Automatic identification method of conglomerate rock combined with imaging logging and logging data

Abstract

This invention provides an automatic identification method for sandstone and conglomerate based on imaging logging and combined logging data, comprising: Step 1, preprocessing imaging logging data; Step 2, constructing sandstone and conglomerate lithology sample data; Step 3, constructing an automatic identification network structure for sandstone and conglomerate based on deep learning-based imaging logging and combined logging data; Step 4, training and validating the automatic identification network for sandstone and conglomerate based on deep learning-based imaging logging and combined logging data; Step 5, conducting application tests of the automatic identification method for sandstone and conglomerate based on deep learning-based imaging logging and combined logging data. This automatic identification method for sandstone and conglomerate based on imaging logging and combined logging data takes one-dimensional logging data and two-dimensional imaging logging data as input, and utilizes the constructed combined identification network model to effectively fuse multi-dimensional logging data, achieving high-precision automatic identification of sandstone and conglomerate lithofacies and lithology.

Description

Technical Field

[0001] This invention relates to the field of exploration geophysics, and in particular to an automatic identification method for sandstone and conglomerate rocks based on imaging logging and logging data fusion. Background Technology

[0002] Conglomerate is an important target for oil and gas exploration. Well logging is a crucial exploration method for identifying promising lithologies in conglomerate. However, due to the large vertical thickness, rapid facies transitions, and complex spatial variations of conglomerate, characterization is challenging. Conventional well logging provides a comprehensive series of measurements, including sonic transit time, density, resistivity, gamma ray, spontaneous potential, and other observational data, offering a multifaceted basis for lithofacies and lithology interpretation. However, its resolution is relatively low, making it unable to subdivide promising lithofacies and lithologies. Imaging logging, using arrayed sensors to acquire wellbore lithological information from all angles, including sonic and electrical imaging logging, can provide more sufficient evidence for lithological classification of conglomerates of different grain sizes. However, it suffers from high acquisition costs, relatively limited observation types, heavy interpretation workload, low efficiency, and strong subjectivity. Therefore, it is necessary to conduct relevant research to combine the advantages of both methods and design an automatic identification technology process for conglomerate that can simultaneously perform conventional and imaging logging.

[0003] Chinese patent application CN201810971743.1 discloses an automatic identification method and apparatus for electrical imaging logging facies. The method includes: acquiring historical electrical imaging logging data; preprocessing the historical data to generate electrical imaging logging images covering the entire wellbore; identifying typical imaging logging facies in the images and determining them as training samples according to their respective imaging logging facies categories; constructing a deep learning model, including an input layer, multiple hidden layers, and an output layer; training the deep learning model using the training samples to obtain a trained deep learning model; and using the trained deep learning model to identify logging facies in the electrical imaging logging images of the well section to be identified. This scheme can automatically and accurately identify the imaging logging facies type of each section, thereby better studying and predicting the distribution patterns of reservoirs. However, this application only uses electrical imaging logging data and does not incorporate information from conventional logging data.

[0004] Chinese patent application CN201810758068.4 discloses a method for fine evaluation of the lithology of conglomerate bodies based on electrical imaging logging. This evaluation method offers more reliable assessment and more accurate results. The method includes the following steps: preprocessing a two-dimensional static microresistivity scan image by filling blank strips; performing edge detection on the conglomerate particles in the preprocessed image and segmenting them from the background; merging and splitting the segmented conglomerate particles according to geological sedimentary constraints; and calculating the grain size spectrum and lithology curve based on a sliding window and particle size. However, this application only addresses the edge detection segmentation method in imaging logging and does not incorporate information from conventional logging data.

[0005] Chinese patent application CN202110556296.5 discloses a seismic lithology prediction method based on integrated deep learning, belonging to the field of petroleum geophysical exploration technology. The method includes: acquiring lithological data corresponding to a calibrated target stratigraphic level and wellbore seismic data; establishing an integrated deep learning model for seismic lithology identification based on the wellbore seismic data and the wellbore lithological data, using the wellbore seismic data as input data and the wellbore lithological data as target data; and inputting actual seismic data of the area to be predicted based on the seismic lithology identification integrated deep learning model to obtain the predicted lithology. This method extracts weak seismic response characteristics between seismic data and lithological data by establishing an integrated deep learning model for seismic lithology identification, enabling simpler and more efficient determination of lithological data volumes of reservoir distribution, solving the problem of reservoir prediction between wells, and providing a reference for oil and gas exploration and development.

[0006] The above-mentioned existing technologies are all quite different from the present invention and have failed to solve the technical problem we want to solve. Therefore, we have invented an automatic identification method for sandstone and conglomerate based on imaging logging and logging data. Summary of the Invention

[0007] The purpose of this invention is to provide a more efficient and reliable automatic identification technology and process for sandstone and conglomerate logging, and to provide technical support for exploration deployment by combining imaging logging and logging data for automatic identification of sandstone and conglomerate.

[0008] The objective of this invention can be achieved through the following technical measures: The automatic identification method for sandstone and conglomerate based on imaging logging and combined logging data includes:

[0009] Step 1: Perform imaging logging data preprocessing;

[0010] Step 2: Construct lithological sample data for sandstone and conglomerate;

[0011] Step 3: Construct an automatic identification network structure for sandstone and conglomerate based on deep learning-based imaging logging and logging data fusion;

[0012] Step 4: Train and validate the automatic identification network for sandstone and conglomerate based on deep learning-based imaging logging and logging data fusion.

[0013] Step 5: Based on deep learning, imaging logging and logging data fusion are used to test the application of the automatic identification method for sandstone and conglomerate.

[0014] The objective of this invention can also be achieved through the following technical measures:

[0015] In step 1, the blank strips of the imaging logging data are filled to obtain imaging logging data covering the entire wellbore.

[0016] In step 1, the image is segmented along the depth direction, and the white stripes of the imaging logging data are interpolated and filled to obtain the preprocessed imaging logging image.

[0017] The strip filling problem is transformed into an optimization problem, with the objective function J as follows:

[0018] J = ||d-RFx img ||2+λ||Fx img ||1

[0019] Where d represents the two-dimensional imaging logging data before filling, R is the mask matrix indicating the strip positions, F is the two-dimensional Fourier transform operator, and x img To fill the desired stripes with data, the ||d-RFx of the objective function img ||2 represents the data fitting term, indicating that the filled imaging data x should fit the original imaging logging data y at the stripe-free location;||Fx img ||1 represents the sparsity constraint term, indicating that the Fourier transform result of the filled imaging data x should have sparsity; λ is the weight of the sparsity constraint term.

[0020] In step 2, lithological interpretation is performed on imaging logging and conventional logging data, and sample feature data and sample label data required for supervised deep learning are constructed.

[0021] In step 2, the sample data consists of two parts: sample feature data as input and sample label data as output; the sample feature data includes two-dimensional imaging logging sample data x. img One-dimensional conventional well logging curve sample characteristic data x log The sample label data d is a discrete integer code representing the lithological interpretation conclusion, which comes from core calibration or well logging interpretation based on expert experience.

[0022] In step 2, to highlight the local features of different depth segments of the data, the two-dimensional imaging logging sample data, the one-dimensional conventional logging curve sample feature data, and the lithological interpretation conclusion sample label data are split along the depth segment using a sliding time window method.

[0023] In step 3, a deep learning-based imaging logging and logging data fusion automatic identification network structure for sandstone and conglomerate is constructed to realize the joint analysis of imaging logging and logging data and jointly carry out automatic identification of sandstone and conglomerate lithology.

[0024] Step 3 includes:

[0025] Step 31: For two-dimensional imaging logging data, an encoding network consisting of two-dimensional convolutional layers and activation layers, and a decoding network consisting of two-dimensional deconvolutional layers and activation layers are constructed. The output of each layer of the encoding network is extended and connected to each layer of the corresponding decoding network along the channel direction. The output of the last layer of the decoding network is convolved along the horizontal direction, keeping the size of the output result unchanged in the depth direction, thereby reducing the dimensionality of the two-dimensional imaging data feature extraction result and obtaining the one-dimensional imaging data feature extraction result.

[0026] Step 32: For one-dimensional conventional logging data, an encoding network consisting of one-dimensional convolutional layers and activation layers, and a decoding network consisting of one-dimensional deconvolutional layers and activation layers are used. The output of each layer of the encoding network is extended and connected to each layer of the corresponding decoding network along the channel direction to obtain the feature extraction results of the one-dimensional conventional logging data.

[0027] Step 33: Extend and connect the feature extraction results of one-dimensional imaging data and the feature extraction results of one-dimensional conventional logging data along the channel direction to obtain a joint feature output result containing imaging data features and conventional logging data features;

[0028] Step 34: Convolve along the channel direction of the joint feature output using a convolutional layer to obtain a one-dimensional joint feature output of the same size as the lithological sample label data.

[0029] In step 4, driven by the sample data, the weight parameters of the constructed sandstone and conglomerate automatic identification network based on deep learning imaging logging and logging data are optimized and solved to obtain a sandstone and conglomerate automatic identification network model that meets the expected results.

[0030] Step 4 includes:

[0031] Step 41: Select a loss function to train the deep learning network for automatic identification of multi-dimensional data conglomerate. The optimization objective can then be expressed as minimizing the loss function L:

[0032]

[0033] Where y represents the true lithological label. The joint feature output of the sandstone and conglomerate automatic identification network based on deep learning-based imaging logging and logging data. min represents minimization, and Loss is the loss calculation function;

[0034] Step 42: Set key parameters such as the number of iterations and the learning rate, and perform iterative optimization calculations through backpropagation to obtain the optimal network weights;

[0035] Step 43: Test the test set using the trained conglomerate lithology network model to evaluate the recognition effect. When the expected effect is met, output the conglomerate lithology recognition network model. Otherwise, readjust the key parameters such as the number of iterations and the learning rate, train repeatedly, and then output the conglomerate lithology recognition network model.

[0036] In step 5, the trained automatic identification network model for sandstone and conglomerate lithology is used to predict the imaging logging data and conventional logging data at the actual well points to obtain the automatic identification results of sandstone and conglomerate lithology, which helps interpreters identify the depth range with favorable lithological development.

[0037] Step 5 includes:

[0038] Step 51: Using the trained sandstone and conglomerate lithology identification network model, segmented predictions are made using imaging logging data and conventional logging data of the well to be predicted.

[0039] Step 52: Select the maximum probability value of the network output as the lithology output result corresponding to the depth sampling point to obtain the final sandstone and conglomerate lithology identification result.

[0040] This invention proposes an automatic identification method for sandstone and conglomerate rocks by combining imaging logging and logging data. Using core descriptions and expert interpretations as sample labels, and one-dimensional logging data and two-dimensional imaging logging data as input, a constructed joint identification network model is used to effectively fuse multi-dimensional logging data, achieving high-precision automatic identification of sandstone and conglomerate lithofacies and lithology. Compared with existing technologies, the main innovations of this invention are:

[0041] (1) A combined imaging logging and conventional logging lithology identification method is proposed for sandstone and conglomerate. Existing lithology identification methods at well points usually only identify based on imaging logging data or conventional logging data. However, this patent proposes a deep learning-based automatic lithology identification method for sandstone and conglomerate that can combine imaging logging data and conventional logging data. Wellbore observation data is more fully utilized, which is more conducive to further improving the accuracy of lithology identification.

[0042] (2) The constructed automatic identification network structure for sandstone and conglomerate based on deep learning imaging logging and logging data fusion is clearer. Traditional professional methods for well logging lithology identification require manual design to obtain spatial variation characteristics of wellbore data, which is difficult to apply to the identification of sandstone and conglomerate with multiple lithology types and complex spatial variation patterns. However, this patent constructs a clear deep neural network structure for the sandstone and conglomerate lithology identification problem. Driven by the fusion of imaging logging data and conventional logging data, it can automatically mine deep features of wellbore data, which helps to improve the accuracy of sandstone and conglomerate lithology identification. At the same time, the application process is simple and the identification efficiency is high. Attached Figure Description

[0043] Figure 1 This is a flowchart of a specific embodiment of the automatic identification method for sandstone and conglomerate based on imaging logging and logging data combination of the present invention;

[0044] Figure 2 This is a diagram showing the effect of filling in imaging logging data strips according to a specific embodiment of the present invention;

[0045] Figure 3 This is a diagram of the automatic identification network structure for sandstone and conglomerate based on imaging logging and logging data integration, as described in a specific embodiment of the present invention.

[0046] Figure 4 This is a test image illustrating the application of the automatic identification method for sandstone and conglomerate based on imaging logging and logging data combined according to a specific embodiment of the present invention.

[0047] Figure 5 This is a test image illustrating the application of the automatic identification method for sandstone and conglomerate based on imaging logging and logging data combined, as described in a specific embodiment of the present invention. Detailed Implementation

[0048] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0049] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, and / or combinations thereof.

[0050] like Figure 1 The flowchart shown is a method for automatic identification of sandstone and conglomerate based on deep learning and imaging logging multi-dimensional data according to the present invention. The specific steps of this embodiment are as follows:

[0051] (1) Imaging logging data preprocessing. The blank strips in the imaging logging data are filled to obtain imaging logging data covering the entire wellbore.

[0052] (2) Construction of lithological sample data for sandstone and conglomerate. Lithological interpretation of imaging logging and conventional logging data is performed, and sample feature data and sample label data required for supervised deep learning are constructed.

[0053] (3) Automatic identification network structure for sandstone and conglomerate based on deep learning imaging logging and logging data fusion. A basic deep neural network structure including two-dimensional image analysis and one-dimensional curve analysis is constructed to realize the fusion analysis of imaging logging and logging data, and to carry out automatic identification of sandstone and conglomerate lithology.

[0054] (4) Training and validation of an automatic sand and conglomerate identification network based on deep learning imaging logging and logging data fusion. Driven by sample data, the weight parameters in the constructed deep neural network for automatic sand and conglomerate identification are optimized to obtain an automatic sand and conglomerate identification network model that meets the expected results.

[0055] (5) Application test of automatic identification method for sandstone and conglomerate based on imaging logging and logging data combination. Using the trained automatic identification network model for sandstone and conglomerate lithology, the imaging logging data and conventional logging data at the actual well points are used to predict the automatic identification results of sandstone and conglomerate lithology, which helps interpreters to identify the depth range of favorable lithology development.

[0056] The following are several specific embodiments of the application of the present invention.

[0057] Example 1

[0058] In a specific embodiment 1 of the present invention, the automatic identification method for sandstone and conglomerate based on imaging logging and logging data includes the following steps:

[0059] Step 1: Fill in the blank strips in the imaging logging data to obtain imaging logging data covering the entire wellbore.

[0060] The image is segmented along the depth direction, and the white stripes of the imaging logging data are interpolated and filled to obtain the preprocessed imaging logging image.

[0061] The strip filling problem is transformed into an optimization problem, with the objective function J as follows:

[0062] J = ||d-RFx img ||2+λ||Fx img ||1

[0063] Where d represents the two-dimensional imaging logging data before filling, R is the mask matrix indicating the strip positions, F is the two-dimensional Fourier transform operator, and ximg Fill the data to obtain the desired stripes. The ||d-RFx|| of the objective function. img ||2 represents the data fitting term, indicating that the filled imaging data x should fit the original imaging logging data y at the stripe-free location;||Fx img ||1 represents the sparsity constraint term, indicating that the Fourier transform result of the filled imaging data x should have sparsity; λ is the weight of the sparsity constraint term.

[0064] Step 2 involves interpreting the lithology of imaging logging and conventional logging data, and constructing sample feature data and sample label data required for supervised deep learning.

[0065] The sample data consists of two parts: sample feature data as input and sample label data as output. The sample feature data includes two-dimensional imaging logging sample data x. img One-dimensional conventional well logging curve sample characteristic data x log The sample label data d is a discrete integer code representing the lithological interpretation conclusion, which comes from core calibration or well logging interpretation based on expert experience.

[0066] To highlight the local characteristics of different depth segments of the data, the two-dimensional imaging logging sample data, the one-dimensional conventional logging curve sample feature data, and the lithological interpretation conclusion sample label data are split along the depth segment using a sliding time window method.

[0067] Step 3: Construct an automatic identification network structure for sandstone and conglomerate based on deep learning-based imaging logging and logging data fusion, so as to realize the joint analysis of imaging logging and logging data and carry out automatic identification of sandstone and conglomerate lithology.

[0068] First, for two-dimensional imaging logging data, an encoding network consisting of two-dimensional convolutional layers and activation layers, and a decoding network consisting of two-dimensional deconvolutional layers and activation layers are used. The output of each layer of the encoding network is extended and connected to the corresponding layer of the decoding network along the channel direction. The output of the last layer of the decoding network is convolved along the horizontal direction, keeping the size of the output result unchanged in the depth direction, thereby achieving dimensionality reduction of the two-dimensional imaging data feature extraction results and obtaining the one-dimensional imaging data feature extraction results.

[0069] Secondly, for one-dimensional conventional logging data, an encoding network consisting of one-dimensional convolutional layers and activation layers, and a decoding network consisting of one-dimensional deconvolutional layers and activation layers are used. The output of each layer of the encoding network is extended and connected to each layer of the corresponding decoding network along the channel direction to obtain the feature extraction results of one-dimensional conventional logging data.

[0070] Next, the feature extraction results of one-dimensional imaging data and the feature extraction results of one-dimensional conventional logging data are extended and connected along the channel direction to obtain a joint feature output result that includes both imaging data features and conventional logging data features.

[0071] Finally, convolution is performed along the channel direction of the joint feature output result using a convolutional layer to obtain a one-dimensional joint feature output result of the same size as the lithological sample label data.

[0072] Step 4: Driven by the sample data, optimize the weight parameters of the constructed sand and conglomerate automatic identification network based on deep learning imaging logging and logging data fusion to obtain a sand and conglomerate automatic identification network model that meets the expected results.

[0073] First, by selecting a loss function for training a deep learning network for automatic identification of sandstone and conglomerate based on multi-dimensional data, the optimization objective can be expressed as minimizing the loss function L:

[0074]

[0075] Where y represents the true lithological label. The joint feature output of the sandstone and conglomerate automatic identification network based on deep learning-based imaging logging and logging data. min represents minimization, and Loss is the loss calculation function.

[0076] Secondly, key parameters such as the number of iterations and the learning rate are set, and iterative optimization calculations are performed through backpropagation to obtain the optimal network weights.

[0077] Finally, the trained conglomerate lithology network model is used to test the test set and evaluate the recognition effect. When the expected effect is met, the conglomerate lithology recognition network model is output. Otherwise, the key parameters such as the number of iterations and learning rate are readjusted, and the model is trained repeatedly before outputting the conglomerate lithology recognition network model again.

[0078] Step 5: Using the trained automatic identification network model for sandstone and conglomerate lithology, predict the imaging logging data and conventional logging data at the actual well points to obtain the automatic identification results of sandstone and conglomerate lithology, which helps interpreters identify the depth ranges with favorable lithological development.

[0079] First, using a trained sandstone and conglomerate lithology identification network model, segmented predictions are made based on imaging logging data and conventional logging data of the well to be predicted.

[0080] Secondly, the maximum probability value of the network output is selected as the lithology output result corresponding to the depth sampling point to obtain the final sandstone and conglomerate lithology identification result.

[0081] Example 2

[0082] In a specific embodiment 2 of the present invention, the present invention will be further described below in conjunction with specific embodiments.

[0083] (1) Preprocessing of imaging logging data. Figure 2 This is a diagram showing the effect of filling the imaging logging data strips as described in the embodiment.

[0084] (2) Construction of lithological sample data for sandstone and conglomerate. Lithological interpretation was performed on imaging logging and conventional logging data, and sample feature data and sample label data required for supervised deep learning were constructed. The lithology included five types: mudstone, sandstone, gravelly sandstone, conglomerate sandstone, and conglomerate, coded with the numbers 1, 2, 3, 4, and 5, respectively. The sample data came from three wells, and the sample data from each well was divided into 600 data sets, each 6m in length.

[0085] (3) Automatic identification network structure for sandstone and conglomerate based on deep learning imaging logging and logging data fusion. A basic deep neural network structure including two-dimensional image analysis and one-dimensional curve analysis is constructed to realize the fusion analysis of imaging logging and logging data, and to carry out automatic identification of sandstone and conglomerate lithology.

[0086] Figure 3 This is a deep learning network diagram for automatic identification of multi-dimensional data conglomerate as described in a specific embodiment.

[0087] First, for 2D imaging logging data, a three-layer encoding network consisting of a 3×3 2D convolutional layer, a ReLU activation layer, and another 3×3 2D convolutional layer with a ReLU activation layer is used. A three-layer decoding network consisting of a 3×3 2D deconvolutional layer, a ReLU activation layer, and another 3×3 2D convolutional layer with a ReLU activation layer is also used. Each layer output of the encoding network is extended and connected to each corresponding layer of the decoding network along the channel direction. The output of the last layer of the decoding network is convolved horizontally, maintaining the depth dimension of the output, thus reducing the dimensionality of the 2D imaging data feature extraction results to obtain the 1D imaging data feature extraction results.

[0088] Secondly, for one-dimensional conventional logging data, a three-layer encoding network consisting of a 3×1 one-dimensional convolutional layer, a ReLU activation layer, and another 3×1 one-dimensional convolutional layer is used; a three-layer decoding network consisting of a 3×1 one-dimensional deconvolutional layer, a ReLU activation layer, and another 3×1 one-dimensional convolutional layer is used; the output of each layer of the encoding network is extended and connected to each layer of the corresponding decoding network along the channel direction to obtain the feature extraction results of the one-dimensional conventional logging data.

[0089] Next, the feature extraction results of one-dimensional imaging data and the feature extraction results of one-dimensional conventional logging data are extended and connected along the channel direction to obtain a joint feature output result that includes both imaging data features and conventional logging data features.

[0090] Finally, a 1×1 convolutional layer is used to perform convolution along the channel direction of the joint feature output result to obtain a one-dimensional joint feature output result of the same size as the lithological sample label data.

[0091] (4) Training and validation of an automatic sand and conglomerate identification network based on deep learning imaging logging and logging data fusion. Driven by sample data, the weight parameters in the constructed deep neural network for automatic sand and conglomerate identification are optimized to obtain an automatic sand and conglomerate identification network model that meets the expected results.

[0092] First, if the Dice coefficient Loss function is selected for training the deep learning network for automatic identification of multi-dimensional data conglomerate, then the optimization objective can be expressed as minimizing the loss function L:

[0093]

[0094] Where y represents the true lithological label. The joint feature output of the sandstone and conglomerate automatic identification network based on deep learning-based imaging logging and logging data. min represents minimization, and Loss is the loss calculation function.

[0095] Secondly, the number of iterations was set to 500 and the learning rate to 0.001. Iterative optimization calculations were performed through backpropagation to obtain the optimal network weights.

[0096] (5) Application test of automatic identification method for sandstone and conglomerate based on imaging logging and logging data combination. Figure 4 This image shows a test application of a deep learning network model for automatic identification of sandstone and conglomerate lithology based on multi-dimensional data. Using the trained network model, predictions were made based on imaging logging data and conventional logging data from actual well locations. The obtained sandstone and conglomerate lithology results showed a high degree of agreement with expert interpretations, effectively assisting interpreters in clarifying different sandstone and conglomerate lithologies.

[0097] Example 3

[0098] In a specific embodiment 3 of the present invention, the present invention will be further described below in conjunction with specific embodiments.

[0099] (1) Preprocessing of imaging logging data. Figure 2 This is a diagram showing the effect of filling the imaging logging data strips as described in the embodiment.

[0100] (2) Construction of lithological sample data for sandstone and conglomerate. Lithological interpretation was performed on imaging logging and conventional logging data, and sample feature data and sample label data required for supervised deep learning were constructed. The lithology included five types: mudstone, sandstone, gravelly sandstone, conglomerate sandstone, and conglomerate, coded with the numbers 1, 2, 3, 4, and 5, respectively. The sample data came from three wells, and the sample data from each well was divided into 600 data sets, each 6m in length.

[0101] (3) Automatic identification network structure for sandstone and conglomerate based on deep learning imaging logging and logging data fusion. A basic deep neural network structure including two-dimensional image analysis and one-dimensional curve analysis is constructed to realize the fusion analysis of imaging logging and logging data, and to carry out automatic identification of sandstone and conglomerate lithology.

[0102] First, for 2D imaging logging data, a three-layer encoding network consisting of a 5×5 2D convolutional layer, a ReLU activation layer, and another 5×5 2D convolutional layer with a ReLU activation layer is used. A three-layer decoding network consisting of a 5×5 2D deconvolutional layer, a ReLU activation layer, and another 5×5 2D convolutional layer with a ReLU activation layer is also used. Each layer output of the encoding network is extended and connected to each corresponding layer of the decoding network along the channel direction. The output of the last layer of the decoding network is convolved horizontally, maintaining the depth dimension of the output, thus reducing the dimensionality of the 2D imaging data feature extraction results to obtain the 1D imaging data feature extraction results.

[0103] Secondly, for one-dimensional conventional logging data, a three-layer encoding network consisting of a 5×1 one-dimensional convolutional layer, a ReLU activation layer, and another 5×1 one-dimensional convolutional layer is used; a three-layer decoding network consisting of a 5×1 one-dimensional deconvolutional layer, a ReLU activation layer, and another 5×1 one-dimensional convolutional layer is used; the output of each layer of the encoding network is extended and connected to each layer of the corresponding decoding network along the channel direction to obtain the feature extraction results of the one-dimensional conventional logging data.

[0104] Next, the feature extraction results of one-dimensional imaging data and the feature extraction results of one-dimensional conventional logging data are extended and connected along the channel direction to obtain a joint feature output result that includes both imaging data features and conventional logging data features.

[0105] Finally, a 1×1 convolutional layer is used to perform convolution along the channel direction of the joint feature output result to obtain a one-dimensional joint feature output result of the same size as the lithological sample label data.

[0106] (4) Training and validation of an automatic sand and conglomerate identification network based on deep learning imaging logging and logging data fusion. Driven by sample data, the weight parameters in the constructed deep neural network for automatic sand and conglomerate identification are optimized to obtain an automatic sand and conglomerate identification network model that meets the expected results.

[0107] First, if the Dice coefficient Loss function is selected for training the deep learning network for automatic identification of multi-dimensional data conglomerate, then the optimization objective can be expressed as minimizing the loss function L:

[0108]

[0109] Where y represents the true lithological label. The joint feature output of the sandstone and conglomerate automatic identification network based on deep learning-based imaging logging and logging data. min represents minimization, and Loss is the loss calculation function.

[0110] Secondly, the number of iterations was set to 1000 and the learning rate to 0.001. Iterative optimization calculations were performed through backpropagation to obtain the optimal network weights.

[0111] (5) Application test of automatic identification method for sandstone and conglomerate based on imaging logging and logging data combination. Figure 5 This image shows a test application of a deep learning network model for automatic identification of sandstone and conglomerate based on multi-dimensional data. Using the trained network model, predictions were made based on imaging logging data and conventional logging data at actual well locations. The obtained sandstone and conglomerate lithology identification results showed a high degree of agreement with expert interpretations.

[0112] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0113] Except for the technical features described in the specification, all other technologies are known to those skilled in the art.

Claims

1. An automatic identification method for sandstone and conglomerate lithology based on the combination of two-dimensional imaging logging data and one-dimensional conventional logging data, characterized in that, The method includes: Step 1: Perform two-dimensional imaging logging data preprocessing; Step 2: Construct lithological sample data for sandstone and conglomerate; Step 3: Construct an automatic identification network structure for sandstone and conglomerate lithology based on deep learning, combining two-dimensional imaging logging data and one-dimensional conventional logging data. This enables joint analysis of the two-dimensional imaging logging data and one-dimensional conventional logging data to conduct automatic identification of sandstone and conglomerate lithology, including: Step 31: For two-dimensional imaging logging data, the network structure includes an encoding network composed of two-dimensional convolutional layers and activation layers, and a decoding network composed of two-dimensional deconvolutional layers and activation layers. The output of each layer of the encoding network is extended and connected to each layer of the corresponding decoding network along the channel direction. The output of the last layer of the decoding network is convolved along the horizontal direction to keep the size of the output result unchanged in the depth direction, thereby reducing the dimensionality of the two-dimensional imaging data feature extraction result and obtaining the one-dimensional imaging data feature extraction result. Step 32: For one-dimensional conventional logging data, the network structure includes an encoding network composed of one-dimensional convolutional layers and activation layers, and a decoding network composed of one-dimensional deconvolutional layers and activation layers. The output of each layer of the encoding network is extended and connected to each layer of the corresponding decoding network along the channel direction to obtain the feature extraction result of the one-dimensional conventional logging data. The one-dimensional conventional logging data is one-dimensional conventional logging curve data, including AC, GR, DEN, SP, RD, RS and R25. Step 33: Extend and connect the feature extraction results of one-dimensional imaging data and the feature extraction results of one-dimensional conventional logging data along the channel direction to obtain a joint feature output result containing features of one-dimensional imaging data and features of one-dimensional conventional logging data. Step 34: Convolve along the channel direction of the joint feature output using a convolutional layer to obtain a one-dimensional joint feature output of the same size as the lithological sample label data. ; Step 4: Train and validate the automatic identification network for sandstone and conglomerate lithology to obtain the trained automatic identification network model for sandstone and conglomerate lithology. Step 5: Using the trained automatic identification network model for sandstone and conglomerate lithology, predict the two-dimensional imaging logging data and one-dimensional conventional logging data of the well to be predicted, and obtain the final sandstone and conglomerate lithology identification results.

2. The method for automatic identification of sandstone and conglomerate lithology by combining two-dimensional imaging logging data and one-dimensional conventional logging data according to claim 1, characterized in that, In step 1, the blank strips of the two-dimensional imaging logging data are filled to obtain two-dimensional imaging logging data covering the entire wellbore.

3. The method for automatic identification of sandstone and conglomerate lithology by combining two-dimensional imaging logging data and one-dimensional conventional logging data according to claim 2, characterized in that, In step 1, the data is segmented along the depth direction, and the white stripes of the two-dimensional imaging logging data are interpolated and filled to obtain the preprocessed two-dimensional imaging logging image. The strip filling problem is transformed into an optimization problem with the objective function. as follows: ； in, This is the two-dimensional imaging logging data before filling. A mask matrix indicating the location of the stripes. For two-dimensional Fourier transform operators, To fill the desired stripes with data, the objective function's The data fitting term represents the filled imaging data. The location without stripes should be consistent with the original imaging logging data. Fitting; The sparse constraint term represents the filled imaging data. The Fourier transform results should be sparsity; The weights of the sparsity constraint terms.

4. The method for automatic identification of sandstone and conglomerate lithology by combining two-dimensional imaging logging data and one-dimensional conventional logging data according to claim 1, characterized in that, In step 2, lithological interpretation is performed on the two-dimensional imaging logging data and the one-dimensional conventional logging data, and sample feature data and sample label data required for supervised deep learning are constructed.

5. The automatic identification method for sandstone and conglomerate lithology based on the combination of two-dimensional imaging logging data and one-dimensional conventional logging data according to claim 4, characterized in that, In step 2, the sample data consists of two parts: sample feature data as input and sample label data as output. The sample feature data includes two-dimensional imaging logging sample data and one-dimensional conventional logging curve sample feature data. Sample label data This is a discrete integer code representing the lithological interpretation conclusions, which are derived from core calibration or well logging interpretations based on expert experience.

6. The method for automatic identification of sandstone and conglomerate lithology by combining two-dimensional imaging logging data and one-dimensional conventional logging data according to claim 5, characterized in that, In step 2, to highlight the local features of different depth segments of the data, the two-dimensional imaging logging sample data, the one-dimensional conventional logging curve sample feature data, and the lithological interpretation conclusion sample label data are split along the depth segment using a sliding time window method.

7. The method for automatic identification of sandstone and conglomerate lithology by combining two-dimensional imaging logging data and one-dimensional conventional logging data according to claim 1, characterized in that, In step 4, driven by the sample data, the weight parameters of the automatic identification network for sandstone and conglomerate lithology, which is constructed by combining deep learning-based two-dimensional imaging logging data and one-dimensional conventional logging data, are optimized and solved to obtain an automatic identification network model for sandstone and conglomerate lithology that meets the expected results.

8. The method for automatic identification of sandstone and conglomerate lithology by combining two-dimensional imaging logging data and one-dimensional conventional logging data according to claim 7, characterized in that, Step 4 includes: Step 41: Select a loss function to train the deep learning network for automatic identification of multi-dimensional data conglomerate. The optimization objective can then be expressed as minimizing the loss function L: ； in, For lithological authenticity labeling, This is the joint feature output of an automatic sandstone and conglomerate lithology identification network that combines deep learning-based two-dimensional imaging logging data with one-dimensional conventional logging data. min represents minimization, and Loss is the loss calculation function; Step 42: Set key parameters such as the number of iterations and the learning rate, and perform iterative optimization calculations through backpropagation to obtain the optimal network weights; Step 43: Test the test set using the trained sandstone and conglomerate lithology automatic identification network model to evaluate the identification effect. When the expected effect is met, output the sandstone and conglomerate lithology automatic identification network model. Otherwise, readjust the key parameters such as the number of iterations and the learning rate, train repeatedly, and then output the sandstone and conglomerate lithology automatic identification network model.

9. The method for automatic identification of sandstone and conglomerate lithology by combining two-dimensional imaging logging data and one-dimensional conventional logging data according to claim 1, characterized in that, In step 5, the trained automatic identification network model for sandstone and conglomerate lithology is used to predict the two-dimensional imaging logging data and one-dimensional conventional logging data at the actual well points to obtain the automatic identification results of sandstone and conglomerate lithology, which helps interpreters identify the depth range with favorable lithological development.

10. The method for automatic identification of sandstone and conglomerate lithology by combining two-dimensional imaging logging data and one-dimensional conventional logging data according to claim 9, characterized in that, Step 5 includes: Step 51: Using the trained automatic identification network model for sandstone and conglomerate lithology, segmented predictions are made on the two-dimensional imaging logging data and one-dimensional conventional logging data of the well to be predicted. Step 52: Select the maximum probability value of the network output as the lithology output result corresponding to the depth sampling point to obtain the final sandstone and conglomerate lithology identification result.

Images