Sedimentary microfacies recognition model training method and device and sedimentary microfacies recognition method

Through joint model and multi-scale data processing, combined with convolutional neural network and recurrent neural network, the problem of low analysis accuracy of a single model is solved, and the accuracy and efficiency of deposition microphase recognition are improved.

CN118940078APending Publication Date: 2024-11-12PETROCHINA CO LTD

Patent Information

Application Number
CN202310518605.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-09
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

In the prior art, a single model is used to analyze a single well log data, resulting in poor accuracy of sedimentary microphase recognition.

Method used

Using a joint model, including feature extraction model and feature classification model, by obtaining two-dimensional data with labels, using median filtered multi-scale data, a deposition microfacial recognition model based on convolutional neural networks and recurrent neural networks is constructed, combined with Multi-classified cross-entropy loss function and exponential loss function are trained.

Benefits of technology

The classification accuracy of sedimentary microfacial recognition is improved, which can better reflect sedimentary characteristics and geological changes, and reduce the uncertainty of identification efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940078A_ABST
    Figure CN118940078A_ABST
Patent Text Reader

Abstract

The invention discloses a sedimentary microfacies recognition model training method and device and a sedimentary microfacies recognition method. The sedimentary microfacies recognition model training method comprises the steps that a training sample is obtained, the training sample comprises two-dimensional data with a label, the two-dimensional data comprises an original logging curve arranged in the transverse direction and multiple median filtered curves, the multiple median filtered curves are obtained by conducting median filtering of different scales on the original logging curve, and the two-dimensional data are marked with labels; the label indicates the sedimentary microfacies category; a joint model is constructed, the joint model comprises a feature extraction model and a feature classification model, the feature extraction model extracts multi-scale spatial features of the two-dimensional data, and the feature classification model extracts vertical features of the two-dimensional data and performs classification based on the spatial features and the vertical features; and training the joint model by using a training sample to obtain a sedimentary microfacies recognition model. According to the invention, the sedimentary microfacies identification precision can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas exploration, and in particular to a sedimentary microfacies identification model training method, device and sedimentary microfacies identification method. Background Art

[0002] Sedimentary microfacies identification based on logging data is an important part of reservoir evaluation and characterization, which reflects the changes in sedimentary transport and material cycles over time. Traditional sedimentary microfacies analysis usually relies on the experience and knowledge of geologists, and divides sedimentary microfacies based on previous research results and regional sedimentary background, combined with rock thickness, grain size and other characteristics. This manual identification process involves a lot of data analysis and comparison, which is very tedious and complicated, and also leads to a certain degree of uncertainty and possible illusions.

[0003] With the development of computer technology, sedimentary microfacies identification based on deep learning methods has become a research hotspot. In theory, well logging data provides the physical and chemical properties of underground strata, reflecting the changes in sedimentary environment and characteristics such as sediment properties. Different sedimentary cycles correspond to different logging curve shapes. Intelligent sedimentary microfacies identification methods divide sedimentary microfacies based on the differences in curve shapes. Various machine learning methods (such as support vector machines, decision trees, random forests, etc.) and network architectures (such as convolutional neural networks, recurrent neural networks, etc.) have been applied to sedimentary microfacies identification and have shown different application effects.

[0004] However, in some solutions of the prior art, a single model is used to extract and classify features of a single well logging data, which makes it impossible to fully grasp the features of the well logging data and results in poor accuracy in identifying sedimentary microfacies. Summary of the invention

[0005] The main purpose of the present invention is to provide a sedimentary microfacies identification model training method, device and sedimentary microfacies identification method to solve the problem of poor sedimentary microfacies identification accuracy caused by using a single model to analyze a single well logging data in the prior art.

[0006] According to one aspect of the present invention, a sedimentary microfacies identification model training method is proposed, comprising:

[0007] Acquire a training sample, wherein the training sample includes two-dimensional data with a label, wherein the two-dimensional data includes an original well logging curve arranged in a transverse direction and a plurality of median filtered curves, wherein the plurality of median filtered curves are obtained by performing median filtering of different scales on the original well logging curve, and the label indicates a sedimentary microfacies category;

[0008] Constructing a joint model, the joint model comprising a feature extraction model and a feature classification model, the feature extraction model extracting multi-scale spatial features of the two-dimensional data, the feature classification model extracting vertical features of the two-dimensional data and performing classification based on the spatial features and the vertical features;

[0009] The joint model is trained using the training samples to obtain the sedimentary microfacies identification model.

[0010] According to one embodiment of the present invention, the feature extraction model is a convolutional neural network model.

[0011] According to one embodiment of the present invention, the feature extraction model is a UNet model.

[0012] According to one embodiment of the present invention, the feature classification model is a ConvLSTM model.

[0013] According to one embodiment of the present invention, the original logging curve is a natural gamma curve; and / or the sedimentary microfacies categories include: channel distributary bay, front sheet sand, underwater distributary channel, estuary bar, and channel side edge.

[0014] According to one embodiment of the present invention, the joint model is trained based on a joint loss function, where the joint loss function is constructed based on a multi-classification cross entropy loss function and an exponential loss function, and the formula is as follows:

[0015]

[0016] Among them, J represents the joint loss function; b represents the label; p represents the predicted probability; K represents the total number of sample categories; N represents the number of samples; b i,k Indicates whether the label of the i-th sample is the k-th sedimentary microfacies, if yes, it is 1, if not, it is 0; p i,k It represents the probability of predicting that the i-th sample is the k-th sedimentary microfacies; λ is a hyperparameter.

[0017] According to another aspect of the present invention, a sedimentary microfacies identification model training device is provided, comprising:

[0018] An acquisition module is configured to acquire training samples, wherein the training samples include two-dimensional data with labels, wherein the two-dimensional data include original well logging curves arranged in a transverse direction and a plurality of median filtered curves, wherein the plurality of median filtered curves are obtained by performing median filtering of different scales on the original well logging curves, and the labels indicate sedimentary microfacies categories;

[0019] A construction module is configured to construct a joint model, wherein the joint model includes a feature extraction model and a feature classification model, wherein the feature extraction model extracts multi-scale spatial features of the two-dimensional data, and the feature classification model extracts vertical features of the two-dimensional data and performs classification based on the spatial features and the vertical features;

[0020] A training module is configured to train the joint model using the training samples to obtain the sedimentary microfacies identification model.

[0021] According to another aspect of the present invention, a computer device is provided, comprising:

[0022] at least one processor; and

[0023] A memory, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the program, the method described in any of the above embodiments is implemented.

[0024] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method described in any one of the above embodiments is implemented.

[0025] According to another aspect of the present invention, a sedimentary microfacies identification method is proposed, comprising: performing identification using a sedimentary microfacies identification model obtained by training according to the method described in any of the above embodiments.

[0026] In the sedimentary microfacies identification model training method according to an embodiment of the present invention, two-dimensional data is obtained as training samples based on the original logging curve and multiple median filtered curves, so that the training samples have multi-scale information and can better reflect the sedimentary characteristics; the feature extraction model and the feature classification model are jointly adopted, so that the spatial characteristics and vertical characteristics of the two-dimensional data can be considered at the same time, thereby giving a higher classification accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only 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.

[0028] Figure 1 A flow chart showing a sedimentary microfacies identification model training method according to an embodiment of the present invention;

[0029] Figure 2 A schematic diagram showing a sedimentary microfacies identification model training method according to an embodiment of the present invention;

[0030] Figure 3 A schematic diagram showing a sedimentary microfacies identification model training device according to an embodiment of the present invention;

[0031] Figure 4 A schematic diagram showing two-dimensional data and labels obtained in Example 1 of the present invention;

[0032] Figure 5 A schematic diagram showing the classification of sedimentary microfacies and the classification of sample data in Example 1 of the present invention;

[0033] Figure 6 The sample distribution characteristics of the training set in Example 1 of the present invention are shown;

[0034] Figure 7 The sample distribution characteristics of the validation set in Example 1 of the present invention are shown;

[0035] Figure 8 Shows the training error curves of different models;

[0036] Fig. 9 Shows the training accuracy curves of different models;

[0037] Fig.10 The sedimentary microfacies identification results on the training set in Example 1 of the present invention are shown;

[0038] Fig.11 The sedimentary microfacies identification results on the validation set in Example 1 of the present invention are shown. DETAILED DESCRIPTION

[0039] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the embodiments of the present invention are further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0040] It should be noted that all expressions using "first" and "second" in the embodiments of the present invention are for distinguishing two non-identical entities with the same name or non-identical parameters. It can be seen that "first" and "second" are only for the convenience of expression and should not be understood as limitations on the embodiments of the present invention. The subsequent embodiments will not explain this one by one.

[0041] refer to Figure 1 The present invention proposes a sedimentary microfacies identification model training method, comprising the following steps:

[0042] S1, obtaining a training sample, wherein the training sample includes two-dimensional data with a label, wherein the two-dimensional data includes an original well logging curve arranged in a transverse direction and a plurality of median filtered curves, wherein the plurality of median filtered curves are obtained by performing median filtering of different scales on the original well logging curve, and the label indicates a sedimentary microfacies category;

[0043] S2, constructing a joint model, the joint model comprising a feature extraction model and a feature classification model, the feature extraction model extracting multi-scale spatial features of the two-dimensional data, the feature classification model extracting vertical features of the two-dimensional data and performing classification based on the spatial features and the vertical features;

[0044] S3, using the training samples to train the joint model to obtain the sedimentary microfacies identification model.

[0045] In an embodiment of the present invention, two-dimensional data is obtained as training samples based on the original logging curve and multiple median filtered curves, so that the training samples have multi-scale information and can better reflect the sedimentary characteristics; the feature extraction model and the feature classification model are jointly adopted, so that the spatial and vertical characteristics of the two-dimensional data can be considered at the same time, thereby giving a higher classification accuracy.

[0046] The logging curve usually extends vertically, and the vertical direction corresponds to the well depth direction, so that the logging curve can indicate the change of the measurement result along the well depth direction. In an embodiment of the present invention, the original logging curve is a natural gamma curve. In other embodiments, the original logging curve can also be any other suitable logging curve. The basic principle of median filtering is: by setting the window value, the output signal of a certain point can be replaced by the statistical median of all signals within the window range of this point. By performing median filtering, the edge characteristics of the data can be well preserved while filtering out the high-frequency noise of the logging data, which is very conducive to the division of the boundaries of different sedimentary microfacies. By performing median filtering of different scales, the characteristics of the logging data at different geological scales can be retained, so that the two-dimensional data including the original logging curve arranged in the horizontal direction and multiple median filtered curves can more richly reflect the geological characteristics. The two-dimensional logging response characteristics constructed by combining the original curve and the filtered curve are not only manifested as the change of the logging curve morphology in the vertical direction, but also present the correlation of the logging curve morphology in the space (i.e., in the horizontal direction). The sedimentary environment and geological lithology characteristics reflected by this joint logging data are extremely important for the accurate division of sedimentary microfacies.

[0047] The logging data can be combined with the original data and the data after median filtering to divide the sedimentary microfacies. The two-dimensional data can be artificially labeled through the interpretation of sedimentary microfacies. Different labels can be assigned to a certain range of the two-dimensional data along the vertical direction to indicate different sedimentary microfacies categories. In some embodiments, the sedimentary microfacies categories include: a low-amplitude micro-toothed river channel distributary bay, a bell-shaped upper and funnel-shaped lower front edge sheet sand, a box-shaped underwater distributary channel, a funnel-shaped estuary bar, and a bell-shaped river channel side edge.

[0048] The present invention combines a feature extraction model and a feature classification model to extract the lateral features and vertical features of two-dimensional data. The feature extraction model can extract multi-scale spatial features, and the feature classification model can inherit the spatial features and extract the vertical features and process the spatial features and vertical features to achieve division. The feature extraction model can be any model that can extract the lateral features of two-dimensional data (the lateral features indicate the spatial features). For example, the feature extraction model can be any suitable type of convolutional neural network model, which extracts the lateral features based on convolution kernels with different weights. In some embodiments, the feature extraction model is a UNet model. The UNet model extracts morphological information of different scales on the well logging curve with convolution kernels of different sizes and different weights. The feature classification model can be any model that can extract the vertical features of two-dimensional data and can simultaneously process spatial features and vertical features. In an embodiment of the present invention, the feature classification model is a ConvLSTM model.

[0049] refer to Figure 2 The present invention proposes a well logging sedimentary microfacies identification framework that combines convolutional neural networks and recurrent neural networks. First, the well logging data is median filtered, and the median filtered data and the original data form two-dimensional data, which are input into the UNet convolutional neural network structure for feature extraction. The extracted features are inherited and processed by the spatiotemporal sequence model ConvLSTM (Convolutional Long Short-Term Memory) network, and finally the probability of each sedimentary microfacies is output, thereby identifying and classifying the well logging microfacies.

[0050] The present invention performs median filtering of different scales on the natural gamma curve to extract the geological and lithological characteristics reflected by the data. At the same time, the filtered data and the original data form two-dimensional training data to ensure that the network can establish a correct mapping relationship and have a certain degree of noise resistance. Different from the traditional intelligent sedimentary microfacies identification method based on the original logging data as input, the present invention uses the combined logging data to more accurately reflect the sedimentary environment and geological and lithological characteristics, which is conducive to the accurate division of sedimentary microfacies.

[0051] The convolutional neural network UNet uses convolution kernels of different sizes and weights to extract morphological information of different scales on the logging curve. The ConvLSTM network inherits and processes the sequence characteristics of the logging curve, and comprehensively utilizes the morphological changes of the same logging curve vertically and different logging curves in space. Compared with the traditional sedimentary microfacies identification method, it can improve the utilization rate of the logging curve and realize the division of the logging sedimentary microfacies more efficiently and accurately. The present invention uses the convolutional neural network UNet as a feature extractor and the ConvLSTM as a feature classifier, and sends the extracted features to the ConvLSTM neural network model, processes the sequence data on the basis of spatial feature extraction, and comprehensively analyzes the morphological characteristics of the logging curve.

[0052] The samples can be divided into training sets and validation sets, and sent to the constructed neural network model to complete the network training and parameter optimization and adjustment processes, establish the category correspondence between the logging curve and the sedimentary microfacies, and ensure the classification effect of the sedimentary microfacies. Finally, the network model test can also be carried out to further verify the reliability and practicality of the technical solution of the present invention, and improve the recognition of sedimentary microfacies identification and geological feature analysis based on logging data.

[0053] In some embodiments, the joint model is trained based on a joint loss function, where the joint loss function is constructed based on a multi-classification cross entropy loss function and an exponential loss function, and the formula is as follows:

[0054]

[0055] Among them, J represents the joint loss function; b represents the label; p represents the predicted probability; K represents the total number of sample categories; N represents the number of samples; b i,k Indicates whether the label of the i-th sample is the k-th sedimentary microfacies, if yes, it is 1, if not, it is 0; p i,k It represents the probability of predicting that the i-th sample is the k-th sedimentary microfacies; λ is a hyperparameter that controls the weight of the exponential loss function, and its value can be determined through experiments.

[0056] The present invention can improve the model training effect by adopting a joint loss function, and use the joint loss function to more accurately evaluate the classification effect of the network to complete the process of network parameter optimization and adjustment. After the training is completed, the category correspondence between the logging curve and the sedimentary microfacies is established, and the optimal network architecture and node parameters and other parameter settings are saved. A comparative test can also be set to analyze and evaluate the influence of a single network structure and a joint neural network framework on the sedimentary microfacies identification effect. The test results of the training set and the validation set can also be evaluated separately to further examine the practical application potential of the training method of the present invention.

[0057] refer to Figure 3The present invention proposes a sedimentary microfacies identification model training device 100, comprising: an acquisition module 10, configured to acquire training samples, the training samples comprising two-dimensional data with labels, the two-dimensional data comprising original well logging curves arranged in a transverse direction and a plurality of median filtered curves, the plurality of median filtered curves being obtained by median filtering the original well logging curves at different scales, and the labels indicating sedimentary microfacies categories; a construction module 20, configured to construct a joint model, the joint model comprising a feature extraction model and a feature classification model, the feature extraction model extracting multi-scale spatial features of the two-dimensional data, the feature classification model extracting vertical features of the two-dimensional data and classifying based on the spatial features and the vertical features; a training module 30, configured to train the joint model using the training samples to obtain the sedimentary microfacies identification model.

[0058] The present invention further proposes a computer device, comprising: at least one processor; and a memory, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the program, the method described in any of the above embodiments is implemented.

[0059] The present invention further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the method described in any of the above embodiments is implemented.

[0060] The present invention also proposes a sedimentary microfacies identification method, comprising: using a sedimentary microfacies identification model trained by the method described in any of the above embodiments for identification.

[0061] In summary, the present invention can at least achieve the following technical effects:

[0062] 1. The present invention proposes a sedimentary microfacies identification method based on deep learning technology. This method does not rely on prior knowledge and geological interpretation results. It mines the internal characteristics of the data based on the network structure and associates it with sedimentary microfacies, which greatly improves the identification efficiency of sedimentary microfacies while ensuring the identification accuracy.

[0063] 2. The present invention comprehensively considers the characteristics of well logging data, and combines the UNet and ConvLSTM network architecture to comprehensively extract the spatial and vertical characteristics of the well logging curve, providing reliable analysis data for the data-driven sedimentary microfacies identification method, and enhancing the ability to reflect sedimentary characteristics and geological changes based on well logging curves. The present invention is simple to operate, stable and effective, and easy to promote and use, providing technical support for the geological interpretation of well logging data and reservoir characterization.

[0064] The following is a description based on specific embodiments.

[0065] Example 1

[0066] Sedimentary microfacies identification was carried out by taking the natural gamma logging curve and sedimentary microfacies interpretation results of a certain study area as an example. The different forms of the natural gamma logging curve correspond to 5 different sedimentary microfacies interpretation results, namely, the low-amplitude micro-toothed river channel distributary bay, the bell-shaped upper and funnel-shaped lower front edge sheet sand, the box-shaped underwater distributary channel, the funnel-shaped estuary bar and the bell-shaped river channel side edge, which correspond to 5 label categories of 0, 1, 2, 3, and 4 respectively. Based on the logging data and sedimentary microfacies classification results, a joint network architecture with convolutional neural network UNet as feature extractor and ConvLSTM as feature classifier was constructed to carry out logging sedimentary microfacies identification based on morphological differences such as vertical and amplitude of logging curves. The specific steps are as follows:

[0067] Step 1: Perform median filtering of different scales on the natural gamma-ray logging curves, and use them together with the original logging curves to guide the interpretation of sedimentary microfacies and obtain the input data and labels of the network.

[0068] The original natural gamma curve contains high-frequency noise. The gamma curve is filtered at different scales by median filtering, which not only filters out the high-frequency noise but also retains the edge features of the logging curve. The original logging curve and the filtered data form a two-dimensional logging response feature. Based on this two-dimensional logging response feature, the logging data is interpreted for sedimentary microfacies. The morphological changes of the logging curve reflect the changes in sedimentary environments such as rock grain size, sediment thickness, and hydrodynamic conditions. Taking delta sedimentation as an example, the overall low value of the gamma curve is usually interpreted as a bay between river diversions, while the overall high value of the gamma curve, which is a box-shaped morphology, is interpreted as an underwater diversion channel. This change in the morphology of the logging curve is an important basis for the interpretation of the logging sedimentary microfacies. The two-dimensional logging response feature constructed by combining the original curve and the filtered curve not only shows the change in the morphology of the logging curve in the vertical direction, but also shows the correlation of the morphology of the logging curve in space. Therefore, it is necessary to construct a neural network architecture that can simultaneously process the lateral (i.e. spatial) and vertical data connections, so as to achieve high utilization of the logging curve and complete extraction of the logging morphology, and provide sufficient basic support for the division of sedimentary microfacies.

[0069] The original gamma curves were subjected to median filtering with windows of 7, 13, 19 and 25, respectively. After filtering, gamma curves of different scales reflected geological features of different levels, and the logging curves removed high-frequency noise and effectively retained edge information. Figure 4 The schematic diagram of a well logging curve and sedimentary microfacies interpretation results in a certain section of the study area is shown, in which the original well logging curve and multiple median filtered curves are arranged horizontally to form two-dimensional data, and the two-dimensional data are given sedimentary microfacies labels. 100,000 and 10,000 gamma curve sample points and the corresponding 5 sedimentary microfacies interpretation results are selected as training sets and validation sets respectively. Figure 5The sedimentary microfacies categories and the division of network training data involved in this embodiment are demonstrated. Figure 6 and Figure 7 The distribution characteristics of each sedimentary microfacies category in the training set and validation set are shown respectively.

[0070] Step 2: Construct a convolutional neural network UNet as a feature extractor to extract the spatial features of the well logging curve from the two-dimensional data.

[0071] The two-dimensional logging response features composed of the original logging data and the logging data after median filtering are the basic data for the interpretation of logging sedimentary microfacies. Due to the complexity and particularity of the formation of geological bodies, the thickness and grain size of geological bodies formed under the same sedimentary environment in the sedimentary cycle are different. This difference is reflected in the logging curve as the difference in the curve amplitude and the number of sampling points in the depth direction of the same sedimentary microfacies. This vertical data correlation can be processed with sequence data, while multi-scale lines with differences such as curve amplitude need to be extracted based on convolution kernels with different weights. Therefore, the extraction of the spatial directional morphological changes of the two-dimensional logging response features is handed over to the convolutional neural network UNet, which includes a downsampling layer, an upsampling layer and a feature extraction layer. The downsampling process extracts features, and the upsampling process performs feature fusion, and the size of the input data remains unchanged. The extracted features reflect the morphological differences such as the amplitude of the logging curve, providing a basis for the division of sedimentary microfacies.

[0072] Step 3: Construct the ConvLSTM network model architecture, inherit the spatial features extracted by UNet and process the vertical connection of the logging curve to form a well logging sedimentary microfacies feature classifier.

[0073] The spatiotemporal sequence model ConvLSTM network was first proposed to solve the problem of precipitation forecasting. Since the simple LSTM neural network model only considers the temporal characteristics of the data and cannot utilize the spatial characteristics of the data, ConvLSTM was proposed to obtain the spatiotemporal relationship of the data. The two-dimensional response characteristics composed of the original logging data and the median filtered data not only reflect the characteristic changes in the sedimentary microfacies space, but also show temporal characteristics. The sudden change in the logging curve response is interpreted as being related to unconformity and sequence boundaries, and the change in the curve morphology reveals the sedimentary cycle and sedimentary background. Based on the logging curve features extracted by UNet, the ConvLSTM neural network model is selected as the feature classifier to further process the previously extracted features by comprehensively considering the differences in spatial morphology such as the amplitude of the logging curve. ConvSTM not only considers the amplitude difference of the logging curve, inherits the spatial features extracted by UNet, and makes full use of the vertical characteristic changes of the logging curve, but also processes this data association of a certain length with sequence features.

[0074] Step 4: Use the median filtered data together with the original logging data to construct training samples, divide the training set and validation set, perform network training and parameter adjustment, construct a joint loss function, combine multi-classification cross entropy and exponential loss, and evaluate the classification effect of the network. Train the network architecture, complete the process of optimizing and adjusting network parameters, establish the category correspondence between logging curves and sedimentary microfacies, and save the optimal network architecture and node parameters and other parameter settings.

[0075] 100,000 gamma curve samples in the study area were used as training data, and 10,000 samples were used as validation sets to evaluate the effectiveness and reliability of the method. In order to ensure the stability of the training process, the input logging data was normalized to 0-1, and the two-dimensional logging response features were divided into a two-dimensional matrix of 128*128 size as the input of the UNet convolutional neural network. After passing through this feature extractor, the extracted spatial features were input into the ConvLSTM model again, the time step was set to 128, the learning rate was set to 0.0001, and every 512 samples were used as a batch to participate in the network training. The input data contained 5 features and iterated 150 times. RMSprop was selected as the optimization algorithm to extract the spatiotemporal correlation of the data, construct a joint loss function, and examine the changes in the network error and accuracy during the training process to optimize the network architecture.

[0076]

[0077] Among them, J represents the joint loss function; b represents the label; p represents the predicted probability; K represents the total number of sample categories; N represents the number of samples; b i,k Indicates whether the label of the i-th sample is the k-th sedimentary microfacies, if yes, it is 1, if not, it is 0; p i,k It represents the probability of predicting that the i-th sample is the k-th sedimentary microfacies; λ is a hyperparameter that controls the weight of the exponential loss function and needs to be determined through hyperparameter experiments.

[0078] Finally, according to the loss value and accuracy changes of the network model, it was determined that the UNet model contained 3 downsampling layers, a feature extraction layer and 3 upsampling layers, and the ConvLSTM network model contained 5 hidden layers. Relu was used as the activation function, and a fully connected layer and a softmax probability layer were used to output the division results of sedimentary microfacies. The results of the network output were normalized by the softmax function, and the classification probability between 0 and 1 was output. Through the network hyperparameter experiment, the weight coefficient λ in the joint loss function was determined to be 0.2. The training process of the network model was stable and efficient, and the network did not have problems such as overfitting. The node weights and hyperparameters of the network were saved to facilitate the subsequent application of complex scenarios such as transfer learning.

[0079] Step 5: Set up a comparative experiment. Analyze and evaluate the effects of the single network structure and the combined neural network framework on the identification of sedimentary microfacies.

[0080] In order to verify the accuracy and effectiveness of the present invention for the classification of well logging sedimentary microfacies, two comparative experiments were designed to evaluate the degree of mining of morphological differences in logging data by different network structures. One is to use the convolutional neural network Unet model as the network architecture to extract the two-dimensional logging response features and perform an identification method for sedimentary microfacies classification; the other is to use ConvLSTM as the network architecture to comprehensively consider the correlation and changes in the amplitude and vertical direction of the logging curve, so as to perform a technical framework for sedimentary microfacies division. Figure 8 and Fig. 9 The loss values ​​and recognition accuracy curves of different classification models are shown respectively. Among them, the joint network architecture with UNet as feature extractor and ConvLSTM as feature classifier has the lowest classification error and the highest recognition accuracy.

[0081] Step 6: Evaluate the test results of the training set and the validation set respectively to further examine the practical application potential of the recognition method of the present invention.

[0082] The trained network structures are applied to the training set and the validation set respectively. Fig.10 The results of sedimentary microfacies identification of different network structures on the training set are shown. From left to right, they are the manually calibrated sedimentary microfacies categories, the identification results of the UNet neural network, the division results of the ConvLSTM network, and the classification effect of the joint network architecture model with UNet as the feature classifier and ConvLSTM as the feature classifier, with accuracy rates of 82.50%, 90.18%, and 93.78%, respectively. This experimental result fully demonstrates the applicability of the joint network architecture for well logging sedimentary microfacies identification scenarios. Compared with the single convolutional neural network UNet and ConvLSTM network models, it greatly improves the recognition accuracy.

[0083] Fig.11The results of sedimentary microfacies recognition on the validation set were shown for different network structures. The recognition accuracy of the convolutional neural network UNet was 71.89%, the recognition accuracy of the ConvLSTM neural network model was 85.18%, and the classification accuracy of the joint network architecture model with UNet as the feature extractor and ConvLSTM as the feature classifier was 91.40%. The experimental results show that the convolutional neural network UNet is most susceptible to the impact of uneven data distribution, which is reflected in the significant reduction in recognition accuracy on the training set and the validation machine. Compared with the convolutional neural network UNet, ConvLSTM gives a higher classification accuracy because it considers both the spatial and vertical features of the logging curve; and the joint network architecture of the joint UNet feature extractor and ConvLSTM feature classifier has achieved good application results on both the training set and the validation set because it considers the spatiotemporal morphological changes of the logging curve at the same time. It is relatively less affected by the uneven distribution of labels and has better generalization. Judging from the sedimentary microfacies recognition effect of the validation set, the introduction of the gamma curve after median filtering improves the influence of data distribution characteristics on the recognition results, has higher recognition accuracy, and provides two-dimensional logging response characteristics, which facilitates the UNet model to mine the spatial morphological changes of two-dimensional logging response characteristics. The convolution operation in ConvLSTM makes good use of the data features provided by UNet, and processes the vertical connection of the data based on this spatial feature, which is conducive to the division of boundaries of different sedimentary microfacies and lays the foundation for efficient and accurate geological interpretation and sedimentary analysis of logging data.

[0084] It should be noted that a person skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods.

[0085] Furthermore, it should be appreciated that the computer-readable storage medium (eg, memory) herein may be either a volatile memory or a nonvolatile memory, or may include both volatile and nonvolatile memory.

[0086] A person skilled in the art should understand that the discussion of any of the above embodiments is only exemplary and is not intended to imply that the scope of the disclosure of the embodiments of the present invention (including the claims) is limited to these examples; under the concept of the embodiments of the present invention, the technical features in the above embodiments or different embodiments can also be combined, and there are many other changes in different aspects of the embodiments of the present invention as described above, which are not provided in detail for the sake of simplicity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present invention should be included in the protection scope of the embodiments of the present invention.

Claims

1. A sedimentary microfacies identification model training method, characterized in that: include: Acquire a training sample, wherein the training sample includes two-dimensional data with a label, wherein the two-dimensional data includes an original well logging curve arranged in a transverse direction and a plurality of median filtered curves, wherein the plurality of median filtered curves are obtained by performing median filtering of different scales on the original well logging curve, and the label indicates a sedimentary microfacies category; Constructing a joint model, the joint model comprising a feature extraction model and a feature classification model, the feature extraction model extracting multi-scale spatial features of the two-dimensional data, the feature classification model extracting vertical features of the two-dimensional data and performing classification based on the spatial features and the vertical features; The joint model is trained using the training samples to obtain the sedimentary microfacies identification model.

2. The method according to claim 1, characterized in that The feature extraction model is a convolutional neural network model.

3. The method according to claim 2, characterized in that The feature extraction model is a UNet model.

4. The method according to claim 1, characterized in that: The feature classification model is a ConvLSTM model.

5. The method according to claim 1, characterized in that The original logging curve is a natural gamma curve; And / or the sedimentary microfacies categories include: channel distributary bay, front edge sheet sand, underwater distributary channel, estuary bar, and channel side edge.

6. The method according to claim 1, characterized in that The joint model is trained based on a joint loss function, which is constructed based on a multi-classification cross entropy loss function and an exponential loss function, and the formula is as follows: Among them, J represents the joint loss function; b represents the label; p represents the predicted probability; K represents the total number of sample categories; N represents the number of samples; b i,k Indicates whether the label of the i-th sample is the k-th sedimentary microfacies, if yes, it is 1, if not, it is 0; p i,k It represents the probability of predicting that the i-th sample is the k-th sedimentary microfacies; λ is a hyperparameter.

7. A sedimentary microfacies identification model training device, characterized in that: include: An acquisition module is configured to acquire training samples, wherein the training samples include two-dimensional data with labels, wherein the two-dimensional data include original well logging curves arranged in a transverse direction and a plurality of median filtered curves, wherein the plurality of median filtered curves are obtained by performing median filtering of different scales on the original well logging curves, and the labels indicate sedimentary microfacies categories; A construction module is configured to construct a joint model, wherein the joint model includes a feature extraction model and a feature classification model, wherein the feature extraction model extracts multi-scale spatial features of the two-dimensional data, and the feature classification model extracts vertical features of the two-dimensional data and performs classification based on the spatial features and the vertical features; A training module is configured to train the joint model using the training samples to obtain the sedimentary microfacies identification model.

8. A computer device comprising: at least one processor; as well as A memory storing a computer program executable on the processor, wherein the processor implements the method according to any one of claims 1 to 6 when executing the program.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

10. A sedimentary microfacies identification method, characterized in that: include: Identification is performed using a sedimentary microfacies identification model trained according to the method described in any one of claims 1 to 6.

Citation Information

Patent Citations

  • Logging sedimentary microfacies identification method based on deep learning

    CN112016477A

  • Logging lithology identification method and system based on LSTM neural network

    CN112989708A

  • Carbonate reservoir sedimentary facies identification method and device, electronic equipment and medium

    CN113627607A

  • Logging sedimentary microfacies division method and device based on target detection

    CN114463610A

  • Lithology recognition method for volcanic rock based on fractal dimensions of electrical imaging logging

    WO2019205725A1

Cited By

  • Sedimentary microfacies identification method, system and equipment and computer storage medium

    CN119293641A

  • A sedimentary microfacies identification method, system, device and computer storage medium

    CN119293641B