An intelligent prediction method and system for reservoir physical property parameters under small sample conditions

Abstract

The present invention relates to an intelligent prediction method and system for reservoir physical property parameters under small sample conditions, which are characterized by including: obtaining actual logging data and performing preprocessing; using a conditional generative adversarial network to perform data augmentation on the preprocessed logging data; synthesizing post-stack seismic data according to the data-augmented logging data; inputting the synthesized post-stack seismic data into a pre-constructed mapping relationship model based on a deep convolutional neural network to calculate the elastic physical property parameter profile of the synthesized post-stack seismic data, and obtaining the predicted value of the reservoir physical property parameters of the actual seismic data. The present invention can provide a feasible way for sample expansion in areas with few wells and can be widely applied in the field of reservoir parameter prediction.

Description

Technical Field

[0001] The present invention relates to the field of reservoir parameter prediction, and in particular to an intelligent prediction method and system for reservoir physical property parameters under small sample conditions. Background Art

[0002] Subtle hydrocarbon reservoirs represented by buried hill hydrocarbon reservoirs have complex structural causes, generally characterized by deep burial and high pressure, and are difficult to identify during the exploration process, being a special type of hydrocarbon reservoir. The reservoirs of subtle hydrocarbon reservoirs are complex and highly heterogeneous, resulting in difficulty for conventional inversion methods to obtain accurate reservoir parameters.

[0003] In recent years, with the development of industrial intelligence, intelligent inversion methods based on convolutional neural networks can be effectively applied to the prediction of reservoir parameters under complex geological conditions. After iterative training of the network, a complex mapping relationship between seismic responses and physical property parameters can be directly obtained, making the whole process completely data-driven and having broad application prospects.

[0004] However, due to the great difficulty and high cost of drilling and logging for subtle hydrocarbon reservoirs, problems such as insufficient logging sample quantity, single or missing logging data over a large range, and data distortion are very common. The generalization ability of the training network under conventional small sample conditions is weak, and the overfitting effect severely restricts the final prediction results. Summary of the Invention

[0005] Aiming at the above problems, the purpose of the present invention is to provide an intelligent prediction method and system for reservoir physical property parameters under small sample conditions, which can solve the problems of insufficient sample set quantity and difficulty in meeting training generalization.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: On the one hand, it provides an intelligent prediction method for reservoir physical property parameters under small sample conditions, including:

[0007] Obtain actual logging data and perform preprocessing;

[0008] Adopt a conditional generative adversarial network to perform data augmentation on the preprocessed logging data;

[0009] Synthesize post-stack seismic data according to the data-augmented logging data;

[0010] Input the synthesized post-stack seismic data into a pre-constructed mapping relationship model based on a deep convolutional neural network to calculate the elastic physical property parameter profile of the synthesized post-stack seismic data, and obtain the predicted value of the reservoir physical property parameters of the actual seismic data.

[0011] Further, the construction process of the mapping relationship model based on the deep convolutional neural network is as follows:

[0012] Obtain logging data and perform preprocessing;

[0013] Use a conditional generative adversarial network to perform data augmentation on the preprocessed logging data;

[0014] Synthesize post-stack seismic data based on the data-augmented logging data;

[0015] Take the synthesized post-stack seismic data as the training input, construct a mapping relationship model based on a deep convolutional neural network, and the output of the mapping relationship model based on the deep convolutional neural network is the elastic physical property parameter profile.

[0016] Furthermore, the preprocessing includes outlier removal and normalization.

[0017] Furthermore, the objective function of the conditional generative adversarial network is:

[0018]

[0019] where is the objective function under this framework; D is the discriminator model, and G is the generator model; is the mathematical expectation of the true logging data distribution; is the mathematical expectation of the hypothetical logging data distribution; P(d fake ) is defined as a simple distribution such as the standard normal distribution; D(d well ) is the discriminator function; G(d fake ) is the generator function; D(d well | reservoir ) and G(d fake | reservoir ) are the posterior probability distributions of the discriminator function D(d well ) and the generator function G(d fake ) under the label condition, respectively.

[0020] Furthermore, the synthesizing of the post-stack seismic data based on the data-augmented logging data includes:

[0021] Based on the Zoeppritz equation, solve the formation reflection coefficient according to the P-wave velocity, S-wave velocity, and density logging curves in the data-augmented logging data;

[0022] Perform forward modeling on the convolution of the formation reflection coefficient and the seismic wavelet to synthesize a seismic trace gather, and the synthesized post-stack seismic data is obtained after full stacking of the seismic trace gather;

[0023] Divide the synthesized post-stack seismic data into a training set and a test set, construct an inversion initial model with the data-augmented logging data as a constraint, and perform physical property parameter inversion to output the elastic physical property parameter profile of the synthesized post-stack seismic data.

[0024] Furthermore, the inversion matrix equation of the formation reflection coefficient is as follows:

[0025]

[0026] where the lower subscript of each term in the above formula is the angle of the common image point gather; nR pp is the P-wave reflection coefficient at the nth angle; is the first-order partial derivative of the P-wave reflection coefficient with respect to the P-wave velocity V p at the nth angle; nR pp, is the first-order partial derivative of the P-wave reflection coefficient with respect to the S-wave velocity V p at the nth angle; nR pp, is the first-order partial derivative of the P-wave reflection coefficient with respect to the density ρ at the nth angle; nR ps is the S-wave reflection coefficient at the nth angle; nR ps,p is the first-order partial derivative of the S-wave reflection coefficient with respect to the P-wave velocity V p at the nth angle; nR ps, is the first-order partial derivative of the S-wave reflection coefficient with respect to the S-wave velocity V s at the nth angle; nR ps, is the first-order partial derivative of the S-wave reflection coefficient with respect to the density ρ at the nth angle, and thus the formation reflection coefficient is solved.

[0027] Furthermore, using the synthetic post-stack seismic data as the training input to construct a mapping relationship model based on a deep convolutional neural network includes:

[0028] Using the synthetic post-stack seismic data as the training input and the inverted elastic physical property parameter profile as the training output, and using a deep convolutional neural network to iteratively train the encoding and decoding of the training input and training output to obtain a mapping relationship model between the synthetic post-stack seismic data and the elastic physical property parameter profile:

[0029] m k = -1 (-(F - I)m k-1 )

[0030] where m k is the model matrix for the kth iterative training; D is the observed data; F is the forward operator; F -1 is the inversion operator; I is the total number of inputs.

[0031] In a second aspect, an intelligent prediction system for reservoir physical property parameters under small sample conditions is provided, including:

[0032] A data acquisition module, configured to acquire actual logging data and perform preprocessing;

[0033] A data augmentation module, which is used to perform data augmentation on the preprocessed logging data by using a conditional generative adversarial network;

[0034] A data synthesis module, which is used to synthesize post-stack seismic data according to the augmented logging data;

[0035] A prediction module, which is used to input the synthesized post-stack seismic data into a pre-constructed mapping relationship model based on a deep convolutional neural network to calculate the elastic physical property parameter profile of the synthesized post-stack seismic data, and obtain the predicted value of the reservoir physical property parameters of the actual seismic data.

[0036] In a third aspect, a processing device is provided, including computer program instructions, wherein when the computer program instructions are executed by the processing device, they are used to implement the steps corresponding to the intelligent prediction method for reservoir physical property parameters under the above small-sample conditions.

[0037] In a fourth aspect, a computer-readable storage medium is provided, on which computer program instructions are stored, wherein when the computer program instructions are executed by a processor, they are used to implement the steps corresponding to the intelligent prediction method for reservoir physical property parameters under the above small-sample conditions.

[0038] Due to the above technical solutions adopted by the present invention, it has the following advantages:

[0039] 1. In the data augmentation process of the present invention, a conditional adversarial generative network is adopted, which can provide a feasible way for sample expansion in areas with few wells.

[0040] 2. The mapping relationship model based on a deep convolutional neural network constructed by the present invention can directly map the relationship between seismic and reservoir physical property parameters, skipping the conventional elastic parameter conversion process, and avoiding cumulative errors to a certain extent.

[0041] 3. The whole process of the present invention does not require much human intervention, and the process is efficient and objective.

[0042] In summary, the present invention can be widely applied to the field of reservoir parameter prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0044] Figure 1 It is a schematic diagram of the construction process of a mapping relationship model based on a deep convolutional neural network provided by an embodiment of the present invention;

[0045] Figure 2It is a schematic structural diagram of a conditional generative adversarial network provided by an embodiment of the present invention;

[0046] Figure 3 It is a schematic diagram of the result of data augmentation provided by an embodiment of the present invention;

[0047] Figure 4 It is a schematic structural diagram of a mapping relationship model based on a deep convolutional neural network provided by an embodiment of the present invention;

[0048] Figure 5 It is a schematic diagram of a training loss function provided by an embodiment of the present invention;

[0049] Figure 6 It is a schematic diagram of a prediction result provided by an embodiment of the present invention, where Figure 6 (a) is the actual seismic profile, Figure 6 (b) is the intelligent predicted physical property parameter (permeability) profile. Detailed implementation manners

[0050] Hereinafter, the exemplary embodiments of the present invention will be described in more detail with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present invention can be more thoroughly understood and the scope of the present invention can be completely conveyed to those skilled in the art.

[0051] It should be understood that the terms used herein are only for the purpose of describing specific exemplary embodiments and are not intended to be limiting. Unless otherwise clearly specified in the context, the singular forms "a", "an", and "the" as used herein may also include the plural forms. The terms "comprising", "including", "containing", and "having" are inclusive and thus specify the presence of the stated features, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not to be construed as necessarily requiring them to be performed in the particular order described or illustrated, unless the execution order is explicitly stated. It should also be understood that additional or alternative steps may be used.

[0052] The "small" in few-shot learning not only refers to a small number of samples, but essentially refers to the deviation between the data space constructed by a small number of sample sets and the actual data space. The biggest problem faced by few-shot learning is that the number of supervised samples is difficult to support the generalization modeling and training of the entire data space. Therefore, using auxiliary information to help the original sample set for data augmentation is a common means to solve the few-shot learning problem. Early data augmentation methods such as sample shifting, rotation, and flipping can quickly obtain a large number of new samples, but these augmentation rules do not have generality and show extremely poor adaptability when dealing with data with complex conditions. And using machine learning to extract sample data features and then generate new data is undoubtedly a new expansion. The quality of this method depends on the quality of the training samples. Therefore, the intelligent prediction method and system for reservoir physical property parameters under few-shot conditions provided by the embodiments of the present invention predict reservoir physical property parameters through a mapping relationship model based on a deep convolutional neural network. The mapping relationship model based on a deep convolutional neural network includes two parts: data augmentation and intelligent prediction. The data augmentation part mainly involves training a conditional generative adversarial network for data augmentation of logging data; the intelligent prediction part mainly involves training a network to synthesize the mapping relationship between seismic records and elastic physical property parameters to achieve intelligent prediction of reservoir parameters.

[0053] Embodiment 1

[0054] This embodiment provides an intelligent prediction method for reservoir physical property parameters under few-shot conditions, including the following steps:

[0055] 1) As Figure 1 shown, pre-construct a mapping relationship model based on a deep convolutional neural network to achieve data augmentation under few-shot conditions, specifically:

[0056] 1.1) Obtain logging data and perform preprocessing.

[0057] Specifically, the logging data includes several types such as compressional wave velocity, shear wave velocity, density logging curve, porosity, acoustic travel time, natural gamma, permeability, compensated neutron, density, resistivity, well inclination, water saturation, and shale content logging.

[0058] Specifically, the preprocessing includes removing outliers and standardization:

[0059]

[0060] where d well is the preprocessed logging data; d is the original logging data; d abnormal is the outlier in the original logging data.

[0061] 1.2) Use a conditional generative adversarial network (CGAN) to perform data augmentation on the preprocessed well logging data.

[0062] Specifically, the data synthesis method based on the conditional generative adversarial network is a representative method for machine learning data augmentation. The idea of realizing data augmentation is as follows: Train the GAN network through the original sample set, and reach a "balanced" state in the game between the generator and the discriminator. At this time, the new samples continuously generated by the generator can be recognized as the original samples by the discriminator to achieve the purpose of data augmentation. The objective function of the GAN network is:

[0063]

[0064] where, is the objective function under this framework, D is the discriminator model, and G is the generator model; is the mathematical expectation of the true well logging data distribution; is the mathematical expectation of the hypothetical (generated) well logging data distribution; P(d well ) is the distribution of the preprocessed data; P(d fake ) is defined as a simple distribution such as the standard normal distribution; D(d well ) is the discriminator function; G(d fake ) is the generator function, which is a binary minimax game model.

[0065] To generate data more characteristic of reservoir features, the present invention introduces the reservoir distribution information under the geological background of the target area as the label condition c reservoir . With the constraint of the label condition, the error tolerance rate of the generated data can be greatly improved, and the GAN thus becomes a supervised model, called the conditional generative adversarial network (CGAN). The objective function of the conditional generative adversarial network is:

[0066]

[0067] where, is the objective function to be solved. When the objective function meets the requirements (for example, iterates to the minimum value), the output at this time is the enhanced well logging data; D(d well | reservoir ) and G(d fake | reservoir ) are the posterior probability distributions of the discriminator function D(d well ) and the generator function G(d fake ) under the label condition, respectively.

[0068] 1.3) Synthesize post-stack seismic data based on the enhanced logging data, and calculate the elastic property parameter profile of the synthesized post-stack seismic data, specifically:

[0069] 1.3.1) Based on the Zoeppritz equation, the formation reflection coefficient is solved according to the P-wave velocity, S-wave velocity and density logging curve in the well logging data after data enhancement.

[0070] Specifically, based on the well logging data after data enhancement, the P-wave velocity, S-wave velocity and density curve are selected and substituted into the Zoeppritz equation. Since the S-wave velocity ratio has little effect on the reflection coefficient and can be ignored, this item is selectively ignored when constructing the equation:

[0071] R(ΔVp, ΔVs, Δρ)=0(4)

[0072] The first-order Taylor expansion of this formula is:

[0073]

[0074] Among them, R is the ratio of longitudinal and transverse wave velocities; Vp is the longitudinal wave velocity; Vs is the transverse wave velocity; ρ is the density.

[0075] After rearranging the above formula (5), we can obtain the inversion matrix equation for solving the formation reflection coefficient:

[0076]

[0077] The left subscripts of the terms in the above formula (6) are the angles of the common imaging point gathers; nR pp is the longitudinal wave reflection coefficient at the nth angle; nR pp,p is the longitudinal wave reflection coefficient at the nth angle to the longitudinal wave velocity V p The first-order partial derivative of nR pp, is the longitudinal wave reflection coefficient at the nth angle to the shear wave velocity V p The first-order partial derivative of nR pp, is the first-order partial derivative of the longitudinal wave reflection coefficient with respect to the density ρ at the nth angle; nR ps is the shear wave reflection coefficient at the nth angle; nRps,p is the shear wave reflection coefficient at the nth angle with respect to the longitudinal wave velocity V p The first-order partial derivative of nR ps, is the shear wave reflection coefficient at the nth angle to the shear wave velocity V s The first-order partial derivative of nR ps, is the first-order partial derivative of the shear wave reflection coefficient with respect to the density ρ at the nth angle. The formation reflection coefficient is solved from this, where the longitudinal wave reflection coefficient and the shear wave reflection coefficient are collectively referred to as the formation reflection coefficient.

[0078] 1.3.2) Forward model the convolution of the formation reflection coefficient and the seismic wavelet to synthesize a seismic trace gather. After fully stacking the seismic trace gather, the synthetic stacked seismic data is obtained. This process can be expressed as:

[0079]

[0080] where S is the synthetic stacked seismic data; N is the total number of seismic traces, n is the seismic trace number; W(t) is the seismic wavelet; R(t) is the formation reflection coefficient; t is the time; T is the number of time sampling points; W(τ) is the value of the seismic wavelet corresponding to the τ-th time sampling point; τ is the τ-th time sampling point.

[0081] 1.3.3) Divide the synthetic stacked seismic data into a training set and a test set. Using the well logging data after data augmentation as a constraint, construct an initial inversion model and perform physical property parameter inversion, and output elastic physical property parameter profiles such as porosity and permeability of the synthetic stacked seismic data.

[0082] 1.4) Use the synthetic stacked seismic data as the training input and the elastic physical property parameter profile as the training output to train the model, and construct a mapping relationship model based on a deep convolutional neural network.

[0083] Specifically, use the synthetic stacked seismic data as the training input and the elastic physical property parameter profiles such as porosity and permeability obtained by inversion as the training output, and use a deep convolutional neural network to perform iterative training on the encoding and decoding of the training input and the training output. Among them, one iteration training in the deep convolutional neural network is expressed as:

[0084]

[0085] where y j is the feature map output by the i-th layer; B is batch normalization; f is the activation function; I is the total number of inputs, i is the input of the i-th layer; W j is the convolution kernel of the j-th layer; x i is the input of the i-th layer; b j is the bias; J is the number of convolution kernels.

[0086] Therefore, the mapping relationship for realizing the intelligent prediction of reservoir physical property parameters can be regarded as an iteratively updated model problem:

[0087] D = Fm(9)

[0088] where D is the observed data; m is the model matrix; F is the forward operator.

[0089] After multiple iterations of training, the mapping relationship between the synthetic stacked seismic data and the elastic physical property parameter profile is:

[0090] m k= -1 (-(F - I)m k-1 )(10)

[0091] where m k is the model matrix for the k - th iterative training; F -1 is the inversion operator.

[0092] 2) Obtain the actual logging data and perform pre - processing, where the pre - processing includes removing outliers and standardization.

[0093] 3) Use a conditional generative adversarial network to perform data augmentation on the pre - processed logging data, and synthesize post - stack seismic data based on the augmented logging data.

[0094] 4) Input the synthesized post - stack seismic data into a pre - constructed mapping relationship model based on a deep convolutional neural network to calculate the elastic physical property parameter profile of the synthesized post - stack seismic data, and obtain the predicted values of the reservoir physical property parameters of the actual seismic data.

[0095] The following takes the logging curve data of a well in a work area in the South China Sea as a specific example of small - sample data augmentation to illustrate in detail the intelligent prediction method of reservoir physical property parameters under small - sample conditions of the present invention:

[0096] The logging data of this well includes several types such as porosity, acoustic travel - time, natural gamma, permeability, compensated neutron, density, resistivity, well deviation, water saturation, and shale content logging. In this embodiment, the water saturation curve is taken as an example. Since there are missing and distorted parts in the logging data, using this part for training will seriously affect the quality of the generated samples. Therefore, it is necessary to remove the outlier part in the water saturation logging data. At the same time, to accelerate the convergence speed of network training, the value range of the logging data is standardized to [-1, 1]. This step is the pre - processing before data augmentation.

[0097] Add the pre - processed water saturation logging data into the conditional generative adversarial network for training. As Figure 3 (a) shows, the logging data is the input data participating in the game training. In this embodiment, the data is cut into training patches with a pixel size of 32, and the patches are moved with a step size of 1 during training. Under the conditional constraint (as Figure 3 (b) shows), logging curve samples that conform to the reservoir labels under the geological background of this work area are generated through the game training of the discriminator and the generator, realizing low - cost and high - quality data augmentation. As Figure 2 shows, it is the network structure of the generator and the discriminator in the conditional generative adversarial network. The generator includes 5 convolutional layers and 5 de - convolutional layers, and the output is excited by the tanh function; the discriminator includes 4 convolutional layers, and finally the output is excited by the ReLU function, where the activation function can be expressed as:

[0098] ReLU(x) = max(0, x) (11)

[0099]

[0100] Wherein, x is the input variable; ReLU(x) is the ReLU activation expression of x; tanh() is the tanh activation expression of x.

[0101] In the game training, the generator model G is first trained, and the water saturation log data d well and the label condition c of the reservoir information in this area reservoir are input. The generator model G generates a water saturation log curve, and then the discriminator model D is trained to learn the label distribution between the water saturation log data d well and the label condition c reservoir among the well curves, and tries to distinguish the connection between the real log curve and the generated log curve. In this embodiment, the objective function of the binary min-max game is adopted.

[0102] The result of log data enhancement is as Figure 3 (c) shown. On the basis of not deviating from the geological background, log data that is more abundant and diverse in information is generated. Using these enhanced log data as the inversion constraint term can better constrain the subsequent prediction results. Thus, the data enhancement work under the small sample condition is completed.

[0103] Based on the enhanced log data, the longitudinal wave velocity, transverse wave velocity, and density curves are selected from them, and these three are substituted into the Zoeppritz equation to solve the formation reflection coefficient. Based on the calculated formation reflection coefficient, in this embodiment, a Ricker wavelet with a frequency of 35 Hz is selected for convolution to synthesize a seismic trace gather. After the full stack of the seismic trace gather, the synthetic post-stack seismic data is obtained. The synthetic post-stack seismic data is divided into a training set and a test set, with a ratio of 5:1. Using the enhanced log data as a constraint, an initial inversion model is constructed, and physical property parameter inversion is carried out to output elastic physical property parameter profiles such as porosity and permeability of the synthetic post-stack seismic data.

[0104] Taking the synthetic post-stack seismic data as the training input and the elastic physical property parameter profile as the training output, a deep convolutional neural network is used to implement iterative training of the training input and the training output. In this embodiment, a seven-layer convolutional neural network is designed, as Figure 4As shown in the figure, it includes three encoding layers, three decoding layers and one intermediate layer. Each layer includes 32 filters with a convolutional kernel of 3×3. The ReLU activation function is used in the first six layers to increase the non-linearity of the mapping, and the tanh activation function is used in the last layer to simulate the diversity of the prediction results. At the same time, to accelerate the convergence speed of the network, max-pooling layers, upsampling layers and batch normalization layers are set in the encoding layer and the three decoding layers.

[0105] Finally, the stacked seismic data and the inverted elastic physical property parameter profiles are cut into small blocks with a pixel size of 32×32. The horizontal and vertical moving step sizes of the small blocks are 4. Finally, 16104 training input small blocks are obtained. The number of training output small blocks is the same as that of the input. In this embodiment, the mean square error is selected as the loss function for training:

[0106]

[0107] where N is the number of samples. At the same time, 100 epochs are set for iterative training. As Figure 5 shown in the figure is the schematic diagram of the loss function in this embodiment.

[0108] The stable convergence of the loss function indicates that the network training is effective, and an accurate mapping relationship between the input seismic response and the output attribute parameters can be obtained.

[0109] Finally, the test set in the synthetic stacked seismic data is input into the mapping relationship model obtained above. After model prediction, the corresponding attribute parameter (such as permeability) profile is output. As Figure 6 shown in the figure, combined with the actual well logging and geological interpretation results, it is known that there are three gas layers developed in this work area as marked in the figure. According to the predicted permeability profile, it can be inferred that the connectivity between Well w2 and Well w3 in Gas Layer 1 and Gas Layer 2 is better, and the connectivity of Gas Layer 3 is worse, which is in good agreement with the actual reservoir development situation.

[0110] Embodiment 2

[0111] This embodiment provides an intelligent prediction system for reservoir physical property parameters under small sample conditions, including:

[0112] A data acquisition module for acquiring actual well logging data and performing preprocessing;

[0113] A data enhancement module for enhancing the preprocessed well logging data by using a conditional generative adversarial network;

[0114] A data synthesis module for synthesizing stacked seismic data according to the enhanced well logging data;

[0115] A prediction module, configured to input the synthetic post-stack seismic data into a pre-constructed mapping relationship model based on a deep convolutional neural network to calculate the elastic physical property parameter profile of the synthetic post-stack seismic data, and obtain the predicted values of the reservoir physical property parameters of the actual seismic data.

[0116] The system provided in this embodiment is used to execute the above method embodiments. For the specific process and detailed content, please refer to the above embodiments and will not be elaborated here.

[0117] Embodiment 3

[0118] This embodiment provides a processing device corresponding to the intelligent prediction method for reservoir physical property parameters under small sample conditions provided in Embodiment 1. The processing device can be a processing device applicable to a client, such as a mobile phone, a laptop computer, a tablet computer, a desktop computer, etc., to execute the method of Embodiment 1.

[0119] The processing device includes a processor, a memory, a communication interface, and a bus. The processor, the memory, and the communication interface are connected through the bus to complete communication with each other. The memory stores a computer program that can run on the processing device. When the processing device runs the computer program, it executes the intelligent prediction method for reservoir physical property parameters under small sample conditions provided in Embodiment 1 of this invention.

[0120] In some implementations, the memory can be a high-speed random access memory (RAM: Random Access Memory), and may also include a non-volatile memory, such as at least one disk memory.

[0121] In other implementations, the processor can be a central processing unit (CPU), a digital signal processor (DSP), or various other types of general-purpose processors, which are not limited here.

[0122] In addition, when the logical instructions in the above-mentioned memory are implemented in the form of software functional units and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical disks, etc., which can store program codes.

[0123] Those skilled in the art can understand that the structure of the above computing device is only a part of the structure related to the solution of this application, and does not constitute a limitation on the computing device to which the solution of this application is applied. The specific computing device may include more or fewer components, or combine some components, or have different component arrangements.

[0124] Embodiment 4

[0125] This embodiment provides a computer program product corresponding to the intelligent prediction method for reservoir physical property parameters under small sample conditions provided in Embodiment 1 of this application. The computer program product may include a computer-readable storage medium, on which computer-readable program instructions for executing the intelligent prediction method for reservoir physical property parameters under small sample conditions described in Embodiment 1 of this application are uploaded.

[0126] The computer-readable storage medium may be a tangible device that holds and stores instructions used by an instruction execution device. The computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination of the above.

[0127] For the computer-readable storage medium provided in the above embodiment, its implementation principle and technical effects are similar to those of the above method embodiment, and will not be elaborated here.

[0128] This application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of this application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing in the process Figure 1 one process or multiple processes and / or blocks Figure 1 a device for the functions specified in one block or multiple blocks.

[0129] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements in the process Figure 1 one process or multiple processes and / or blocks Figure 1 a device for the functions specified in one block or multiple blocks.

[0130] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable apparatus to generate a computer-implemented process, thereby providing instructions for implementing the process Figure 1 one process or a plurality of processes and / or blocks Figure 1 steps for the functions specified in one block or a plurality of blocks.

[0131] The above embodiments are only used to illustrate the present invention. The structures, connection manners, manufacturing processes, etc. of the components can all be changed. Any equivalent transformation and improvement made on the basis of the technical solution of the present invention should not be excluded from the protection scope of the present invention.

Claims

1. An intelligent prediction method for reservoir physical property parameters under small sample conditions, characterized in that Including: Obtain actual logging data and perform preprocessing; Adopt a conditional generative adversarial network to perform data augmentation on the preprocessed logging data; Synthesize post-stack seismic data based on the data-augmented logging data; Input the synthesized post-stack seismic data into a pre-constructed mapping relationship model based on a deep convolutional neural network to calculate the elastic physical property parameter profile of the synthesized post-stack seismic data, and obtain the predicted reservoir physical property parameter values of the actual seismic data; The construction process of the mapping relationship model based on the deep convolutional neural network is as follows: Use the synthesized post-stack seismic data as the training input to construct a mapping relationship model based on the deep convolutional neural network, and the output of the mapping relationship model based on the deep convolutional neural network is the elastic physical property parameter profile; The objective function of the conditional generative adversarial network is as follows: Among them, is the objective function; is the discriminator model, is the generator model; is the mathematical expectation of the true logging data distribution; is the mathematical expectation of the hypothetical logging data distribution; is defined as a simple distribution such as the standard normal distribution; is the discriminator function; is the generator function; and are the discriminator functions and generator functions under the label conditions, respectively, The synthesizing of the post-stack seismic data according to the data-augmented logging data includes: Based on the Zoeppritz equation, solve the formation reflection coefficient according to the P-wave velocity, S-wave velocity, and density logging curves in the data-augmented logging data; Perform forward modeling on the convolution of the formation reflection coefficient and the seismic wavelet to synthesize a seismic trace gather, and obtain the synthesized post-stack seismic data after full stacking of the seismic trace gather; Divide the synthesized post-stack seismic data into a training set and a test set, construct an inversion initial model with the data-augmented logging data as the constraint, perform physical property parameter inversion, and output the elastic physical property parameter profile of the synthesized post-stack seismic data; The using of the synthesized post-stack seismic data as the training input to construct a mapping relationship model based on the deep convolutional neural network includes: Use the synthesized post-stack seismic data as the training input, use the inversely obtained elastic physical property parameter profile as the training output, and adopt a deep convolutional neural network to perform iterative training on the encoding and decoding of the training input and the training output to obtain a mapping relationship model between the synthesized post-stack seismic data and the elastic physical property parameter profile; Among them, is the model matrix for the th iterative training; is the observed data; is the forward operator; is the inverse operator; is the total number of inputs.

2. The intelligent prediction method for reservoir physical property parameters under small sample conditions according to claim 1, characterized in that, The preprocessing includes removing outliers and normalizing.

3. The intelligent prediction method for reservoir physical property parameters under small sample conditions according to claim 1, characterized in that, The inversion matrix equation of the formation reflection coefficient is: Among them, the lower subscript of each term in the above equation is the angle of the common imaging point gather; is the P-wave reflection coefficient at the th angle; is the first-order partial derivative of the P-wave reflection coefficient with respect to the P-wave velocity at the th angle; is the first-order partial derivative of the P-wave reflection coefficient with respect to the S-wave velocity at the th angle; is the first-order partial derivative of the P-wave reflection coefficient with respect to the density at the th angle; is the S-wave reflection coefficient at the th angle; is the first-order partial derivative of the S-wave reflection coefficient with respect to the P-wave velocity at the th angle; is the first-order partial derivative of the S-wave reflection coefficient with respect to the S-wave velocity at the th angle; is the first-order partial derivative of the S-wave reflection coefficient with respect to the density at the th angle. Thus, the formation reflection coefficient is solved.

4. An intelligent prediction system for reservoir physical property parameters under small sample conditions, characterized in that, Including: A data acquisition module for obtaining actual logging data and performing preprocessing; A data augmentation module for performing data augmentation on the preprocessed logging data by using a conditional generative adversarial network; A data synthesis module for synthesizing post-stack seismic data according to the data-augmented logging data; A prediction module for inputting the synthesized post-stack seismic data into a pre-constructed mapping relationship model based on the deep convolutional neural network to calculate the elastic physical property parameter profile of the synthesized post-stack seismic data, and obtaining the predicted reservoir physical property parameter values of the actual seismic data; The construction process of the mapping relationship model based on the deep convolutional neural network is as follows: Use the synthesized post-stack seismic data as the training input to construct a mapping relationship model based on the deep convolutional neural network, and the output of the mapping relationship model based on the deep convolutional neural network is the elastic physical property parameter profile; The objective function of the conditional generative adversarial network is as follows: Among them, is the objective function; is the discriminator model, is the generator model; is the mathematical expectation of the true logging data distribution; is the mathematical expectation of the imaginary logging data distribution; is defined as a simple distribution such as the standard normal distribution; is the discriminator function; is the generator function; and are the discriminator functions and the generator functions under the label conditions, respectively, of the posterior probability distribution; The synthesizing of the post-stack seismic data according to the data-augmented logging data includes: Based on the Zoeppritz equation, solve the formation reflection coefficient according to the P-wave velocity, S-wave velocity, and density logging curves in the data-augmented logging data; Perform forward modeling on the convolution of the formation reflection coefficient and the seismic wavelet to synthesize a seismic trace gather, and obtain the synthesized post-stack seismic data after full stacking of the seismic trace gather; Divide the synthetic post-stack seismic data into a training set and a test set. Using the well logging data after data augmentation as a constraint, construct an initial inversion model and perform physical property parameter inversion to output the elastic physical property parameter profile of the synthetic post-stack seismic data; Use the synthetic post-stack seismic data as the training input to construct a mapping relationship model based on a deep convolutional neural network, including: Use the synthetic post-stack seismic data as the training input and the elastic physical property parameter profile obtained by inversion as the training output. Use a deep convolutional neural network to iteratively train the encoding and decoding of the training input and training output to obtain a mapping relationship model between the synthetic post-stack seismic data and the elastic physical property parameter profile: Among them, is the model matrix for the th iterative training; is the observed data; is the forward operator; is the inverse operator; is the total number of inputs.

5. A processing device, characterized in that, It includes computer program instructions. Among them, when the computer program instructions are executed by a processing device, they are used to implement the steps corresponding to the intelligent prediction method of reservoir physical property parameters under small sample conditions described in any one of claims 1-3.

6. A computer-readable storage medium, characterized in that, Computer program instructions are stored on the computer-readable storage medium. Among them, when the computer program instructions are executed by a processor, they are used to implement the steps corresponding to the intelligent prediction method of reservoir physical property parameters under small sample conditions described in any one of claims 1-3.

Images