A multi-well lithofacies identification method, device and medium

Abstract

The application discloses a kind of multi-well lithofacies identification method, device and medium, by the original well logging data of acquisition data pre-processing is carried out to obtain training data set, based on training data set to initial identification model is trained, to obtain target identification model, the well channel data of target well channel that needs to be identified, spontaneous potential data and well logging project data are respectively input into the corresponding branch model in target identification model, and the output each branch feature is weighted and fused, to obtain corresponding target fusion feature, based on the mid-frequency feature of target fusion feature, by the global discriminator and local discriminator of target identification model after the alignment training of confrontation are output identification result, therefore the geological feature of each branch model identification based on different angles in the present application, improve the reliability of identification, and by the mid-frequency feature of target fusion feature is aligned, retain the high frequency and low frequency information reflecting real geological difference, to realize to improve the identification accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of oil and gas exploration and development, and particularly relates to a multi-well lithofacies identification method, device and medium. BACKGROUND

[0002] In the process of oil and gas exploration and development, lithofacies identification is the basis for reservoir evaluation, reserve calculation and development plan formulation.

[0003] For the identification of lithofacies, the existing scheme is usually based on a machine learning scheme and logging data of a single depth point, or deep learning of deep features of a logging curve to achieve lithofacies identification; however, the scheme of identifying lithofacies based on logging data of a single depth point through a machine learning scheme ignores the continuity and spatial correlation of stratum deposition, and thus has low identification accuracy; the scheme of identifying lithofacies by deep learning to capture deep features of a logging curve has a serious cross-well generalization problem, and thus has low identification reliability.

[0004] Therefore, the existing scheme has the defects of low accuracy and reliability of lithofacies identification results. SUMMARY

[0005] The technical problem to be solved by the present application is the defects of low accuracy and reliability of the existing lithofacies identification scheme, and the purpose is to provide a multi-well lithofacies identification method, device and medium, which solves the problems of low accuracy and reliability in the existing lithofacies identification scheme.

[0006] The present application is achieved by the following technical scheme:

[0007] In a first aspect, the present application provides a multi-well lithofacies identification method, comprising:

[0008] Obtaining original logging data and pre-processing the original logging data to obtain a training data set, wherein the training data set includes multiple groups of track data, the track data includes logging item data at different depth positions of each track, and facies class labels corresponding to the depth positions, the facies class labels are used to indicate expert-annotated lithofacies classes;

[0009] Training an initial identification model based on the training data set to obtain a target identification model;

[0010] Obtaining target track data of a target track to be identified, and inputting the target track data into an adaptive wavelet transform branch model in the target identification model to obtain corresponding multi-scale features output by the adaptive wavelet transform branch model;

[0011] Input the spontaneous potential data in the logging project data of the target shaft data into a spontaneous potential transformation branch model in the target recognition model to obtain a sedimentary rhythm feature corresponding to a spontaneous potential transformation branch model output;

[0012] Input a key project logging curve pair in the target shaft data into a logging curve analysis branch model in the target recognition model to obtain a rock physical relationship feature corresponding to a logging curve analysis branch model output, wherein the key project logging curve pair is used to indicate a corresponding relationship between different two project data in the logging project data;

[0013] Weighted fusion is performed on the multi-scale feature, the sedimentary rhythm feature and the rock physical relationship feature to obtain a corresponding target fusion feature;

[0014] Based on the medium frequency feature of the target fusion feature, an identification result is output by a global discriminator and a local discriminator in the target recognition model after adversarial alignment training, wherein the identification result includes a lithofacies classification result of each depth point in the target shaft and a lithofacies conversion boundary position.

[0015] In a possible design, the preprocessing of the original logging data to obtain a training data set includes:

[0016] Data cleaning is performed on the original logging data;

[0017] Based on the logging data after data cleaning, normalization processing is performed, and the logging data after normalization processing is organized as a three-dimensional vector and a two-dimensional vector, and the training data set is obtained according to the three-dimensional vector and the two-dimensional vector;

[0018] The three-dimensional vector is used to indicate logging project data at different depth positions of the shaft, and the two-dimensional vector is used to indicate facies labels at the corresponding depth positions, and the two-dimensional vector is associated with the corresponding three-dimensional vector according to the same depth position.

[0019] In a possible design, the training data set is obtained according to the three-dimensional vector and the two-dimensional vector, including:

[0020] A sliding window with a preset width is constructed;

[0021] The three-dimensional vector and the corresponding two-dimensional vector are slidingly sampled along the direction of the shaft depth indicated by the three-dimensional vector to obtain a plurality of window data;

[0022] The plurality of window data is used as the training data set.

[0023] In a possible design, the obtaining the multi-scale feature corresponding to the output of the adaptive wavelet transform branch model comprises:

[0024] According to different depth positions of each logging item data in the target well track data, an input index corresponding to the logging item data is established;

[0025] According to the input index of each logging item data, the odd and even samples are obtained by separating each logging item data in the target well track data.

[0026] Based on the odd and even samples, a plurality of detail coefficients and approximation coefficients are calculated and obtained by using the learnable prediction operator weight and the update operator weight in the adaptive wavelet transform branch model, and the plurality of detail coefficients and the plurality of approximation coefficients are used as the multi-scale feature.

[0027] In a possible design, the obtaining the sedimentary rhythm feature corresponding to the output of the spontaneous potential transform branch model comprises:

[0028] According to the spontaneous potential data, a spontaneous potential curve is established;

[0029] According to the spontaneous potential curve and the preset mother wavelet, the continuous wavelet transform is performed by using the spontaneous potential transform branch model, so as to extract the corresponding sedimentary rhythm feature.

[0030] In a possible design, the corresponding relationship between different two item data in the logging item data comprises:

[0031] A first corresponding relationship, the first corresponding relationship is used to indicate the relationship between shale content and porosity;

[0032] A second corresponding relationship, the second corresponding relationship is used to indicate the relationship between skeleton density and propagation velocity;

[0033] A third corresponding relationship, the third corresponding relationship is used to indicate the difference between deep resistivity and shallow resistivity;

[0034] A fourth corresponding relationship, the fourth corresponding relationship is used to indicate the relationship between permeability and shale content.

[0035] In a possible design, the outputting the identification result by using the global discriminator and the local discriminator trained by the adversarial alignment in the target identification model comprises:

[0036] The target fusion feature is decomposed to obtain a low-frequency feature, the medium-frequency feature, and a high-frequency feature, wherein the low-frequency feature is used to indicate large-scale geological background information, the medium-frequency feature is used to indicate instrument response difference information, and the high-frequency feature is used to indicate rock texture detail information.

[0037] The first loss function of the global discriminator and the second loss function of the local discriminator are used to realize the adversarial training of the global discriminator and the local discriminator, wherein the first loss function is used to eliminate the instrument response difference between different wells, and the second loss function is used to retain the lithofacies classification features of the target track.

[0038] In a second aspect, the present application provides a multi-well lithofacies identification device, comprising:

[0039] An acquisition module is configured to acquire original logging data and pre-process the original logging data to obtain a training data set, wherein the training data set includes multiple groups of track data, the track data includes logging item data at different depth positions of each track, and facies class labels corresponding to the depth positions, the facies class labels are used to indicate expert-annotated lithofacies classes, and an initial identification model is trained based on the training data set to obtain a target identification model;

[0040] A first processing module is configured to acquire target track data of a target track to be identified, input the target track data into an adaptive wavelet transform branch model in the target identification model to obtain corresponding multi-scale features output by the adaptive wavelet transform branch model, and input spontaneous potential data in logging item data of the target track data into a spontaneous potential transform branch model in the target identification model to obtain corresponding sedimentary rhythm features output by the spontaneous potential transform branch model;

[0041] A second processing module is configured to input logging item data of the target track data into a logging curve analysis branch model in the target identification model to obtain corresponding petrophysical relationship features output by the logging curve analysis branch model, wherein the key logging curves are used to indicate the corresponding relationship between different two item data in the logging item data;

[0042] A first output module is configured to weight and fuse the multi-scale features, the sedimentary rhythm features and the petrophysical relationship features to obtain corresponding target fusion features;

[0043] A second output module is configured to output an identification result based on the medium-frequency features of the target fusion features through a global discriminator and a local discriminator in the target identification model after adversarial alignment training, wherein the identification result includes lithofacies classification results of each depth point in the target track and lithofacies conversion boundary positions.

[0044] In a third aspect, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by a processor to implement the multi-well lithofacies identification method.

[0045] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0046] The multi-well lithofacies identification method, device and medium provided by the present application perform data preprocessing on the obtained original logging data to obtain a training data set, train an initial identification model based on the training data set to obtain a target identification model, input well data, spontaneous potential data and logging project data of a target well to be identified into corresponding branch models of the target identification model, weight and fuse multi-scale features, sedimentary rhythm features and rock physical relationship features output by the branch models to obtain corresponding target fusion features, and output an identification result including lithofacies classification results of each depth point and lithofacies conversion boundary positions based on the medium frequency features of the target fusion features through a global discriminator and a local discriminator in the target identification model after adversarial alignment training, so that the present application identifies geological features based on branch models of different angles, improves the identification reliability, aligns through the medium frequency features of the target fusion features, retains high frequency and low frequency information reflecting real geological differences, and thus improves the identification accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the example embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor. In the drawings:

[0048] Figure 1 Flowchart of the multi-well lithofacies identification method provided by the present application Figure One ;

[0049] Figure 2 Flowchart of the multi-well lithofacies identification method provided by the present application Figure Two ;

[0050] Figure 3 Extraction diagram of the sliding window provided by the present application

[0051] Figure 4 Sedimentary microfacies diagram provided by the present application

[0052] Figure 5A wavelet transform time-frequency domain feature extraction schematic diagram provided for an embodiment of the present application;

[0053] Figure 6 A lithofacies prediction and real result comparison schematic diagram provided for an embodiment of the present application;

[0054] Figure 7 A structure schematic diagram of a multi-well lithofacies identification device. DETAILED DESCRIPTION

[0055] The exemplary embodiments will be described in detail herein with reference to the attached drawings. In the following description, the same numbers refer to the same or similar elements throughout the drawings. The implementation described in the following exemplary embodiments is not meant to represent all implementations consistent with the present application.

[0056] Identifying lithofacies types is an important link in the process of oil and gas exploration and development. In identifying lithofacies, various logging project data in logging curves are usually analyzed, such as identifying lithofacies types by analyzing GR (Gamma Ray), RT (Resistivity Tool), DEN (density), NPHI (Neutron Porosity Log) and other logging curves.

[0057] To improve the efficiency of lithofacies identification, existing schemes usually adopt intelligent identification schemes, such as implementing lithofacies identification based on machine learning schemes, deep learning schemes and transfer learning schemes.

[0058] For traditional machine learning schemes such as support vector machines and random forests, statistical features of logging curves are extracted for classification. However, such schemes treat the logging response of each depth point as an independent sample, ignoring the continuity and spatial correlation of strata deposition. Lithofacies features have different performances at different scales, and existing schemes mostly use fixed-scale feature extraction, which cannot comprehensively capture the multi-scale characteristics of lithofacies.

[0059] For deep learning schemes such as convolutional neural networks and long short-term memory networks, deep features of logging curves can be automatically learned. However, such schemes perform well in single wells or regions, but have serious cross-well generalization problems, that is, when the trained model is applied to a new well, the identification accuracy usually decreases. This is because the logging response of different wells is affected by non-geological factors such as instrument calibration, mud properties and wellbore conditions, resulting in different logging features of the same lithofacies in different wells.

[0060] For the transfer learning scheme such as domain adaptation neural network, conditional domain adaptation network and other network models, the data distribution difference between different wells is reduced through adversarial training. However, this scheme usually aligns all features indiscriminately, resulting in the loss of important geological information. And for multi-channel logging curves, the existing method usually uses simple splicing or averaging, without considering the relative importance of different curves in different geological environments, and the pure data-driven method does not incorporate geological prior knowledge such as sedimentary rhythm, lithofacies combination rule, etc., resulting in the identification result may violate the basic geological law, causing low reliability of identification.

[0061] Therefore, the present application provides a multi-well lithofacies identification method, by performing data preprocessing on the obtained original logging data to obtain a training data set, training an initial identification model based on the training data set to obtain a target identification model, then inputting the well log data, spontaneous potential data and logging project data of the target well to be identified into the corresponding branch model of the target identification model, and weighting and fusing the multi-scale features, sedimentary rhythm features and rock physical relationship features output by each branch model to obtain the corresponding target fusion features, and based on the medium frequency features of the target fusion features, outputting the identification result including the lithofacies classification result of each depth point and the lithofacies conversion boundary position through the global discriminator and the local discriminator after the adversarial alignment training in the target identification model. Therefore, the present application identifies geological features based on different angle branch models, improves the identification reliability, and aligns through the medium frequency features of the target fusion features, retains the high frequency and low frequency information reflecting the real geological difference, thereby improving the identification accuracy.

[0062] The technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present application will be described below with reference to the accompanying drawings.

[0063] Embodiments

[0064] Figure 1 Flowchart of multi-well lithofacies identification method provided by embodiments of the present application Figure One As Figure 1 shown, the method comprises:

[0065] S101, obtaining original logging data and preprocessing the original logging data to obtain a training data set.

[0066] Specifically, the well logging data of each well track is collected as original well logging data, wherein the well logging data of each well track includes a plurality of different logging items such as neutron porosity, gamma ray, micro-shallow resistivity, acoustic time difference, spontaneous potential, density, density correction and deep induction resistivity, and the original well logging data further includes facies labels indicating the lithofacies categories labeled by experts for each depth point.

[0067] Further, when the original well logging data is preprocessed, data cleaning and normalization processing need to be performed on the original well logging data to obtain the preprocessed training data set.

[0068] S102, based on the training data set, an initial identification model is trained to obtain a target identification model.

[0069] Specifically, after obtaining the preprocessed training data set, the initial identification model is trained based on the training data set, and the initial identification model after training is taken as the target identification model, wherein the target identification model includes a plurality of branch models such as an adaptive wavelet transform branch model, a spontaneous potential transform branch model and a well logging curve analysis branch model.

[0070] S103, target well track data of a target well track to be identified is obtained, and the target well track data is input into the adaptive wavelet transform branch model in the target identification model to obtain corresponding multi-scale features output by the adaptive wavelet transform branch model.

[0071] Specifically, after obtaining the trained target identification model, the target well track data of the target well track to be identified is input into the adaptive wavelet transform branch model in the target identification model, wherein the target well track data includes logging item data at different depth positions of the target well track, and after the adaptive wavelet transform branch model separates, predicts and updates the logging item data of the target well track, a plurality of detail coefficients and approximation coefficients are obtained, and the plurality of detail coefficients and the plurality of approximation coefficients are taken as corresponding multi-scale features.

[0072] S104, the spontaneous potential data in the logging item data of the target well track data is input into the spontaneous potential transform branch model in the target identification model to obtain corresponding sedimentary rhythm features output by the spontaneous potential transform branch model.

[0073] Specifically, the spontaneous potential data in the target well track data, i.e., the spontaneous potential, is used to generate a corresponding spontaneous potential curve, wherein the spontaneous potential curve is used to indicate the variation trend of the spontaneous potential of each depth point in the continuous well track depth direction, and based on a preset wavelet such as a Morlet wavelet and the spontaneous potential of each depth point in the spontaneous potential curve, corresponding sedimentary rhythm features are extracted.

[0074] S105, input the logging project data of the target shaft data into the logging curve analysis branch model in the target recognition model to obtain the corresponding petrophysical relationship characteristics output by the logging curve analysis branch model.

[0075] Specifically, two selected project data in the logging project data are determined as a corresponding correlation coefficient pair, and a corresponding key logging curve pair is determined according to the correlation coefficient pair, wherein the key logging curve pair includes a selected first logging project curve and a second logging project curve. The corresponding time-delay physical relationship characteristics are extracted by the logging curve analysis branch model according to the project data of each depth point indicated by the first logging project curve and the second logging project curve.

[0076] S106, weighting and fusing the multi-scale features, the sedimentary rhythm features and the petrophysical relationship features to obtain corresponding target fusion features.

[0077] Specifically, after obtaining the multi-scale features, the sedimentary rhythm features and the petrophysical relationship features output by the branch model in the target recognition model, the multi-scale features, the sedimentary rhythm features and the petrophysical relationship features are weighted and fused by using a wavelet energy spectrum driven attention mechanism to obtain corresponding target fusion features.

[0078] S107, outputting a recognition result by the global discriminator and the local discriminator in the target recognition model after the adversarial alignment training based on the medium frequency features of the target fusion features.

[0079] Specifically, after obtaining the fused target fusion features, the target fusion features are feature-decomposed to obtain medium frequency features indicating instrument response differences. According to the decomposed medium frequency features, a first loss function about the global discriminator and a second loss function about the local discriminator are established according to the medium frequency features, so as to realize the adversarial alignment training of the global discriminator and the local discriminator. The target fusion features are output by the global discriminator and the local discriminator after the adversarial alignment training to output a recognition result, wherein the recognition result includes the lithofacies classification result of each depth point in the target shaft and the lithofacies conversion boundary position.

[0080] The multi-well lithofacies identification method provided by the embodiment of the application comprises the following steps: data preprocessing is performed on obtained original logging data to obtain a training data set, an initial identification model is trained based on the training data set to obtain a target identification model, well data, spontaneous potential data and logging project data of a target well to be identified are respectively input into corresponding branch models in the target identification model, multi-scale features, sedimentary rhythm features and rock physical relationship features output by the branch models are weighted and fused to obtain corresponding target fusion features, and based on the medium frequency features of the target fusion features, a global discriminator and a local discriminator after adversarial alignment training in the target identification model output an identification result including lithofacies classification results of each depth point and a lithofacies conversion boundary position. Therefore, the embodiment of the application identifies geological features based on branch models at different angles, improves the identification reliability, and aligns through the medium frequency features of the target fusion features, retains high-frequency and low-frequency information reflecting real geological differences, and thus improves the identification accuracy.

[0081] Embodiment

[0082] Figure 2 Flowchart of the multi-well lithofacies identification method provided by the embodiment of the application Figure Two , Figure 3 Extraction schematic diagram of the sliding window provided by the embodiment of the application, as shown in Figure 2 and Figure 3 , the method comprises the following steps:

[0083] S201, obtaining original logging data, and performing data cleaning on the original logging data.

[0084] Specifically, after the original logging data is collected, data cleaning is performed on the original logging data, wherein the data cleaning comprises removing null values in the well section, deleting duplicate records, and removing abnormal values such as jump values and physically unreasonable intervals.

[0085] S202, based on the logging data after data cleaning, performing normalization processing, and organizing the logging data after normalization processing into three-dimensional vectors and two-dimensional vectors.

[0086] Specifically, after the data cleaning of the logging data is completed, normalization processing is performed on each logging project data in the logging data, and each logging curve corresponding to the logging project data is mapped to a standard numerical range, and the logging data after normalization and standardization is organized into three-dimensional vectors, as shown in the following formula (1):

[0087] (1);

[0088] wherein, is the number of wells, is the number of depth sampling points of each well, For the number of logging projects, each facies class label in the logging data is organized into a corresponding two-dimensional vector, as shown in the following formula (2):

[0089] (2);

[0090] wherein, is the total number of lithofacies classifications, is the facies class label of the depth point in the well, the three-dimensional vector is used to indicate the logging project data of different depth positions of each well track, and the two-dimensional vector is used to indicate the facies class label of the corresponding depth position. According to the same depth position, the two-dimensional vector is associated with the corresponding three-dimensional vector.

[0091] S203, obtaining a training data set according to the three-dimensional vector and the two-dimensional vector.

[0092] Specifically, as shown in Figure 3 , a sliding window with a preset width is constructed, and the sliding window is sampled along the direction indicating the depth of the well track in the three-dimensional vector and the corresponding two-dimensional vector.

[0093] Further, a sliding window with a fixed width is used to sample in the depth direction with a step size of 1, and a training sample sequence is constructed. The size of the sliding window can be adjusted in the range of 20-200 sampling points according to specific geological characteristics and sampling rates. For each depth position , the window data is constructed, as shown in the following formula (3):

[0094] (3);

[0095] wherein, is the fixed width of the sliding window, is the depth position, and the facies class label corresponding to each window data is as shown in the following formula (4):

[0096] (4);

[0097] wherein, is the corresponding depth position. After obtaining a plurality of window data, the plurality of window data is used as a training data set.

[0098] S204, training an initial identification model based on the training data set to obtain a target identification model.

[0099] S205, obtaining target well track data of a target well track to be identified.

[0100] Specifically, target borehole data of a target borehole currently in need of lithofacies identification is acquired, where the target borehole data includes multiple different logging items.

[0101] In S206, each logging item data in the target borehole data is separated into odd samples and even samples through an adaptive wavelet transform branch model in the target identification model.

[0102] Specifically, the target borehole data is input into the adaptive wavelet transform branch model in the target identification model. In the adaptive wavelet transform branch model, an input index corresponding to each logging item data is established according to different depth positions of the logging item data in the target borehole data, and the logging item data is separated into odd samples and even samples according to odd input indexes and even input indexes, through the following formula (5):

[0103] (5);

[0104] wherein, is the even sample, is the odd sample.

[0105] In S207, a plurality of detail coefficients and approximation coefficients are calculated based on the odd samples and the even samples through learnable prediction operator weights and update operator weights in the adaptive wavelet transform branch model, and the plurality of detail coefficients and the plurality of approximation coefficients are taken as the multi-scale features.

[0106] Specifically, after the odd samples and the even samples are acquired, the predicted odd samples are calculated through the learnable prediction operator weights in the adaptive wavelet transform branch model, and the difference between the actual odd sample value and the predicted odd sample value is taken as the detail coefficient, through the following formula (6):

[0107] (6);

[0108] wherein, is the learnable prediction operator weight, is the detail coefficient, i.e., the difference between the actual odd value and the predicted odd sample value;

[0109] Further, the actual even sample value and the predicted even sample are added as the corresponding approximation coefficient through the learnable update operator weight in the adaptive wavelet transform branch model and the detail coefficient at this time, through the following formula (7):

[0110] (7);

[0111] wherein, The learnable update operator weight, The approximation coefficient, after obtaining the detail coefficient and the approximation coefficient of the first layer at this time, the approximation coefficient of the first layer is taken as the input of the parity separation of the second layer, and the above prediction and update steps are repeated based on the approximation coefficient of the first layer to obtain the corresponding detail coefficient and approximation coefficient of the second layer, and the calculation is repeated for a predetermined number of layers such as three to five layers to obtain a plurality of detail coefficients and a plurality of approximation coefficients as a multi-scale feature .

[0112] S208, through the spontaneous potential transformation branch model in the target identification model, the spontaneous potential curve corresponding to the spontaneous potential data and the preset mother wavelet are continuously wavelet transformed to extract the corresponding sedimentary rhythm feature.

[0113] Specifically, the spontaneous potential curve corresponding to the spontaneous potential data in the target well data is continuously wavelet transformed to extract the sedimentary rhythm feature, and the following formula (8) is used:

[0114] (8);

[0115] Among them, is the mother wavelet function, is the complex conjugate of the mother wavelet, a is the scale parameter controlling the wavelet stretching, b is the translation parameter controlling the position of the wavelet, and for the mother wavelet function, the following formula (9) is used:

[0116] (9);

[0117] Further, is the mother wavelet function, by transforming the scale parameter a in a predetermined range such as 2 0 to 2 8 , the sedimentary rhythm features of different scales can be obtained.

[0118] S209, input the key project logging curve pair of the target well data into the logging curve analysis branch model in the target identification model to obtain the rock physical relationship feature corresponding to the output of the logging curve analysis branch model.

[0119] Specifically, two project data are determined as key logging project data in the multiple logging project data of the target well data, and the logging curves corresponding to the two key logging project data are taken as the key project logging curve pair, wherein the key logging curve pair includes the selected first logging project curve and the second logging project curve. According to the project data of each depth point indicated by the first logging project curve and the second logging project curve, the rock physical relationship feature is extracted, and the following formula (10) is used:

[0120] (10);

[0121] wherein, is the correlation coefficient of two logging curves at depth position i, i.e., the petrophysical relationship characteristics, W is the size of the sliding window, is the logging value of the first logging project curve X at depth position i+j, is the average value of the first logging project curve X in the sliding window, is the logging value of the second logging project curve Y at depth position i+j, is the average value of the second logging project curve Y in the sliding window.

[0122] Further, the key logging curves are used to indicate the corresponding relationship between different two project data in the logging project data, and the corresponding relationship includes a first corresponding relationship used to indicate the relationship between the shale content and the porosity, a second corresponding relationship used to indicate the relationship between the skeleton density and the propagation velocity, a third corresponding relationship used to indicate the difference between the deep resistivity and the shallow resistivity, and a fourth corresponding relationship used to indicate the relationship between the permeability and the shale content.

[0123] S210, the multi-scale features, the sedimentary rhythm features and the petrophysical relationship features are weighted and fused to obtain corresponding target fusion features.

[0124] Specifically, after obtaining the multi-scale features, the sedimentary rhythm features and the petrophysical relationship features output by the branch model in the target recognition model, the multi-scale features, the sedimentary rhythm features and the petrophysical relationship features are mapped to a unified dimension through a full connection layer, and the mapping is realized through the following formula (11):

[0125] (11);

[0126] wherein, is the original feature vector of the i th branch, is the learnable mapping matrix of the i th branch, is the original feature dimension, is the unified mapping dimension; is the bias vector of the i th branch.

[0127] Further, according to the multi-scale features, the sedimentary rhythm features and the petrophysical relationship features mapped to the unified dimension, the energy score weight corresponding to each feature is calculated, and the calculation is realized through the following formula (12):

[0128] (12);

[0129] wherein, is a learnable energy evaluation vector used to measure the importance of each branch feature, is a scalar, is an energy fraction weight of the i-th branch, used to indicate the information richness of the corresponding branch feature.

[0130] Further, according to the respective features mapped to the unified dimension and the corresponding energy fraction weight, weighted fusion is performed to obtain a target fusion feature, which is calculated by the following formula (13):

[0131]

[0132] wherein, is a Hadamard product, which broadcasts the attention weight to the same dimension and then performs element-wise multiplication.

[0133] S211, decompose the target fusion feature to obtain a low-frequency feature, a medium-frequency feature and a high-frequency feature.

[0134] Specifically, after obtaining the target fusion feature, the target fusion feature is decomposed into a low-frequency feature, a medium-frequency feature and a high-frequency feature, which is shown by the following formula (14):

[0135] (14);

[0136] wherein, is a low-frequency feature, is a medium-frequency feature, is a high-frequency feature, the low-frequency feature is used to indicate large-scale geological background information, the medium-frequency feature is used to indicate instrument response difference information, and the high-frequency feature is used to indicate rock texture detail information.

[0137] S212, realize the adversarial training of the global discriminator and the local discriminator through the first loss function of the global discriminator and the second loss function of the local discriminator.

[0138] Specifically, in order to eliminate the systematic difference caused by instrument response between different wellbores while retaining the real geological information, the medium-frequency feature representing the instrument response difference information is subjected to adversarial domain alignment, and when the global discriminator is subjected to adversarial training, for the first loss function, it is shown by the following formula (15):

[0139] (15);

[0140] wherein, N is the total number of well logging, is a data set of the i-th well logging, is a multi-classification network, outputting an N-dimensional probability vector wherein, ​represents the probability that the input feature comes from the i-th well log; the first loss value is minimized The discriminator learns to recognize the unique instrument response pattern of each well, while the feature extractor maximizes this first loss value through the gradient reversal layer, making the mid-frequency features indistinguishable between different wells, thus eliminating the systematic differences between wells caused by instrument responses.

[0141] Further, in the adversarial training of the local discriminator, the lithofacies classification features of the target track are retained through the second loss function. For the second loss function, the following formula (16) is used:

[0142]

[0143] where C is the total number of lithofacies classes, is the data subset in the i-th well belonging to the c-th lithofacies class, is the one-hot encoding vector representing the c-th class, is the element-wise Hadamard product, The purpose is to selectively retain the c-th class features through the general mask, and the rest of the features are set to zero. is the local discriminator of the c-th lithofacies class. Unlike the global discriminator, each lithofacies class is provided with an independent local discriminator, considering that different lithofacies classes may have different domain shift patterns between different wells.

[0144] Further, according to the first A distance of the global distribution corresponding to the global discriminator and the second A distance of the local distribution corresponding to the local discriminator, a balance coefficient between the first loss value and the second loss value is calculated. For the balance coefficient, the following formula (17) is used:

[0145] (17);

[0146] where C is the total number of lithofacies classes, is the first A distance, is the second A distance, the first A distance is used to indicate the average A-distance between all source wells and the target track, and the second A distance is used to indicate the average A-distance between the source wells and the target track within each lithofacies class.

[0147] Further, for the A-distance as a standard measure of domain adaptation, which is used to measure the difference between the global distribution and the local distribution, the following formula (18) is used:

[0148] (18);

[0149] where, is a newly trained binary classifier to distinguish the error rate of the two domains; the larger the A-distance, the greater the difference between the two domains, and the stronger the domain adaptation is required.

[0150] S213, based on the medium frequency feature of the target fusion feature, outputting a recognition result through the global discriminator and the local discriminator after the adversarial alignment training in the target recognition model.

[0151] Specifically, the target fusion feature is input into the global discriminator and the local discriminator after the adversarial alignment training in the target recognition model, and based on the first loss value and the second loss value of the global discriminator and the local discriminator, a total loss function is constructed, and for the total loss function, the following formula (19) is used:

[0152] (19);

[0153] wherein, is the lithofacies classification loss of each depth point, is the boundary detection loss; based on the total loss function, the recognition result is output through the trained target recognition model, wherein the recognition result includes the lithofacies classification result of each depth point in the target well and the lithofacies conversion boundary position.

[0154] The multi-well lithofacies recognition method provided by the embodiments of the present application acquires the training data set by performing data preprocessing on the acquired original logging data, trains the initial recognition model based on the training data set, and then inputs the well data, the spontaneous potential data and the logging project data of the target well to be recognized into the corresponding branch model of the target recognition model, and weights the multi-scale feature, the sedimentary rhythm feature and the rock physical relationship feature output by each branch model to obtain the corresponding target fusion feature. Based on the medium frequency feature of the target fusion feature, the global discriminator and the local discriminator after the adversarial alignment training in the target recognition model output the recognition result including the lithofacies classification result of each depth point and the lithofacies conversion boundary position. Therefore, the embodiments of the present application recognize geological features based on different angle branch models, improve the recognition reliability, and align through the medium frequency feature of the target fusion feature, retain high frequency and low frequency information reflecting real geological differences, thereby improving the recognition accuracy.

[0155] The present application provides specific embodiments, Figure 4 is a sedimentary microfacies diagram provided by the embodiments of the present application, Figure 5 is a wavelet transform time-frequency domain feature extraction diagram provided by the embodiments of the present application, Figure 6 is a lithofacies prediction and real result comparison diagram provided by the embodiments of the present application, which is used to further illustrate the above embodiments.

[0156] This embodiment collects 43 vertical wells (i.e. N = 43), the well depth interval covers the main reservoir section, the logging data sampling interval is 0.1m, and the average number of depth points per well is about 1000. 11 logging curves (i.e. C = 11) are collected at each depth point as original logging data, including depth (DEPT), neutron porosity (NPHI), gamma ray (GR), micro-shallow resistivity (MSFL), acoustic time (DT), acoustic time correction (DT_CALI), spontaneous potential (SP), caliper (CALI), density (RHOB), density correction (RHOB_CALI), and deep induction resistivity (ILD) and other conventional logging channels.

[0157] As shown in Figure 4 Three geology interpretation experts (E = 3) respectively interpreted the sedimentary microfacies of the whole well section, and the label space was divided into R = 6 facies categories, namely Outer Ramp, Distal Mid-Ramp, Proximal Mid-Ramp, High Shoal, Lagoon, Basin, and Medium-low shal.

[0158] The original logging data is preprocessed and the sample is constructed.

[0159] Specifically, the null values and abnormal jump values in the logging curves are removed; the data of the caliper abnormal expansion section (CALI>15inches) is marked; and the z-score standardization processing is used to process each curve, so that the mean value is 0 and the standard deviation is 1.

[0160] Further, the sliding window size W = 101 is set, that is, it corresponds to a depth range of 10.1 meters; with a step size of 1, the context features are constructed for each depth point to obtain the training data set.

[0161] Further, the training data set is divided, 30 wells are selected as the source domain training set, 8 wells are selected as the validation set, and 5 wells are reserved as the target domain test set.

[0162] Feature extraction is performed through the adaptive wavelet transform branch model, the spontaneous potential transform branch model and the logging curve analysis branch model in the target recognition model.

[0163] Specifically, for the adaptive wavelet transform branch model, 4-layer lifting scheme decomposition is used; the prediction operator and the update operator are realized by 1D convolution, and the kernel size is 7.

[0164] Further, as shown in Figure 5 For the spontaneous potential transform branch model, Morlet wavelet transform is performed on the spontaneous potential curve; the scale parameter a is selected Extract the energy distribution at each scale to form a 128-dimensional feature vector.

[0165] Furthermore, for the well logging curve analysis branch model, the sliding correlation coefficients of 8 key curve combinations are calculated, such as GR-NPHI (shale-porosity relationship), RHOB-DT (density-sonic relationship), oILD-MSFL (depth-shallow resistivity difference), SP-GR (permeability-shale relationship), RHOB-NPHI (density-neutron relationship), DT-NPHI (sonic-neutron relationship), CALI-GR (wellbore-shale relationship), and ILD-SP (resistivity-spontane potential relationship).

[0166] Attention fusion and selective domain adaptation training are performed on the features extracted from each branch model.

[0167] Specifically, the features of each branch are fused, and the three branch features are mapped to a 256-dimensional unified space through a fully connected layer; attention weights are calculated based on wavelet energy spectrum; typical weight distributions: homogeneous layer segment (0.25, 0.35, 0.40), thin interlayer segment (0.45, 0.30, 0.25); a 3-layer wavelet decomposition is used to separate the fused features into low, medium, and high frequency components, only for medium frequency features such as the detail coefficients d of the second layer. 2 Conduct domain confrontation,

[0168] Furthermore, dynamic weight adjustment was performed, with an initial ω=0.5, which was automatically adjusted according to A-distance during training. The final convergence value was ω≈0.68, indicating that local alignment is more important in this dataset.

[0169] When training an object recognition model, the boundary detection task weights of the overall loss function. =0.5, adversarial loss weight =0.1.

[0170] As shown in the figure, Figure 6 As shown, the identification results on 5 test wells show that, for cross-well performance comparison, the method of this invention has an average accuracy of 87.3% and a performance decrease of only 12.7%; the version without domain adaptation has an average accuracy of 72.5% and a performance decrease of 27.5%; the classic machine learning method has an average accuracy of 61.2% and a performance decrease of 38.8%; and the expert interpretation consistency is approximately 82.6%.

[0171] For the phase class performance, for High Shoal: recognition accuracy 92.1%, boundary detection accuracy 89.3%; Proximal Mid-Ramp: recognition accuracy 88.7%, boundary detection accuracy 86.2%; Distal Mid-Ramp: recognition accuracy 85.4%, boundary detection accuracy 83.1%; Lagoon: recognition accuracy 86.9%, boundary detection accuracy 84.7%; OuterRamp: recognition accuracy 84.2%, boundary detection accuracy 81.5%; Basin: recognition accuracy 85.5%, boundary detection accuracy 82.8%.

[0172] The embodiments of the present application can divide the functional modules of the electronic device or the host device according to the above method examples. For example, each functional module can be divided according to each function, or two or more functions can be integrated in one processing unit. The integrated unit can be realized in the form of hardware or in the form of a software functional module. It should be noted that the division of the modules in the embodiments of the present application is illustrative, and is only a logical functional division. When actually implemented, another division method can be used.

[0173] Figure 7 A structural schematic diagram of a multi-well lithofacies recognition device provided by an embodiment of the present application is shown in FIG. 7, which includes: Figure 7

[0174] The acquisition module 701 is configured to acquire original logging data, and pre-process the original logging data to obtain a training data set. The training data set includes multiple groups of track data. The track data includes logging item data at different depth positions of each track, and a facies class label corresponding to the depth position. The facies class label is used to indicate an expert-labeled lithofacies class. Based on the training data set, an initial recognition model is trained to obtain a target recognition model.

[0175] The first processing module 702 is configured to acquire target track data of a target track to be recognized, and input the target track data into an adaptive wavelet transform branch model in the target recognition model to obtain corresponding multi-scale features output by the adaptive wavelet transform branch model. The spontaneous potential data in the logging item data of the target track data is input into a spontaneous potential transform branch model in the target recognition model to obtain corresponding sedimentary rhythm features output by the spontaneous potential transform branch model.

[0176] ​The second processing module 703 is configured to input the well logging item data of the target shaft data into a well logging curve analysis branch model in the target identification model, to obtain a rock physical relationship feature corresponding to an output of the well logging curve analysis branch model, and the key well logging curve is used to indicate a corresponding relationship between different two item data in the well logging item data.

[0177] The first output module 704 is configured to weight and fuse the multi-scale feature, the sedimentary rhythm feature and the rock physical relationship feature, to obtain a corresponding target fusion feature.

[0178] The second output module 705 is configured to output an identification result based on a medium frequency feature of the target fusion feature, through a global discriminator and a local discriminator trained by the target identification model in an adversarial alignment manner, wherein the identification result includes a lithofacies classification result of each depth point in the target shaft and a lithofacies conversion boundary position.

[0179] Further, the acquisition module 701 is specifically configured to perform data cleaning on the original well logging data, perform normalization processing based on the well logging data after data cleaning, organize the well logging data after normalization processing into a three-dimensional vector and a two-dimensional vector, and acquire the training data set according to the three-dimensional vector and the two-dimensional vector.

[0180] The three-dimensional vector is used to indicate well logging item data at different depth positions of each shaft, and the two-dimensional vector is used to indicate facies labels at corresponding depth positions. According to the same depth position, the two-dimensional vector is associated with the corresponding three-dimensional vector.

[0181] Further, the acquisition module 701 is specifically configured to construct a sliding window with a preset width, slide the sliding window along a direction indicating the shaft depth in the three-dimensional vector, and slide sample the three-dimensional vector and the corresponding two-dimensional vector, to obtain a plurality of window data, and use the plurality of window data as the training data set.

[0182] Further, the first processing module 702 is specifically configured to establish an input index corresponding to each well logging item data in the target shaft data according to different depth positions of the well logging item data.

[0183] According to the input index of each well logging item data, the even and odd samples are separated from each well logging item data in the target shaft data.

[0184] Based on the odd sample and the even sample, a plurality of detail coefficients and approximation coefficients are calculated by the prediction operator weight and the update operator weight which are learnable in the adaptive wavelet transform branch model, and the plurality of detail coefficients and the plurality of approximation coefficients are taken as the multi-scale features.

[0185] Further, the first processing module 702 is specifically configured to establish a spontaneous potential curve according to the spontaneous potential data, and perform continuous wavelet transform on the spontaneous potential curve and a preset mother wavelet through the spontaneous potential transform branch model to extract the corresponding sedimentary rhythm features.

[0186] Further, the second processing module 703 is configured to determine a corresponding relationship between different two project data in the logging project data, including: a first corresponding relationship, the first corresponding relationship being used to indicate a relationship between shale content and porosity; a second corresponding relationship, the second corresponding relationship being used to indicate a relationship between skeleton density and propagation velocity; a third corresponding relationship, the third corresponding relationship being used to indicate a difference between deep resistivity and shallow resistivity; and a fourth corresponding relationship, the fourth corresponding relationship being used to indicate a relationship between permeability and shale content.

[0187] Further, the second output module 705 is specifically configured to decompose the target fusion features to obtain low-frequency features, the medium-frequency features and high-frequency features, wherein the low-frequency features are used to indicate large-scale geological background information, the medium-frequency features are used to indicate instrument response difference information, and the high-frequency features are used to indicate rock texture detail information.

[0188] The global discriminator and the local discriminator are subjected to adversarial training through a first loss function of the global discriminator and a second loss function of the local discriminator, wherein the first loss function is used to eliminate instrument response differences between different wells, and the second loss function is used to retain facies classification features of the target track.

[0189] The present application also provides a computer readable storage medium, wherein computer execution instructions are stored in the computer readable storage medium, and when a processor executes the computer execution instructions, the multi-well facies recognition method is realized.

[0190] The above specific embodiments further explain the purpose, technical solutions and advantages of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A multi-well lithofacies identification method, characterized in that, The method includes: Acquire raw logging data and preprocess the raw logging data to obtain a training dataset, wherein the training dataset includes multiple sets of wellbore data, the wellbore data includes logging project data at different depths of each wellbore, and facies labels at the corresponding depths, the facies labels being used to indicate the lithofacies categories marked by experts; Based on the training dataset, the initial recognition model is trained to obtain the target recognition model; Acquire target well data for the target well to be identified, and input the target well data into the adaptive wavelet transform branch model in the target identification model to obtain the multi-scale features corresponding to the output of the adaptive wavelet transform branch model. The spontaneous potential data from the logging project data of the target wellbore data is input into the spontaneous potential transformation branch model in the target identification model to obtain the sedimentary rhythm features corresponding to the output of the spontaneous potential transformation branch model; The key item logging curve pairs in the target wellbore data are input into the logging curve analysis branch model in the target identification model to obtain the corresponding rock physical relationship characteristics output by the logging curve analysis branch model. The key item logging curve pairs are used to indicate the correspondence between two different items in the logging data. The correspondence between two different items in the logging data includes: a first correspondence, indicating the relationship between clay content and porosity; a second correspondence, indicating the relationship between skeleton density and propagation velocity; a third correspondence, indicating the difference in resistivity between shallow and deep wells; and a fourth correspondence, indicating the relationship between permeability and clay content. The multi-scale features, the sedimentary rhythm features, and the petrophysical relationship features are weighted and fused to obtain the corresponding target fused features; Based on the mid-frequency features of the target fusion features, the recognition results are output by the global discriminator and local discriminator after adversarial alignment training in the target recognition model. The recognition results include the lithofacies classification results of each depth point in the target wellbore, as well as the location of the lithofacies transition boundary.

2. The multi-well lithofacies identification method according to claim 1, characterized in that, The preprocessing of the raw well logging data to obtain a training dataset includes: The raw logging data is cleaned. Based on the cleaned logging data, normalization processing is performed, and the normalized logging data is organized into three-dimensional vectors and two-dimensional vectors. The training dataset is obtained based on the three-dimensional vectors and the two-dimensional vectors. The three-dimensional vector is used to indicate logging data at different depths in each wellbore, and the two-dimensional vector is used to indicate the corresponding label at the same depth. Based on the same depth, the two-dimensional vector is associated with the corresponding three-dimensional vector.

3. The multi-well lithofacies identification method according to claim 2, characterized in that, The step of obtaining the training dataset based on the three-dimensional vector and the two-dimensional vector includes: Create a sliding window of a preset width; The sliding window is used to slide and sample the three-dimensional vector and the corresponding two-dimensional vector along the direction indicating the well depth in the three-dimensional vector to obtain multiple window data. Multiple window data sets are used as the training dataset.

4. The multi-well lithofacies identification method according to claim 1, characterized in that, The step of obtaining the multi-scale features corresponding to the output of the adaptive wavelet transform branch model includes: Based on the different depth locations of each logging item data in the target wellbore data, an input index corresponding to the logging item data is established. Based on the input index of each logging item data, the logging item data in the target wellbore data is separated into odd and even numbers to obtain odd and even samples. Based on the odd and even samples, multiple detail coefficients and approximation coefficients are calculated and obtained through the learnable prediction operator weights and update operator weights in the adaptive wavelet transform branch model, and the multiple detail coefficients and multiple approximation coefficients are used as the multi-scale features.

5. The multi-well lithofacies identification method according to claim 1, characterized in that, The step of obtaining the sedimentary rhythm features corresponding to the output of the spontaneous potential transformation branch model includes: Based on the aforementioned natural potential data, a natural potential curve is established; Based on the natural potential curve and the preset mother wavelet, continuous wavelet transform is performed through the natural potential transform branch model to extract the corresponding sedimentary rhythm features.

6. The multi-well lithofacies identification method according to claim 1, characterized in that, The output of the recognition result by the global discriminator and local discriminator trained on adversarial alignment in the target recognition model includes: The target fusion feature is decomposed to obtain low-frequency features, mid-frequency features and high-frequency features, wherein the low-frequency features are used to indicate large-scale geological background information, the mid-frequency features are used to indicate instrument response difference information, and the high-frequency features are used to indicate rock texture detail information; The adversarial training of the global discriminator and the local discriminator is achieved through the first loss function of the global discriminator and the second loss function of the local discriminator, wherein the first loss function is used to eliminate the differences in instrument response between different wells, and the second loss function is used to preserve the lithofacies classification characteristics of the target well.

7. A multi-well facies identification device, characterized in that, include: The acquisition module is used to acquire raw logging data and preprocess the raw logging data to obtain a training dataset. The training dataset includes multiple sets of wellbore data, which include logging project data at different depths in each wellbore and facies labels at the corresponding depths. The facies labels are used to indicate the lithofacies categories marked by experts. Based on the training dataset, the initial recognition model is trained to obtain a target recognition model. The first processing module is used to acquire target well data of the target well to be identified, input the target well data into the adaptive wavelet transform branch model in the target identification model to obtain the multi-scale features corresponding to the output of the adaptive wavelet transform branch model, and input the spontaneous potential data in the logging project data of the target well data into the spontaneous potential transform branch model in the target identification model to obtain the sedimentary rhythm features corresponding to the output of the spontaneous potential transform branch model. The second processing module is used to input the logging project data of the target wellbore data into the logging curve analysis branch model in the target identification model, so as to obtain the rock physical relationship characteristics corresponding to the output of the logging curve analysis branch model, wherein the key project logging curve pair is used to indicate the correspondence between two different project data in the logging project data; The correspondence between two different data items in the well logging data includes: a first correspondence, which indicates the relationship between clay content and porosity; a second correspondence, which indicates the relationship between skeleton density and propagation velocity; a third correspondence, which indicates the difference in resistivity between shallow and deep areas; and a fourth correspondence, which indicates the relationship between permeability and clay content. The first output module is used to perform weighted fusion of the multi-scale features, the sedimentary rhythm features and the rock physical relationship features to obtain the corresponding target fused features; The second output module is used to output the recognition result based on the mid-frequency features of the target fusion features through the global discriminator and local discriminator trained by the adversarial alignment in the target recognition model. The recognition result includes the lithofacies classification result of each depth point in the target wellbore, as well as the location of the lithofacies transition boundary.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the multi-well facies identification method as described in any one of claims 1 to 6.

Images