Information processing device, survey device and program

The information processing device addresses the challenge of representing actual underground structures by generating learning data with multiple layers and noise simulation, enhancing the accuracy of resistivity structure estimation and enabling real-time analysis.

JP2024168640AActive Publication Date: 2024-12-05WASEDA UNIV
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
JP2023085483
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-24
Publication Date
2024-12-05
Estimated Expiration
2043-05-24

AI Technical Summary

Technical Problem

Existing methods using a simple horizontal multilayer structure for resistivity learning data fail to accurately represent actual underground structures, leading to potential local solutions during electromagnetic exploration analysis.

Method used

An information processing device that generates learning data by setting up an underground structure with multiple layers, dividing them into groups, assigning resistivity using random numbers, smoothing resistivity values, and combining them to create a more realistic resistivity structure for machine learning, incorporating noise simulation to enhance accuracy.

Benefits of technology

This approach allows for the generation of learning data that accurately represents various underground structures, improving the estimation of resistivity structures with high accuracy and reducing the risk of local solutions, enabling real-time analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024168640000001_ABST
    Figure 2024168640000001_ABST
Patent Text Reader

Abstract

To provide an information processing device for generating learning data having a resistivity structure that can correspond to actual various underground structures.SOLUTION: According to the present invention, an information processing device is provided, including a layer setting part for setting an underground structure composed of a plurality of layers, classification part for classifying the plurality of layers into one or more groups composed of one layer or two or more continuous layers, a resistivity setting part for determining resistivity by a random number in each one or more groups, and setting the determined resistivity to each group in each layer included in the group, a smoothing part for performing smoothing processing using resistivity of one layer among a plurality of layers with resistivity set and resistivity of a layer adjacent to the one layer in each layer to set smoothing resistivity in each layer, and a learning resistivity structure data generation part for generating learning resistivity structure data obtained by specifying an underground structure on the basis of a plurality of layers with resistivity or smoothing resistivity set.SELECTED DRAWING: Figure 2
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to an information processing device, a search device, and a program. [Background technology]

[0002] A method using machine learning is known as an optimization (inverse analysis) method for electromagnetic exploration. A horizontal multi-layer structure is often used for the resistivity structure of the exploration target that serves as the learning data.

[0003] Non-Patent Document 1 discloses an inverse analysis method that uses a horizontal multi-layer structure with a relatively small number of layers as learning data. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Xin Wu, 5 others, "A Deep Learning Estimation of the Earth Resistivity Model for the Airborne Transient Electromagnetic Observation", [online], February 25, 2020, American Geophysical Union, [Retrieved April 12, 2020], Internet<URL:https: / / doi.org / 10.1029 / 2021JB023185> Summary of the Invention [Problem to be solved by the invention]

[0005] However, such a simple horizontal multi-layer structure is significantly different from the actual underground structure, and there was a risk of falling into a local solution during the analysis process.

[0006] The present invention has been made in consideration of the above circumstances, and provides an information processing device that generates learning data having a resistivity structure that can correspond to various actual underground structures. [Means for solving the problem]

[0007] According to the present invention, the following inventions are provided. [1] An information processing device comprising a layer setting unit, a division unit, a resistivity setting unit, a smoothing unit, and a learning resistivity structure data generation unit, wherein the layer setting unit sets an underground structure consisting of a plurality of layers, the division unit divides the plurality of layers into one or more groups consisting of one layer or two or more consecutive layers, the resistivity setting unit determines a resistivity for each of the one or more groups using a random number, and sets the resistivity determined for each group to each layer included in each group, the smoothing unit performs a smoothing process for each layer using the resistivity of one of the plurality of layers in which the resistivity is set and the resistivity of a layer adjacent to the one layer, and sets a smoothed resistivity to each of the layers, and the learning resistivity structure data generation unit generates learning resistivity structure data in which an underground structure is defined based on the resistivity or the plurality of layers in which the smoothed resistivity is set. [2] An information processing device as described in [1], further comprising a synthesis unit, wherein the synthesis unit performs synthesis processing for each layer of the resistivity and a smoothed resistivity of a layer corresponding to the layer in which the resistivity is set, and sets a composite resistivity for each layer, and the learning resistivity structure data generation unit generates learning resistivity structure data in which an underground structure is defined based on the multiple layers in which the composite resistivity is set. [3] The information processing device according to [1] or [2], wherein the depth of each of the plurality of layers is determined by a quadratic function. [4] An information processing device according to any one of [1] to [3], further comprising a learning data generation unit, a machine learning unit, an input processing unit, and an output control unit, wherein the learning data generation unit generates learning data by associating the learning resistivity structure data with artificial response data calculated based on the learning resistivity structure data, the machine learning unit has a machine learning model trained by machine learning using the learning data, the input processing unit inputs measurement response data obtained by geophysical exploration measurements of the underground at a measurement site to the machine learning model, and the output control unit outputs estimated resistivity structure data of the underground at the measurement site from which the measurement response data was obtained, which is output from the machine learning model. [5] An information processing device according to [4], wherein the artificial response data and the measured response data are voltage data or magnetic field data. [6] An information processing device according to the invention described in [4] or [5], wherein the geophysical exploration measurement is an electromagnetic exploration measurement. [7] The information processing device according to any one of [4] to [6], wherein the machine learning unit has the machine learning model trained by machine learning using the noise-added artificial response data. [8] An information processing device according to [7], wherein the noise is empirically calculated noise or noise calculated based on environmental noise present at the time the measurement response data was measured. [9] An exploration device comprising an information processing device according to any one of [1] to [8].

[10] A program causing a computer to execute a layer setting process, a division process, a resistivity setting process, a smoothing process, and a learning resistivity structure data generating process, in which the layer setting process sets an underground structure consisting of multiple layers, the division process divides the multiple layers into one or more groups consisting of one layer or two or more consecutive layers, the resistivity setting process generates resistivity for each of the one or more groups using random numbers, and sets the resistivity generated for each group to each layer included in each group, the smoothing process performs a smoothing process for each layer using the resistivity of one of the multiple layers in which the resistivity is set and the resistivity of a layer adjacent to the one layer, and sets a smoothed resistivity to each of the layers, and the learning resistivity structure data generating process generates learning resistivity structure data in which an underground structure is defined based on the resistivity or the multiple layers in which the smoothed resistivity is set.

[0008] According to the present invention, by generating learning data using both the initial resistivity and the smoothed resistivity, it is possible to deal with a variety of underground structures. [Brief description of the drawings]

[0009] [Figure 1] 1 is a block diagram showing a hardware configuration of an information processing device 1 according to a first embodiment of the present invention. [Diagram 2] 1 is a block diagram showing a functional configuration of an information processing device 1 according to a first embodiment of the present invention. [Diagram 3] 1 is a diagram showing an image of a division of an underground structure divided by a division unit 20b of an information processing device 1 according to a first embodiment of the present invention. [Figure 4] 2 is an example of learning resistivity structure data generated as a response variable by the information processing device 1 according to the first embodiment of the present invention. [Diagram 5] FIG. 1 is a flow diagram of data processing using an information processing device 1 according to a first embodiment of the present invention. [Figure 6] FIG. 2 is a flow diagram of a learning data generation step S2 of data processing using the information processing device 1 according to the first embodiment of the present invention. [Figure 7] FIG. 11 is a block diagram showing a hardware configuration of a exploration device 3 according to a second embodiment of the present invention. [Figure 8] 4 is an example of learning resistivity structure data used in the data processing of the present invention. [Figure 9] 1 is an example of learning resistivity structure data used in data processing according to a conventional method. [Figure 10] 4 shows the learning curve and validation error for the data processing method of the present invention. [Figure 11] 1 shows the learning curve and validation error for a conventional data processing method. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, embodiments of the present invention will be described. Various features shown in the following embodiments can be combined with each other. Also, each feature can be an invention independently.

[0011] 1. First embodiment The inverse analysis system according to the present embodiment is a system capable of estimating the resistivity structure of the underground at the location where the measurement was performed from the measurement data obtained by the electromagnetic exploration measurement. The inverse analysis system according to the present embodiment includes an information processing device 1.

[0012] 1.1 Hardware configuration of information processing device 1 1 shows a hardware configuration of an information processing device 1 according to this embodiment. The information processing device 1 according to this embodiment includes a control unit 10, a storage unit 11, a communication unit 12, an input unit 13, and an output unit 14. The control unit 10, the storage unit 11, the communication unit 12, the input unit 13, and the output unit 14 are connected to each other by a communication bus 40.

[0013] The control unit 10 may include a processor such as a central processing unit (CPU), a digital signal processor, a microprocessor, a microcontroller, an application specific integrated circuit (ASIC), a programmable logic device (PLD), or the like.

[0014] The storage unit 11 may include a volatile memory or a non-volatile memory. Examples of the volatile memory include a RAM (Random Access Memory) and a DRAM (Dynamic Random Access Memory), which are used as a work area when the control unit 10 executes processes based on various programs. Examples of the non-volatile memory include a ROM (Read Only Memory) and a flash memory, which are used to store various data and programs used in the processes of the control unit 10. The storage unit 11 may also include an external storage device such as an HDD (Hard Disk Drive).

[0015] The programs stored in the memory unit 11 include, for example, an OS (Operating System) for realizing the basic functions of the information processing device 1, drivers for controlling various hardware, programs for realizing various functions, etc., and include the computer program related to this embodiment.

[0016] The communication unit 12 is configured to be capable of transmitting and receiving, and has a function of connecting to a communication line. For example, the communication unit 12 may be configured to be capable of communicating with an external probe device, in which case the measurement response data can be acquired via the communication unit 12. The communication unit 12 may have, for example, a NIC (Network Interface Controller), a function of connecting to a wireless LAN (Local Area Network), a function of connecting to a wireless WAN (Wide Area Network), a function of enabling short-distance wireless communication such as Bluetooth (registered trademark), infrared communication, or the like, or may have a function of enabling a wired connection.

[0017] The input unit 13 can input various parameters set by the user in a parameter setting step S1 described below, and measurement response data. Here, the input measurement response data is data acquired by geophysical exploration measurement, particularly electromagnetic exploration measurement. The measurement response data is voltage data (induced electromotive force data) or magnetic field data. The input unit 13 can include one or more of, for example, a keyboard, a keypad, a mouse, a microphone, a touch screen, a button, and the like.

[0018] The output unit 14 can display the measurement response data and the estimated resistivity structure data output from the output control unit 22 of the information processing device 1. The output unit 14 can be, for example, any display.

[0019] 1.2 Functional configuration of information processing device 1 2 shows a functional configuration of the information processing device 1. A control unit 10 of the information processing device 1 includes a parameter setting unit (not shown), a learning data generating unit 20, an input processing unit 21, an output control unit 22, and a machine learning unit .

[0020] 1.2.1 Parameter setting section The parameter setting unit sets various parameters. The parameter setting unit may set some or all of the necessary parameters based on values ​​received by the input unit 13 of the information processing device 1 from the user, or may set some or all of the necessary parameters based on conditions or random numbers set in advance by the parameter setting unit. The parameters to be set include the number of learning data to be generated (hereinafter, the number of generated learning data), the deepest depth D of the underground structure, the number of divided layers N, the layer thickness setting coefficient t, the range of generated resistivity ρ (the maximum resistivity ρ max ≦ρ≦Minimum resistivity ρ min ), the resistivity types and their ratios used for the learning data, as well as settings for the measurement system, signal source, and measurement site information used to obtain the measurement response data to be analyzed.

[0021] 1.2.2 Learning data generation unit 20 The learning data generating unit 20 can generate learning data by associating the learning resistivity structure with the artificial response data. In the learning data generated in this embodiment, the objective variable is the resistivity structure, and the explanatory variable is the artificial response data. The learning data generating unit 20 can include a layer setting unit 20a, a division unit 20b, a resistivity setting unit 20c, a smoothing unit 20d, a synthesis unit 20e, a learning resistivity structure data generating unit 20f, a resistivity used determining unit 20g, a response data calculating unit 20h, and a noise adding unit 20i.

[0022] (1) Layer setting unit 20a The layer setting unit 20a can set a one-dimensional underground structure composed of multiple layers. Specifically, the layer structure setting unit can set the boundary depth of each underground layer by, for example, calculating the following formula.

[0023]

number

[0024] Here, D is the deepest depth input by the user via the input unit 13, N is the number of divided layers input by the user via the input unit 13, n is a natural number (n=1, 2, . . . N+1), and t is a layer thickness setting coefficient (-1≦t≦1) input by the user via the input unit 13. The method of determining the layer structure is configured to be adjustable depending on the value of the layer thickness setting coefficient t. For example, the layer structure (layer thickness) is determined exponentially when t=-1, linearly when t=0, and quadratically when t=1.

[0025] Generally, when determining a layer structure, only an exponential function is used, but in a layer structure determined only by an exponential function, the difference in layer thickness between the shallow part and the deep part is large (the layer thickness in the shallow part is small, and the layer thickness in the deep part is large). On the other hand, in a layer structure determined by the method including the quadratic function element according to the present embodiment, the difference in layer thickness between the shallow part and the deep part is small compared to a layer structure determined by an exponential function. Therefore, it is possible to generate a more detailed layer structure, i.e., a more realistic underground structure.

[0026] (2) Separation part 20b The division unit 20b can divide the layers set by the layer setting unit 20a into one or more groups each consisting of one layer or two or more consecutive layers. The division unit 20b performs division by, for example, selecting the boundary depth of the layer set by the layer setting unit 20a. For example, the number of boundaries selected is preferably 0 to 4 when the number of layers is 30, 0 to 6 when the number of layers is 50, and 0 to 11 when the number of layers is 100.

[0027] FIG. 3 shows an image of the division of the underground structure divided by the division unit 20b when the number of boundaries is 0, 1, and 2. For example, when the number of boundaries is 0, a group G1 including all layers is generated. When the number of boundaries is one, that is, when the boundary B1 is selected, the group is divided into a group G1 including layers located above the boundary B1 and a group G2 including layers located below the boundary B1. When the number of boundaries is two, that is, when the boundary B1 and the boundary B2 are selected, the group is divided into a group G1 including layers located above the boundary B1, a group G2 including layers located below the boundary B1 and above the boundary B2, and a group G3 including layers located below the boundary B2.

[0028] The division unit 20b can divide the layers by any method, but for example, it may select a boundary depth by a random number to determine the layers to be included in each group, or it may accept a boundary depth selection by a user and determine the layers to be included in each group based on the selection. In the following description, it is assumed that the method of selecting a boundary depth by a random number is used.

[0029] (3) Specific resistance setting section 20c The resistivity setting unit 20c can determine the resistivity of one or more divided groups using random numbers, and set the resistivity determined for each group to each layer included in the group. Specifically, the resistivity setting unit 20c determines the resistivity of the group based on the following formula, for example.

[0030]

number

[0031] where i is the number of groups, ρ max is the maximum resistivity set by the user, ρ min is the minimum resistivity set by the user, ρ i is the resistivity of the i-th group, and θ is a pseudorandom number. In addition, θ is determined for each group according to a uniform distribution from 0 to 1. An example of a learning resistivity structure generated using the resistivity set by the resistivity setting unit 20c is shown in the left diagram of FIG. 4 (before smoothing). In the example of FIG. 4, the number of groups i is 4, and each layer has a resistivity corresponding to the resistivity of the group determined by the above formula.

[0032] (4) Smoothing section 20d The smoothing unit 20d can smooth the resistivity of each layer (hereinafter, referred to as the resistivity before smoothing) set by the resistivity setting unit 20c and set the smoothed resistivity. Specifically, for example, the smoothing process by moving average is performed for each layer using the resistivity before smoothing of a total of three layers, that is, one layer among the multiple layers set by the layer setting unit 20a and two layers adjacent to the one layer. For the top layer and the bottom layer, the smoothing process is performed using a total of three smoothed resistivities, that is, one pre-smoothed resistivity of the layer adjacent to the top layer / bottom layer and two pre-smoothed resistivities of the top layer / bottom layer. The central diagram (after smoothing) in FIG. 4 shows an example of a learning resistivity structure generated using the smoothed resistivity set by the smoothing unit 20d.

[0033] (5) Synthesis section 20e The synthesis unit 20e can perform synthesis processing of the unsmoothed resistivity and the smoothed resistivity of the layer corresponding to the layer in which the unsmoothed resistivity is set for each layer, and set a synthetic resistivity for each layer. Here, synthesis processing refers to performing a weighted average between the unsmoothed resistivity and the smoothed resistivity of each layer. Specifically, the synthesis unit 20e performs the weighted average based on, for example, the following formula.

[0034]

number

[0035] Here, j is the number of layers, ρ' is the resistivity before smoothing, ρ'' is the smoothed resistivity, and α is a weight. α is determined for each underground structure based on a uniform distribution from 0 to 1. The degree of synthesis of the resistivity and the smoothed resistivity is determined by α, and the closer α is to 0, the closer it is to the structure before smoothing, and the closer α is to 1, the closer it is to the structure after smoothing. The right diagram (synthesis) in Figure 4 shows an example of a learning resistivity structure generated using the synthesized resistivity set by the synthesis unit 20e. In the example of Figure 4, the weight α is 0.02, and a synthesized resistivity structure that is relatively close to the structure before smoothing is displayed.

[0036] (6) Learning resistivity structure data generation unit 20f The learning resistivity structure data generating unit 20f can generate learning resistivity structure data in which an underground structure is defined based on a plurality of layers in which the unsmoothed resistivity, the smoothed resistivity, or the composite resistivity is set. Fig. 4 shows examples of learning resistivity structure data generated based on the unsmoothed resistivity, the smoothed resistivity, and the composite resistivity, respectively.

[0037] (7) Use specific resistance determining section 20g The resistivity usage determination unit 20g can determine the number of pieces of learning resistivity structure data to be used for the learning data, the number of pieces of learning resistivity structure data to be used for the learning data, based on the resistivity type and ratio to be used for the learning data input by the user via the input unit 13. Specifically, the resistivity usage determination unit 20g can determine the learning resistivity structure data to be used for the learning data, based on one or more types of learning resistivity structure data to be used as the learning data input by the user, and the ratio at which each of the one or more types of learning resistivity structure data is used among the number of generated learning data.

[0038] In addition, when the user does not set the resistivity type and ratio to be used for the learning data, the resistivity use determination unit 20g can determine the learning resistivity structure using the composite resistivity (hereinafter, composite learning resistivity structure data) as the learning resistivity structure data to be used for the learning data. In addition, when the user does not set the resistivity type and ratio to be used for the learning data, the resistivity use determination unit 20g can determine the learning resistivity structure data for the number of generated learning data to be used for the learning data based on the information of the measurement site input by the user via the input unit 13. For example, when it is input that the underground structure of the measurement site is relatively simple, the resistivity use determination unit 20g can determine to use the learning resistivity structure data using at least the resistivity before smoothing (hereinafter, unsmoothed learning resistivity structure data) among the learning resistivity structure data for the number of generated learning data to be used as the learning data, or to increase the ratio of unsmoothed learning resistivity structure data. For example, when it is input that the underground structure of the measurement site is complex, the resistivity determination unit 20g can decide to increase the proportion of learning resistivity structure data in which at least smoothed resistivity is used (hereinafter referred to as smoothed learning resistivity structure data) or smoothed ratio learning resistivity structure data among the learning resistivity structure data equal to the number of generated learning data to be used as learning data.

[0039] The types of learning resistivity structure data used may be all of pre-smoothing learning resistivity structure data, smoothed learning resistivity structure data, and synthetic learning resistivity structure data. The types of learning resistivity structure data used may be a combination of pre-smoothing learning resistivity structure data and smoothed learning resistivity structure data, a combination of pre-smoothing learning resistivity structure data and synthetic learning resistivity structure data, or a combination of smoothed learning resistivity structure data resistivity and synthetic learning resistivity structure data. Also, the types of learning resistivity structure data used may be only pre-smoothing learning resistivity structure data, only smoothed learning resistivity structure data, or only synthetic learning resistivity structure data.

[0040] (8) Response data calculation unit 20h The response data calculation unit 20h performs forward analysis based on the generated learning resistivity structure data, calculates artificial response data associated with the learning resistivity structure data, and generates learning data. Specifically, based on parameters of the measurement system and signal source set by the user when the measurement response data to be analyzed is acquired, the Maxwell equation corresponding to the parameters is solved as a partial differential equation of a cylindrical coordinate system for each learning resistivity structure data to calculate theoretical response data (= artificial response data). The artificial response data is voltage data or magnetic field data.

[0041] The response data calculation unit 20h can calculate the transmitter types including electric and magnetic dipoles, circular loops, rectangular loops, and ground wires. The response data calculation unit 20h can calculate the received components at any position, including three electric field components and three magnetic field components. The response data calculation unit 20h can also handle measurement systems placed in the air, on the ground, underground, and underwater, for both transmission and reception.

[0042] (9) Noise adding unit 20i The noise adding unit 20i can add empirically calculated noise or environmental noise present during geophysical exploration measurement to the learning data generated by the response data calculation unit 20h, calculate noise-added artificial response data associated with the learning resistivity structure data, and generate noise-added learning data. Preferably, the noise adding unit 20i adds empirically calculated noise.

[0043] Specifically, the noise adding unit 20i calculates the standard deviation of the measurement response data obtained by the geophysical exploration measurement, and sets this as the noise level. When adding empirically calculated noise, the noise adding unit 20i adds Gaussian distribution noise to the artificial response data of the learning data. The mean of the Gaussian distribution is 0, and the standard deviation is the standard deviation of the measurement response data described above.

[0044] When adding environmental noise, the noise adding unit 20i uniformly adds the standard deviation of the above-mentioned measured response data to the artificial response data of the learning data.

[0045] Measurement response data from geophysical exploration contains noise due to vibrations of the equipment during measurement and surrounding environmental noise (electromagnetic noise from metal objects, power lines, etc.), so by training using data with noise added, it is possible to obtain a more accurate machine learning network.

[0046] The learning data generated by the learning data generating unit 20 according to this embodiment uses different types of resistivity, such as unsmoothed resistivity, smoothed resistivity, and composite resistivity, making it possible to generate various resistivity structures. As a result, the true underground structure can be estimated more accurately when this learning data is used than when conventional learning data generated using only unsmoothed resistivity is used. Here, "high accuracy" means that the error function when analysis is performed using verification data and the relative percentage error between the artificial response data calculated from the estimated resistivity structure and the measured response data are small.

[0047] 1.2.3 Input processing unit 21 The input processing unit 21 can input measurement response data obtained from an external exploration device via the communication unit 12, or measurement response data directly input via the input unit 13, into a machine learning model possessed by the machine learning unit 23.

[0048] 1.2.4 Output control section 22 The output control unit 22 can output the estimated resistivity structure data of the geodesic underground from which the measurement response data input by the input processing unit 21 was obtained, which is output from the machine learning model possessed by the machine learning unit 23, and display it on the output unit 14.

[0049] 1.2.5 Machine Learning Department 23 The machine learning unit 23 learns based on the learning data generated by the learning data generation unit 20, thereby constructing a machine learning model that outputs an estimated resistivity structure underground at a measurement site where the measurement response data input by the input processing unit 21 was measured, and holds the constructed machine learning model. The machine learning model can be a model constructed by machine learning the artificial response data associated with the learning resistivity structure data as the learning data. Also, the machine learning model can be a model constructed by machine learning the noise-added artificial response data associated with the learning resistivity structure data as the noise-added learning data.

[0050] The internal structure of the machine learning model is arbitrary, but it can be composed of, for example, a DNN (Deep Neural Network), a CNN (Convolutional Neural Network), an RNN (Recurrent Neural Network), or an LSTM (Long Short-Term Memory).

[0051] 1.3 Data processing procedure using the information processing device 1 A data processing method using the information processing device 1 according to this embodiment will be described with reference to Fig. 5. The data processing method includes a parameter setting step S1, a learning data generating step S2, a network model constructing step S3, a measurement response data reading step S4, an inverse analysis step S5, and an estimated resistivity structure output step S6.

[0052] (1) Parameter setting process S1 In the parameter setting step S1, the parameter setting unit sets various parameters. The parameter setting unit may set some or all of the necessary parameters based on values ​​received by the input unit 13 of the information processing device 1 from the user, or the parameter setting unit may set some or all of the necessary parameters based on preset conditions, random numbers, etc. The parameters to be set include the number of generated learning data, the deepest depth D of the underground structure which is the objective variable of the learning data, the number of divided layers N, the layer thickness setting coefficient t, the range of generated resistivity ρ (maximum resistivity ρ max ≦ρ≦Minimum resistivity ρ min ), resistivity types and their ratios used in the learning data, and the measurement system, signal source, and measurement value information used to obtain the measurement response data to be analyzed. An example of the procedure for setting each parameter is shown below.

[0053] The number of pieces of generated learning data can be any number. The number of pieces of generated learning data can be set by the parameter setting unit based on a value received by the input unit 13 of the information processing device 1 from the user. The number of pieces of generated learning data is preferably 40,000 to 1,200,000, and the parameter setting unit may set the number of pieces of generated learning data within this range using random numbers or the like.

[0054] The parameter setting unit calculates a DOI (Depth of Investigation) that applies the concept of skin depth, and the maximum depth D can be set based on the calculated value. Alternatively, the calculated value may be presented to the user via the output unit 14 of the information processing device 1. In this case, the input unit 13 of the information processing device 1 can accept input from the user, and the maximum depth D can be set based on the accepted value. Since the DOI varies depending on the measurement system and the underground resistivity of the measured value, the maximum depth D input by the user can be empirically set by the user based on a combination of various conditions such as the measurement system and measurement site of the analysis target.

[0055] The number of divided layers N can be any number. The number of divided layers N can be set by the user based on a value received by the input unit 13 of the information processing device 1. The number of divided layers N is often set to 10 to 100, and particularly often set to 20 to 50. The parameter setting unit may set the number of divided layers N within this range using a random number or the like.

[0056] The layer thickness setting coefficient t can be set by the user based on a value received by the input unit 13 of the information processing device 1. The layer thickness setting coefficient t may be determined by the parameter setting unit using a random number or the like between −1 and 1, but the layer thickness setting coefficient t is preferably determined to be 1.

[0057] The range of the generation resistivity ρ can be set to any range. The range of the generation resistivity ρ can be set by the user based on a value received by the input unit 13 of the information processing device 1. The range of the generation resistivity ρ is 10 -2 ≦ρ≦10 4 Therefore, the parameter setting unit may set the range of the generation resistivity ρ to this range in advance.

[0058] The resistivity types and their ratios to be used for the learning data can be set arbitrarily. Specifically, one or more types of learning resistivity structure data to be used as learning data and the ratios of the one or more types of learning resistivity structure data to be used in the generated learning data can be set by the user based on values ​​received by the input unit 13 of the information processing device 1.

[0059] When a reanalysis is performed, which will be described later, in the parameter setting step S1, the parameter setting unit sets the maximum depth D, the number of divided layers N, the layer thickness setting coefficient t, and the range of the generated resistivity ρ (the maximum resistivity ρ max ≦ρ≦Minimum resistivity ρ min ), and the parameters can be reset based on the resistivity types and their ratios used in the learning data. The various parameters input by the user can be adjusted by the user based on the results of the previous analysis.

[0060] (2) Learning data generation process S2 When the parameter setting process S1 is completed, the learning data generation process S2 is started. In the learning data generation process S2, the learning data generation unit 20 generates learning data in which the resistivity structure and the artificial response data are associated with each other. The learning data generation process S2 includes a layer setting process S21, a division process S22, a resistivity setting process S23, a smoothing process S24, a synthesis process S25, a learning resistivity structure data generation process S26, a response data calculation process S28, and a noise addition process S29. Figure 6 shows the flow of the learning data generation process S2.

[0061] In the layer setting step S21, the layer setting unit 20a sets an underground structure consisting of multiple layers based on the deepest depth D, the number of divided layers N, and the layer thickness setting coefficient t set in the parameter setting step S1. In the layer setting step S21, the layer setting unit 20a sets the underground structure for the number of generated learning data. When the layer setting step S21 is completed, the process proceeds to the division step S22.

[0062] In the division step S22, the division unit 20b divides the layers set in the layer setting step S21 in one underground structure into one or more groups each consisting of one layer or two or more consecutive layers, and performs this process for the number of generated learning data. After the division step S22 is completed, the process proceeds to the resistivity setting step S23.

[0063] In the resistivity setting step S23, the resistivity setting unit 20c determines the resistivity of one or more groups in one underground structure divided by the division step S22, and sets the resistivity (pre-smoothing resistivity) of one or more layers included in each group. In the resistivity setting step S23, the resistivity setting unit 20c sets resistivities for the number of generated learning data. When the resistivity setting step S23 is completed, the process proceeds to the smoothing step S24.

[0064] In the smoothing step S24, the smoothing unit 20d smooths the unsmoothed resistivity of each layer in one underground structure set by the resistivity setting unit 20c, and sets the smoothed resistivity. In the smoothing step S24, the smoothing unit 20d sets the smoothed resistivity for the number of generated learning data. When the smoothing step S24 is completed, the process proceeds to the synthesis step S25.

[0065] In the synthesis step S25, the synthesis unit 20e performs synthesis processing using the unsmoothed resistivity and the smoothed resistivity in one underground structure to set a synthetic resistivity. In the synthesis step S25, the synthesis unit 20e sets synthetic resistivities for the number of generated learning data. When the synthesis step S25 is completed, the process proceeds to the learning resistivity structure data generation step S26.

[0066] In the learning resistivity structure data generating step S26, the learning resistivity structure data generating unit 20f generates learning resistivity structure data in which an underground structure is defined based on a plurality of layers in one underground structure in which unsmoothed resistivity, smoothed resistivity, or composite resistivity is set, and this is performed for the number of generated learning data. When the learning resistivity structure data generating step S26 is completed, the process proceeds to the resistivity to be used determination step S27.

[0067] In the resistivity determination step S27, the resistivity determination unit 20g determines the resistivity structure data for learning, the number of which corresponds to the number of generated learning data to be used for the learning data, based on the resistivity type and ratio or measurement value information to be used for the learning data set in the parameter setting step S1. Specifically, the resistivity determination unit 20g determines the resistivity structure data for learning, which is used for the learning data and is configured as any one of the following (1) to (7): (1) Training resistivity structure data using unsmoothed training resistivity structure data, smoothed training resistivity structure data, and synthetic training resistivity structure data (2) Training resistivity structure data using unsmoothed training resistivity structure data and smoothed training resistivity structure data (3) Training resistivity structure data using unsmoothed training resistivity structure data and synthetic training resistivity structure data (4) Smoothed learning resistivity structure data Learning resistivity structure data using resistivity and synthetic learning resistivity structure data (5) Training resistivity structure data using unsmoothed training resistivity structure data (6) Training resistivity structure data using smoothed training resistivity structure data (7) Training resistivity structure data using synthetic training resistivity structure data When the use resistivity determination step S27 is completed, the process proceeds to a response data generation step.

[0068] In the response data calculation step S28, the response data calculation unit 20h performs a forward analysis based on one of the generated learning resistivity structure data among the learning resistivity structure data determined in the resistivity determination step S27, calculates artificial response data associated with the learning resistivity structure data, and generates learning data. In the response data calculation step S28, the response data calculation unit 20h calculates artificial response data for all the learning resistivity structure data determined in the resistivity determination step S27, thereby generating learning data for the number of generated learning data. When the response data calculation step S28 is completed, if noise is to be added, the process proceeds to the noise addition step S29, and if noise is not to be added, the learning data generation step S2 is completed.

[0069] In the noise adding step S29, the noise adding unit 20i adds empirically calculated noise to the artificial response data of the learning data generated by the response data calculation unit 20h to generate noise-added learning data. When the noise adding step S29 is completed, the learning data generation step S2 is completed.

[0070] (3) Network model construction process S3 When the learning data generation step S2 is completed, the network model construction step S3 is started. In the network model construction step S3, the machine learning unit 23 performs machine learning using the learning data generated in the learning data generation step S2 or the noise-added learning data generated in the noise-adding step S29 to construct a machine learning model that outputs an estimated resistivity structure of the input measurement response data. Specifically, in the network model construction step S3, supervised learning is performed in which the artificial response data or the noise-added artificial response data is input (explanatory variable), resistivity structure data is output (target variable), and the learning resistivity structure data associated with the artificial response data or the noise-added artificial response data is the correct answer label. By constructing such a machine learning model, it is possible to estimate the resistivity structure corresponding to the input measurement response data.

[0071] (4) Measurement response data reading process S4 The measurement response data reading step S4 may be performed at any timing as long as it can be performed before the inverse analysis step S5 is started. In Fig. 5, the measurement response data reading step S4 is shown to start after the network model construction step S3 is completed, and the following description will be given assuming that the steps are performed in this order. In the measurement response data reading step S4, the input processing unit 21 inputs the measurement response data into the machine learning model constructed by the network model construction step S3.

[0072] (5) Back analysis process S5 When the measurement response data reading step S4 is completed, the inverse analysis step S5 is started. In the inverse analysis step S5, the machine learning unit 23 performs an analysis using the measurement response data input to the machine learning unit 23, and outputs estimated resistivity structure data.

[0073] (6) Estimated resistivity structure output process S6 When the inverse analysis step S5 is completed, the estimated resistivity structure output step S6 starts. In the estimated resistivity structure output step S6, the output control unit 22 displays on the output unit 14 the estimated resistivity structure data output in the inverse analysis step S5.

[0074] When the estimated resistivity structure output step S6 is completed, the data processing according to the present invention is completed. If the desired analysis result is not obtained based on the analysis result output from the output unit 14, the process returns to the parameter setting step S1, and the above steps are repeated to perform reanalysis.

[0075] The reanalysis may be configured, for example, to calculate artificial response data based on the resistivity structure output in the estimated resistivity structure output process S6, and to be performed when the relative percentage error between the artificial response data and the analyzed measured response data is a predetermined value or more (e.g., 50% or more) across all measurement points.

[0076] In the data processing method according to the present embodiment, analysis using machine learning is performed, which makes it possible to shorten the analysis time compared to conventional optimization methods such as the nonlinear least squares method. In particular, when an actual electromagnetic exploration measurement is performed, the analysis takes time using conventional analysis methods, so the measurement is evaluated using a hypothetical resistivity structure such as apparent resistivity. Data processing using the present method is advantageous because it makes it possible to estimate the resistivity structure in real time.

[0077] 2. Second embodiment The inverse analysis system according to this embodiment includes an exploration device 3. The inverse analysis system according to this embodiment differs from the first embodiment in that the functions of the information processing device 1 are incorporated in the exploration device 3. The following description will focus on the differences from the first embodiment.

[0078] 2.1 Hardware configuration of exploration device 3 7 shows the hardware configuration of the exploration device 3 according to this embodiment. The exploration device 3 includes an electromagnetic field control unit 30, a transmission unit 31, a reception unit 32, an information processing control unit 33, a storage unit 34, an input unit 36, and an output unit 37. The electromagnetic field control unit 30, the transmission unit 31, the reception unit 32, the information processing control unit 33, the storage unit 34, the input unit 36, and the output unit 37 are connected to each other by a communication bus 40. The exploration device 3 may further include a communication unit 35.

[0079] The electromagnetic field control unit 30 may include a processor such as a central processing unit (CPU), a digital signal processor, a microprocessor, a microcontroller, an application specific integrated circuit (ASIC), a programmable logic device (PLD), or the like.

[0080] The transmitter 31 generates a magnetic field determined by the electromagnetic field control unit 30. The transmitter 31 may include any transmitter that can be used as a transmitter for electromagnetic exploration. The transmitter 31 may be, for example, a coil, a grounded electric wire, or a loop electric wire.

[0081] The receiving unit 32 can receive secondary magnetic field data or voltage data, which is response data from underground. The receiving unit 32 can include any receiver that can be used as a receiver for electromagnetic exploration. The receiving unit 32 can be, for example, a coil when measuring voltage data, or a magnetic field sensor such as a fluxgate magnetometer, an MI magnetometer, or a SQUID magnetometer when measuring magnetic field data.

[0082] The information processing control unit 33, the storage unit 34, the communication unit 35, and the output unit 37 may have the same configuration as the control unit 10, the storage unit 11, the communication unit 12, and the output unit 14 of the information processing device 1.

[0083] In addition to various parameters set by the user in the parameter setting step S1 of the first embodiment, a current value, frequency, voltage, etc. used in the measurement can be input to the input unit 36. The input unit 36 ​​can include, for example, one or more of a keyboard, a keypad, a mouse, a microphone, a touch screen, a button, etc.

[0084] 2.2 Functional configuration of exploration device 3 The electromagnetic field control unit 30 of the exploration device 3 has a function of controlling the transmission unit 31 based on the current value, frequency, voltage, etc. input by the user via the input unit 36.

[0085] The information processing control unit 33 of the exploration device 3 can have a functional configuration similar to the control unit 10 of the information processing device 1, but the input processing unit can input the measurement response data acquired by the receiving unit 32 into a machine learning model possessed by the machine learning unit.

[0086] 3. Other embodiments 3.1 Computer Programs A computer program according to one embodiment of the present invention is a program for causing a computer to execute the parameter setting step S1, learning data generating step S2, network model constructing step S3, measurement response data reading step S4, inverse analysis step S5, and estimated resistivity structure output step S6, which are performed by the information processing device 1 in the data processing method of the first embodiment. The program for the learning data generating step S2 executed by the computer also includes a layer setting step S21, a division step S22, a resistivity setting step S23, a smoothing step S24, a synthesis step S25, a learning resistivity structure data generating step S26, a resistivity to be used determining step S27, a response data calculating step S28, and a noise adding step S29.

[0087] 3.2 Information processing device 1 The information processing device 1 can also be realized as a computer-readable non-transitory recording medium storing a program relating to the information processing functions of each embodiment, or a computer-readable recording medium having the above-mentioned program recorded therein.

[0088] 3.3 Data processing methods A data processing method according to an embodiment of the present invention can include a machine learning setting step after the learning data generating step S2 in the data processing method of the first embodiment. In the machine learning setting step, a user can select a machine learning framework to be used in data processing and set a network.

[0089] 3.4 Inverse analysis system The inverse analysis system according to one embodiment of the present invention can calculate a two-dimensional or three-dimensional estimated resistivity structure. In this case, it can be realized by expanding the data generated by the learning data generating unit 20 in the first embodiment as one of the grid data to two-dimensional or three-dimensional learning data.

[0090] 4. Working Example The following describes the results of verification of a network model using the information processing device 1 according to the present invention.

[0091] 4.1 Overview In this verification, an airborne frequency domain electromagnetic survey method towed by a helicopter was assumed, and the constructed network model was evaluated using verification data. In addition, a comparison was made between the case where the data processing method of the present invention was used and the case where a conventional data processing method using only a simple structure was used, using an error function (mean square error of the objective variable (resistivity structure)). In the data processing method of the present invention, two types of resistivity structure data for learning, namely unsmoothed resistivity structure data and smoothed resistivity structure data, were used to generate noisy learning data with empirically calculated noise added.

[0092] The parameter settings used in this verification are shown in Table 1. An example of the learning resistivity structure data used in the data processing method according to the present invention is shown in Fig. 8, and an example of the learning resistivity structure data used in the conventional data processing method is shown in Fig. 9. [Table 1]

[0093] 4.2 Results The learning curve and validation error for the data processing method of the present invention are shown in FIG. 10, and the learning curve and validation error for the conventional data processing method are shown in FIG. 11. From FIG. 10, it was confirmed that the mean square error of the validation data decreased with learning in the network model constructed by the data processing method of the present invention, and overlearning did not occur. In addition, in terms of comparison with the conventional method, the mean square error in the present invention was 0.16 and the standard deviation was 0.18, while the mean square error in the conventional method was 0.21 and the standard deviation was 0.31. Therefore, it was confirmed that the data processing method of the present invention resulted in a smaller error function than the conventional method, and a more accurate network model was constructed. [Explanation of symbols]

[0094] 1: information processing device, 3: exploration device, 10: control unit, 11: storage unit, 12: communication unit, 13: input unit, 14: output unit 20: learning data generation unit, 21: input processing unit, 22: output control unit, 23: machine learning unit, 20a: layer setting section, 20b: division section, 20c: resistivity setting section, 20d: smoothing section, 20e: synthesis section, 20f: learning resistivity structure data generation section, 20g: resistivity used determination section, 20h: response data calculation section, 20i: noise addition section 30: electromagnetic field control unit, 31: transmission unit, 32: reception unit, 33: information processing control unit, 34: storage unit, 35: communication unit, 36: input unit, 37: output unit, 40: Communication bus B1, B2: Boundary, G1, G2, G3: Group S1: Parameter setting process, S2: Learning data generation process, S3: Network model construction process, S4: Measurement response data reading process, S5: Inverse analysis process, S6: Estimated resistivity structure output process, S21: Layer setting process, S22: Division process, S23: Resistivity setting process, S24: Smoothing process, S25: Synthesis process, S26: Learning resistivity structure data generation process, S27: Resistivity to be used determination process, S28: Response data calculation process, S29: Noise addition process

Claims

1. An information processing device, The method includes a layer setting unit, a division unit, a resistivity setting unit, a smoothing unit, and a learning resistivity structure data generating unit, The layer setting unit sets an underground structure composed of a plurality of layers, The dividing unit divides the plurality of layers into one or more groups each composed of one layer or two or more consecutive layers, the resistivity setting unit determines a resistivity for each of the one or more groups using a random number, and sets the resistivity determined for each group to each layer included in the group; the smoothing unit performs a smoothing process for each layer using a resistivity of one layer among the plurality of layers in which the resistivity is set and a resistivity of a layer adjacent to the one layer, and sets a smoothed resistivity for each layer; The learning resistivity structure data generation unit generates learning resistivity structure data in which an underground structure is defined based on the plurality of layers in which the resistivity or the smoothed resistivity is set, in an information processing device.

2. 2. The information processing device according to claim 1, Further comprising a synthesis unit, the synthesis unit performs synthesis processing for each layer of the resistivity and a smoothed resistivity of a layer corresponding to the layer in which the resistivity is set, and sets a synthetic resistivity for each layer; The learning resistivity structure data generation unit generates learning resistivity structure data in which an underground structure is defined based on the plurality of layers in which the composite resistivity is set, in an information processing device.

3. 3. The information processing device according to claim 1, An information processing apparatus, wherein the depth of each of the plurality of layers is determined by a quadratic function.

4. 4. The information processing device according to claim 3, The apparatus further includes a learning data generating unit, a machine learning unit, an input processing unit, and an output control unit. the learning data generating unit generates learning data by associating the learning resistivity structure data with artificial response data calculated based on the learning resistivity structure data; The machine learning unit has a machine learning model that has been machine-learned using the learning data, The input processing unit inputs measurement response data acquired by geophysical exploration measurements of the underground of a measurement site into the machine learning model, The output control unit is an information processing device that outputs estimated resistivity structure data of the underground of the measurement site from which the measurement response data was obtained, which is output from the machine learning model.

5. 5. The information processing device according to claim 4, The artificial response data and the measured response data are voltage data or magnetic field data.

6. 5. The information processing device according to claim 4, The information processing device, wherein the geophysical exploration measurement is electromagnetic exploration measurement.

7. 5. The information processing device according to claim 4, The information processing device, wherein the machine learning unit has the machine learning model that is machine-learned using the noise-added artificial response data.

8. The information processing device according to claim 7, The noise is empirically calculated noise or noise calculated based on environmental noise present at the time of measurement of the measurement response data.

9. 1. A detection device, comprising: An exploration device comprising the information processing device according to claim 1.

10. A program, causing a computer to execute a layer setting step, a division step, a resistivity setting step, a smoothing step, and a learning resistivity structure data generating step; In the layer setting step, an underground structure composed of a plurality of layers is set, In the dividing step, the plurality of layers are divided into one or a plurality of groups each composed of one layer or two or more consecutive layers; In the resistivity setting step, a resistivity is generated for each of the one or more groups by using random numbers, and the resistivity generated for each group is set for each layer included in each group; In the smoothing step, a smoothing process is performed for each layer using a resistivity of one of the plurality of layers in which the resistivity is set and a resistivity of a layer adjacent to the one layer, and a smoothed resistivity is set for each layer; In the learning resistivity structure data generating step, learning resistivity structure data is generated in which a subsurface structure is defined based on the plurality of layers in which the resistivity or the smoothed resistivity is set. program.

Citation Information

Patent Citations

  • Magnetotelluric data based resistivity and magnetic susceptibility inversion method and system

    CN104102814A

  • Fill dam control system by specific resistance tomography method and its control method

    JP2000338258A

  • Underground electromagnetic survey device

    JP2018194454A

  • Electromagnetic exploration device, Electromagnetic exploration method and Computer-Readable Medium recording program

    KR1020090036340A

  • Adaptive time-lapse sub-surface electrical resistivity monitoring

    US20150006081A1