Deep learning sample set determination method and system based on full-space transient electromagnetic method

Abstract

The application provides a deep learning sample set determination method and system based on full-space transient electromagnetic method. The method comprises the following steps: acquiring time-induction electromotive force data of each transient electromagnetic measured point in a well at each time in a preset period, and pre-processing the data to obtain corrected induction electromotive force data; determining the formation layer thickness and formation resistivity of a plurality of model measuring points according to the corrected induction electromotive force data; performing one-dimensional forward processing on the full-space layered medium model to obtain time-induction electromotive force response data of the plurality of model measuring points, and extracting time-domain features; taking the time-domain features as an input set, taking the formation layer thickness and formation resistivity of the plurality of model measuring points as an output set, and constructing a deep learning sample set. The technical scheme provided by the application can eliminate the deviation between the measured data and the forward data, make the data set more representative, and further ensure the accuracy of subsequent model training.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of coal mine water-bearing geological body detection, and particularly relates to a deep learning sample set determination method and system based on full-space transient electromagnetic method. BACKGROUND

[0002] The mine full-space transient electromagnetic method adopts a multi-turn small loop to perform transmission and reception in a mine roadway, generates a primary field by applying a current to the transmission loop, and then turns off the current, measures the secondary field induced by the electrical heterogeneity in the coal and rock mass over time during the transmission interval, and is mainly used to solve the problems of coal mine underground water-bearing geological anomaly detection, water disaster prediction and other geological problems.

[0003] At present, the underground transient electromagnetic data acquisition mainly measures the secondary field induced electromotive force after the pulse current is normalized. The induced electromotive force data in the logarithmic time sequence is a curve that decays over time in the two-dimensional coordinate system. The stratum depth and resistivity data can be obtained by data processing. The current large-scale data processing method applied in coal mines mainly includes linear methods such as smoke ring inversion and time-depth conversion algorithm. Various nonlinear inversion algorithms such as neural networks have attracted more and more attention because they can mine deep features contained in the data. The coal mine underground geophysical inversion method based on deep learning has been rapidly developed in recent years. However, there are still some problems in applying it to full-space transient electromagnetic detection: 1) Deep learning has high requirements for data, and usually requires a large number of sample data for training and testing. However, the geological conditions in coal mines are complex, and it is difficult to obtain relatively accurate layer thickness division and resistivity data of layered coal and rock mass. It is difficult to obtain a large number of effective samples based on measured data, and the data set is difficult to construct. 2) The measured data contains complex human factors such as metal bodies and power in the mine roadway. In addition, due to the multi-turn loop of the underground transient electromagnetic method, the wire frame and inductance effect will affect the shape of the induced electromotive force curve, which deviates from the theoretical data, and also indirectly causes the deviation between the measured data and the forward data set, and the forward data is not representative. SUMMARY

[0004] The present application provides a deep learning sample set determination method and system based on full-space transient electromagnetic method to at least solve the technical problems of difficult construction of data sample set and lack of representativeness.

[0005] The first aspect embodiment of the present application provides a deep learning sample set determination method based on full-space transient electromagnetic method, which comprises:

[0006] The time-induction electromotive force data of each transient electromagnetic measured measuring point in the well at each time in the preset period is obtained, and the time-induction electromotive force data of each transient electromagnetic measured measuring point in the well at each time in the preset period is preprocessed to obtain the corrected induction electromotive force data of each transient electromagnetic measured measuring point in the well at each time in the preset period.

[0007] The formation layer thickness and formation resistivity of a plurality of model measuring points in the full-space layered medium model are determined according to the corrected induction electromotive force data of the measured measuring point, and the formation layer thickness and formation resistivity of the plurality of model measuring points are taken as parameters of the full-space layered medium model.

[0008] The full-space layered medium model is subjected to one-dimensional forward processing to obtain time-induction electromotive force response data of a plurality of model measuring points in the full-space layered medium model.

[0009] Time-domain features are extracted based on the time-induction electromotive force response data of the plurality of model measuring points to obtain induction electromotive force, apparent resistivity, induction electromotive force change rate, and apparent resistivity change rate time-domain features.

[0010] The induction electromotive force, apparent resistivity, induction electromotive force change rate, and apparent resistivity change rate time-domain features of the plurality of model measuring points are taken as an input set, and the formation layer thickness and formation resistivity of the plurality of model measuring points are taken as an output set to form a deep learning sample set.

[0011] Preferably, the pre-processing of the time-induction electromotive force data of each transient electromagnetic measured measuring point in the well at each time in the preset period to obtain the corrected induction electromotive force data of each transient electromagnetic measured measuring point in the well at each time in the preset period comprises:

[0012] The time-induction electromotive force data of each transient electromagnetic measured measuring point in the well at each time in the preset period is normalized to obtain normalized induction electromotive force data of each transient electromagnetic measured measuring point in the well at each time in the preset period.

[0013] The corrected induction electromotive force data of each transient electromagnetic measured measuring point in the well at each time in the preset period is determined based on the normalized induction electromotive force data of the measured measuring point.

[0014] Further, the calculation formula of the corrected induction electromotive force data of each transient electromagnetic measured measuring point in the well at each time in the preset period is as follows:

[0015] D ij =V ij ·α i

[0016] In the formula, D ijVij(i) is the corrected induction electromotive force data of the jth transient electromagnetic measured point in the well at the ith time in the preset period ij Vij(i) is the normalized induction electromotive force data of the jth transient electromagnetic measured point in the well at the ith time in the preset period i is the correction coefficient at the ith time.

[0017] Preferably, the method further comprises:

[0018] performing inversion calculation on the corrected induction electromotive force data of the measured points to obtain the layer thickness and the corresponding layer resistivity of each measured point;

[0019] determining the standard deviation and expectation of the layer thickness and the standard deviation and expectation of the corresponding layer resistivity of each measured point according to the layer thickness and the corresponding layer resistivity of each measured point;

[0020] determining the normal distribution probability density function of the layer thickness and the normal distribution probability density function of the corresponding layer resistivity of each measured point according to the standard deviation and expectation of the layer thickness and the standard deviation and expectation of the corresponding layer resistivity of each measured point;

[0021] randomly generating the layer thickness and the corresponding layer resistivity of M model points based on the normal distribution probability density function of the layer thickness and the normal distribution probability density function of the corresponding layer resistivity.

[0022] Further, the calculation formula of the standard deviation of the layer thickness of each measured point is as follows:

[0023]

[0024] In the formula, σ h is the standard deviation of the layer thickness of the measured point, n is the total number of measured points, l is the number of layers of the full-space layered medium model, h jk is the layer thickness of the kth layer corresponding to the jth measured point, is the layer thickness mean value;

[0025] The calculation formula of the expectation of the layer thickness of each measured point is as follows:

[0026]

[0027] In the formula, μ h is the expectation of the layer thickness of the measured point;

[0028] The calculation formula of the standard deviation of the layer resistivity of each measured point is as follows:

[0029]

[0030] wherein σ ρ is the standard deviation of the formation resistivity of the measured point, ρ jk is the formation resistivity of the kth layer corresponding to the jth measured point, is the mean value of the formation resistivity.

[0031] The calculation formula of the expected formation resistivity of the measured point is as follows:

[0032]

[0033] wherein μ ρ is the expected formation resistivity of the measured point.

[0034] Further, the calculation formula of the normal distribution probability density function of the formation thickness of the measured point is as follows:

[0035]

[0036] wherein f(h) is the normal distribution probability density function of the formation thickness, and h is the formation thickness.

[0037] The calculation formula of the normal distribution probability density function of the formation resistivity is as follows:

[0038]

[0039] wherein f(ρ) is the normal distribution probability density function of the formation resistivity, and ρ is the formation resistivity.

[0040] Further, the time-domain features are extracted based on the time-induced electromotive force response data of the plurality of model measuring points, and the induced electromotive force, the apparent resistivity, the induced electromotive force change rate, and the apparent resistivity change rate time-domain features are obtained, including:

[0041] The time-induced electromotive force response data of the plurality of model measuring points are taken as input data, and the corresponding formation thickness and resistivity model data of the plurality of model measuring points are taken as output data to form an initial deep learning sample set.

[0042] The induced electromotive force, the apparent resistivity, the induced electromotive force change rate, and the apparent resistivity change rate time-domain features are extracted in the initial deep learning sample set to obtain the induced electromotive force, the apparent resistivity, the induced electromotive force change rate, and the apparent resistivity change rate time-domain features.

[0043] The second aspect embodiment of the present application proposes a deep learning sample set determination system based on full-space transient electromagnetic method, and the system comprises:

[0044] The pre-processing module is configured to acquire time-induction electromotive force data of each transient electromagnetic measured measuring point in a well at each time in a preset period, and pre-process the time-induction electromotive force data of each transient electromagnetic measured measuring point in the well at each time in the preset period to obtain corrected induction electromotive force data of each transient electromagnetic measured measuring point in the well at each time in the preset period.

[0045] The first determining module is configured to determine stratum layer thickness and stratum resistivity of a plurality of model measuring points in the full-space layered medium model according to the corrected induction electromotive force data of the measured measuring point, and take the stratum layer thickness and the stratum resistivity of the plurality of model measuring points as parameters of the full-space layered medium model.

[0046] The first processing module is configured to perform one-dimensional forward processing on the full-space layered medium model to obtain time-induction electromotive force response data of the plurality of model measuring points in the full-space layered medium model.

[0047] The feature extraction module is configured to extract time-domain features based on the time-induction electromotive force response data of the plurality of model measuring points to obtain induction electromotive force, apparent resistivity, induction electromotive force change rate, and apparent resistivity change rate time-domain features.

[0048] The constituting module is configured to take the induction electromotive force, the apparent resistivity, the induction electromotive force change rate, and the apparent resistivity change rate time-domain features of the plurality of model measuring points as an input set, and take the stratum layer thickness and the stratum resistivity of the plurality of model measuring points as an output set to constitute a deep learning sample set.

[0049] Preferably, the pre-processing module is further configured to:

[0050] The pre-processing module is further configured to:

[0051] The pre-processing module is further configured to:

[0052] Further, the feature extraction module comprises:

[0053] The initial construction unit is configured to take the time-induction electromotive force response data of the plurality of model measuring points as input data, and take corresponding stratum layer thickness and resistivity model data of the plurality of model measuring points as output data to constitute an initial deep learning sample set.

[0054] The extraction unit is configured to extract the time-domain characteristics of the induced electromotive force, apparent resistivity, induced electromotive force change rate and apparent resistivity change rate in the initial deep learning sample set, and obtain the time-domain characteristics of the induced electromotive force, apparent resistivity, induced electromotive force change rate and apparent resistivity change rate.

[0055] The technical scheme provided by the embodiment of the present application at least brings the following beneficial effects:

[0056] The present application provides a deep learning sample set determination method and system based on full-space transient electromagnetic method, wherein the method comprises: obtaining time-induced electromotive force data of each transient electromagnetic measured point in a well at each time in a preset period, and preprocessing the time-induced electromotive force data of each transient electromagnetic measured point in a well at each time in the preset period to obtain corrected induced electromotive force data of each transient electromagnetic measured point in a well at each time in the preset period; determining the formation layer thickness and formation resistivity of a plurality of model measuring points in a full-space layered medium model according to the corrected induced electromotive force data of the measured points, and taking the formation layer thickness and formation resistivity of the plurality of model measuring points as parameters of the full-space layered medium model; performing one-dimensional forward processing on the full-space layered medium model to obtain time-induced electromotive force response data of the plurality of model measuring points in the full-space layered medium model; extracting time-domain characteristics based on the time-induced electromotive force response data of the plurality of model measuring points to obtain induced electromotive force, apparent resistivity, induced electromotive force change rate and apparent resistivity change rate time-domain characteristics; taking the induced electromotive force, apparent resistivity, induced electromotive force change rate and apparent resistivity change rate time-domain characteristics of the plurality of model measuring points as an input set, taking the formation layer thickness and formation resistivity of the plurality of model measuring points as an output set, and constructing a deep learning sample set. The technical scheme provided by the present application can eliminate the deviation between measured data and forward data, make the data set more representative, and thus ensure the accuracy of subsequent model training.

[0057] The additional aspects and advantages of the present application will be partially given in the following description, partially become obvious from the following description, or be known by the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0058] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:

[0059] Figure 1 A flowchart of a deep learning sample set determination method based on full-space transient electromagnetic method according to an embodiment of the present application is provided;

[0060] Figure 2 A data structure diagram of a full-space layered medium model according to an embodiment of the present application is provided;

[0061] Figure 3 This is a structural diagram of a deep learning sample set determination system based on the full-space transient electromagnetic method according to an embodiment of this application;

[0062] Figure 4 This is a structural diagram of a feature extraction module provided according to an embodiment of this application. Detailed Implementation

[0063] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0064] This application proposes a method and system for determining a deep learning sample set based on the full-space transient electromagnetic method. The method includes: acquiring time-induced electromotive force (EMF) data from downhole transient electromagnetic measurement points at each moment within a preset time period; preprocessing the time-induced EMF data from these measurement points to obtain corrected EMF data; determining the formation thickness and resistivity of multiple model measurement points in a full-space layered medium model based on the corrected EMF data; and setting the formation thickness and resistivity of the multiple model measurement points... The parameters of the full-space layered medium model are used as follows: One-dimensional forward modeling is performed on the full-space layered medium model to obtain time-induced electromotive force (EMF) response data at multiple model measurement points; based on the time-induced EMF response data of the multiple model measurement points, time-domain features are extracted to obtain time-domain features of induced EMF, apparent resistivity, rate of change of induced EMF, and rate of change of apparent resistivity; using the time-domain features of induced EMF, apparent resistivity, rate of change of induced EMF, and rate of change of apparent resistivity at multiple model measurement points as the input set, and the formation thickness and formation resistivity at multiple model measurement points as the output set, a deep learning sample set is constructed. The technical solution proposed in this application can eliminate the deviation between measured data and forward modeling data, making the dataset more representative, thereby ensuring the accuracy of subsequent model training.

[0065] The following describes, with reference to the accompanying drawings, a method and system for determining a deep learning sample set based on the full-space transient electromagnetic method, according to embodiments of this application.

[0066] Example 1

[0067] Figure 1 This is a flowchart illustrating a method for determining a deep learning sample set based on the full-space transient electromagnetic method, according to an embodiment of this application. Figure 1As shown, the method includes:

[0068] Step 1: Obtain the time-induced electromotive force data of each transient electromagnetic measurement point downhole at each moment within a preset time period, and preprocess the time-induced electromotive force data of each transient electromagnetic measurement point downhole at each moment within the preset time period to obtain the corrected induced electromotive force data of each transient electromagnetic measurement point downhole at each moment within the preset time period.

[0069] In this embodiment of the disclosure, the preprocessing of the time-induced electromotive force data of each transient electromagnetic measurement point downhole at each moment within the preset time period to obtain the corrected induced electromotive force data of each transient electromagnetic measurement point downhole at each moment within the preset time period includes:

[0070] The time-induced electromotive force data of each transient electromagnetic measurement point in the well at each moment within the preset time period are normalized to obtain the normalized induced electromotive force data of each transient electromagnetic measurement point in the well at each moment within the preset time period.

[0071] Based on the normalized induced electromotive force data of the measured points, the corrected induced electromotive force data of each transient electromagnetic measured point in the well at each moment within the preset time period are determined.

[0072] The calculation formula for the induced electromotive force data after correction at each downhole transient electromagnetic measurement point at each moment within the preset time period is as follows:

[0073] D ij =V ij ·α i

[0074] In the formula, D ij V represents the induced electromotive force data corrected for the j-th transient electromagnetic measurement point downhole at the i-th moment within a preset time period. ij For the normalized induced electromotive force (EMF) data of the j-th transient electromagnetic measurement point at the i-th moment in the preset time period, i.e., the normalized induced EMF data after eliminating the turn-off effect, α i Let be the correction coefficient at time i.

[0075] It should be noted that V ij =V ij ' / NIS, V ij 'This refers to the induced electromotive force data of the j-th transient electromagnetic measurement point downhole at the i-th moment within the preset time period, i.e., the induced electromotive force data measured by the field instruments and equipment. N is the number of turns of the transmitting coil, I is the transmitting current, and S is the area of ​​the transmitting coil.'

[0076] d i Let be the uniform full-space forward response value at time i. It represents the average value of the induced electromotive force at all measuring points in a single measurement at the same location within the i-th time window.

[0077] Step 2: Determine the formation thickness and formation resistivity of multiple model measurement points in the full-space layered medium model based on the induced electromotive force data after correction of the measured measurement points, and use the formation thickness and formation resistivity of the multiple model measurement points as parameters of the full-space layered medium model;

[0078] In this embodiment of the disclosure, determining the formation thickness and formation resistivity of multiple model measuring points in the full-space layered medium model based on the induced electromotive force data corrected from the measured points includes:

[0079] The induced electromotive force data after correction at the measured points are inverted and calculated to obtain the formation thickness and corresponding formation resistivity at each measured point.

[0080] Based on the formation thickness and corresponding formation resistivity of each measured point, determine the standard deviation and expected value of the formation thickness and the standard deviation and expected value of the corresponding formation resistivity.

[0081] Based on the standard deviation and expected value of the formation thickness at the measured points, and the standard deviation and expected value of the corresponding formation resistivity, determine the normal distribution probability density function of the formation thickness at the measured points and the normal distribution probability density function of the corresponding formation resistivity.

[0082] Based on the normal distribution probability density function of the formation thickness and the normal distribution probability density function of the corresponding formation resistivity, M model measurement points are randomly generated to obtain the formation thickness and corresponding formation resistivity.

[0083] Among them, the stratigraphic thickness h at the M model measurement points p Formation resistivity ρ p p = 1 - M is the index number of the total number of model parameters.

[0084] Furthermore, the formula for calculating the standard deviation of the stratigraphic thickness at the measured points is as follows:

[0085]

[0086] In the formula, σ h Here, n is the standard deviation of the stratigraphic layer thickness at the measured points, l is the total number of measured points, h is the number of layers in the full-space layered medium model, and ρ is the standard deviation of the stratigraphic layer thickness at the measured points. jk Let the thickness of the k-th stratum be the measured thickness at the j-th measurement point. This represents the average layer thickness.

[0087] The formula for calculating the expected stratigraphic thickness at the measured points is as follows:

[0088]

[0089] In the formula, μ h The expected stratigraphic thickness at the measured point;

[0090] The formula for calculating the standard deviation of the formation resistivity at the measured points is as follows:

[0091]

[0092] In the formula, σ ρ ρ represents the standard deviation of the formation resistivity at the measured points. jk Let J be the resistivity of the k-th stratum corresponding to the j-th measured point. The mean resistivity of the formation;

[0093] The formula for calculating the expected formation resistivity at the measured points is as follows:

[0094]

[0095] In the formula, μ ρ This represents the expected formation resistivity at the measured points.

[0096] Furthermore, the normal distribution probability density function of the stratigraphic thickness at the measured points is calculated as follows:

[0097]

[0098] In the formula, f(h) is the normal distribution probability density function of the stratum thickness, and h is the stratum thickness;

[0099] The formula for calculating the normal distribution probability density function of the formation resistivity is as follows:

[0100]

[0101] In the formula, f(ρ) is the normal distribution probability density function of the formation resistivity, and ρ is the formation resistivity.

[0102] It should be noted that, as Figure 2 The data structure diagram of the full-space layered medium model is shown below. k = 1 - l is the sequence number, l is the total number of stratigraphic layers in the model, and the stratigraphic thickness is h. k The resistivity is ρ k Where h1=∞, h l =∞, the initial stratum thickness on both sides of the emission source is 5-10m, the thickness difference between adjacent strata is no more than 50%, and the resistivity of adjacent strata is not equal.

[0103] Step 3: Perform one-dimensional forward modeling on the full-space layered medium model to obtain time-induced electromotive force response data of multiple model measurement points in the full-space layered medium model;

[0104] It should be noted that forward modeling is performed on the generated K model measurement point samples to calculate the time-induced electromotive force response data of the K model measurement point samples, where K = M / l is the number of model measurement point samples, and the value of M is determined according to the performance of the deep learning model and the required number of samples.

[0105] Step 4: Extract time-domain features based on the time-induced electromotive force response data of the multiple model measurement points to obtain the time-domain features of induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity.

[0106] In this embodiment of the disclosure, step 4 specifically includes:

[0107] Step 4-1: Use the time-induced electromotive force response data of the multiple model measurement points as input data, and use the corresponding formation thickness and resistivity model data of the multiple model measurement points as output data to form an initial deep learning sample set.

[0108] Step 4-2: Extract the time-domain features of induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity from the initial deep learning sample set to obtain the time-domain features of induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity.

[0109] It should be noted that the induced electromotive force was extracted from the initial deep learning sample set, and the apparent resistivity ρ i ΔV was obtained based on the formula for calculating late apparent resistivity. i , Δρ i According to t i V i ρ i The sequence is obtained by difference calculation, referring to the formula:

[0110] Feature extraction was performed on the sample set according to Table 1, and the data was normalized using the Z-score standardization method to generate a deep learning sample set feature extraction data table.

[0111] Table 1. Data on Deep Learning Sample Feature Extraction

[0112]

[0113] Step 5: Use the time-domain features of induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity from multiple model measurement points as the input set, and use the formation thickness and formation resistivity from multiple model measurement points as the output set to form a deep learning sample set.

[0114] In summary, this embodiment proposes a method for determining a deep learning sample set based on the full-space transient electromagnetic method. This method completes the construction and feature extraction of a deep learning sample set based on the full-space transient electromagnetic method. Furthermore, through the data processing of this invention, the deviation between measured data and forward modeling data can be eliminated, making the dataset more representative and ensuring the effectiveness and robustness of the inversion model. The model can be trained and tested during non-measurement idle time, and the characteristics of deep learning can be utilized during measurement. By using the calculated inversion model, a near real-time high-precision inversion effect can be achieved.

[0115] Example 2

[0116] Figure 3 Here is a structural diagram of a deep learning sample set determination system based on the full-space transient electromagnetic method according to an embodiment of this application, as shown below. Figure 3 As shown, the system includes:

[0117] The preprocessing module 100 is used to acquire the time-induced electromotive force data of each transient electromagnetic measurement point downhole at each moment within a preset time period, and to preprocess the time-induced electromotive force data of each transient electromagnetic measurement point downhole at each moment within the preset time period to obtain the corrected induced electromotive force data of each transient electromagnetic measurement point downhole at each moment within the preset time period.

[0118] The first determining module 200 is used to determine the formation thickness and formation resistivity of multiple model measuring points in the full-space layered medium model based on the induced electromotive force data after correction of the measured measuring points, and to use the formation thickness and formation resistivity of the multiple model measuring points as parameters of the full-space layered medium model.

[0119] The first processing module 300 is used to perform one-dimensional forward modeling on the full-space layered medium model to obtain time-induced electromotive force response data of multiple model measurement points in the full-space layered medium model.

[0120] The feature extraction module 400 is used to extract time-domain features based on the time-induced electromotive force response data of the multiple model measurement points, and obtain the time-domain features of induced electromotive force, apparent resistivity, induced electromotive force change rate, and apparent resistivity change rate.

[0121] The module 500 is used to take the induced electromotive force, apparent resistivity, induced electromotive force change rate, and apparent resistivity change rate time domain features of multiple model measurement points as input set, and the formation layer thickness and formation resistivity of multiple model measurement points as output set, to form a deep learning sample set.

[0122] In this embodiment of the disclosure, the preprocessing module 100 is further configured to:

[0123] The time-induced electromotive force data of each transient electromagnetic measurement point in the well at each moment within the preset time period are normalized to obtain the normalized induced electromotive force data of each transient electromagnetic measurement point in the well at each moment within the preset time period.

[0124] Based on the normalized induced electromotive force data of the measured points, the corrected induced electromotive force data of each transient electromagnetic measured point in the well at each moment within the preset time period are determined.

[0125] Furthermore, the calculation formula for the induced electromotive force data after correction at each transient electromagnetic measurement point downhole at each moment within the preset time period is as follows:

[0126] D ij =V ij ·α i

[0127] In the formula, D ij V represents the induced electromotive force data corrected for the j-th transient electromagnetic measurement point downhole at the i-th moment within a preset time period. ij α represents the normalized induced electromotive force data of the j-th transient electromagnetic measurement point downhole at the i-th moment within a preset time period. i Let be the correction coefficient at time i.

[0128] Furthermore, the first determining module 200 is also used for:

[0129] The induced electromotive force data after correction at the measured points are inverted and calculated to obtain the formation thickness and corresponding formation resistivity at each measured point.

[0130] Based on the formation thickness and corresponding formation resistivity of each measured point, determine the standard deviation and expected value of the formation thickness and the standard deviation and expected value of the corresponding formation resistivity.

[0131] Based on the standard deviation and expected value of the formation thickness at the measured points, and the standard deviation and expected value of the corresponding formation resistivity, determine the normal distribution probability density function of the formation thickness at the measured points and the normal distribution probability density function of the corresponding formation resistivity.

[0132] Based on the normal distribution probability density function of the formation thickness and the normal distribution probability density function of the corresponding formation resistivity, M model measurement points are randomly generated to obtain the formation thickness and corresponding formation resistivity.

[0133] The formula for calculating the standard deviation of the stratigraphic thickness at the measured points is as follows:

[0134]

[0135] In the formula, σ hHere, n is the standard deviation of the stratigraphic layer thickness at the measured points, l is the total number of measured points, h is the number of layers in the full-space layered medium model, and ρ is the standard deviation of the stratigraphic layer thickness at the measured points. jk Let the thickness of the k-th stratum be the measured thickness at the j-th measurement point. This represents the average layer thickness.

[0136] The formula for calculating the expected stratigraphic thickness at the measured points is as follows:

[0137]

[0138] In the formula, μ h The expected stratigraphic thickness at the measured point;

[0139] The formula for calculating the standard deviation of the formation resistivity at the measured points is as follows:

[0140]

[0141] In the formula, σ ρ ρ represents the standard deviation of the formation resistivity at the measured points. jk Let J be the resistivity of the k-th stratum corresponding to the j-th measured point. The mean resistivity of the formation;

[0142] The formula for calculating the expected formation resistivity at the measured points is as follows:

[0143]

[0144] In the formula, μ ρ This represents the expected formation resistivity at the measured points.

[0145] Furthermore, the normal distribution probability density function of the stratigraphic thickness at the measured points is calculated as follows:

[0146]

[0147] In the formula, f(h) is the normal distribution probability density function of the stratum thickness, and h is the stratum thickness;

[0148] The formula for calculating the normal distribution probability density function of the formation resistivity is as follows:

[0149]

[0150] In the formula, f(ρ) is the normal distribution probability density function of the formation resistivity, and ρ is the formation resistivity.

[0151] Furthermore, such as Figure 4 As shown, the feature extraction module 400 includes:

[0152] The initial construction unit 401 is used to take the time-induced electromotive force response data of the multiple model measurement points as input data and the corresponding stratigraphic thickness and resistivity model data of the multiple model measurement points as output data to form an initial deep learning sample set.

[0153] Extraction unit 402 is used to extract time-domain features of induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity from the initial deep learning sample set, thereby obtaining time-domain features of induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity.

[0154] In summary, the deep learning sample set determination system based on the full-space transient electromagnetic method proposed in this embodiment can eliminate the deviation between measured data and forward modeling data, making the dataset more representative and thus ensuring the accuracy of subsequent model training.

[0155] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0156] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0157] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for determining a deep learning sample set based on the all-space transient electromagnetic method, characterized in that, The method includes: The time-induced electromotive force data of each transient electromagnetic measurement point in the well at each moment within a preset time period are obtained, and the time-induced electromotive force data of each transient electromagnetic measurement point in the well at each moment within the preset time period are preprocessed to obtain the corrected induced electromotive force data of each transient electromagnetic measurement point in the well at each moment within the preset time period. The induced electromotive force data after correction at the measured points are inverted and calculated to obtain the formation thickness and corresponding formation resistivity at each measured point. Based on the formation thickness and corresponding formation resistivity of each measured point, determine the standard deviation and expected value of the formation thickness and the standard deviation and expected value of the corresponding formation resistivity. Based on the standard deviation and expected value of the formation thickness at the measured points, and the standard deviation and expected value of the corresponding formation resistivity, determine the normal distribution probability density function of the formation thickness at the measured points and the normal distribution probability density function of the corresponding formation resistivity. Based on the normal distribution probability density function of the formation thickness and the normal distribution probability density function of the corresponding formation resistivity, the formation thickness and corresponding formation resistivity of multiple model measurement points in the full-space layered medium model are randomly generated, and the formation thickness and formation resistivity of the multiple model measurement points are used as parameters of the full-space layered medium model. One-dimensional forward modeling was performed on the full-space layered medium model to obtain time-induced electromotive force response data of multiple model measurement points in the full-space layered medium model; Based on the time-induced electromotive force response data of the multiple model measurement points, time domain features are extracted to obtain the time domain features of induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity. The induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity at multiple model measurement points are used as the input set, and the formation thickness and formation resistivity at multiple model measurement points are used as the output set to form a deep learning sample set.

2. The method as described in claim 1, characterized in that, The process of preprocessing the time-induced electromotive force data of each transient electromagnetic measurement point downhole at each moment within the preset time period to obtain the corrected induced electromotive force data of each transient electromagnetic measurement point downhole at each moment within the preset time period includes: The time-induced electromotive force data of each transient electromagnetic measurement point in the well at each moment within the preset time period are normalized to obtain the normalized induced electromotive force data of each transient electromagnetic measurement point in the well at each moment within the preset time period. Based on the normalized induced electromotive force data of the measured points, the corrected induced electromotive force data of each transient electromagnetic measured point in the well at each moment within the preset time period are determined.

3. The method as described in claim 2, characterized in that, The calculation formula for the induced electromotive force data of each transient electromagnetic measurement point downhole at each time point within the preset time period after correction is as follows: In the formula, This refers to the induced electromotive force data corrected for the j-th transient electromagnetic measurement point downhole at the i-th moment within a preset time period. This refers to the normalized induced electromotive force data of the j-th transient electromagnetic measurement point downhole at the i-th moment within a preset time period. Let be the correction coefficient at time i.

4. The method as described in claim 1, characterized in that, The formula for calculating the standard deviation of the stratigraphic thickness at the measured points is as follows: In the formula, This represents the standard deviation of the stratigraphic thickness at the measured points. This represents the total number of measured points. The number of layers in the full-space layered medium model. Let the thickness of the k-th stratum be the measured thickness at the j-th measurement point. This represents the average layer thickness. The formula for calculating the expected stratigraphic thickness at the measured points is as follows: In the formula, The expected stratigraphic thickness at the measured point; The formula for calculating the standard deviation of the formation resistivity at the measured points is as follows: In the formula, This represents the standard deviation of the formation resistivity at the measured points. Let J be the resistivity of the k-th stratum corresponding to the j-th measured point. The mean resistivity of the formation; The formula for calculating the expected formation resistivity at the measured points is as follows: In the formula, This represents the expected formation resistivity at the measured points.

5. The method as described in claim 4, characterized in that, The formula for calculating the normal distribution probability density function of the stratigraphic thickness at the measured points is as follows: In the formula, Let be the normal distribution probability density function of the stratigraphic thickness. The thickness of the strata; The formula for calculating the normal distribution probability density function of the formation resistivity is as follows: In the formula, Let be the probability density function of the normal distribution of formation resistivity. The resistivity is the formation resistivity.

6. The method as described in claim 1, characterized in that, The extraction of time-domain features from the time-induced electromotive force response data of the multiple model measurement points yields the time-domain features of the induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity, including: The time-induced electromotive force response data of the multiple model measurement points are used as input data, and the corresponding formation thickness and resistivity model data of the multiple model measurement points are used as output data to form an initial deep learning sample set. In the initial deep learning sample set, the time-domain features of induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity are extracted to obtain the time-domain features of induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity.

7. A deep learning sample set determination system based on the all-space transient electromagnetic method, characterized in that, The system includes: The preprocessing module is used to acquire the time-induced electromotive force data of each transient electromagnetic measurement point downhole at each moment within a preset time period, and to preprocess the time-induced electromotive force data of each transient electromagnetic measurement point downhole at each moment within the preset time period to obtain the corrected induced electromotive force data of each transient electromagnetic measurement point downhole at each moment within the preset time period. The first determining module is used to perform inversion calculations on the induced electromotive force data after correction of the measured points to obtain the formation thickness and corresponding formation resistivity of each measured point; determine the standard deviation and expected value of the formation thickness and the standard deviation and expected value of the corresponding formation resistivity of the measured points based on the formation thickness and expected value of the measured points; determine the normal distribution probability density function of the formation thickness and the normal distribution probability density function of the corresponding formation resistivity based on the standard deviation and expected value of the formation thickness and the corresponding formation resistivity of the measured points; randomly generate the formation thickness and corresponding formation resistivity of multiple model points in the full-space layered medium model based on the normal distribution probability density function of the formation thickness and the normal distribution probability density function of the corresponding formation resistivity, and use the formation thickness and formation resistivity of the multiple model points as parameters of the full-space layered medium model; The first processing module is used to perform one-dimensional forward modeling on the full-space layered medium model to obtain time-induced electromotive force response data of multiple model measurement points in the full-space layered medium model. The feature extraction module is used to extract time-domain features based on the time-induced electromotive force response data of the multiple model measurement points, and obtain the time-domain features of induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity. The module is used to take the time-domain features of induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity from multiple model measurement points as the input set, and the formation thickness and formation resistivity from multiple model measurement points as the output set, thus forming a deep learning sample set.

8. The system as described in claim 7, characterized in that, The preprocessing module is also used for: The time-induced electromotive force data of each transient electromagnetic measurement point in the well at each moment within the preset time period are normalized to obtain the normalized induced electromotive force data of each transient electromagnetic measurement point in the well at each moment within the preset time period. Based on the normalized induced electromotive force data of the measured points, the corrected induced electromotive force data of each transient electromagnetic measured point in the well at each moment within the preset time period are determined.

9. The system as described in claim 8, characterized in that, The feature extraction module includes: The initial construction unit is used to take the time-induced electromotive force response data of the multiple model measurement points as input data and the corresponding stratigraphic thickness and resistivity model data of the multiple model measurement points as output data to form an initial deep learning sample set. The extraction unit is used to extract the time-domain features of induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity from the initial deep learning sample set, thereby obtaining the time-domain features of induced electromotive force, apparent resistivity, rate of change of induced electromotive force, and rate of change of apparent resistivity.

Images