-excitation and self-reception time of seismic waves in a layered medium. , Let be the absorption factor and velocity of the i-th layer, respectively.
2. The reservoir porosity prediction method according to claim 1, characterized in that, The step of optimizing the closed-loop porosity prediction network based on the first error value and the second error value to determine the functional mapping relationship between the pre-stack seismic data and the predicted porosity includes: Based on the first error value and the second error value, a loss value is obtained, and it is determined whether the loss value meets a preset condition. In response to the loss value not meeting the preset condition, the training parameters are adjusted, and the first temporal convolutional network and the second temporal convolutional network are retrained based on the pre-stack seismic gather, the labeled porosity parameter, and the adjusted training parameters to optimize the closed-loop porosity prediction network, and the step of determining whether the loss value meets the preset condition is returned. In response to the loss value satisfying the preset condition, the mapping relationship obtained by training the first temporal convolutional network is used as a functional mapping relationship between the pre-stack seismic data and the predicted porosity.
3. The reservoir porosity prediction method according to claim 2, characterized in that, The step of obtaining the loss value based on the first error value and the second error value includes: The loss value is calculated based on the loss function, the first error value, and the second error value. The loss function is expressed as follows: ； In the formula, This represents the label porosity parameter. This indicates the predicted porosity. This indicates the first error value. This refers to the pre-stack seismic gather. This refers to the synthetic seismic data. This represents the second error value. For hyperparameters, The loss value is given.
4. The reservoir porosity prediction method according to claim 1, characterized in that, The step of inputting the pre-stack seismic gathers into the first temporal convolutional network to train and obtain predicted porosity includes: The pre-stack seismic gathers are normalized, and the normalized pre-stack seismic gathers are divided into several one-dimensional vector data of a preset size. All the corresponding one-dimensional vector data are input into the first temporal convolutional network to train and obtain the predicted porosity.
5. The reservoir porosity prediction method according to claim 1, characterized in that, The step of obtaining the porosity parameters corresponding to the well point sample points as label porosity parameters includes: Obtain historical logging data for a preset number of well point sample points; The historical logging data are interpreted to obtain the porosity parameters of each well point sample point, and the porosity parameters are used as tag porosity parameters.
6. The reservoir porosity prediction method according to claim 2, characterized in that, The training parameters include hyperparameters and network structure parameters of the closed-loop porosity prediction network.
7. A reservoir porosity prediction system, characterized in that, include: The acquisition unit is used to acquire pre-stack seismic data of a preset number of well point sample points to obtain pre-stack seismic gathers, and to acquire the porosity parameters corresponding to each well point sample point as label porosity parameters. The pre-stack seismic data is acquired based on the historical seismic data of each well point sample point and forward modeling using the reflectivity method. The construction unit is used to construct a closed-loop porosity prediction network including a first temporal convolutional network and a second temporal convolutional network. The pre-stack seismic gathers are input to the first temporal convolutional network to train and obtain predicted porosity, and the predicted porosity is input to the second temporal convolutional network to train and obtain synthetic seismic data. The first temporal convolutional network and the second temporal convolutional network use one-dimensional causal convolution and extended convolution as standard convolutional layers, and encapsulate every two standard convolutional layers with an identity mapping into a residual module, which is formed by stacking multiple residual modules. An optimization unit is configured to compare the predicted porosity with the corresponding label porosity parameter to obtain a first error value, compare the synthetic seismic data with the corresponding pre-stack seismic gather to obtain a second error value, and optimize the closed-loop porosity prediction network based on the first error value and the second error value to determine the functional mapping relationship between the pre-stack seismic data and the predicted porosity. The prediction unit is used to collect actual seismic data and predict the target porosity of the actual seismic data based on the function mapping relationship and the closed-loop porosity prediction network. The acquisition unit is also used for: calculating the reflection and transmission coefficient matrices of different layers; calculating the displacement solution of the full wavefield in the frequency-ray parameter domain; calculating the seismic record in the time-space domain; the displacement solution in the frequency-ray parameter domain is expressed as: ； In the formula, The expression for the displacement below the detector is given. This is the portion that radiates downwards from the earthquake source. The reflection and transmission coefficient matrix represents the wave propagation response received by the receiver from below. The superscripts in the matrix denote different layers: F represents the free surface, S represents the source, R represents the receiver, and N represents the bottom interface. The subscripts in the matrix denote different traveling waves: U represents an upward-traveling wave, and D represents a downward-traveling wave. This is the reflection coefficient matrix that propagates downwards between the detector point and the bottom interface. This is the transmission coefficient matrix propagating downwards between the source and the receiver. The prediction unit is also used to: analyze the formation absorption related to the shot-receiver distance in the actual seismic data; compensate the actual seismic data according to the formation absorption; use the compensated actual seismic data as the target seismic data for predicting the target porosity; input the target seismic data into the closed-loop porosity prediction network that has determined the function mapping relationship to obtain the target porosity; and decompose the formation absorption into the shot-receiver distance direction and the depth direction, wherein the equivalent absorption factor in the shot-receiver distance direction is: ； in, For mean square root and velocity, For the vertical two-way propagation travel of the i-th layer, The self-excitation and self-reception time of seismic waves in a layered medium. , Let be the absorption factor and velocity of the i-th layer, respectively.
8. A computer device, comprising: At least one processor; as well as A memory storing a computer program executable on the processor, characterized in that the processor executes the program and performs the steps of the method as described in any one of claims 1 to 6.
9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it performs the steps of the method as described in any one of claims 1 to 6.