Underground physical characteristic estimation method and device

Abstract

According to the present disclosure, provided are an underground physical characteristic estimation method and apparatus, the underground physical characteristic estimation method comprising: a step of preprocessing logging record data and elastic wave data to generate a training data set; a step of generating an estimation model, the estimation model comprising: an encoder model for generating a potential space reflecting the characteristics of the formation of the input data using the training data; a decoder model based on the potential spatial reconstruction as a first factor of a subsurface physical property; and a regression model based on the potential spatial estimation as a second factor of subsurface physical properties; and a step of inputting elastic wave data into the estimation model to estimate a second factor that is a subsurface physical characteristic. According to the underground physical property estimation method disclosed by the invention, the data for judging whether the reservoir exists or not can be obtained by using the elastic wave exploration result with relatively low cost.

Description

Technical Field

[0001] The present invention relates to a method and device for estimating underground physical properties. Background Art

[0002] Oil or natural gas may exist in the reservoir of the formation. It takes a lot of time and effort to explore the buried location of oil or natural gas. Since the method of obtaining well-log data by drilling a well in the formation requires a lot of equipment and cost, a method of obtaining elastic wave data by seismic exploration can be used. Since it is difficult to fully estimate the underground physical properties only by elastic wave data, the well-log data is analyzed together. The process of estimating the underground physical properties by analyzing well-log data and elastic wave data must rely on the experience and knowledge of experts, so the analysis results may be biased depending on the experts and may take a lot of time.

[0003] Prior art literature

[0004] [Patent Document]

[0005] (Patent Document 1) US 2023 / 0026857 A1 Summary of the invention

[0006] 1. Technical issues to be resolved

[0007] The present disclosure provides a method and device for estimating underground physical properties using an artificial intelligence model.

[0008] (II) Technical solution

[0009] According to the present disclosure, the underground physical property estimation method is executed in a computer device, which includes one or more processors, and the processors are configured to run computer-readable instructions included in a storage unit. The underground physical property estimation method may include: a step of preprocessing well logging data and elastic wave data to generate a training data set; a step of generating an estimation model, the estimation model including: an encoder model, which uses the training data to generate a latent space reflecting the characteristics of the formation of the input data; a decoder model, which reconstructs a first factor as the underground physical property based on the latent space; and a regression model, which estimates a second factor as the underground physical property based on the latent space; and a step of inputting elastic wave data into the estimation model to estimate the second factor as the underground physical property.

[0010] According to one embodiment, the step of generating an estimation model may include: a step of generating an encoder model, wherein the encoder model uses the training data set to extract features of the strata included in the training data set in a self-supervised training method to generate a latent space reflecting the features of the strata; a step of generating a decoder model, wherein the decoder model reconstructs a first factor based on the latent space; a step of generating a regression model, wherein the regression model estimates a second factor based on the latent space; and a step of applying a loss function weight to simultaneously train the encoder model and the decoder model to simultaneously minimize the reconstruction error of the first factor and the estimation error of the second factor.

[0011] According to one embodiment, in the step of the encoder model, a BYOL (Bootstrap Your Own Latent) method as self-supervised training can be used to train the encoder part of BYOL that extracts the characteristics of the stratum from the training data included in the training data set, and the encoder part is extracted to obtain the encoder model.

[0012] According to one embodiment, in the step of generating a decoder model, the decoder model can be formed to reconstruct the first factor based on the latent space generated by the encoder model, and the decoder model can be trained using a reconstruction error, wherein the reconstruction error is obtained by comparing the first factor output by the decoder model with the first factor of the first label data included in the training dataset when the training data included in the training dataset is input into the encoder model.

[0013] According to one embodiment, in the step of generating a regression model, a regression equation may be generated to estimate the second factor based on the latent space generated by the encoder model.

[0014] According to one embodiment, in the step of simultaneously training the encoder model and the decoder model, when the training data of the training data set is input into the encoder model, the reconstruction error of the first factor output by the decoder model and the estimation error of the second factor output by the regression model can be obtained, and the encoder model and the decoder model can be trained based on the reconstruction error of the first factor, and the encoder model can be trained based on the estimation error of the second factor.

[0015] According to one embodiment, the loss function used to simultaneously train the encoder model and the decoder model may be weighted to reflect that the estimation error of the second factor is greater than the reconstruction error of the first factor.

[0016] According to one embodiment, the step of generating a training data set may include: a data preprocessing step, in which the well logging data is converted into a time domain form, the well logging data is sampled so that the resolution of the well logging data is the same as that of the elastic wave data, and the first factor as the reconstruction object is smoothed in the horizontal direction of the formation; a step of extracting cube-shaped data of the width, length and height of each factor from the elastic wave data to generate training data; a step of extracting a first factor of the width, length and height corresponding to the cube of the training data from the elastic wave data to generate first label data; and a step of extracting a second factor corresponding to one of the width, length and height of the cube of the training data from the well logging data to generate second label data.

[0017] The underground physical property estimation device according to the present disclosure may include: one or more processors; and a storage unit connected to the processor, capable of sending and receiving data, and storing program codes executed by the processor to execute the underground physical property estimation method.

[0018] The non-transitory computer-readable recording medium according to the present disclosure may include a program code executed by a processor to perform any one of the steps of the underground physical property estimation method.

[0019] Features and advantages of the present disclosure will become more apparent from the following detailed description based on the accompanying drawings.

[0020] It should be noted that the terms or words used in this specification and claims should not be interpreted as general meanings or dictionary meanings, but should be interpreted as meanings and concepts that are consistent with the technical ideas of the present disclosure based on the principle that the inventor can appropriately define the concepts of the terms in order to explain his own invention in the best way.

[0021] (III) Beneficial effects

[0022] According to the present disclosure, artificial intelligence models trained using elastic wave data and well log data can be used to estimate subsurface physical properties with high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a flow chart showing the steps of a method for estimating subsurface physical properties according to one embodiment.

[0024] Figure 2 is a diagram showing an underground physical property estimation apparatus according to one embodiment.

[0025] Figure 3 is a diagram illustrating the structure of a training dataset according to one embodiment.

[0026] Figure 4 is a diagram illustrating an estimation model according to one embodiment.

[0027] Figure 5 is a diagram illustrating the training of an encoder model according to one embodiment.

[0028] Figure 6 is a diagram illustrating the training of a decoder model according to one embodiment.

[0029] Figure 7 is a diagram illustrating simultaneous training of an encoder model and a decoder model according to one embodiment.

[0030] Description of reference numerals:

[0031] 1: Underground physical properties estimation device

[0032] 10: Processor

[0033] 20: Storage unit

[0034] 30: Communication unit

[0035] 40: Input / output unit

[0036] 100: Estimation Model

[0037] 110: Encoder Model

[0038] 120: Latent Space

[0039] 130: Decoder Model

[0040] 140: Regression Model

[0041] TDS: Training Dataset

[0042] TDS1: The first training dataset

[0043] TDS2: Second training data set

[0044] TD: Training Data

[0045] LD1: First label data

[0046] LD2: Second label data DETAILED DESCRIPTION

[0047] The purpose, advantages and features of the present disclosure will become more clear through the following detailed description and preferred embodiments related to the drawings, but the present disclosure is not limited thereto. In addition, when describing the present disclosure, if it is judged that the detailed description of the relevant known technology may unnecessarily obscure the main purpose of the present disclosure, its detailed description will be omitted.

[0048] It should be noted that when assigning reference numerals to components in the drawings, the same reference numerals are assigned as much as possible even if the same components are shown in different drawings, and similar reference numerals are assigned to similar components.

[0049] The terms used to describe one embodiment of the present disclosure are not intended to limit the present disclosure.It should be noted that unless the context clearly indicates otherwise, a singular expression includes a plural expression.

[0050] For the purpose of illustrating the embodiments, the drawings may be shown schematically or exaggeratedly.

[0051] In this specification, expressions such as “having”, “may have”, “including” or “may include” indicate the existence of corresponding features (for example, values, functions, operations or components such as parts), and do not exclude the existence of additional features.

[0052] The terms “one”, “other”, “another”, “first”, “second”, etc. are used to distinguish one component from other components, and the components are not limited by the above terms.

[0053] The embodiments and drawings described in this specification are not intended to limit the present disclosure to specific embodiments. It should be understood that the present disclosure includes various modifications, equivalents and / or alternatives of the embodiments.

[0054] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

[0055] Figure 1 is a flow chart showing the steps of a method for estimating subsurface physical properties according to one embodiment.

[0056] According to an embodiment, a method for estimating underground physical properties can be executed on a computer device. The computer device may include one or more processors 10, which are configured to execute computer-readable instructions included in a storage unit 20. The method for estimating underground physical properties may include a step (S10) of receiving well logging data and elastic wave data by the computer device. The method for estimating underground physical properties may include: a step (S20) of preprocessing the well logging data and elastic wave data to generate a training data set (Train Data Set, TDS); a step (S30) of generating an estimation model 100, wherein the estimation model 100 includes: an encoder model 110, which uses the training data (Train Data, TD) to generate a latent space 120 reflecting the characteristics of the strata of the input data; a decoder model 130, which reconstructs a first factor as an underground physical property based on the latent space 120; and a regression model 140, which estimates a second factor as an underground physical property based on the latent space 120; and a step (S40) of inputting the elastic wave data into the estimation model 100 to estimate the second factor as an underground physical property.

[0057] Well-log data are values ​​of physical properties of the subsurface measured directly by forming a well in the formation and inserting a sensor into the well. Well-log data may further include values ​​calculated using the directly measured values. Well-log data may include factors of various subsurface physical properties.

[0058] Subsurface properties refer to the properties of a formation. Factors of subsurface properties are items that represent the properties of a formation. Factors of subsurface properties include information about the properties of a formation. Factors of subsurface properties may include measured depth, borehole inner diameter (caliper), gamma ray, resistivity, bulk density, eutronporosity, photoelectric factor, volume of shale, volume of cabonate, volume of sand, porosity, water saturation, bulk modulus, P-wave velocity, S-wave velocity, P-wave impedance, S-wave impedance, etc. Factors of subsurface properties may be obtained by forming a borehole in the formation and measuring directly, or by calculating using the measured values.

[0059] Seismic data can be obtained through seismic exploration. When seismic exploration is land exploration, seismic exploration is a method of using explosives to impact the strata and obtaining data through sensors set on the ground. Seismic exploration can calculate factors through impedance inversion. Seismic data can include factors such as P-wave impedance, S-wave impedance, and Vp / Vs.

[0060] Among the factors of various underground physical properties, the first factor can be used as label data (Label Data) for training the encoder model 110 and the decoder model 130 of the estimation model 100. Among the factors of various underground physical properties, the second factor is a factor that is highly correlated with the possibility of the existence of an underground reservoir. In one embodiment, the first factor is P-wave impedance and the second factor is porosity. Different from one embodiment, other factors can be used as the first factor and the second factor.

[0061] Figure 2 FIG. 1 is a diagram showing an underground physical property estimation apparatus 1 according to an embodiment. Figure 1 and Figure 2 Provide explanation.

[0062] The underground physical property estimation device 1 may be a computer device. The underground physical property estimation device 1 may include: one or more processors 10; and a storage unit 20, connected to the processor 10, capable of sending and receiving data, and storing program codes executed by the processor 10 to perform the underground physical property estimation method. The underground physical property estimation device 1 may further include: a communication unit 30, connected to the processor 10, capable of sending and receiving data, and connected to a wired or wireless network to send and receive data; and an input / output unit 40, a user inputs data or instructions, and displays information to the user in a visual or auditory manner.

[0063] The program code is a set of instructions executed by the processor 10 to perform the underground physical property estimation method. The processor 10 can read the program code stored in the storage unit 20 and execute the program code to perform the underground physical property estimation method. The processor 10 is a computing device that performs information processing functions. The processor 10 may include a microprocessor 10, a programmable integrated circuit, a CPU, a GPU, an AP, or other semiconductor chips capable of processing information.

[0064] The storage unit 20 may include a ROM, RAM, flash memory, hard disk, cloud storage, etc. capable of storing program codes. The storage unit 20 may store not only program codes but also data for executing the method for estimating underground physical properties. The storage unit 20 may store well logging data, elastic wave data, cubic index, filter, estimation module, BYOL structure, etc.

[0065] The communication unit 30 can connect the underground physical property estimation device with an external computer device or sensor, so that the underground physical property estimation device can send and receive data with the external computer device or sensor. The communication unit 30 can be connected to wired and wireless networks. The communication unit 30 can use various communication methods, such as Ethernet, LAN, WAN, IP4, IP6, LTE, 5G, 6G, zigbee, wi-fi, Bluetooth, etc.

[0066] The input / output unit 40 may include input devices such as a keyboard, a mouse, and a touch pad, and output devices such as a monitor, a speaker, and a printer. A user may input instructions or data to the underground physical property estimating device 1 through the input / output unit 40. A user may confirm the result of estimating the underground physical property by the underground physical property estimating device 1 through the input / output unit 40.

[0067] The underground physical property estimation device 1 may be a PC, a server computer, a notebook computer, a tablet computer, or other devices capable of processing information.

[0068] The underground physical property estimation method may perform a step (S10) in which a computer device receives well logging data and elastic wave data. The underground physical property estimation device may receive the well logging data and elastic wave data through a wired or wireless network, or may read the well logging data and elastic wave data from an external storage device (memory card, hard disk, etc.). The step (S10) of receiving the well logging data and elastic wave data may include a process of receiving and storing data measured by a sensor inserted into a borehole, and receiving and storing data measured by a sensor used for elastic wave exploration.

[0069] When the well logging data and the elastic wave data are received, the method for estimating underground physical properties may perform a step (S20) of generating a training data set (TDS). In the step (S20) of generating a training data set (TDS), the apparatus for estimating underground physical properties may pre-process the well logging data and the elastic wave data, extract part of the data, and set them in a predetermined form to generate the training data set (TDS).

[0070] The step (S20) of generating a training data set (TDS) may include a data preprocessing step, in which the well logging data is converted into a time domain form, the well logging data is sampled so that the resolution of the well logging data is the same as the resolution of the elastic wave data, and the first factor as the reconstruction object is smoothed in the horizontal direction of the formation.

[0071] The elastic wave data includes the value of a factor measured by a sensor installed on the ground with the time point when the impact is applied to the formation as a reference over time. That is, the elastic wave data is obtained in the form of time domain.

[0072] Well logging data includes the values ​​of factors measured by the sensor at the depth of the descent through the borehole. That is, well logging data is obtained in the depth domain. Compared with elastic wave data, well logging data can be obtained with higher accuracy. Here, high accuracy means high resolution that can distinguish differences in underground physical properties in three-dimensional space.

[0073] In the data preprocessing step, the underground physical property estimation device 1 converts the logging data in the depth domain form into the time domain form. In addition, the logging data is resampled (resampling) so that the resolution of the logging data is the same as the resolution of the elastic wave data. Then, the first factor is smoothed to reflect the continuous geological characteristics of the formation. Smoothing can be performed by applying a filter to the data of the first factor. The filter can use a 3D gaussian filter. The data of the smoothed first factor can be in a state of improved geological continuity in the horizontal direction. For example, if the data of the P-wave impedance as the first factor is smoothed, the P-wave impedance can be in a state of improved continuity in the horizontal direction.

[0074] After performing the data preprocessing step, the well logging data and elastic wave data have the same state (time domain, resolution). After performing the data preprocessing step, the training data (TD) and label data of the training data set (TDS) can be generated.

[0075] Figure 3 is a diagram illustrating the structure of a training data set (TDS) according to one embodiment.

[0076] The step (S20) of generating a training data set (TDS) may include the step of extracting data in the form of a cube (CB) of width, length and height of each factor from the elastic wave data to generate training data (TD). The step (S20) of generating a training data set (TDS) may include the step of extracting a first factor of width, length and height corresponding to the cube (CB) of the training data (TD) from the elastic wave data to generate first label data (LD1). The step (S20) of generating a training data set (TDS) may include the step of extracting a second factor corresponding to one of the width, length and height of the cube (CB) of the training data (TD) from the well logging data to generate second label data (LD2).

[0077] The training data set (TDS) may include training data (TD) and label data corresponding to the training data (TD). The label data may include first label data (LD1) and second label data (LD2).

[0078] The training data (TD) may be extracted in the form of a three-dimensional (3D) cube (CB) of a predetermined size. The cube may be a 3D matrix structure having a width, length, and height of a predetermined size. In one embodiment, the cube (CB) may be a 3D matrix structure (7x7x7) having a width, length, and height of 7 grids, respectively, and each grid may contain a value of a factor. The training data (TD) and the first label data (LD1) may be extracted in the form of a cube (CB) of a 3D matrix structure.

[0079] Any cube (CB) can contain factor values ​​for any part of the formation. Well logging data and elastic wave data can divide the measured formation into multiple cubes (CB). In well logging data or elastic wave data, the part of the cube (CB) without null data can be used as training data (TD). Training data (TD) can be extracted from well logging data or elastic wave data. In order to facilitate data processing (handling), the extracted cubes (CB) can be assigned serial numbers and indexed.

[0080] In the step of generating training data (TD), the underground physical property estimation device 1 can extract a cube (CB) containing factor values ​​of a predetermined number of factors at a specific formation location to generate training data (TD) at the specific formation location. One training data (TD) may include multiple cubes (CB) with different factor values ​​at the same underground location. According to one embodiment, the training data (TD) may include a cube (CB) with values ​​of S-wave impedance factors, a cube (CB) with values ​​of P-wave impedance factors, and a cube (CB) with values ​​of Vs / Vp factors, etc. According to one embodiment, the training data (TD) may include cubes (CB) of eight different factors at a specific formation location. The training data (TD) may be expressed in the form of "number of factors x (cube (CB) size)", and according to one embodiment, the training data (TD) may be expressed in the form of "8x(7x7x7)".

[0081] The first label data (LD1) is the value of the first factor. In one embodiment, the first factor is P-wave impedance, and thus the first label data (LD1) is the P-wave impedance value included in the elastic wave data. The first label data (LD1) is a cube (CB) (7x7x7 matrix structure) having the P-wave impedance value of the same formation position as the cube (CB) of the training data (TD). In one embodiment, the training data set (TDS) may include training data (TD) of 8 cubes (CB) with different factors and first label data (LD1) of one cube (CB) with one first factor. The first label data (LD1) may be expressed in the form of "number of factors x (cube (CB) size)", and according to one embodiment, the first label data (LD1) may be expressed in the form of "1x(7x7x7)".

[0082] In the step of generating the first label data (LD1), the underground physical property estimation device 1 may extract a cube (CB) having a first factor value at the same position as the cube (CB) of the training data (TD) to generate the first label data (LD1). When generating the first label data (LD1), the first label data (LD1) corresponding to the training data (TD) extracted from the elastic wave data may be extracted from the elastic wave data.

[0083] The second label data (LD2) is the value of the second factor. In one embodiment, the second factor is porosity, and thus the second label data (LD2) is the porosity value included in the well logging data. The second label data (LD2) is a data array (DA) having porosity values ​​of the same formation position as the cube (CB) of the training data (TD). The data array (DA) is a structure in which the factor values ​​are arranged in one of the width, length, and height of the cube (CB). For example, the second label data (LD2) is a data array (DA) arranged in one of the width, length, and height according to the porosity values ​​of a specific formation position. The second label data (LD2) can be expressed in the form of "number of factors x (array length)", and according to one embodiment, the second label data (LD2) can be expressed in the form of "1x(1x1x7)".

[0084] Porosity is the factor most related to the presence or absence of a reservoir. Since porosity cannot be measured by elastic wave exploration, porosity does not exist in elastic wave data. Porosity exists in well logging data. Therefore, the second label data (LD2) is extracted from the well logging data.

[0085] In one embodiment, the training data set (TDS) may include training data (TD) of eight cubes (CB) having different factors and second label data (LD2) of a data array (DA) having one second factor.

[0086] In the step of generating the second label data (LD2), the underground physical property estimation device 1 may extract a cube (CB) having a second factor value at the same position as the cube (CB) of the training data (TD) to generate the second label data (LD2). When generating the second label data (LD2), the second label data (LD2) corresponding to the training data (TD) extracted from the well logging data may be extracted from the well logging data.

[0087] The training data set (TDS) may include: a first training data set (TDS1), including training data (TD) and first label data (LD1); and a second training data set (TDS2), including training data (TD), first label data (LD1) and second label data (LD2). The eight different factors included in the training data (TD) may select factors present in both elastic wave data and well logging data. The first label data (LD1) is data for reconstruction error using the encoder model 110 and the decoder model 130, and thus can only be extracted from the elastic wave data. Since the second label data (LD2) is porosity, it can only be extracted from the well logging data.

[0088] Obtaining well logging data requires a lot of cost and effort, so the amount of data is small, while obtaining elastic wave data requires relatively less cost and effort than obtaining well logging data, so the amount of data is relatively large. Therefore, the number of first training data sets (TDS1) that can be obtained from both well logging data and elastic wave data is large, while the number of second training data sets (TDS2) that can be obtained only from well logging data is relatively small. However, according to an embodiment of the method for estimating underground physical properties, an estimation model 100 can be trained, and the estimation model 100 can estimate the factor that only exists in the well logging data, namely, porosity, with high accuracy.

[0089] Figure 4 1 is a diagram illustrating an estimation model 100 according to one embodiment. Figure 1 and Figure 4 Provide explanation.

[0090] The estimation model 100 may include an encoder model 110, a decoder model 130, and a regression model 140. The encoder model 110 may receive data in a cube form and extract features of a stratum to generate a latent space 120. The latent space 120 is formed by combining the latent features of the stratum extracted by the encoder model 110. The decoder model 130 reconstructs a first factor based on the latent space 120. The decoder model 130 may reconstruct a factor of underground physical properties based on the features of the stratum reflected in the latent space 120. The regression model 140 outputs a second factor based on the latent space 120. The regression model 140 estimates the second factor based on the features of the stratum reflected in the latent space 120 using a regression analysis method.

[0091] A ResNet structure may be applied to the encoder model 110, the regression model 140, and the latent space 120. The encoder model 110, the latent space 120, and the decoder model 130 may be formed in a variational auto encoder (VAE) structure.

[0092] The step (S30) of generating the estimation model 100 may include: a step of generating an encoder model 110, wherein the encoder model 110 uses the training data set (TDS) to extract the characteristics of the strata included in the training data set (TDS) in a self-supervised training method to generate a latent space 120 reflecting the characteristics of the strata; a step of generating a decoder model 130, wherein the decoder model 130 reconstructs the first factor based on the latent space 120; a step of generating a regression model 140, wherein the regression model 140 estimates the second factor based on the latent space 120; and a step of applying a loss function weight to simultaneously train the encoder model 110 and the decoder model 130 to simultaneously minimize the reconstruction error of the first factor and the estimation error of the second factor.

[0093] Figure 5 is a diagram illustrating the training of the encoder model 110 according to one embodiment.

[0094] The step of generating the encoder model 110 may be performed using a BYOL (Bootstrap Your Own Latent) method in a self-supervised learning method. In the step of generating the encoder model 110, the encoder part of BYOL that extracts the characteristics of the formation from the training data (TD) included in the training data set (TDS) is trained using the self-supervised learning BYOL method, and the encoder part is extracted to obtain the encoder model 110.

[0095] In order to generate the encoder model 110 in a BYOL method, a teacher network and a student network are prepared. The teacher network and the student network may include an augmentation, an encoder, and a projector, and the student network may further include a predictor. The augmentor receives data in a cube form and outputs a view v, v' as a result of the augmented data. The encoder extracts features of the augmented data and outputs representations y, y'. The projector projects the representation and outputs projections z, z'. The predictor projects the projection again and outputs a prediction q(z).

[0096] The teacher network is pre-trained using training data (TD) and connected to the student network to start training. When the training data (TD) is input to the teacher network and the teacher network generates the final output sg(z') of the teacher network, the training data (TD) is input to the student network to generate the final output q(z). The parameters of the encoder and projector of the student network are trained so that the difference between the final output of the student network and the final output of the teacher network is minimized. Then, the parameters of the encoder and projector of the teacher network can be reset to the exponential moving average of the parameters of the encoder and projector of the student network. By repeating the process of training the student network using multiple training data (TD) and updating the parameters of the teacher network, the performance of the encoder of the student network in extracting features can be improved.

[0097] When the difference between the final output of the student network and the final output of the teacher network is minimized to a predetermined level, the training is stopped, and the encoder of the student network is extracted and set in the encoder model 110 of the estimation model 100. In the step of generating the encoder model 110, the encoder model 110 can be generated only using the training data (TD) using a BYOL method of self-supervised training.

[0098] Figure 6 is a diagram illustrating the training of the decoder model 130 according to one embodiment.

[0099] In the step of generating the decoder model 130, the decoder model 130 is generated to reconstruct the first factor based on the latent space 120 generated by the encoder model 110, and the decoder model 130 is trained using a reconstruction error, wherein the reconstruction error is obtained by comparing the first factor output by the decoder model 130 with the first factor of the first label data (LD1) included in the training data set (TDS) when the training data (TD) included in the training data set (TDS) is input into the encoder model 110.

[0100] The decoder model 130 is a generative model that reconstructs a first factor based on the characteristics of the formation included in the latent space 120 .

[0101] The step of generating the decoder model 130 is a process of connecting the decoder model 130 to the encoder generated in the step of generating the encoder model 110 and training the decoder model 130 using VAE. The encoder model 110 is trained to extract the characteristics of the stratum using the training data (TD) to generate the latent space 120. The decoder model 130 is formed to reconstruct the first factor based on the characteristics of the stratum included in the latent space 120 generated by the encoder model 110, and the decoder model 130 is connected to the encoder model 110.

[0102] When the training data (TD) is input to the encoder model 110, the encoder model 110 extracts the features of the strata of the training data (TD) to generate the latent space 120, and the decoder model 130 reconstructs the first factor based on the features of the strata included in the latent space 120. The reconstruction error is calculated by comparing the first factor reconstructed by the decoder model 130 with the first factor of the first label data (LD1), and the parameters of the decoder model 130 are trained to minimize the reconstruction error. When the reconstruction error reaches a predetermined level, the training is stopped to obtain the decoder model 130 that has been trained.

[0103] The decoder model 130 trains parameters of the decoder model 130 by reconstructing only the first factor instead of all factors of the data input to the encoder model 110. The decoder model 130 trained in this manner can be regarded as an initial state before the entire encoder model 110 and the decoder model 130 are trained together.

[0104] Figure 7 is a diagram illustrating simultaneous training of the encoder model 110 and the decoder model 130 according to one embodiment.

[0105] In the step of generating the regression model 140, a regression equation is generated to estimate the second factor based on the latent space 120 generated by the encoder model 110. The regression model 140 may select a predetermined number of factors from the factors included in the latent space 120 in order of large correlation with the second factor and apply weights to generate a regression equation that can best estimate the second factor. When the training data (TD) is input to the encoder model 110, the encoder model 110 may generate the latent space 120, and the regression model 140 estimates the second factor according to the regression equation based on the latent space 120. The regression model 140 may be generated by comparing the second factor estimated by the regression model 140 with the second factor of the second label data (LD2) and updating the regression loss into the regression equation.

[0106] In the estimation model 100 , the output of the encoder model 110 is connected to the input of the decoder model 130 , and the output of the encoder model 110 is connected to the input of the regression model 140 .

[0107] In the step of simultaneously training the encoder model 110 and the decoder model 130, when the training data (TD) of the training data set (TDS) is input into the encoder model 110, the reconstruction error of the first factor output by the decoder model 130 and the estimation error of the second factor output by the regression model 140 are obtained, and the encoder model 110 and the decoder model 130 are trained based on the reconstruction error of the first factor, and the encoder model 110 is trained based on the estimation error of the second factor. At this time, the decoder model 130 starts training from the initial state trained in the step of generating the decoder model 130.

[0108] In a state where the trained encoder model 110, the trained decoder model 130, and the regression model 140 are connected, when the training data (TD) is input to the encoder model 110, the decoder model 130 reconstructs and outputs the first factor, and the regression model 140 estimates and outputs the second factor. The reconstruction error can be obtained by comparing the first factor output by the decoder model 130 with the first factor of the first label data (LD1). The encoder model 110 and the decoder model 130 can be trained to accurately reconstruct the first factor using the reconstruction error. The estimation error can be obtained by comparing the second factor output by the regression model 140 with the second factor of the second label data (LD2). The encoder model 110 can be trained to accurately estimate the second factor using the estimation error. Training the encoder model 110 and the decoder model 130 using the reconstruction error and training the encoder model 110 using the estimation error can be performed simultaneously. To this end, the two errors (reconstruction error and estimation error) can be combined into one loss function to train simultaneously to minimize the two errors at the same time.

[0109] Here, the first factor is in a cubic form (7x7x7) and the second factor is in a data array form (1x1x7), so the amount of data associated with the first factor is greater. Therefore, when the reconstruction error and the estimation error are reflected in the loss function in the same proportion, the encoder model 110 and the decoder model 130 may be trained to minimize the reconstruction error greater than the estimation error. However, since the underground physical property estimation method according to an embodiment is used to explore the strata where reservoirs may exist, it is important to accurately estimate the performance of the second factor related to the presence of the reservoir. Therefore, the loss function for simultaneously training the encoder model 110 and the decoder model 130 may be weighted to reflect more the estimation error of the second factor than the reconstruction error of the first factor. The loss function may include a reconstruction error part and an estimation error part, and the weight multiplied by the reconstruction error part may be determined to be less than the weight multiplied by the estimation error part. Training the encoder model 110 using a loss function that adjusts the weights can improve the performance of the encoder model 110 in extracting the characteristics of the strata, thereby simultaneously improving the performance of reconstructing the first factor and the performance of estimating the second factor.

[0110] By performing such a process, an estimation model 100 including an encoder model 110, a decoder model 130, and a regression model 140 may be generated.

[0111] Refer again Figure 1 .

[0112] In the step (S40) of estimating the second factor, unseen data is input to the estimation model 100 trained using the training data set (TDS), and the estimation model 100 estimates and outputs the second factor corresponding to the input unseen data. The unseen in the unseen data is data that is not used for training in the process of generating the estimation model 100. That is, the unseen data refers to data that the estimation model 100 has not seen during training.

[0113] The unseen data may have the same form as the training data (TD). For example, the unseen data is in the form of a 3D cube having eight factors and may be formed into a structure represented by 8x(7x7x7). The unseen data may be obtained from elastic wave data obtained by elastic wave exploration of a formation in order to determine whether a reservoir exists.

[0114] The operation cost of drilling in the formation is very high, while the cost of elastic wave exploration is relatively low. Elastic wave exploration is performed in an area where it is desired to confirm whether a reservoir exists, and unseen data is obtained from the obtained elastic wave data and input into the estimation model 100, and the estimation model 100 can estimate and output the second factor most related to whether a reservoir exists at the formation position. Therefore, according to an underground physical property estimation method of an embodiment, data for determining whether a reservoir exists can be obtained using relatively low-cost elastic wave exploration results.

[0115] A non-transitory computer-readable recording medium may include a program code executed by the processor 10 to perform any one of the steps of the method for estimating underground physical properties according to an embodiment. The non-transitory computer-readable recording medium may include an electrical recording medium such as a memory card and a USB memory, a magnetic recording medium such as a hard disk drive and a magnetic tape, an optical recording medium such as a CD, a DVD, and a Blu-ray Disc, and a cloud storage, etc.

[0116] The present disclosure is described in detail above through specific embodiments. The embodiments are intended to illustrate the present disclosure in detail, but the present disclosure is not limited thereto. It is obvious to those skilled in the art that modifications or improvements can be made within the technical concept of the present disclosure.

[0117] Simple modifications or changes to the present disclosure all belong to the scope of the present disclosure, and the specific protection scope of the present disclosure will be determined by the appended claims.

Claims

1. A method for estimating a subsurface physical property, the method being performed on a computer device, the computer device comprising one or more processors, the processors being configured to execute computer-readable instructions contained in a storage unit, The underground physical property estimation method comprises: The step of preprocessing the well log data and elastic wave data to generate a training data set; The step of generating an estimation model, the estimation model comprising: an encoder model that generates a latent space reflecting characteristics of a stratum of input data using the training data; a decoder model that reconstructs a first factor as a physical characteristic of the subsurface based on the latent space; and a regression model that estimates a second factor as a physical characteristic of the subsurface based on the latent space; and The step of inputting elastic wave data into the estimation model to estimate a second factor as a subsurface physical property.

2. The underground physical property estimation method according to claim 1, wherein: The steps to generate the estimation model include: The step of generating an encoder model, wherein the encoder model uses the training data set to extract features of the strata included in the training data set in a self-supervised training method to generate a latent space reflecting the features of the strata; a step of generating a decoder model, wherein the decoder model reconstructs the first factor based on the latent space; a step of generating a regression model, the regression model estimating the second factor based on the latent space; and The step of applying a loss function weight to simultaneously train the encoder model and the decoder model to simultaneously minimize the reconstruction error of the first factor and the estimation error of the second factor.

3. The underground physical property estimation method according to claim 2, wherein: In the step of generating the encoder model, an encoder part of BYOL that extracts features of the formation from training data included in the training data set is trained using the BYOL method as self-supervised training, and the encoder part is extracted to obtain an encoder model.

4. The underground physical property estimation method according to claim 2, wherein: In the step of generating a decoder model, the decoder model is generated to reconstruct the first factor based on the latent space generated by the encoder model, and the decoder model is trained using a reconstruction error, wherein the reconstruction error is obtained by comparing the first factor output by the decoder model with the first factor of the first label data included in the training dataset when the training data included in the training dataset is input into the encoder model.

5. The underground physical property estimation method according to claim 2, wherein: In the step of generating a regression model, a regression equation is generated to estimate a second factor based on a latent space generated by the encoder model.

6. The underground physical property estimation method according to claim 2, wherein: In the step of simultaneously training the encoder model and the decoder model, when the training data of the training data set is input into the encoder model, a reconstruction error of a first factor output by the decoder model and an estimation error of a second factor output by the regression model are obtained, and the encoder model and the decoder model are trained based on the reconstruction error of the first factor, and the encoder model is trained based on the estimation error of the second factor, The loss function used to simultaneously train the encoder model and the decoder model is weighted to reflect that the estimation error of the second factor is greater than the reconstruction error of the first factor.

7. The underground physical property estimation method according to claim 1, wherein: The step of generating a training data set comprises: a data preprocessing step, in which the logging data is converted into a time domain form, the logging data is sampled so that the resolution of the logging data is the same as the resolution of the elastic wave data, and the first factor as a reconstruction object is smoothed in the horizontal direction of the formation; A step of extracting cube-shaped data of width, length and height of each factor from the elastic wave data to generate training data; A step of extracting first factors of width, length and height corresponding to the cube of the training data from the elastic wave data to generate first label data; and A step of extracting a second factor corresponding to one of the width, length and height of the cube of the training data from the well log data to generate second label data.

8. An apparatus for estimating underground physical properties, comprising: More than one processor; as well as a storage unit connected to the processor, capable of sending and receiving data, and storing a program code executed by the processor to perform a method for estimating underground physical properties, The underground physical property estimation method comprises: The step of preprocessing the well log data and elastic wave data to generate a training data set; The step of generating an estimation model, the estimation model comprising: an encoder model that generates a latent space reflecting characteristics of a stratum of input data using the training data; a decoder model that reconstructs a first factor as a physical characteristic of the subsurface based on the latent space; and a regression model that estimates a second factor as a physical characteristic of the subsurface based on the latent space; and The elastic wave data is input into the estimation model to estimate the step of a second factor as a subsurface physical property.

9. The underground physical property estimation device according to claim 8, wherein: The steps to generate the estimation model include: The step of generating an encoder model, wherein the encoder model uses the training data set to extract features of the strata included in the training data set in a self-supervised training method to generate a latent space reflecting the features of the strata; a step of generating a decoder model, wherein the decoder model reconstructs the first factor based on the latent space; a step of generating a regression model, the regression model estimating the second factor based on the latent space; and The step of applying a loss function weight to simultaneously train the encoder model and the decoder model to simultaneously minimize the reconstruction error of the first factor and the estimation error of the second factor.

10. The underground physical property estimation device according to claim 9, wherein: In the step of generating the encoder model, an encoder part of BYOL that extracts features of the formation from training data included in the training data set is trained using the BYOL method as self-supervised training, and the encoder part is extracted to obtain an encoder model.

11. The underground physical property estimation device according to claim 9, wherein: In the step of generating a decoder model, the decoder model is generated to reconstruct the first factor based on the latent space generated by the encoder model, and the decoder model is trained using a reconstruction error, wherein the reconstruction error is obtained by comparing the first factor output by the decoder model with the first factor of the first label data included in the training dataset when the training data included in the training dataset is input into the encoder model.

12. The underground physical property estimation device according to claim 9, wherein: In the step of generating a regression model, a regression equation is generated to estimate a second factor based on a latent space generated by the encoder model.

13. The underground physical property estimation device according to claim 9, wherein: In the step of simultaneously training the encoder model and the decoder model, when the training data of the training data set is input into the encoder model, a reconstruction error of a first factor output by the decoder model and an estimation error of a second factor output by the regression model are obtained, and the encoder model and the decoder model are trained based on the reconstruction error of the first factor, and the encoder model is trained based on the estimation error of the second factor, The loss function used to simultaneously train the encoder model and the decoder model is weighted to reflect that the estimation error of the second factor is greater than the reconstruction error of the first factor.

14. The underground physical property estimation device according to claim 9, wherein: The step of generating a training data set comprises: a data preprocessing step, in which the logging data is converted into a time domain form, the logging data is sampled so that the resolution of the logging data is the same as the resolution of the elastic wave data, and the first factor as a reconstruction object is smoothed in the horizontal direction of the formation; A step of extracting cube-shaped data of width, length and height of each factor from the elastic wave data to generate training data; A step of extracting first factors of width, length and height corresponding to the cube of the training data from the elastic wave data to generate first label data; and A step of extracting a second factor corresponding to one of the width, length and height of the cube of the training data from the well log data to generate second label data.

15. A non-transitory computer-readable recording medium comprising a program code executed by a processor to perform the subsurface physical property estimation method according to any one of claims 1 to 7.

Images