Training method and device for oil and gas reservoir quality classification model

Abstract

The invention discloses a training method and device of an oil and gas reservoir quality classification model, which can be applied to the technical field of oil and gas engineering and can train the oil and gas reservoir quality classification model from four aspects of reservoir rock types, physical properties, longitudinal distribution and fluid types. Therefore, the oil and gas reservoir quality classification model can evaluate reservoir classification more comprehensively, reservoir quality classification can be carried out more comprehensively and accurately through comprehensive utilization of the deep learning technology and geological and geophysical knowledge, scientific basis and technical support are provided for oil and gas exploration and development, and an oil and gas reservoir development scheme is optimized.

Description

Technical Field

[0001] The present application relates to the technical field of oil and gas engineering, and particularly to a method and device for training an oil and gas reservoir quality classification model. Background Art

[0002] Reservoir quality classification refers to the process of comprehensively analyzing the physical properties of a reservoir (such as porosity, permeability, oil saturation, etc.) and dividing the reservoir into different quality grades. Reservoir quality classification is a key link in oil and gas exploration and development, which directly affects the development strategy, well pattern design and economic benefits of oil and gas fields.

[0003] Currently, there is a lack of an effective method for classifying reservoir quality. Therefore, how to classify reservoir quality is a technical problem that needs to be urgently solved by those skilled in the art. Summary of the Invention

[0004] The present application provides a method for training an oil and gas reservoir quality classification model to classify reservoir quality, and the present application also provides a device for training an oil and gas reservoir quality classification model.

[0005] In a first aspect, the present application provides a method for training an oil and gas reservoir quality classification model, including: Classifying the reservoir quality of the target interval according to the reservoir physical property parameters of the target interval and encoding the obtained classification labels to obtain label data of reservoir categories, where the reservoir physical property parameters include: acoustic impedance, lithology type, sedimentary facies type, porosity, permeability, oil saturation, vertical reservoir thickness, and fluid type; Obtaining seismic interval data of the target interval and extracting seismic trace data beside known wells from the seismic interval data; Statistically analyzing the value range distribution of reservoir physical property parameters of different categories of known wells, generating a reservoir geological model according to the value range distribution of the reservoir physical property parameters, and forward modeling to generate corresponding seismic data to obtain a reservoir template data set; Training a pre-constructed oil and gas reservoir quality classification model according to the label data, the seismic trace data, and the reservoir template data set.

[0006] Optionally, the model includes A convolutional blocks, B ResNet blocks, a fully connected layer, and a softmax layer, and the method further includes: Obtaining non-adjacent well seismic data; Inputting the non-adjacent well seismic data into the trained oil and gas reservoir quality classification model for prediction.

[0007] Optionally, classifying the reservoir quality of the target interval according to the reservoir physical property parameters of the target interval and encoding the obtained classification labels to obtain the label data of the reservoir category, including: Classifying the reservoir quality of the target interval by using the reservoir physical property parameters and the reservoir quality classification standard of the target interval to obtain the reservoir quality classification label; Performing one-hot encoding on the reservoir quality classification label to obtain the label data of the reservoir category, where the data dimension of the label data is (N, M), N is the number of wells, and M is the reservoir category.

[0008] Optionally, obtaining the seismic interval data of the target interval and extracting the seismic trace data beside the known wells from the seismic interval data, including: Using the top and bottom seismic horizons of the target interval as constraints to generate the seismic interval data S1 of the target interval; Performing feature extraction on S1 to obtain the seismic feature data S2 of the target interval; Extracting the seismic trace data S3 beside the known wells from S2 according to the coordinates of the known well points.

[0009] Optionally, the value range distribution of the reservoir physical property parameters includes: the value range distribution of wave impedance, lithology type, sedimentary facies type, porosity, permeability, oil saturation, vertical reservoir thickness, and fluid type. Generating a reservoir geological model according to the value range distribution of the reservoir physical property parameters and forward modeling to generate corresponding seismic data to obtain the reservoir template data, including: Determining the normal distribution function of each reservoir physical property parameter according to the mean and variance of the value range distribution of the reservoir physical property parameters; Generating multiple groups of reservoir physical property parameters according to the normal distribution function of each reservoir physical property parameter to obtain the reservoir template data.

[0010] Optionally, the method further includes: Inputting the wave impedance parameter of the reservoir template into the convolution model to obtain the seismic waveforms corresponding to multiple groups of reservoir templates; Training the pre-constructed oil and gas reservoir quality classification model according to the label data, the seismic trace data, and the reservoir template data set, including: Training the pre-constructed oil and gas reservoir quality classification model according to the label data, the seismic trace data, the reservoir template data set, and the seismic waveforms In a second aspect, the present application further provides a training device for an oil and gas reservoir quality classification model, and the device includes: An encoding unit, configured to classify the reservoir quality of the target interval according to the reservoir physical property parameters of the target interval and encode the obtained classification labels to obtain label data of reservoir categories, where the reservoir physical property parameters include: acoustic impedance, lithology type, sedimentary facies type, porosity, permeability, oil saturation, vertical reservoir thickness, and fluid type; An extraction unit, configured to obtain seismic interval data of a target interval and extract seismic trace data near known wells from the seismic interval data; A generation unit, configured to statistically analyze the value range distributions of reservoir physical property parameters of different categories of known wells, generate a reservoir geological model according to the value range distributions of the reservoir physical property parameters, and forward model to generate corresponding seismic data to obtain a reservoir template data set; A training unit, configured to train a pre-constructed oil and gas reservoir quality classification model according to the label data, the seismic trace data, and the reservoir template data set.

[0011] Optionally, the oil and gas reservoir quality classification model includes A convolutional blocks, B ResNet blocks, a fully connected layer, and a softmax layer, and the apparatus further includes: An acquisition unit, configured to acquire non-adjacent well seismic data; A prediction unit, configured to input the non-adjacent well seismic data into the trained oil and gas reservoir quality classification model for prediction.

[0012] Optionally, the encoding unit is specifically configured to: Classify the reservoir quality of the target interval by using the reservoir physical property parameters and the reservoir quality classification standard of the target interval to obtain reservoir quality classification labels; Perform one-hot encoding on the reservoir quality classification labels to obtain label data of reservoir categories, where the data dimension of the label data is (N, M), N is the number of wells, and M is the number of reservoir categories.

[0013] Optionally, the extraction unit is specifically configured to: Use the top and bottom seismic horizons of the target interval as constraints to generate seismic interval data S1 of the target interval; Extract features from S1 to obtain seismic feature data S2 of the target interval; Extract seismic trace data S3 near known wells from S2 according to the coordinates of known well points Optionally, the value range distributions of reservoir physical property parameters include: value range distributions of acoustic impedance, lithology type, sedimentary facies type, porosity, permeability, oil saturation, vertical reservoir thickness, and fluid type, and the generation unit is specifically configured to: Determine the normal distribution function of each reservoir physical property parameter according to the mean and variance of the value range distributions of the reservoir physical property parameters; According to the normal distribution function of each reservoir physical property parameter, multiple groups of reservoir physical property parameters are generated to obtain reservoir template data.

[0014] Optionally, the device further includes: An input unit, configured to input the wave impedance parameter of the reservoir template into the convolution model to obtain seismic waveforms corresponding to multiple groups of reservoir templates; A training unit, specifically configured to: Train a pre-constructed oil and gas reservoir quality classification model according to the label data, the seismic trace data, the reservoir template data set, and the seismic waveforms.

[0015] In a third aspect, an embodiment of the present application provides a device, which includes a memory and a processor. The memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the method described in the foregoing first aspect.

[0016] In a fourth aspect, an embodiment of the present application provides a computer storage medium, in which codes are stored. When the codes are run, the device running the codes implements the method described in the foregoing first aspect. Description of the Drawings

[0017] To more clearly illustrate the technical solutions in the embodiments or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a flowchart of a method for training an oil and gas reservoir quality classification model provided by an embodiment of the present application; Figure 2 It is a reservoir plane distribution map provided by an embodiment of the present application; Figure 3 It is a schematic structural diagram of a specific implementation manner of a training device for an oil and gas reservoir quality classification model provided by the present application. Detailed Embodiments

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the protection scope of the present application.

[0020] It should be noted that a training method and device for an oil and gas reservoir quality classification model provided in this application are used in the application scenario of reservoir quality classification prediction. The above is only an example and does not limit the application field of the method and device names provided in this application.

[0021] Reservoir quality classification refers to the process of comprehensively analyzing the physical properties of a reservoir (such as porosity, permeability, oil saturation, etc.) and dividing the reservoir into different quality grades. Reservoir quality classification is a key link in oil and gas exploration and development, which directly affects the development strategy, well pattern design and economic benefits of oil and gas fields.

[0022] Currently, there is a lack of an effective method for classifying reservoir quality. Therefore, how to classify reservoir quality is a technical problem that needs to be urgently solved by those skilled in the art.

[0023] In view of this, this application proposes a training method for an oil and gas reservoir quality classification model, including: classifying the reservoir quality of the target interval according to the reservoir physical property parameters of the target interval and encoding the obtained classification labels to obtain label data of reservoir categories, where the reservoir physical property parameters include: wave impedance, lithology type, sedimentary facies type, porosity, permeability, oil saturation, vertical reservoir thickness, and fluid type; obtaining the seismic interval data of the target interval, extracting the seismic trace data near known wells from the seismic interval data, statistically analyzing the value range distribution of reservoir physical property parameters of different categories of known wells, generating a reservoir geological model according to the value range distribution of reservoir physical property parameters and forward modeling to generate corresponding seismic data, obtaining a reservoir template data set, and training a pre-constructed oil and gas reservoir quality classification model according to the label data, seismic trace data, and reservoir template data set. In this application, the oil and gas reservoir quality classification model can be trained from four aspects: reservoir rock type, physical properties, vertical distribution, and fluid type, so that the oil and gas reservoir quality classification model can more comprehensively evaluate the classification of reservoirs. By comprehensively using deep learning technology and geological and geophysical knowledge, reservoir quality classification can be carried out more comprehensively and accurately, providing a scientific basis and technical support for oil and gas exploration and development, and optimizing the oil and gas reservoir development plan.

[0024] The method provided in the embodiments of this application can be executed by software on a computing device. The computing device can be, for example, a mobile phone, a tablet computer, a computer, or other devices.

[0025] In order to enable those skilled in the art to better understand the solution of this application, the following further detailed description of this application will be given in conjunction with the drawings and specific implementation manners. The following takes the method provided in the embodiments of this application being executed by a computing device as an example for illustration.

[0026] Figure 1The flowchart of a method for training an oil and gas reservoir quality classification model provided by an embodiment of this application. This embodiment can be referred to as Embodiment 1. As Figure 1 shown, the method includes: S101: Classify the reservoir quality of the target interval according to the reservoir physical property parameters of the target interval, and encode the obtained classification labels to obtain the label data of the reservoir categories.

[0027] The computing device can obtain the reservoir physical property parameters of the target interval by comprehensively interpreting the known well curves (the reservoir physical property parameters are the parameters describing the petrophysical properties of the reservoir rocks, and these parameters directly reflect the quality of the reservoir and the oil and gas storage capacity). Then, it can classify the reservoir quality of the target interval according to the obtained reservoir physical property parameters and encode the obtained classification labels to obtain the label data of the reservoir categories. In this embodiment, the number of known wells can be 148.

[0028] In some possible implementation manners, the reservoir physical property parameters may include at least multiple of: acoustic impedance (acoustic impedance is the product of the density of the rock and the seismic wave velocity when the seismic wave propagates in the rock medium), lithology type (lithology type refers to the type of rock, such as sandstone, mudstone, carbonate rock, etc.), sedimentary facies type (sedimentary facies type refers to the type of sedimentary environment, such as delta facies, fluvial facies, lacustrine facies, marine facies, etc.), porosity (porosity is the percentage of the volume of pores in the rock in the total volume of the rock, reflecting the storage space of fluids in the rock), permeability (permeability is the ability of fluids to flow in the rock, reflecting the conductivity of fluids in the rock), oil saturation (oil saturation is the percentage of the volume of oil in the pores in the rock, reflecting the oil content in the rock), vertical reservoir thickness (vertical reservoir thickness refers to the thickness of the reservoir in the vertical direction, reflecting the scale of the reservoir), and fluid type (fluid type refers to the type of fluid in the reservoir, such as oil, gas, water). In the embodiment of this application, the seismic data includes all of the above as an example for introduction.

[0029] The reservoir quality of the target interval of each well can be classified by using reservoir physical property parameters and the reservoir quality classification criteria of the target interval (here, the reservoir quality classification is taken as an example of three categories: type I reservoir, type II reservoir, and type III reservoir), and a reservoir quality classification label is obtained. Among them, the reservoir quality classification label is N, and the one-hot encoding is performed on the reservoir quality classification label to obtain the label data of the reservoir category. The data dimension of the label data is (N, M), where N is the number of wells and M is the reservoir category. That is, the reservoir category labels at the well points of the target interval are processed by the one-hot encoding method for the three types of reservoirs, and are converted from the string category (type I reservoir, type II reservoir, type III reservoir) to the integer format: {"type I reservoir": [1, 0, 0], "type II reservoir": [0, 1, 0], "type II reservoir": [0, 0, 1]}, and the output format is (N, 3).

[0030] S102: Obtain the seismic interval data of the target interval, and extract the seismic trace data beside the known wells from the seismic interval data.

[0031] The computing device can obtain the seismic interval data of the target interval, and then extract the seismic trace data beside the known wells from the seismic interval data.

[0032] In some possible implementations, the computing device can use the top and bottom seismic horizons of the target interval as constraints to generate the seismic interval data S1 of the target interval, extract the seismic feature data S2 of the target interval by using the wavelet transform method, and extract the seismic trace data S3 beside the known wells from S2 according to the coordinates of the known well points. Among them, the data dimension of S1 can be (T, H, W), the data dimension of S2 is (F, T, H, W), and the data dimension of S3 is (F, T, M).

[0033] Exemplarily, the data dimension of S1 can be (40, 400, 600), the data dimension of S2 can be (102, 40, 400, 600), and the data dimension of S3 can be (120, 40, 148).

[0034] Exemplarily, the formula (1) can be used to process each single seismic trace of the data S1. The processing of each seismic trace is completed in a loop to obtain the seismic data features of the target interval. The output data S2, that is, the CWT(a, b) format is (F, T, H, W), and is named Seismic_F.

[0035] (1) Wherein: b: Used to locate time. a: Used to locate frequency. f(t): Represents the original seismic waveform. It represents a wavelet basis. In the wavelet basis, the variable a represents the scale, corresponding to the frequency. The translation amount b represents the time.

[0036] S103: Statistically analyze the value range distribution of physical property parameters of different types of reservoirs in known wells, generate a reservoir geological model based on the value range distribution of the reservoir physical property parameters, and forward model to generate corresponding seismic data to obtain a reservoir template data set.

[0037] The computing device can statistically analyze the value range distribution of physical property parameters of different types of reservoirs in known wells, generate a reservoir geological model based on the value range distribution of the physical property parameters, and forward model to generate corresponding seismic data to obtain a reservoir template data set.

[0038] Exemplarily, based on 148 known wells, the value range distributions of 8 reservoir physical property parameters, namely wave impedance, lithology type, sedimentary facies type, porosity, permeability, oil saturation, vertical reservoir thickness, and fluid type, of 3 types of reservoirs can be statistically analyzed. Then, based on the statistical results of the 8 parameters of the three types of reservoirs, 8 reservoir physical property parameter values of the three types of reservoirs can be randomly generated, different types of reservoir geological models can be made, and corresponding seismic data can be forward modeled. Finally, a reservoir template data set is generated.

[0039] For one type of reservoir, the average wave impedance can be 6000 - 7000 (kg / m^3·m / s); the lithology type can mainly be medium - fine grained sandstone with relatively developed pores; the sedimentary facies type can mainly be distributary channels, mouth bars, etc.; the average porosity can be 25 - 30%; the average permeability can be 1000 - 3000 mD; the average oil saturation can be 60 - 80% with a small variance; the average vertical reservoir thickness can be 15 - 30 meters; the fluid type can mainly be light oil.

[0040] For the second type of reservoir, the average wave impedance can be 5000 - 6000 (kg / m^3·m / s); the lithology type can mainly be fine - grained sandstone and siltstone; the sedimentary facies type can mainly be inter - distributary bays, frontal sand bars, etc.; the average porosity can be 15 - 25%; the average permeability can be 100 - 1000 mD; the average oil saturation can be 40 - 60%; the average vertical reservoir thickness can be 10 - 20 meters; the fluid type can mainly be medium - quality oil.

[0041] For the third type of reservoir, the average wave impedance can be 4000 - 5000 (kg / m^3·m / s); the lithology type can mainly be siltstone and argillaceous sandstone; the sedimentary facies type can mainly be flood plains, lake facies, etc.; the porosity: the average can be 5 - 15%; the average permeability can be 10 - 100 mD; the average oil saturation can be 20 - 40%; the average vertical reservoir thickness can be 5 - 10 meters; the fluid type can mainly be heavy oil or aquifer.

[0042] Specifically, the normal distribution function of each reservoir physical property parameter can be determined according to the mean and variance of the value range distribution of the physical property parameters. Then, based on the normal distribution function of each reservoir physical property parameter, multiple groups (taking 10,000 groups as an example here) of reservoir physical property parameters are generated to obtain reservoir template data.

[0043] The determination of the normal distribution function of each reservoir physical property parameter according to the mean and variance of the reservoir physical property parameter can be specifically shown in Formula (2): (2) Wherein, : represents the mean of different reservoir properties. : represents the variance of different reservoir properties. x: represents different reservoir property parameters, The normalization constant ensures that the integral of the entire distribution (i.e., the total probability) is 1, , the exponential term, determines the shape of the distribution. It represents the deviation degree of x from the mean , and is scaled by the standard deviation .

[0044] By making a reservoir geological model and generating a reservoir template data set, additional data with prior knowledge is provided for model training. These data can help the model learn knowledge in aspects such as seismic forward modeling and reservoir division criteria, and enhance the model's understanding and generalization ability of reservoir quality classification.

[0045] In some possible implementation manners, the computing device can also use the acoustic impedance parameters of the reservoir template, and utilize the convolution model, namely Formula (3), to obtain multiple groups (taking 10,000 groups as an example in this embodiment) of seismic waveforms S4 corresponding to the reservoir templates.

[0046] (3) Wherein: : The reservoir template generates a seismic waveform. : The Ricker wavelet. : The reflection coefficient is obtained by the acoustic impedance difference of the reservoir template parameters.

[0047] S104: Train a pre-constructed oil and gas reservoir quality classification model according to the label data, the seismic trace data, and the reservoir template data set.

[0048] First, an introduction can be made to the structure of the pre - constructed oil and gas reservoir quality classification model, which may include 2 convolutional blocks and 3 ResNet blocks. The first convolutional block and the second convolutional block of the reservoir quality classification model can be respectively composed of a convolutional layer with a convolution kernel of 1*3, a BatchNormal layer, and a ReLU activation layer. The 3 ResNet blocks of the reservoir quality classification model can be respectively composed of 2 convolutional layers with a convolution kernel of 1*1, a convolutional layer with a convolution kernel of 1*3, a ReLU activation function, and a skip connection structure, as shown in formula (4): (4) Where: : Standard ReLU activation function. BN: Represents the normalization layer BatchNormal.

[0049] : Represents the learnable parameters of the three convolutional layers. x: Represents the input seismic data features. y: Represents the output seismic data features.

[0050] The last layer of the reservoir quality classification model uses a fully - connected layer, and the activation function finally used is softmax, as shown in formula (5). The cross - entropy loss function can be used as the optimization objective function of the reservoir quality classification model, as shown in formula (6): (5) : Represents the i - th component in the input vector. : Represents the probability corresponding to each category.

[0051] (6) Where: Represents the sign function (0 or 1); it takes 1 if the true category of sample point i is equal to c, otherwise it takes 0. Represents the predicted probability that the observed sample point i belongs to category c. M represents the number of categories. N represents the number of samples.

[0052] In the embodiments of the present application, the above - mentioned oil and gas reservoir quality classification model can be trained according to the label data, the seismic trace data, and the reservoir template data set.

[0053] Specifically, the training of the oil and gas reservoir quality classification model can be carried out in two steps. First, the oil and gas reservoir quality classification model (hereinafter referred to as the model for short) can be trained using the reservoir template dataset. In this step of training the model, prior knowledge about seismic forward modeling and reservoir division criteria is mainly added to the model. The hyperparameters used when training the reservoir quality classification model can be that the learning rate is set to 0.001, the epoch is set to 1000, the optimization function is set to the Adam optimization function, and the batch is set to 64.

[0054] In some possible implementations, during the first-step training process, multiple groups (10,000 groups are taken as an example in this embodiment) of seismic waveform S4 data corresponding to the reservoir template can also be added to the model. That is, adding this part of the data to train the model can enable the model to more accurately identify and classify the reservoir quality and enhance the generalization ability of the model.

[0055] The second step can use the seismic trace data beside the known wells to train the oil and gas reservoir quality classification model. This step is to prepare the downstream task model more specifically according to the specific experimental work area. The hyperparameters used when training the downstream task M1 layer reservoir category model can be that the learning rate is set to 0.00001, the epoch is set to 2000, the optimization function is set to the Adam optimization function, and the batch is set to 64. In this way, the training of the oil and gas reservoir quality classification model can be achieved.

[0056] In some possible implementations, non-wellside seismic data can be obtained. Among them, non-wellside seismic data refers to the seismic data obtained in the undrilled area, and these data can be used to predict and evaluate the quality of the reservoir. Input the non-wellside seismic data into the trained oil and gas reservoir quality classification model, cycle to complete the reservoir category prediction of all seismic traces, and finally combine the reservoir categories corresponding to the seismic traces according to the positions of the seismic traces to form a reservoir plane distribution map of the M1 layer, such as Figure 3 shown, green - type I reservoir, gray - type II reservoir, blue - type III reservoir.

[0057] In this embodiment, the reservoir quality of the target interval can be classified according to the reservoir physical property parameters of the target interval, and the classification labels obtained by classification are encoded to obtain the label data of the reservoir category. Among them, the reservoir physical property parameters include: wave impedance, lithology type, sedimentary facies type, porosity, permeability, oil saturation, vertical reservoir thickness, and fluid type; obtain the seismic interval data of the target interval, extract the seismic trace data beside the known wells from the seismic interval data, statistically analyze the value range distribution of the reservoir physical property parameters of different categories of known wells, generate a reservoir geological model according to the value range distribution of the reservoir physical property parameters, and forward model to generate the corresponding seismic data to obtain a reservoir template data set, and train the pre-constructed oil and gas reservoir quality classification model according to the label data, seismic trace data, and reservoir template data set. In this application, the oil and gas reservoir quality classification model can be trained from four aspects: reservoir rock type, physical properties, vertical distribution, and fluid type, so that the oil and gas reservoir quality classification model can more comprehensively evaluate the classification of the reservoir. By comprehensively using deep learning technology and geological and geophysical knowledge, the reservoir quality can be classified more comprehensively and accurately, providing a scientific basis and technical support for oil and gas exploration and development, and optimizing the oil and gas reservoir development plan.

[0058] The above are some specific implementation manners of the training method of the oil and gas reservoir quality classification model provided by the embodiments of the present application. Based on this, the present application also provides a corresponding device. The device provided by the embodiments of the present application will be introduced from the perspective of functional modularization below. The device corresponds to the training method of the oil and gas reservoir quality classification model described above and can be referred to each other.

[0059] Figure 3 It is a structural block diagram of the training device of the oil and gas reservoir quality classification model provided by the embodiments of the present invention, which is called the second specific implementation manner. Refer to Figure 3 The device may include: An encoding unit 300, configured to classify the reservoir quality of the target interval according to the reservoir physical property parameters of the target interval, and encode the classification labels obtained by classification to obtain the label data of the reservoir category. The reservoir physical property parameters include: wave impedance, lithology type, sedimentary facies type, porosity, permeability, oil saturation, vertical reservoir thickness, and fluid type; An extraction unit 310, configured to obtain the seismic interval data of the target interval, and extract the seismic trace data beside the known wells from the seismic interval data; A generation unit 320, configured to statistically analyze the value range distribution of the reservoir physical property parameters of different categories of known wells, generate a reservoir geological model according to the value range distribution of the reservoir physical property parameters, and forward model to generate the corresponding seismic data to obtain a reservoir template data set; A training unit 330, configured to train a pre-constructed oil and gas reservoir quality classification model according to the label data, the seismic trace data, and the reservoir template data set.

[0060] Optionally, the oil and gas reservoir quality classification model includes A convolutional blocks, B ResNet blocks, a fully connected layer, and a softmax layer. The apparatus further includes: An acquisition unit, configured to acquire non-offset well seismic data; A prediction unit, configured to input the non-offset well seismic data into the trained oil and gas reservoir quality classification model for prediction.

[0061] Optionally, the encoding unit is specifically configured to: Classify the reservoir quality of the target interval by using the reservoir physical property parameters and the reservoir quality classification standard of the target interval, to obtain reservoir quality classification labels; Perform one-hot encoding on the reservoir quality classification labels, to obtain label data of reservoir categories, where the data dimension of the label data is (N, M), N is the number of wells, and M is the number of reservoir categories.

[0062] Optionally, the extraction unit is specifically configured to: Use the top and bottom seismic horizons of the target interval as constraints to generate seismic interval data S1 of the target interval; Extract features from S1 to obtain seismic feature data S2 of the target interval; Extract known well-side seismic trace data S3 from S2 according to the coordinates of known well points Optionally, the value range distribution of reservoir physical property parameters includes: value range distributions of wave impedance, lithology type, sedimentary facies type, porosity, permeability, oil saturation, vertical reservoir thickness, and fluid type. The generation unit is specifically configured to: Determine the normal distribution function of each reservoir physical property parameter according to the mean and variance of the value range distribution of the reservoir physical property parameters; Generate multiple groups of reservoir physical property parameters according to the normal distribution function of each reservoir physical property parameter, to obtain reservoir template data.

[0063] Optionally, the apparatus further includes: An input unit, configured to input the wave impedance parameter of the reservoir template into a convolution model to obtain seismic waveforms corresponding to multiple groups of reservoir templates; The training unit is specifically configured to: Train a pre-constructed oil and gas reservoir quality classification model according to the label data, the seismic trace data, the reservoir template data set, and the seismic waveforms.

[0064] The embodiments of the present application also provide corresponding devices and computer storage media for implementing the solutions provided by the embodiments of the present application.

[0065] Among them, the device includes a memory and a processor. The memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the method described in any embodiment of the present application.

[0066] The computer storage medium stores codes. When the codes are run, the device running the codes implements the method described in any embodiment of the present application.

[0067] In the embodiments of the present application, the "first", "second" (if any) in names such as "first" and "second" are only used as name identifiers and do not represent the first and second in order.

[0068] From the description of the above embodiments, those skilled in the art can clearly understand that all or part of the steps in the above embodiment methods can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product. The computer software product can be stored in a storage medium, such as a read-only memory (ROM) / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in various embodiments or some parts of the embodiments of the present application.

[0069] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, they are described relatively simply, and the relevant parts can be referred to the partial description of the method embodiments. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative work.

[0070] The above is only an exemplary embodiment of the present application and is not used to limit the protection scope of the present application.

Claims

1. A method for training an oil and gas reservoir quality classification model, characterized in that: include: Classifying the reservoir quality of the target layer segment according to the reservoir physical property parameters of the target layer segment and encoding the classification labels obtained by classification to obtain label data of the reservoir category, wherein the reservoir physical property parameters include: wave impedance, lithology type, sedimentary facies type, porosity, permeability, oil saturation, vertical reservoir thickness and fluid type; Obtain the seismic layer data of the target layer, and extract the seismic layer data to obtain the seismic trace data near the known well; Statistically analyzing the distribution of reservoir physical property parameters of different categories in known wells, generating a reservoir geological model based on the distribution of reservoir physical property parameters, and forward modeling corresponding seismic data to obtain a reservoir template data set; The pre-constructed oil and gas reservoir quality classification model is trained according to the label data, the seismic trace data and the reservoir template data set.

2. The method according to claim 1, characterized in that The oil and gas reservoir quality classification model includes multiple convolution blocks, multiple ResNet blocks, a fully connected layer and a softmax layer. The method also includes: Acquire non-adjacent well seismic data; The non-adjacent well seismic data is input into a trained oil and gas reservoir quality classification model for prediction.

3. The method according to claim 1, characterized in that The step of classifying the reservoir quality of the target layer section according to the reservoir physical property parameters of the target layer section and encoding the classification labels obtained by classification to obtain label data of the reservoir category includes: Classifying the reservoir quality of the target layer section by using the reservoir physical property parameters and the reservoir quality classification standard of the target layer section to obtain a reservoir quality classification label; The reservoir quality classification label is one-hot encoded to obtain label data of the reservoir category, and the data dimension of the label data is (N, M), where N is the number of wells and M is the reservoir category.

4. The method according to claim 1, characterized in that: The method of obtaining the seismic layer segment data of the target layer segment and extracting the seismic layer segment data to obtain the seismic trace data near the known well includes: Using the top and bottom seismic layers of the target layer segment as restrictions, generating the seismic layer segment data S1 of the target layer segment; Extract features from S1 to obtain seismic feature data S2 of the target layer segment; According to the coordinates of the known well point, the seismic trace data S3 near the known well is extracted from S2.

5. The method according to claim 1, characterized in that The reservoir physical property parameter value range distribution includes: wave impedance, lithology type, sedimentary phase type, porosity, permeability, oil saturation, vertical reservoir thickness and fluid type value range distribution. According to the reservoir physical property parameter value range distribution, a reservoir geological model is generated and corresponding seismic data is forward modeled to obtain reservoir template data, including: Determining a normal distribution function of each reservoir physical property parameter according to the mean and variance of the reservoir physical property parameter range distribution; According to the normal distribution function of each reservoir physical property parameter, multiple groups of reservoir physical property parameters are generated to obtain reservoir template data.

6. The method according to claim 1, characterized in that The method further comprises: Input the wave impedance parameters of the reservoir template into the convolution model to obtain multiple groups of seismic waveforms corresponding to the reservoir templates; The training of the pre-constructed oil and gas reservoir quality classification model according to the label data, the seismic trace data and the reservoir template data set includes: A pre-constructed oil and gas reservoir quality classification model is trained according to the label data, the seismic trace data, the reservoir template data set and the seismic waveform.

7. A training device for an oil and gas reservoir quality classification model, characterized in that: include: An encoding unit, used for classifying the reservoir quality of the target layer segment according to the reservoir physical property parameters of the target layer segment and encoding the classification labels obtained by classification to obtain label data of the reservoir category, wherein the reservoir physical property parameters include: wave impedance, lithology type, sedimentary facies type, porosity, permeability, oil saturation, vertical reservoir thickness and fluid type; An extraction unit is used to obtain the seismic layer segment data of the target layer segment, and extract the seismic layer segment data to obtain the seismic trace data beside the known well; A generating unit, used for statistically analyzing the range distribution of reservoir physical property parameters of different categories in known wells, generating a reservoir geological model according to the range distribution of reservoir physical property parameters, and forward modeling corresponding seismic data to obtain a reservoir template data set; A training unit is used to train a pre-constructed oil and gas reservoir quality classification model according to the label data, the seismic trace data and the reservoir template data set.

8. The device according to claim 7, characterized in that The oil and gas reservoir quality classification model includes multiple convolution blocks, multiple ResNet blocks, a fully connected layer and a softmax layer, and the device also includes: An acquisition unit, used for acquiring non-adjacent well seismic data; The prediction unit is used to input the non-adjacent well seismic data into a trained oil and gas reservoir quality classification model for prediction.

9. The device according to claim 7, characterized in that The encoding unit is specifically used for: Classifying the reservoir quality of the target layer section by using the reservoir physical property parameters and the reservoir quality classification standard of the target layer section to obtain a reservoir quality classification label; The reservoir quality classification label is one-hot encoded to obtain label data of the reservoir category, and the data dimension of the label data is (N, M), where N is the number of wells and M is the reservoir category.

10. The device according to claim 7, characterized in that The extraction unit is specifically used for: Using the top and bottom seismic layers of the target layer segment as restrictions, generating the seismic layer segment data S1 of the target layer segment; Extract features from S1 to obtain seismic feature data S2 of the target layer segment; According to the coordinates of the known well point, the seismic trace data S3 near the known well is extracted from S2.

Images