Model training method and stratigraphic division conclusion extraction method

Abstract

The invention provides a model training method and a stratigraphic division conclusion extraction method, and relates to the technical field of logging curve data analysis. The model training method comprises the following steps: acquiring a first well logging curve of a target well logging, wherein the first well logging curve respectively indicates depth, natural gamma, natural potential, interval transit time and dual lateral resistivity; preprocessing the plurality of first logging curves to obtain corresponding second logging curves; performing multi-scale wavelet decomposition on the plurality of second logging curves to obtain corresponding multi-scale components; and inputting the target fragment length of each second logging curve and the corresponding multi-scale component into a multi-stage bidirectional long-short-term memory network to train the network, and outputting a first stratigraphic division conclusion corresponding to each second logging curve by the trained network. According to the method, the logging curves and the corresponding multi-scale components obtained through wavelet decomposition are utilized to train the multi-stage bidirectional long-short-term memory network, and therefore the stratigraphic division conclusion based on the different logging curves is automatically obtained.

Description

Technical Field

[0001] The present application relates to the technical field of well logging curve data analysis, and in particular to a model training method and a method for extracting stratigraphic division conclusions. Background Art

[0002] Oil, as an important strategic material, is closely related to social and economic development, and plays a vital role in ensuring the normal operation of society and economic development. Oil exploration technology aims to explore and analyze potential oil and gas reservoirs through geological, geophysical and other related methods. Well logging technology, as one of the core technologies in oil exploration, mainly collects data on the physical, chemical and other properties of rock formations by lowering logging instruments into the well during or after drilling. Stratigraphic division plays a key role in oil exploration: Stratigraphic division can provide necessary design and operation information for logging technology; the rock and sediment characteristics of different strata will affect the location and distribution of oil and gas reservoirs. Through detailed stratigraphic division, the reserves, production and economic value of oil and gas reservoirs can be analyzed; at the same time, stratigraphic division can help determine the location and distribution of oil and gas reservoirs. By analyzing the sedimentation environment, lithology changes and geological structure of the strata, possible oil and gas reservoir areas can be identified and their specific locations underground can be determined. Therefore, it is crucial to study stratigraphic division in oil exploration.

[0003] In the prior art, stratigraphic division is based on scientific methods and evidence, but the stratigraphic division conclusion drawn on this basis is artificial, that is, this method requires researchers to draw stratigraphic division conclusions based on personal experience and scientific evidence. Therefore, the stratigraphic division conclusions drawn using the existing methods are subjective, and different researchers may obtain different results, resulting in poor repeatability of stratigraphic division. Summary of the invention

[0004] The present application provides a model training method and a method for extracting stratigraphic division conclusions, so as to solve the problem of low repeatability caused by obtaining stratigraphic division conclusions through existing technologies.

[0005] A first aspect of an embodiment of the present invention provides a model training method, comprising:

[0006] Acquire multiple first logging curves of the target well; the multiple first logging curves are used to indicate depth, natural gamma, natural potential, acoustic time difference and dual lateral resistivity respectively;

[0007] Preprocessing the plurality of the first well logging curves to obtain a second well logging curve corresponding to each of the first well logging curves;

[0008] Performing multi-scale wavelet decomposition on target segment lengths of a plurality of the second well logging curves to obtain a multi-scale component corresponding to each of the second well logging curves;

[0009] The sampling point data of the target segment length of each second well logging curve and the multi-scale components corresponding to each second well logging curve are input into a pre-constructed multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network; wherein the multi-level bidirectional long short-term memory network is constructed by a plurality of bidirectional long short-term memory networks with different network parameters; the trained multi-level bidirectional long short-term memory network is used to output the first stratigraphic division conclusion corresponding to each second well logging curve.

[0010] In a possible design, the second target well logging curve is any one of a plurality of the second well logging curves;

[0011] For the second target well logging curve, performing multi-scale wavelet decomposition on the target segment lengths of the plurality of second well logging curves to obtain a multi-scale component corresponding to each of the second well logging curves includes:

[0012] According to the pre-stored formation division accuracy, the target segment length, target decomposition wavelet basis and target decomposition level of the second target well logging curve are obtained;

[0013] performing at least one wavelet decomposition on a target segment length of the second target well logging curve according to a target decomposition level and a target decomposition wavelet basis of the second target well logging curve;

[0014] A multi-scale component corresponding to the second target logging curve is obtained according to the low-frequency component obtained by each wavelet decomposition; wherein the multi-scale component corresponding to the second target logging curve refers to a set of at least one of the low-frequency components.

[0015] In a possible design, performing at least one wavelet decomposition on the target segment length of the second target well logging curve according to the target decomposition level and the target decomposition wavelet basis of the second target well logging curve includes:

[0016] According to the target decomposition wavelet basis of the second target well logging curve, performing at least one wavelet decomposition on the target segment length of the second target well logging curve until the number of wavelet decompositions reaches the target decomposition level of the second target well logging curve;

[0017] The target wavelet decomposition is any one of the at least one wavelet decompositions. After performing the target wavelet decomposition on the target segment length of the second target well logging curve according to the target decomposition wavelet basis of the second target well logging curve, the method further includes:

[0018] The target segment length is updated to the low-frequency component obtained by the target-th wavelet decomposition.

[0019] In a possible design, after inputting the sampling point data of the target segment length of each second well logging curve and the multi-scale components corresponding to each second well logging curve into a pre-constructed multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network, the method further includes:

[0020] Input the sampling point data of the target segment length of each second well logging curve and the multi-scale component corresponding to each second well logging curve into a pre-constructed long short-term memory network to obtain the second formation division conclusion corresponding to each second well logging curve output by the long short-term memory network;

[0021] According to the plurality of the first stratigraphic division conclusions and the plurality of the second stratigraphic division conclusions, a first stratigraphic division fusion conclusion obtained by fusing the plurality of the first stratigraphic division conclusions and a second stratigraphic division fusion conclusion obtained by fusing the plurality of the second stratigraphic division conclusions are obtained through a pre-stored adaptive fusion network;

[0022] Obtaining the stratigraphic division accuracy of each of a plurality of the first stratigraphic division conclusions, a plurality of the second stratigraphic division conclusions, the first stratigraphic division fusion conclusion, and the second stratigraphic division fusion conclusion;

[0023] According to the multiple stratum division accuracy rates, the training result of the multi-level bidirectional long short-term memory network after the training is completed is obtained; wherein, the training result is successful training, which is used to indicate that the stratum division accuracy rate of the multi-level bidirectional long short-term memory network after the training is completed is higher than the stratum division accuracy rate of the long short-term memory network.

[0024] In a possible design, the adaptive fusion network pre-stores a first fusion network formula and a second fusion network formula;

[0025] When the adaptive fusion network is a two-layer partially connected network, the method of obtaining a first stratum division fusion conclusion obtained by fusing the plurality of first stratum division conclusions and a second stratum division fusion conclusion obtained by fusing the plurality of second stratum division conclusions through a pre-stored adaptive fusion network according to the plurality of first stratum division conclusions and the plurality of second stratum division conclusions includes:

[0026] Through the first fusion network formula, according to the plurality of first stratum division conclusions, a first prediction probability is obtained; the first fusion network formula is:

[0027]

[0028] Among them, the is the first predicted probability, the ReLU() is a linear correction function, and the G j is a fusion conclusion obtained based on a plurality of the first stratigraphic division conclusions, wherein W k is a partially connected layer with k nodes, the b k is the bias of the partially connected layer, W1 is the output layer, and b1 is the bias of the output layer;

[0029] Obtaining the first stratum division fusion conclusion according to the first prediction probability;

[0030] Through the second fusion network formula, according to the plurality of the second stratum division conclusions, the second prediction probability is obtained; the second fusion network formula is:

[0031]

[0032] Among them, the is the second predicted probability, the G m A fusion conclusion based on a plurality of the second stratigraphic division conclusions;

[0033] The second stratum division fusion conclusion is obtained according to the second prediction probability.

[0034] In a possible design, the multi-level bidirectional long short-term memory network includes n+1 network layers, and the target segment length of the second target well logging curve is subjected to n wavelet decompositions;

[0035] The step of inputting the sampling point data of the target segment length of each second well logging curve and the multi-scale components corresponding to each second well logging curve into a pre-constructed multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network comprises:

[0036] Input the sampling point data of the target segment length of each second well logging curve into the first network layer of the multi-level bidirectional long short-term memory network;

[0037] The low-frequency component obtained by the i-th wavelet decomposition is input into the i+1-th network layer of the multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network; wherein i is a positive integer less than or equal to n.

[0038] In a possible design, the first target well logging curve is any one of the plurality of the first well logging curves;

[0039] For the first target well logging curve, preprocessing the plurality of the first well logging curves to obtain a second well logging curve corresponding to each of the first well logging curves includes:

[0040] performing outlier processing on the first target logging curve;

[0041] The sampling point data of the first logging curve after outlier processing is normalized by a normalization processing formula to obtain the sampling point data of the second logging curve corresponding to the first target logging curve.

[0042] A second aspect of an embodiment of the present invention provides a method for extracting a stratum division conclusion, comprising:

[0043] Acquire multiple third logging curves of the well to be tested; the multiple third logging curves are used to indicate depth, natural gamma, natural potential, acoustic time difference and dual lateral resistivity respectively;

[0044] Preprocessing the plurality of third well logging curves to obtain a fourth well logging curve corresponding to each of the third well logging curves;

[0045] Performing multi-scale wavelet decomposition on target segment lengths of a plurality of the fourth well logging curves to obtain a multi-scale component corresponding to each of the fourth well logging curves;

[0046] The sampling point data of the target segment length of each of the fourth logging curves and the multi-scale components corresponding to each of the fourth logging curves are input into a pre-stored multi-level bidirectional long short-term memory network to obtain a third stratigraphic division conclusion corresponding to each of the fourth logging curves; wherein the multi-level bidirectional long short-term memory network is a model obtained by the model training method provided by the first aspect of the embodiment of the present invention.

[0047] In a possible design, after inputting the sampling point data of the target segment length of each of the fourth well logging curves and the multi-scale components corresponding to each of the fourth well logging curves into a pre-stored multi-level bidirectional long short-term memory network to obtain the third formation division conclusion corresponding to each of the fourth well logging curves, the method further includes:

[0048] According to the plurality of third stratum division conclusions, a third stratum division fusion conclusion obtained by fusing the plurality of third stratum division conclusions is obtained through a pre-stored adaptive fusion network.

[0049] In a possible design, a third fusion network formula is pre-stored in the adaptive fusion network;

[0050] When the adaptive fusion network is a two-layer partially connected network, the third stratum division fusion conclusion obtained by fusing the plurality of third stratum division conclusions through the pre-stored adaptive fusion network according to the plurality of third stratum division conclusions includes:

[0051] Through the third fusion network formula, according to the plurality of third stratum division conclusions, a third prediction probability is obtained; the fusion network formula is:

[0052]

[0053] Among them, the is the third predicted probability, the ReLU() is a linear correction function, and the G n is a fusion conclusion obtained based on a plurality of the third stratum division conclusions, wherein W k is a partially connected layer with k nodes, the b k is the bias of the partially connected layer, W1 is the output layer, and b1 is the bias of the output layer;

[0054] The third stratigraphic division fusion conclusion is obtained according to the third prediction probability.

[0055] A third aspect of an embodiment of the present invention provides a model training device, including:

[0056] An acquisition module, used for acquiring a plurality of first logging curves of a target well; the plurality of first logging curves are used for indicating depth, natural gamma, natural potential, acoustic time difference and dual lateral resistivity respectively;

[0057] A preprocessing module, used for preprocessing a plurality of the first well logging curves to obtain a second well logging curve corresponding to each of the first well logging curves;

[0058] A wavelet decomposition module, used for performing multi-scale wavelet decomposition on target segment lengths of a plurality of the second well logging curves to obtain a multi-scale component corresponding to each of the second well logging curves;

[0059] A training module is used to input the sampling point data of the target segment length of each second well logging curve and the multi-scale components corresponding to each second well logging curve into a pre-constructed multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network; wherein the multi-level bidirectional long short-term memory network is constructed by a plurality of bidirectional long short-term memory networks with different network parameters; the trained multi-level bidirectional long short-term memory network is used to output the first stratigraphic division conclusion corresponding to each second well logging curve.

[0060] A fourth aspect of an embodiment of the present invention provides a stratum division conclusion extraction device, comprising:

[0061] An acquisition module is used to acquire a plurality of third logging curves of the well to be tested; the plurality of third logging curves are used to indicate depth, natural gamma, natural potential, acoustic time difference and dual lateral resistivity respectively;

[0062] Preprocessing module: used for preprocessing a plurality of the third well logging curves to obtain a fourth well logging curve corresponding to each of the third well logging curves;

[0063] A wavelet decomposition module, used for performing multi-scale wavelet decomposition on the target segment lengths of the plurality of fourth well logging curves to obtain a multi-scale component corresponding to each of the fourth well logging curves;

[0064] The extraction module is used to input the sampling point data of the target segment length of each of the fourth well logging curves and the multi-scale components corresponding to each of the fourth well logging curves into a pre-stored multi-level bidirectional long short-term memory network to obtain the third stratigraphic division conclusion corresponding to each of the fourth well logging curves.

[0065] A fifth aspect of an embodiment of the present invention provides an electronic device, including: a processor and a memory communicatively connected to the processor;

[0066] The memory stores computer-executable instructions;

[0067] The processor executes the computer-executable instructions stored in the memory to implement the model training method provided in the first aspect of the embodiment of the present invention or the stratigraphic division conclusion extraction method provided in the second aspect of the embodiment of the present invention.

[0068] The sixth aspect of the embodiment of the present invention provides a computer-readable storage medium, which stores computer execution instructions to implement the model training method provided in the first aspect of the embodiment of the present invention or the stratigraphic division conclusion extraction method provided in the second aspect of the embodiment of the present invention.

[0069] The seventh aspect of the embodiment of the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the model training method provided in the first aspect of the embodiment of the present invention or the stratigraphic division conclusion extraction method provided in the second aspect of the embodiment of the present invention.

[0070] The present application provides a model training method and a method for extracting formation division conclusions, the model training method comprising: obtaining multiple first well logging curves of a target well logging, the multiple first well logging curves being used to indicate depth, natural gamma, natural potential, acoustic time difference and dual lateral resistivity respectively; preprocessing the multiple first well logging curves to obtain second well logging curves corresponding to each first well logging curve; performing multi-scale wavelet decomposition on the target segment lengths of the multiple second well logging curves to obtain multi-scale components corresponding to each second well logging curve; inputting sampling point data of the target segment length of each second well logging curve and the multi-scale components corresponding to each second well logging curve into a pre-constructed multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network, wherein the multi-level bidirectional long short-term memory network is constructed by multiple bidirectional long short-term memory networks with different network parameters, and the trained multi-level bidirectional long short-term memory network is used to output the first formation division conclusion corresponding to each second well logging curve. Based on the above method, the following technical effects are achieved: multi-scale wavelet decomposition is performed on the logging curve, and the time series characteristics of the logging data are extracted at multiple scales, so as to obtain richer features and facilitate more accurate stratigraphic division; the traditional wavelet transform is combined with the multi-level bidirectional long short-term memory network in deep learning, and the time series components of the logging curve can be automatically learned from multiple scales without the need for manual analysis of the features obtained after the wavelet transform, thereby automatically drawing stratigraphic division conclusions based on different curves. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0072] Figure 1 Schematic diagram of the model training method provided in the embodiment of the present application Figure 1 ;

[0073] Figure 2 Schematic diagram of the model training method provided in the embodiment of the present application Figure 2 ;

[0074] Figure 3 Schematic diagram of the process of extracting the stratigraphic division conclusion provided in the embodiment of the present application Figure 1 ;

[0075] Figure 4 Schematic diagram of the process of extracting the stratigraphic division conclusion provided in the embodiment of the present application Figure 2 ;

[0076] Figure 5 A schematic diagram of the structure of a model training device provided in an embodiment of the present application;

[0077] Figure 6 A schematic diagram of the structure of a device for extracting stratum division conclusions provided in an embodiment of the present application;

[0078] Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0079] In the embodiments of the present application, words such as "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. Those skilled in the art will understand that words such as "first" and "second" do not limit the quantity and execution order, and words such as "first" and "second" do not necessarily limit the difference. It should be noted that in the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "exemplary" or "for example" in this application should not be interpreted as being more preferred or more advantageous than other embodiments or design. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way. In the embodiments of the present application, "at least one" refers to one or more, and "more" refers to two or more.

[0080] It should be noted that the "at..." in the embodiment of the present application can be the instant when a certain situation occurs, or it can be a period of time after a certain situation occurs, and the embodiment of the present application does not specifically limit this. In addition, a model training method provided in the embodiment of the present application is only used as an example, and the model training method can also include more or less content.

[0081] In order to clearly describe the technical solutions of the embodiments of the present application, some terms and technologies involved in the embodiments of the present application are briefly introduced below:

[0082] Long Short-Term Memory Network: A special recursive neural network designed to solve the gradient vanishing and gradient exploding problems encountered by traditional recurrent neural networks when processing long sequences. The long short-term memory network can save information over a long time span by introducing memory units, thereby solving the memory loss problem of traditional recurrent neural networks in long sequence processing; the long short-term memory network controls the flow of information through input gates, forget gates, and output gates. Each gate has its own independent neural network, which can decide which information should be retained or forgotten.

[0083] Wavelet basis: Wavelet basis is a set of wavelet functions that can effectively decompose and represent signals and are mainly used in signal processing, data compression, denoising and feature extraction. These wavelet functions are the basic elements of signals and can be used to represent and analyze different characteristics of signals.

[0084] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0085] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0086] In order to clearly understand the technical solution of the present application, the solution of the prior art is first introduced in detail.

[0087] In the prior art, the stratigraphic division process is based on scientific methods and evidence, but the stratigraphic division conclusion drawn on this basis is artificial, that is, this method requires researchers to draw stratigraphic division conclusions based on personal experience and scientific evidence. Therefore, the stratigraphic division conclusions drawn using the existing methods are subjective, and different researchers may obtain different results, resulting in poor repeatability of stratigraphic division.

[0088] Therefore, in order to solve the problem of poor repeatability of stratigraphic division using existing technologies, the study found that ① the commonly used logging curves of the target well are obtained; ② the commonly used logging curves are pre-processed and then subjected to multi-scale wavelet transform to obtain the corresponding multi-scale components; ③ based on the commonly used logging curves and their corresponding multi-scale components, a multi-level bidirectional long short-term memory network is trained.

[0089] Specifically, a plurality of first logging curves of the target well are obtained; the plurality of first logging curves are respectively used to indicate depth, natural gamma, natural potential, acoustic time difference and dual lateral resistivity;

[0090] Preprocessing a plurality of first well logging curves to obtain a second well logging curve corresponding to each first well logging curve;

[0091] Performing multi-scale wavelet decomposition on target segment lengths of multiple second well logging curves to obtain multi-scale components corresponding to each second well logging curve;

[0092] The sampling point data of the target segment length of each second logging curve and the multi-scale components corresponding to each second logging curve are input into a pre-constructed multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network; wherein the multi-level bidirectional long short-term memory network is constructed by multiple bidirectional long short-term memory networks with different network parameters; the trained multi-level bidirectional long short-term memory network is used to output the first stratigraphic division conclusion corresponding to each second logging curve.

[0093] The model training method of the embodiment of the present invention is based on the principle of the above-mentioned working condition preset conditions. The present invention combines wavelet transform with a multi-level bidirectional long short-term memory network, and uses commonly used logging curves and their corresponding multi-scale components obtained after wavelet transform to train the multi-level bidirectional long short-term memory network, so as to automatically extract the stratigraphic division conclusions corresponding to the commonly used logging curves, thereby avoiding the problem of poor repeatability caused by human subjective judgment.

[0094] Based on the above creative findings, the technical solution of the present application is proposed.

[0095] The following introduces the application scenarios of the model training method provided by the embodiments of the present invention.

[0096] The model training method of this solution is applicable to a variety of scenarios: in the exploration of oil and gas, accurate identification and classification of strata are crucial to determining the location and scale of oil and gas reservoirs. The model trained using the model training method of this application can automatically extract the required stratigraphic information from seismic data, drilling data and core samples to provide more accurate stratigraphic divisions; by automatically extracting stratigraphic division conclusions, the stability of the geological structure can be obtained to help formulate disaster prevention measures; in engineering construction, automated stratigraphic division conclusion analysis can provide accurate geological information to support environmental impact analysis and engineering design.

[0097] The embodiments of the present application are introduced below in conjunction with the drawings in the specification.

[0098] Figure 1 Schematic diagram of the model training method provided in the embodiment of the present application Figure 1 , the model training method provided in this embodiment includes the following steps:

[0099] S101, obtaining a plurality of first logging curves of a target well; the plurality of first logging curves are used to indicate depth, natural gamma, natural potential, acoustic time difference and dual lateral resistivity respectively.

[0100] In this embodiment, firstly, the logging curve data of the same oil field block are obtained, and the coordinate information of the well is used to select the well that is relatively close to the well to be logged and has similar characteristics as the target logging; secondly, multiple first logging curves, namely, depth curve (Depth), natural gamma ray curve (Gamma Ray Curve, GR), spontaneous potential curve (Spontaneous Potential Curve, SP), acoustic velocity curve (Acoustic Velocity Curve, AC) and dual lateral resistivity curves (Dual Laterolog Resistivity Curves, RT1, RT2) are obtained from the logging curve data of the target logging as research data for feature learning.

[0101] S102: Preprocess multiple first logging curves to obtain a second logging curve corresponding to each first logging curve.

[0102] In this embodiment, the second well logging curve corresponding to each first well logging curve is a plurality of well logging curves obtained after preprocessing the plurality of first well logging curves. The purpose of preprocessing the plurality of first well logging curves is to remove abnormal values ​​and features with a large numerical range in the sampling point data of the plurality of first well logging curves, so as to avoid certain features with a large numerical range in the sampling point data dominating the model effect, causing the model to ignore those features with small numerical values ​​but important.

[0103] S103, performing multi-scale wavelet decomposition on the target segment lengths of the plurality of second well logging curves to obtain a multi-scale component corresponding to each second well logging curve.

[0104] In this embodiment, the wavelet basis is selected to perform multi-scale decomposition on the target segment length of the second logging curve, that is, the target segment length of the multiple logging curves after preprocessing, and the temporal characteristics of the logging data are extracted at multiple scales, which can obtain richer features and is conducive to obtaining more accurate stratigraphic division; through multi-scale decomposition, the noise and useful information in the sampling point data of the logging curve can be separated. High-frequency components often contain noise components, while low-frequency components can better reflect actual geological information. Therefore, the gradual decomposition of the logging curve can extract useful geological information.

[0105] S104. Input the sampling point data of the target segment length of each second logging curve and the multi-scale components corresponding to each second logging curve into a pre-constructed multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network; wherein the multi-level bidirectional long short-term memory network is constructed by a plurality of bidirectional long short-term memory networks with different network parameters; the trained multi-level bidirectional long short-term memory network is used to output the first stratigraphic division conclusion corresponding to each second logging curve.

[0106] In this embodiment, the sampling point data of the target segment length of a single preprocessed well logging curve and the multi-scale components obtained after multi-level wavelet decomposition are input into the constructed multi-level bidirectional long short-term memory network for training, and the training parameters are set. After the training is completed, the stratigraphic division result based on the well logging curve is given on the test set. By constructing and training a multi-level bidirectional long short-term memory network, the stratigraphic division conclusion based on the well logging curve can be automatically extracted.

[0107] The present application provides a model training method and a method for extracting formation division conclusions, the model training method comprising: obtaining multiple first well logging curves of a target well logging, the multiple first well logging curves being used to indicate depth, natural gamma, natural potential, acoustic time difference and dual lateral resistivity respectively; preprocessing the multiple first well logging curves to obtain second well logging curves corresponding to each first well logging curve; performing multi-scale wavelet decomposition on the target segment lengths of the multiple second well logging curves to obtain multi-scale components corresponding to each second well logging curve; inputting sampling point data of the target segment length of each second well logging curve and the multi-scale components corresponding to each second well logging curve into a pre-constructed multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network, wherein the multi-level bidirectional long short-term memory network is constructed by multiple bidirectional long short-term memory networks with different network parameters, and the trained multi-level bidirectional long short-term memory network is used to output the first formation division conclusion corresponding to each second well logging curve. Based on the above method, the following technical effects are achieved: multi-scale wavelet decomposition is performed on the logging curve, and the time series characteristics of the logging data are extracted at multiple scales, so as to obtain richer features and facilitate more accurate stratigraphic division; the traditional wavelet transform is combined with the multi-level bidirectional long short-term memory network in deep learning, and the time series components of the logging curve can be automatically learned from multiple scales without the need for manual analysis of the features obtained after the wavelet transform, thereby automatically drawing stratigraphic division conclusions based on different logging curves.

[0108] Figure 2 Schematic diagram of the model training method provided in the embodiment of the present application Figure 2 .like Figure 2 As shown, the model training method provided in this embodiment is a further refinement of the model training method provided in the previous embodiment of the present invention. The model training method provided in this embodiment includes the following steps:

[0109] S201, obtaining a plurality of first logging curves of a target well; the plurality of first logging curves are used to indicate depth, natural gamma, natural potential, acoustic time difference and dual lateral resistivity respectively.

[0110] In this embodiment, the effect of S201 is similar to the effect of 101 in the previous embodiment of the present invention, and will not be repeated here.

[0111] S202: Preprocess multiple first logging curves to obtain a second logging curve corresponding to each first logging curve.

[0112] In this embodiment, optionally, the first target logging curve is any one of a plurality of first logging curves;

[0113] For the first target logging curve, performing outlier processing on the first target logging curve;

[0114] The sampling point data of the first logging curve after outlier processing is normalized by a normalization processing formula to obtain the sampling point data of the second logging curve corresponding to the first target logging curve.

[0115] Specifically, the normalization formula is:

[0116]

[0117] Among them, x i is the i-th sampling point data of the first target logging curve after outlier processing, is the i-th sampling point data of the second logging curve corresponding to the first target logging curve, x min is the minimum value of the sampling point data of the first target logging curve after outlier processing, x max It is the maximum value in the sampling point data of the first target logging curve after outlier processing.

[0118] In this embodiment, the effect of S202 is similar to the effect of S102 in the previous embodiment of the present invention, and will not be repeated here.

[0119] S203. Obtaining a target segment length, a target decomposition wavelet basis, and a target decomposition level of a second target logging curve according to a pre-stored formation division accuracy rate.

[0120] In this embodiment, the second target logging curve is any one of the multiple second logging curves, that is, the second target logging curve corresponds to any one of the preprocessed logging curves. The target segment length, target decomposition wavelet basis and target decomposition level of each logging curve are obtained based on the accuracy of formation division. For the selection of the target segment length of each logging curve, the number of network layers of the constructed multi-level bidirectional long short-term memory network is set to 2, and the number of hidden layer nodes is 128. When the target segment length of the logging curve is 48, 96, 128 and 192 meters, the corresponding actual depths are 6, 12, 16 and 24 meters respectively.

[0121] When performing wavelet decomposition, selecting appropriate wavelet bases and decomposition levels is of great significance for fully exploring the multi-scale time series characteristics in the logging curve. The present invention conducts a detailed analysis of the logging curve and selects a series of wavelet bases for testing to determine which wavelet transform is most suitable for a specific logging curve. The use of multiple series of wavelet bases is conducive to selecting the wavelet base that is most suitable for time-frequency analysis of the logging curve, thereby improving the comprehensiveness and reliability of the experiment. The target decomposition wavelet base is selected from several candidate wavelet bases, namely the classic Haar wavelet base, the Db1 and Db4 wavelet bases in the Daubechies series, the sym4 wavelet base in the Symlets series, the coif1 wavelet base in the Coiflets series, and the bior2.2 wavelet base in the Biorthogonal series. The classic Haar wavelet base and the Db1 and Db4 wavelet bases in the Daubechies series are often used to process signals with mutation characteristics, and may have better analysis effects for some curves with mutation values. When applied to other logging data sets, more wavelet bases can be selected for comparative experiments.

[0122] As shown in Table 1, taking the gamma curve as an example, after taking different wavelet bases to perform wavelet decomposition on it (that is, the target decomposition level is 1 at this time), various indicators of stratigraphic division based on the gamma curve are observed at this time.

[0123] Table 1

[0124] Db1 Db4 Haar sym4 coif1 bior2.2 Accuracy 58.371 58.815 59.215 59.680 58.343 56.480 Recall 47.141 47.407 48.283 48.849 48.264 45.336 F1 coefficient 46.815 47.544 48.430 48.956 47.882 44.649 Time per epoch 136.051 100.519 94.713 94.079 92.594 94.276

[0125] As shown in Table 1, when the target decomposition level is 1, when the sym4 wavelet basis is used, the various indicators of stratum division based on the gamma curve are higher than those of stratum division based on other wavelet bases, and the performance of the Haar wavelet basis is close to that of the sym4 wavelet basis. Therefore, when the target decomposition level of the gamma curve is 1, the sym4 wavelet basis is taken as its target decomposition wavelet basis.

[0126] For each well logging curve, the target decomposition wavelet basis and the target decomposition level of each well logging curve are selected through experiments. As shown in Table 2, the target decomposition wavelet basis and the target decomposition level of each well logging curve are shown.

[0127] Table 2

[0128] GR SP AC RT1 RT2 Target decomposition wavelet basis sym4 Db1 coif1 Db1 Db4 Target decomposition level 1 2 1 2 1

[0129] S204. Perform at least one wavelet decomposition on the target segment length of the second target well logging curve according to the target decomposition wavelet basis of the second target well logging curve, until the number of wavelet decompositions reaches the target decomposition level of the second target well logging curve.

[0130] In this embodiment, the target decomposition level is usually less than 4, because excessive decomposition will introduce noise.

[0131] S205 , after performing a target-th wavelet decomposition on the target segment length of the second target logging curve according to the target decomposition wavelet basis of the second target logging curve, the target segment length is updated to a low-frequency component obtained by the target-th wavelet decomposition.

[0132] In this embodiment, the target wavelet decomposition is any one of at least one wavelet decomposition. After the target segment length, target decomposition wavelet basis and target decomposition level of the preprocessed logging curve are selected through experiments, the preprocessed logging curve is subjected to multi-scale wavelet decomposition. The preprocessed logging curve is subjected to the first wavelet decomposition to obtain the first-level low-frequency component and the first-level high-frequency component of the logging curve, and the first-level high-frequency component is incorporated into the multi-scale database, and the first-level low-frequency component is used as the new target segment length; the first-level low-frequency component as the new target segment length is subjected to the second wavelet decomposition to obtain the second-level low-frequency component and the second-level high-frequency component of the logging curve, and the second-level high-frequency component is incorporated into the multi-scale database, and the second-level low-frequency component is used as the new target segment length, and so on.

[0133] S206. Obtain a multi-scale component corresponding to the second target logging curve according to the low-frequency component obtained by each wavelet decomposition; wherein the multi-scale component corresponding to the second target logging curve refers to a set of at least one low-frequency component.

[0134] In this embodiment, the multi-scale components of the pre-processed logging curve are a set of low-frequency components obtained after wavelet decomposition is performed on the logging curve each time.

[0135] S207. Input the sampling point data of the target segment length of each second logging curve into the first network layer of the multi-level bidirectional long short-term memory network, and input the low-frequency component obtained by the i-th wavelet decomposition into the i+1-th network layer of the multi-level bidirectional long short-term memory network.

[0136] In this embodiment, the purpose of step S207 is to train a multi-level bidirectional long short-term memory network, wherein the multi-level bidirectional long short-term memory network includes n+1 network layers, and the target segment length of the second target logging curve is subjected to n wavelet decompositions, and i is a positive integer less than or equal to n.

[0137] The multi-level bidirectional long short-term memory network is constructed by multiple bidirectional long short-term memory networks with different network parameters, so as to obtain richer feature representation of logging data from multiple scales.

[0138] The trained multi-level bidirectional long short-term memory network is used to output the first stratigraphic division conclusion corresponding to each second well logging curve.

[0139] The network level is determined by the target decomposition level of the logging curve. If the target decomposition level of the logging curve is 1, the network level is 2; if the target decomposition level of the logging curve is 2, the network level is 3, and so on. The number of layers and nodes of each level of the network is determined by experiments, and the parameters of this network are determined after multiple experiments.

[0140] The data input to the multi-level bidirectional long short-term memory network includes the target segment length of the well logging curve after preprocessing and the multi-scale components obtained after wavelet decomposition of the target segment length of the well logging curve. Taking the target segment length of the well logging curve before wavelet decomposition as an example, for the well logging curve segment with a target segment length of L, the data input to the multi-level bidirectional long short-term memory network is {x1, x2, x3, ..., x L}, where x i (i is 1, 2, 3, ..., L) is the i-th sampling point in the well logging curve segment, and the sampling points are input into the network layer A of the multi-level bidirectional long short-term memory network in the order of the sequence; the first-level low-frequency component obtained after the first wavelet decomposition of the target segment length of the well logging curve, and the second-level low-frequency component obtained after the second wavelet decomposition, are respectively input into the network layer B and network layer C of the multi-level bidirectional long short-term memory network. Among them, the network layer A, network layer B and network layer C of the multi-level bidirectional long short-term memory network have almost the same network structure, the difference is that the number of input layer nodes is different, because the size of the low-frequency component of the well logging curve segment obtained after wavelet decomposition is smaller than the size of the well logging curve segment before decomposition.

[0141] The five logging curves except the depth curve in the selected block of the present invention are sliced ​​according to the alternative length, and then divided into a training set and a verification set at a ratio of 9:1, and input into the constructed multi-level bidirectional long short-term memory network for stratigraphic division, that is, 90% of the sliced ​​data is used for model training, and 10% of the data is used to verify the performance of the model.

[0142] The relevant training parameters of the multi-level bidirectional long short-term memory network are set. The number of training iterations is set to 50, the training batch size is set to 1024, the initial learning rate of the network is 0.01, and the learning rate decays to 0.9 of the previous one after each training iteration cycle. The cross entropy loss function is selected as the loss function. After the training is completed, the stratigraphic division results based on each logging curve are given on the test set.

[0143] S208. Input the sampling point data of the target segment length of each second logging curve and the multi-scale components corresponding to each second logging curve into a pre-constructed long short-term memory network to obtain the second stratigraphic division conclusion corresponding to each second logging curve output by the long short-term memory network.

[0144] In this embodiment, the sampling point data of the target segment length of each second well logging curve and the multi-scale components corresponding to each second well logging curve are input into the pre-constructed long short-term memory network to obtain the second stratigraphic division conclusion based on the long short-term memory network and each second well logging curve. The multi-level bidirectional long short-term memory network constructed in this application is experimentally compared with the ordinary long short-term memory network to prove the superiority of constructing a multi-level bidirectional long short-term memory network.

[0145] S209. Based on the multiple first stratigraphic division conclusions and the multiple second stratigraphic division conclusions, a pre-stored adaptive fusion network is used to obtain a first stratigraphic division fusion conclusion obtained by fusing the multiple first stratigraphic division conclusions and a second stratigraphic division fusion conclusion obtained by fusing the multiple second stratigraphic division conclusions.

[0146] In this embodiment, a first stratum division fusion conclusion and a second stratum division fusion conclusion which fuse a plurality of first stratum division conclusions and a plurality of second stratum division conclusions can be obtained through a pre-stored adaptive fusion network.

[0147] On the one hand, an experimental comparison is made between the first stratigraphic division fusion conclusion obtained by the multi-level bidirectional long short-term memory network constructed by the present application and the second stratigraphic division fusion conclusion obtained by the ordinary long short-term memory network, which can prove the superiority of the multi-level bidirectional long short-term memory network constructed by the present application; on the other hand, the first stratigraphic division fusion conclusion and the second stratigraphic division fusion conclusion are respectively compared with multiple first stratigraphic division conclusions and multiple second stratigraphic division conclusions, which can prove the superiority of the first stratigraphic division fusion conclusion and the second stratigraphic division fusion conclusion.

[0148] The constructed adaptive fusion network consists of two multi-layer partially connected networks. In this embodiment, taking the two-layer partially connected network as an example, the first layer is a hidden layer with k nodes, and the second layer is an output layer with 1 node.

[0149] In this embodiment, optionally, a first fusion network formula and a second fusion network formula are pre-stored in the adaptive fusion network. When the adaptive fusion network is a two-layer partially connected network, the first prediction probability is obtained according to the first fusion network formula and the plurality of first stratum division conclusions; the first fusion network formula is:

[0150]

[0151] in, is the first predicted probability, ReLU() is the linear correction function, G j is the fusion conclusion obtained based on multiple first stratum division conclusions, W k is a partially connected layer with k nodes, b kis the bias of the partially connected layer, W1 is the output layer, and b1 is the bias of the output layer;

[0152] Obtaining a first stratum division fusion conclusion according to the first prediction probability;

[0153] Through the second fusion network formula, according to the multiple second stratum division conclusions, the second prediction probability is obtained; the second fusion network formula is:

[0154]

[0155] in, is the second predicted probability, G m It is a fusion conclusion based on multiple second stratigraphic division conclusions;

[0156] A second stratum division fusion conclusion is obtained according to the second prediction probability.

[0157] The decisions between different formations do not interfere with each other and are only related to the prediction capabilities of multiple logging curves for the formation. The prediction of each logging curve for other formations does not interfere with the prediction results for this formation.

[0158] This embodiment provides a specific formula, which combines multiple first stratigraphic division conclusions and multiple second stratigraphic division conclusions through a pre-stored adaptive fusion network to obtain a first stratigraphic division fusion conclusion obtained by fusing multiple first stratigraphic division conclusions, and a second stratigraphic division fusion conclusion obtained by fusing multiple second stratigraphic division conclusions.

[0159] S210, obtaining the stratigraphic division accuracy of each of a plurality of first stratigraphic division conclusions, a plurality of second stratigraphic division conclusions, a first stratigraphic division fusion conclusion, and a second stratigraphic division fusion conclusion.

[0160] In this embodiment, multiple first stratigraphic division conclusions, multiple second stratigraphic division conclusions, first stratigraphic division fusion conclusions and second stratigraphic division fusion conclusions are compared with their respective actual conclusions to obtain their respective stratigraphic division accuracies, and then compared based on the stratigraphic division accuracies to obtain the superiority of the constructed multi-level bidirectional long short-term memory network and the superiority of the stratigraphic division conclusions that integrate a single well logging curve.

[0161] S211. According to the accuracy of multiple stratum divisions, the training result of the multi-level bidirectional long short-term memory network after training is obtained.

[0162] In this embodiment, the training result is successful training, which is used to indicate that the stratum division accuracy of the multi-level bidirectional long short-term memory network after the training is completed is higher than the stratum division accuracy of the ordinary long short-term memory network.

[0163] As shown in Table 3, when the multi-level bidirectional LSTM network constructed by the present invention is applied to well logging curves, compared with the ordinary LSTM network, the stratigraphic division accuracy of each well logging curve is improved, among which the improvement for the GR curve is the most obvious, which is increased by about 3.5%.

[0164] After extracting features from each logging curve, the accuracy of formation division through adaptive fusion is much higher than the results of formation division prediction based on each logging curve alone.

[0165] Based on a multi-level bidirectional long short-term memory network, the accuracy of stratum division without depth curve and with depth curve can reach 84.952% and 89.916% respectively, which are 1.9% and 1.3% higher than the ordinary long short-term memory network. After each logging curve is extracted using an ordinary long short-term memory network, it is sent to the adaptive fusion network constructed by the present application. The stratum division accuracy of the output stratum division conclusion is also higher than the stratum division accuracy of the stratum division conclusion of a single logging well. Therefore, the adaptive fusion network constructed by the present invention has a good performance in multi-curve fusion.

[0166] Table 3

[0167]

[0168] The present application provides a model training method and a method for extracting stratigraphic division conclusions. Based on this method, the following technical effects are achieved: multi-scale wavelet decomposition is performed on the logging curve, and the time series characteristics of the logging data are extracted at multiple scales, so that richer characteristics can be obtained, which is conducive to obtaining more accurate stratigraphic division; the traditional wavelet transform is combined with the multi-level bidirectional long short-term memory network in deep learning, and the time series components of the logging curve can be automatically learned from multiple scales without the need for manual analysis of the characteristics obtained after the wavelet transform, so as to automatically obtain stratigraphic division conclusions based on different logging curves; the stratigraphic division conclusion based on a single curve is adaptively weighted and then the stratigraphic division is predicted, which is conducive to improving the accuracy of the conclusions of the logging interpretation task.

[0169] Figure 3 Schematic diagram of the process of extracting the stratigraphic division conclusion provided in the embodiment of the present application Figure 1 ,like Figure 3 As shown, the method for extracting stratum division conclusions provided in this embodiment includes the following steps:

[0170] S301, obtaining a plurality of third logging curves of the well to be tested; the plurality of third logging curves are used to indicate depth, natural gamma, natural potential, acoustic time difference and dual lateral resistivity respectively.

[0171] In this embodiment, the plurality of third logging curves refer to depth curves, natural gamma curves, natural potential curves, acoustic time difference curves and dual lateral resistivity curves. The depth curves, natural gamma curves, natural potential curves, acoustic time difference curves and dual lateral resistivity curves of the well to be tested are used as the basis for extracting the conclusion of formation division.

[0172] S302: Preprocess multiple third logging curves to obtain a fourth logging curve corresponding to each third logging curve.

[0173] In this embodiment, the fourth logging curve corresponding to each third logging curve refers to the preprocessed depth curve, natural gamma curve, natural potential curve, acoustic time difference curve and dual lateral resistivity curve. The depth curve, natural gamma curve, natural potential curve, acoustic time difference curve and dual lateral resistivity curve of the well to be tested are preprocessed, and the preprocessing method can adopt the method in S202 or other methods.

[0174] In this embodiment, the effect of the preprocessing is similar to the effect of the preprocessing in S102 in the first embodiment of the present invention, and will not be described in detail here.

[0175] S303 , performing multi-scale wavelet decomposition on the target segment lengths of multiple fourth well logging curves to obtain a multi-scale component corresponding to each fourth well logging curve.

[0176] In this embodiment, the effect of the multi-scale wavelet decomposition in S303 is similar to the effect of the multi-scale wavelet decomposition in S103 in the first embodiment of the present invention, and will not be described in detail here.

[0177] S304. Input the sampling point data of the target segment length of each fourth logging curve and the multi-scale components corresponding to each fourth logging curve into a pre-stored multi-level bidirectional long short-term memory network to obtain the third stratigraphic division conclusion corresponding to each fourth logging curve.

[0178] In this embodiment, the multi-level bidirectional long short-term memory network is a model obtained by training in the above-mentioned embodiment 1 or embodiment 2. The sampling point data of the target segment length of the pre-processed depth curve, natural gamma curve, natural potential curve, acoustic time difference curve and dual lateral resistivity curve, and the multi-scale components corresponding thereto, are input into the trained multi-level bidirectional long short-term memory network, and the stratigraphic division conclusion of the depth curve, natural gamma curve, natural potential curve, acoustic time difference curve and dual lateral resistivity curve can be automatically extracted.

[0179] Figure 4 Schematic diagram of the process of extracting the stratigraphic division conclusion provided in the embodiment of the present application Figure 2 .like Figure 4As shown, the stratigraphic division conclusion extraction method provided in this embodiment is a further refinement of the stratigraphic division conclusion extraction method provided in the previous embodiment of the present invention. The stratigraphic division conclusion extraction method provided in this embodiment includes the following steps:

[0180] S401, obtaining a plurality of third logging curves of the well to be tested; the plurality of third logging curves are respectively used to indicate depth, natural gamma, natural potential, acoustic time difference and dual lateral resistivity.

[0181] In this embodiment, the effect of S401 is similar to the effect of S301 in the previous embodiment of the present invention, and will not be described in detail here.

[0182] S402: Preprocess multiple third logging curves to obtain a fourth logging curve corresponding to each third logging curve.

[0183] In this embodiment, the effect of S402 is similar to the effect of S302 in the previous embodiment of the present invention, and will not be repeated here.

[0184] S403 , performing multi-scale wavelet decomposition on target segment lengths of multiple fourth well logging curves to obtain multi-scale components corresponding to each fourth well logging curve.

[0185] In this embodiment, the effect of the multi-scale wavelet decomposition in S403 is similar to the effect of the multi-scale wavelet decomposition in S103 in the first embodiment of the present invention, and will not be described in detail here.

[0186] S404, input the sampling point data of the target segment length of each fourth logging curve and the multi-scale components corresponding to each fourth logging curve into a pre-stored multi-level bidirectional long short-term memory network to obtain the third stratigraphic division conclusion corresponding to each fourth logging curve.

[0187] In this embodiment, the multi-level bidirectional long short-term memory network is a model obtained by training in the above-mentioned embodiment 1 or embodiment 2.

[0188] The method of extracting the stratum division conclusion in S404 is similar to the method of extracting the stratum division conclusion in S304 in the previous embodiment of the present invention, and will not be repeated here.

[0189] S405. According to the plurality of third stratum division conclusions, a third stratum division fusion conclusion obtained by fusing the plurality of third stratum division conclusions is obtained through a pre-stored adaptive fusion network.

[0190] In this embodiment, optionally, a third fusion network formula is pre-stored in the adaptive fusion network. When the adaptive fusion network is a two-layer partially connected network, the third prediction probability is obtained according to the third fusion network formula based on multiple third stratum division conclusions; the third fusion network formula is:

[0191]

[0192] in, is the third prediction probability, ReLU() is the linear correction function, G n is the fusion conclusion obtained based on multiple third-stratum division conclusions, W k is a partially connected layer with k nodes, b k is the bias of the partially connected layer, W1 is the output layer, and b1 is the bias of the output layer;

[0193] The third stratigraphic division fusion conclusion is obtained according to the third prediction probability.

[0194] The embodiment of the present invention can divide the electronic device or the main control device into functional modules according to the above method example. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiment of the present invention is schematic and is only a logical function division. There may be other division methods in actual implementation.

[0195] Figure 5 This is a schematic diagram of the structure of the model training device provided in the embodiment of the present application. Figure 5 As shown, in this embodiment, the model training device 500 can be located in an electronic device. The model training device 500 includes:

[0196] Acquisition module 501: used to acquire multiple first logging curves of target logging; the multiple first logging curves are used to indicate depth, natural gamma, natural potential, acoustic time difference and dual lateral resistivity respectively;

[0197] Preprocessing module 502: used to preprocess a plurality of first well logging curves to obtain a second well logging curve corresponding to each first well logging curve;

[0198] The wavelet decomposition module 503 is used to perform multi-scale wavelet decomposition on the target segment lengths of the plurality of second well logging curves to obtain a multi-scale component corresponding to each second well logging curve;

[0199] The training module 504 is used to input the sampling point data of the target segment length of each second logging curve and the multi-scale components corresponding to each second logging curve into a pre-constructed multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network; wherein the multi-level bidirectional long short-term memory network is constructed by multiple bidirectional long short-term memory networks with different network parameters; the trained multi-level bidirectional long short-term memory network is used to output the first stratigraphic division conclusion corresponding to each second logging curve.

[0200] The model training device provided in this embodiment can be executed Figure 1 The technical solution of the model training method embodiment shown in the figure has the same implementation principle and technical effect as Figure 1 The model training method embodiments shown are similar and will not be described in detail here.

[0201] At the same time, the model training device provided by the present invention further refines the model training device 500 based on the model training device provided in the previous embodiment.

[0202] Optionally, in this embodiment, the preprocessing module 502 is specifically used for:

[0203] The first target logging curve is any one of the plurality of first logging curves, and for the first target logging curve, outlier processing is performed on the first target logging curve;

[0204] The sampling point data of the first logging curve after outlier processing is normalized by a normalization processing formula to obtain the sampling point data of the second logging curve corresponding to the first target logging curve.

[0205] Optionally, in this embodiment, the wavelet decomposition module 503 is specifically used for:

[0206] The second target logging curve is any one of the plurality of second logging curves. For the second target logging curve, a target segment length, a target decomposition wavelet basis and a target decomposition level of the second target logging curve are obtained according to a pre-stored formation division accuracy rate.

[0207] performing at least one wavelet decomposition on a target segment length of the second target well logging curve according to a target decomposition level and a target decomposition wavelet basis of the second target well logging curve;

[0208] A multi-scale component corresponding to the second target logging curve is obtained according to the low-frequency component obtained by each wavelet decomposition; wherein the multi-scale component corresponding to the second target logging curve refers to a set of at least one low-frequency component.

[0209] Optionally, when the wavelet decomposition module 503 performs at least one wavelet decomposition on the target segment length of the second target well logging curve according to the target decomposition level and the target decomposition wavelet basis of the second target well logging curve, the wavelet decomposition module 503 performs at least one wavelet decomposition on the target segment length of the second target well logging curve according to the target decomposition wavelet basis of the second target well logging curve until the number of wavelet decompositions reaches the target decomposition level of the second target well logging curve;

[0210] The target wavelet decomposition is any one of the at least one wavelet decompositions. After performing the target wavelet decomposition on the target segment length of the second target well logging curve according to the target decomposition wavelet basis of the second target well logging curve, the method further includes:

[0211] The target segment length is updated to the low-frequency component obtained by the target-th wavelet decomposition.

[0212] Optionally, in this embodiment, the training module 504 is specifically used for:

[0213] The multi-level bidirectional long short-term memory network contains n+1 network layers, and the target segment length of the second target well logging curve is subjected to n wavelet decompositions;

[0214] Input the sampling point data of the target segment length of each second logging curve into the first network layer of the multi-level bidirectional long short-term memory network;

[0215] The low-frequency component obtained by the i-th wavelet decomposition is input into the i+1-th network layer of the multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network; wherein i is a positive integer less than or equal to n.

[0216] Optionally, in this embodiment, the model training device 500 includes, in addition to the above modules, a fusion module 505, which is specifically used for:

[0217] Input the sampling point data of the target segment length of each second well logging curve and the multi-scale components corresponding to each second well logging curve into a pre-constructed long short-term memory network to obtain the second formation division conclusion corresponding to each second well logging curve output by the long short-term memory network;

[0218] According to the plurality of first stratigraphic division conclusions and the plurality of second stratigraphic division conclusions, a first stratigraphic division fusion conclusion obtained by fusing the plurality of first stratigraphic division conclusions and a second stratigraphic division fusion conclusion obtained by fusing the plurality of second stratigraphic division conclusions are obtained through a pre-stored adaptive fusion network;

[0219] Obtaining the stratigraphic division accuracy of each of a plurality of first stratigraphic division conclusions, a plurality of second stratigraphic division conclusions, a first stratigraphic division fusion conclusion, and a second stratigraphic division fusion conclusion;

[0220] According to multiple stratigraphic division accuracy rates, the training result of the multi-level bidirectional long short-term memory network after training is obtained; wherein the training result is successful training, which is used to indicate that the stratigraphic division accuracy rate of the multi-level bidirectional long short-term memory network after training is higher than the stratigraphic division accuracy rate of the long short-term memory network.

[0221] Optionally, when the fusion module 505 obtains the first stratum division fusion conclusion obtained by fusing the multiple first stratum division conclusions and the second stratum division fusion conclusion obtained by fusing the multiple second stratum division conclusions through a pre-stored adaptive fusion network according to the multiple first stratum division conclusions and the multiple second stratum division conclusions, the first fusion network formula and the second fusion network formula are pre-stored in the adaptive fusion network;

[0222] When the adaptive fusion network is a two-layer partially connected network, the first prediction probability is obtained according to the first fusion network formula based on the multiple first stratum division conclusions; the first fusion network formula is:

[0223]

[0224] in, is the first predicted probability, ReLU() is the linear correction function, G j is the fusion conclusion obtained based on multiple first stratum division conclusions, W k is a partially connected layer with k nodes, b k is the bias of the partially connected layer, W1 is the output layer, and b1 is the bias of the output layer;

[0225] Obtaining a first stratum division fusion conclusion according to the first prediction probability;

[0226] Through the second fusion network formula, according to the multiple second stratum division conclusions, the second prediction probability is obtained; the second fusion network formula is:

[0227]

[0228] in, is the second predicted probability, G m It is a fusion conclusion based on multiple second stratigraphic division conclusions;

[0229] A second stratum division fusion conclusion is obtained according to the second prediction probability.

[0230] The model training device provided in this embodiment can be executed Figure 1 and Figure 2 The technical solution of the model training method embodiment shown in the figure has the same implementation principle and technical effect as Figure 1 and Figure 2 The model training method embodiments shown are similar and will not be described in detail here.

[0231] Figure 6 This is a schematic diagram of the structure of the stratigraphic division conclusion extraction device provided in the embodiment of the present application. Figure 6 As shown, in this embodiment, the stratigraphic division conclusion extraction device 600 can be located in an electronic device. The model training device 600 includes:

[0232] Acquisition module 601: used to acquire multiple third logging curves of the well to be tested; the multiple third logging curves are used to indicate depth, natural gamma, natural potential, acoustic wave time difference and dual lateral resistivity respectively;

[0233] Preprocessing module 602: used to preprocess multiple third logging curves to obtain a fourth logging curve corresponding to each third logging curve;

[0234] The wavelet decomposition module 603 is used to perform multi-scale wavelet decomposition on the target segment lengths of the plurality of fourth well logging curves to obtain a multi-scale component corresponding to each fourth well logging curve;

[0235] The extraction module 604 is used to input the sampling point data of the target segment length of each fourth logging curve and the multi-scale components corresponding to each fourth logging curve into a pre-stored multi-level bidirectional long short-term memory network to obtain the third stratigraphic division conclusion corresponding to each fourth logging curve.

[0236] The stratum division conclusion extraction device provided in this embodiment can be executed Figure 3 The technical solution of the embodiment of the method for extracting the conclusion of stratigraphic division is shown in FIG. Figure 3 The embodiments of the method for extracting the stratigraphic division conclusion are similar and will not be described in detail here.

[0237] At the same time, the stratum division conclusion extraction device provided by the present invention further refines the stratum division conclusion extraction device 600 based on the stratum division conclusion extraction device provided in the previous embodiment.

[0238] Optionally, in this embodiment, in addition to the above modules, the stratum division conclusion extraction device 600 further includes a fusion module 605, which is specifically used for:

[0239] According to the plurality of third stratum division conclusions, a third stratum division fusion conclusion obtained by fusing the plurality of third stratum division conclusions is obtained through a pre-stored adaptive fusion network.

[0240] Optionally, when the fusion module 605 obtains a third stratum division fusion conclusion obtained by fusing multiple third stratum division conclusions through a pre-stored adaptive fusion network according to multiple third stratum division conclusions, a third fusion network formula is pre-stored in the adaptive fusion network;

[0241] When the adaptive fusion network is a two-layer partially connected network, the third prediction probability is obtained according to the third fusion network formula based on multiple third stratum division conclusions; the fusion network formula is:

[0242]

[0243] in, is the third prediction probability, ReLU() is the linear correction function, G n is the fusion conclusion obtained based on multiple third-stratum division conclusions, W k is a partially connected layer with k nodes, b k is the bias of the partially connected layer, W1 is the output layer, and b1 is the bias of the output layer;

[0244] The third stratigraphic division fusion conclusion is obtained according to the third prediction probability.

[0245] The stratum division conclusion extraction device provided in this embodiment can be executed Figure 3 and Figure 4 The technical solution of the embodiment of the method for extracting the conclusion of stratigraphic division is shown in FIG. Figure 3 and Figure 4 The stratigraphic division conclusion extraction method embodiments shown are similar and will not be described in detail here.

[0246] Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device is intended to be used in various electronic devices that can execute the model training method or the stratum division conclusion extraction method, such as a microcomputer, a single-chip microcomputer, and other suitable computers. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0247] like Figure 7 As shown, the electronic device 70 includes: at least one processor 701 and a memory 702. The electronic device 70 also includes a communication component 703. The processor 701, the memory 702 and the communication component 703 are connected via a bus 704.

[0248] In the specific implementation process, at least one processor 701 executes the computer execution instructions stored in the memory 702, so that at least one processor 701 executes the model training method or the stratigraphic division conclusion extraction method executed by the electronic device side as above.

[0249] The specific implementation process of the processor 701 can refer to the above-mentioned model training method embodiment or the stratigraphic division conclusion extraction method embodiment, and the implementation principles and technical effects are similar, so this embodiment will not be repeated here.

[0250] In the above embodiment, it should be understood that the processor 701 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. The general-purpose processor 701 can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the invention can be directly embodied as being executed by a hardware processor, or executed by a combination of hardware and software modules in the processor.

[0251] The memory 702 may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk storage.

[0252] The bus 704 may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus 704 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus 704 in the drawings of the present application is not limited to only one bus or one type of bus.

[0253] The above-mentioned functions implemented by the electronic device and the main control device introduce the scheme provided by the embodiment of the present invention. It can be understood that in order to implement the above-mentioned functions, the electronic device or the main control device includes a hardware structure and / or software module corresponding to the execution of each function. In combination with the units and algorithm steps of each example described in the embodiment disclosed in the embodiment of the present invention, the embodiment of the present invention can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the technical solution of the embodiment of the present invention.

[0254] The present application also provides a computer-readable storage medium, in which computer execution instructions are stored. When a processor executes the computer execution instructions, the above model training method or stratigraphic division conclusion extraction method is implemented.

[0255] The computer-readable storage medium mentioned above can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0256] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. The readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). The processor and the readable storage medium can also exist as discrete components in an electronic device or a main control device.

[0257] The memory 702 is a non-transient computer-readable storage medium provided by the present invention. The non-transient computer-readable storage medium of the present invention stores computer instructions, which are used to enable a computer to execute the model training method or the stratum division conclusion extraction method provided by the present invention.

[0258] The memory 702 is a non-transient computer-readable storage medium that can be used to store non-transient software programs, non-transient computer executable programs and modules, such as program instructions / modules corresponding to the model training method or the stratigraphic division conclusion extraction method in the embodiment of the present invention (for example, Figure 5 The acquisition module 501, the preprocessing module 502, the wavelet decomposition module 503, the training module 504 or Figure 6 The processor 701 executes various functional applications and data processing by running the non-transient software programs, instructions and modules stored in the memory 702, that is, the model training method or the stratigraphic division conclusion extraction method in the above method embodiment is realized.

[0259] At the same time, this embodiment also provides a computer program product. When the instructions in the computer program product are executed by a processor, the model training method or the stratigraphic division conclusion extraction method of the above-mentioned embodiment can be executed.

[0260] Those skilled in the art will readily appreciate other implementations of the embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the embodiments of the present invention that follow the general principles of the embodiments of the present invention and include common knowledge or customary technical means in the art that are not disclosed in the embodiments of the present invention. The specification and examples are to be regarded as exemplary only, and the true scope and spirit of the embodiments of the present invention are indicated by the claims.

[0261] It should be understood that the embodiments of the present invention are not limited to the precise structures described above and shown in the drawings, and various modifications and changes may be made without departing from the scope thereof. The scope of the embodiments of the present invention is limited only by the appended claims.

Claims

1. A model training method, characterized in that: include: Acquire a plurality of first well logging curves of the target well logging; The plurality of first logging curves are used to indicate depth, natural gamma, natural potential, acoustic time difference and dual lateral resistivity respectively; Preprocessing the plurality of the first well logging curves to obtain a second well logging curve corresponding to each of the first well logging curves; Performing multi-scale wavelet decomposition on target segment lengths of a plurality of the second well logging curves to obtain a multi-scale component corresponding to each of the second well logging curves; The sampling point data of the target segment length of each second well logging curve and the multi-scale components corresponding to each second well logging curve are input into a pre-constructed multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network; wherein the multi-level bidirectional long short-term memory network is constructed by a plurality of bidirectional long short-term memory networks with different network parameters; the trained multi-level bidirectional long short-term memory network is used to output the first stratigraphic division conclusion corresponding to each second well logging curve.

2. The model training method according to claim 1, characterized in that: The second target well logging curve is any one of the plurality of the second well logging curves; For the second target well logging curve, performing multi-scale wavelet decomposition on the target segment lengths of the plurality of second well logging curves to obtain a multi-scale component corresponding to each of the second well logging curves includes: According to the pre-stored formation division accuracy, the target segment length, target decomposition wavelet basis and target decomposition level of the second target well logging curve are obtained; performing at least one wavelet decomposition on a target segment length of the second target well logging curve according to a target decomposition level and a target decomposition wavelet basis of the second target well logging curve; A multi-scale component corresponding to the second target logging curve is obtained according to the low-frequency component obtained by each wavelet decomposition; wherein the multi-scale component corresponding to the second target logging curve refers to a set of at least one of the low-frequency components.

3. The model training method according to claim 2, characterized in that: The step of performing at least one wavelet decomposition on the target segment length of the second target well logging curve according to the target decomposition level and the target decomposition wavelet basis of the second target well logging curve comprises: According to the target decomposition wavelet basis of the second target well logging curve, performing at least one wavelet decomposition on the target segment length of the second target well logging curve until the number of wavelet decompositions reaches the target decomposition level of the second target well logging curve; The target wavelet decomposition is any one of the at least one wavelet decompositions. After performing the target wavelet decomposition on the target segment length of the second target well logging curve according to the target decomposition wavelet basis of the second target well logging curve, the method further includes: The target segment length is updated to the low-frequency component obtained by the target-th wavelet decomposition.

4. The model training method according to claim 1, characterized in that: After inputting the sampling point data of the target segment length of each second well logging curve and the multi-scale components corresponding to each second well logging curve into a pre-constructed multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network, the method further includes: Input the sampling point data of the target segment length of each second well logging curve and the multi-scale component corresponding to each second well logging curve into a pre-constructed long short-term memory network to obtain the second formation division conclusion corresponding to each second well logging curve output by the long short-term memory network; According to the plurality of the first stratigraphic division conclusions and the plurality of the second stratigraphic division conclusions, a first stratigraphic division fusion conclusion obtained by fusing the plurality of the first stratigraphic division conclusions and a second stratigraphic division fusion conclusion obtained by fusing the plurality of the second stratigraphic division conclusions are obtained through a pre-stored adaptive fusion network; Obtaining the stratigraphic division accuracy of each of the plurality of the first stratigraphic division conclusions, the plurality of the second stratigraphic division conclusions, the first stratigraphic division fusion conclusion, and the second stratigraphic division fusion conclusion; According to the multiple stratum division accuracy rates, the training result of the multi-level bidirectional long short-term memory network after the training is completed is obtained; wherein, the training result is successful training, which is used to indicate that the stratum division accuracy rate of the multi-level bidirectional long short-term memory network after the training is completed is higher than the stratum division accuracy rate of the long short-term memory network.

5. The model training method according to claim 4, characterized in that: The adaptive fusion network has a first fusion network formula and a second fusion network formula pre-stored therein; When the adaptive fusion network is a two-layer partially connected network, the method of obtaining a first stratum division fusion conclusion obtained by fusing the plurality of first stratum division conclusions and a second stratum division fusion conclusion obtained by fusing the plurality of second stratum division conclusions through a pre-stored adaptive fusion network according to the plurality of first stratum division conclusions and the plurality of second stratum division conclusions includes: Through the first fusion network formula, according to the plurality of first stratum division conclusions, a first prediction probability is obtained; the first fusion network formula is: Among them, the is the first predicted probability, the ReLU() is a linear correction function, and the G j is a fusion conclusion obtained based on a plurality of the first stratigraphic division conclusions, wherein W k is a partially connected layer with k nodes, the b k is the bias of the partially connected layer, W1 is the output layer, and b1 is the bias of the output layer; Obtaining the first stratum division fusion conclusion according to the first prediction probability; Through the second fusion network formula, according to the plurality of the second stratum division conclusions, the second prediction probability is obtained; the second fusion network formula is: Among them, the is the second predicted probability, the G m A fusion conclusion based on a plurality of the second stratigraphic division conclusions; The second stratum division fusion conclusion is obtained according to the second prediction probability.

6. The model training method according to claim 3, characterized in that: The multi-level bidirectional long short-term memory network includes n+1 network layers, and the target segment length of the second target well logging curve is subjected to n wavelet decompositions; The step of inputting the sampling point data of the target segment length of each second well logging curve and the multi-scale components corresponding to each second well logging curve into a pre-constructed multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network comprises: Input the sampling point data of the target segment length of each second well logging curve into the first network layer of the multi-level bidirectional long short-term memory network; The low-frequency component obtained by the i-th wavelet decomposition is input into the i+1-th network layer of the multi-level bidirectional long short-term memory network to train the multi-level bidirectional long short-term memory network; wherein i is a positive integer less than or equal to n.

7. The model training method according to claim 1, characterized in that: The first target well logging curve is any one of the plurality of the first well logging curves; For the first target well logging curve, preprocessing the plurality of the first well logging curves to obtain a second well logging curve corresponding to each of the first well logging curves includes: performing outlier processing on the first target logging curve; The sampling point data of the first logging curve after outlier processing is normalized by a normalization processing formula to obtain the sampling point data of the second logging curve corresponding to the first target logging curve.

8. A method for extracting stratigraphic division conclusions, characterized in that: include: Acquire multiple third logging curves of the well to be tested; The plurality of third logging curves are respectively used to indicate depth, natural gamma, natural potential, acoustic wave time difference and dual lateral resistivity; Preprocessing the plurality of third well logging curves to obtain a fourth well logging curve corresponding to each of the third well logging curves; Performing multi-scale wavelet decomposition on target segment lengths of a plurality of the fourth well logging curves to obtain a multi-scale component corresponding to each of the fourth well logging curves; The sampling point data of the target segment length of each of the fourth logging curves and the multi-scale components corresponding to each of the fourth logging curves are input into a pre-stored multi-level bidirectional long short-term memory network to obtain a third stratigraphic division conclusion corresponding to each of the fourth logging curves; wherein the multi-level bidirectional long short-term memory network is a model obtained by the model training method described in any one of claims 1 to 7.

9. The method for extracting stratigraphic division conclusion according to claim 8, characterized in that: After inputting the sampling point data of the target segment length of each of the fourth well logging curves and the multi-scale components corresponding to each of the fourth well logging curves into a pre-stored multi-level bidirectional long short-term memory network to obtain the third formation division conclusion corresponding to each of the fourth well logging curves, the method further includes: According to the plurality of third stratum division conclusions, a third stratum division fusion conclusion obtained by fusing the plurality of third stratum division conclusions is obtained through a pre-stored adaptive fusion network.

10. The method for extracting stratigraphic division conclusion according to claim 9, characterized in that: The adaptive fusion network has a third fusion network formula pre-stored therein; When the adaptive fusion network is a two-layer partially connected network, the third stratum division fusion conclusion obtained by fusing the plurality of the third stratum division conclusions through the pre-stored adaptive fusion network according to the plurality of the third stratum division conclusions includes: Through the third fusion network formula, according to the plurality of third stratum division conclusions, a third prediction probability is obtained; the fusion network formula is: Among them, the is the third predicted probability, the ReLU() is a linear correction function, and the G n is a fusion conclusion obtained based on a plurality of the third stratum division conclusions, wherein W k is a partially connected layer with k nodes, the b k is the bias of the partially connected layer, W1 is the output layer, and b1 is the bias of the output layer; The third stratigraphic division fusion conclusion is obtained according to the third prediction probability.

Images