Total organic carbon and free hydrocarbon content prediction method, medium, equipment and product
By constructing a cross-scale evolutionary regression network model, combining multi-scale convolutional modules and fully connected layers, dynamic pseudo-label screening, and geological prior constraints, the reliance on high-quality labeled data in shale oil exploration is resolved. This enables efficient and stable prediction of total organic carbon and free hydrocarbon content, adapts to strongly heterogeneous strata, and improves the accuracy and efficiency of shale oil resource evaluation.
Patent Information
- Application Number
- CN202511266415.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-05
- Publication Date
- 2026-02-03
AI Technical Summary
Existing technologies rely heavily on large amounts of high-quality labeled data and have low data utilization efficiency in shale oil exploration, resulting in insufficient accuracy and generalization ability in predicting shale oil content parameters, especially in highly heterogeneous geological conditions where efficient prediction is difficult to achieve.
A cross-scale evolutionary regression network model is constructed, which combines multi-scale convolutional modules and fully connected layers. Through dynamic pseudo-label selection and geological prior constraints, the scale of the convolutional kernel is adaptively adjusted. The model is optimized using a genetic algorithm, and pseudo-labels are generated and expanded by combining prediction confidence and geological priors to achieve semi-supervised learning.
It improves prediction accuracy and stability under small sample conditions, reduces reliance on high-quality labeled data, enhances data utilization efficiency, adapts to strongly heterogeneous strata, and improves the reliability and efficiency of shale oil resource evaluation.
Smart Images
Figure CN121459983A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of geological exploration, and in particular to a total organic carbon and free hydrocarbon content prediction method, medium, equipment and product. BACKGROUND
[0002] Total organic carbon (TOC) and free hydrocarbon (S1) are important parameters for shale oil resource evaluation and prediction, and their accurate prediction becomes more and more important under strong heterogeneous geological conditions. Based on logging data, a network model for predicting TOC and S1 content is constructed, which provides an economical, efficient and generalizable technical path for shale oil resource evaluation.
[0003] Traditional methods often establish the relationship between logging data and predicted parameters based on mathematical models. However, shale oil reservoirs generally have characteristics such as ultra-low porosity and permeability, strong heterogeneity, and complex micro-pore structure. The traditional method relying on limited measured data and statistical models has obvious limitations in identifying "sweet spot" and reservoir evaluation.
[0004] The current earth science field is transforming towards data-driven and intelligentization, and various artificial intelligence methods have shown significant potential in unconventional oil and gas reservoir geological parameter prediction. Supervised learning methods such as recurrent neural networks (RNN) and convolutional neural networks (CNN) have shown great potential in shale reservoir identification and parameter prediction due to their excellent feature extraction capabilities. Semi-supervised learning methods can train by integrating a small amount of labeled samples and a large amount of unlabeled data, thereby significantly reducing the labeling cost while improving the generalization ability of the model. However, the dependence of supervised learning methods on a large amount of high-quality labeled data still restricts their practical application. The acquisition of labeled data usually relies on experimental testing and geological interpretation, which is costly, time-consuming, and difficult to support large-scale modeling needs.
[0005] Under the conditions of limited training data, weak correlation between input features and prediction targets, and strong heterogeneity of reservoirs, existing semi-supervised learning methods still have great room for improvement in prediction accuracy and data utilization in shale oil content prediction. SUMMARY
[0006] The purpose of the present application is to solve the problem of dependence on a large amount of high-quality labeled data and low data utilization efficiency in existing shale oil exploration technology. A total organic carbon and free hydrocarbon content prediction method is proposed, which includes the following steps: S1, obtain logging curve data and perform preprocessing, and use the preprocessed logging curve data and corresponding measured total organic carbon and free hydrocarbon content labels as training set and test set, and use the unannotated preprocessed logging curve data as unannotated data set; S2, a cross-scale evolutionary regression network model is constructed, the model comprising a multi-scale convolution module and a fully connected layer; the multi-scale convolution module comprises a plurality of parallel convolution branches, each convolution branch comprising a plurality of consecutive convolution layers and a maximum pooling layer; S3, the model is trained using the training set, the model is optimized in combination with a loss function and a geological prior constraint; the unannotated data set is input into the trained model, dynamic pseudo-label screening is performed in combination with a prediction confidence distribution and a geological prior constraint, and the pseudo-label samples screened are expanded to the training set; S4, based on the expanded training set, the model is self-trained, the self-trained model is tested using the test set, and steps S3 and S4 are repeated until the finally trained model meets the requirements; S5, the logging curve data to be predicted is input into the model meeting the requirements, and total organic carbon and free hydrocarbon content prediction values are obtained.
[0007] Further, the convolution kernel scale and the number of channels of the multi-scale convolution layer are dynamically adjusted by a genetic algorithm update mechanism.
[0008] Further, in the model training process of step S3, the loss function is:
[0009]
[0010] wherein L is the total loss, is the mean square error loss, is a regularization term based on a geological prior constraint, is a weight factor.
[0011] Further, the geological prior constraint comprises the following three kinds: (1) an empirical upper limit and lower limit constraint based on total organic carbon and free hydrocarbon content; For pseudo-label samples , if it exceeds the empirical range, it is truncated:
[0012]
[0013] wherein and are the total organic carbon and free hydrocarbon content values predicted by the pseudo-label, and are the empirical lower limit and upper limit of the total organic carbon content value, respectively, and are the empirical lower limit and upper limit of the free hydrocarbon content, respectively; (2) a correlation constraint based on logging curves; When generating pseudo-labels, the correlation coefficient between the pseudo-labels and the key logging curves must meet the following requirements:
[0014] in, The Pearson correlation coefficient is used. These are the predicted values for pseudo-labels. For the first well logging curves, The threshold value set for the lower limit of correlation; (3) Smoothness constraints based on the variation law of stratum thickness; For adjacent depth points , The pseudo-label prediction is required to satisfy the smoothness constraint:
[0015] in, For depth point The predicted value, For depth point The predicted value, This indicates the maximum permissible rate of change of formation properties in the thickness direction.
[0016] Furthermore, dynamic pseudo-label screening is performed by combining the predicted confidence distribution and geological prior constraints, as detailed below: (1) Calculate the prediction results and confidence levels for unlabeled samples; For unlabeled samples The model outputs predicted values. Its confidence level is defined as:
[0017] in, for The confidence level value, This represents the i-th unlabeled sample. Indicates the prediction variance. The model input is The predicted value at that time; (2) Determine the dynamic threshold by combining the confidence distribution and geological prior constraints; In the In the next iteration, the dynamic threshold The calculation formula is:
[0018] in, For the current iteration round, For the confidence set quantiles, As a geological a priori constraint correction factor, For dynamic thresholds; (3) Gradually increase the proportion of pseudo-labeled samples with each iteration; Pseudo-label ratio Increment by the following formula:
[0019] in, For the first The proportion of pseudo-labels selected during each iteration. This represents the initial ratio of pseudo-tags. The maximum proportion of pseudo-tags, For growth rate control parameters, For iteration rounds; (4) Add confidence samples that meet the screening criteria as pseudo-labels to the training set; like > Then pseudo-tags are generated:
[0020] And expand the training set:
[0021] in, For the first Pseudo-labels generated from unlabeled samples For the first The training set for each round of iterations, For the expanded training set, only when the confidence level is greater than the dynamic threshold... Only then will the predicted value be adopted as a pseudo-label.
[0022] Furthermore, the following formula is used to determine whether the final trained model meets the requirements:
[0023] in, This represents the overall score. , , As a weighting factor, satisfying + + =1, The coefficient of determination for the training set. The coefficient of determination for the test set. It serves as a predictive consistency metric for pseudo-labeled samples across multiple iterations.
[0024] The present invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting total organic carbon and free hydrocarbon content.
[0025] The present invention also proposes an electronic device, including a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program including computer-readable instructions, and the processor is configured to invoke the computer-readable instructions to execute the above-described method for predicting total organic carbon and free hydrocarbon content.
[0026] The present invention also proposes a computer program product, including a computer program / instruction that, when executed by a processor, implements the steps of the above-described method for predicting total organic carbon and free hydrocarbon content.
[0027] The beneficial effects of the technical solution provided by this invention are: This invention uses a cross-scale evolutionary regression network model to process well logging data within a multi-scale convolutional feature extraction framework. It extracts local response features and overall structural features of well logging curves through multi-branch convolution and pooling mechanisms. The convolution kernel scale is adaptively adjusted to accommodate different well logging sampling rates and formation thickness characteristics. Combined with geological prior constraints, the prediction results are ensured to be reasonable in terms of geological regularity. Based on a semi-supervised training strategy using dynamic pseudo-label learning, unlabeled well logging data is incorporated into the model iteration optimization process. Dynamic screening is performed by combining prediction confidence and geological prior constraints, generating high-quality pseudo-label samples round by round to expand the training set. In self-training, the model parameters and pseudo-label thresholds are updated collaboratively. This avoids the overfitting and prediction instability problems that traditional methods easily encounter under small sample conditions, reduces reliance on large amounts of high-quality labeled data, and achieves high data utilization efficiency. Attached Figure Description
[0028] Figure 1 This is a flowchart of a method for predicting the total organic carbon and free hydrocarbon content according to an embodiment of the present invention; Figure 2 This is a structural diagram of the cross-scale evolutionary regression network model constructed in this invention; Figure 3 This is a comparison chart of the predicted and actual total organic carbon content values of the model (SCARNet) of this invention and RNN, SSRNN, and CNN models; Figure 4 This is a comparison chart of the predicted and actual values of free hydrocarbon content from the model (SCARNet) of this invention and RNN, SSRNN, and CNN models. Figure 5 This is a block diagram of an electronic device according to an exemplary embodiment of the present invention. Detailed Implementation
[0029] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described below with reference to the accompanying drawings.
[0030] The flowchart of the method for predicting the total organic carbon and free hydrocarbon content in this invention is as follows: Figure 1 Specifically, it includes the following steps: S1. Acquire well logging curve data and perform preprocessing, including outlier removal, vertical depth correction of deviated wells, and well logging data standardization. Use the preprocessed well logging curve data and the corresponding labels of measured total organic carbon and free hydrocarbon content as training and testing sets, respectively. Use the unlabeled preprocessed well logging curve data as the unlabeled dataset.
[0031] S2. Construct a cross-scale evolutionary regression network model, which includes a multi-scale convolutional module and a fully connected layer; the multi-scale convolutional module includes multiple parallel convolutional branches, each of which includes multiple consecutive convolutional layers and a max pooling layer.
[0032] Specifically, the model input is multidimensional logging data, and the multi-scale convolutional layer includes multiple parallel convolutional branches with a kernel scale range of [2,4] to capture geological feature patterns at different scales. The parameter update of this invention adaptively adjusts the kernel scale to adapt to different logging sampling rates and formation thickness characteristics. This invention dynamically adjusts the kernel scale and channel number of the multi-scale convolutional layer using a genetic algorithm update mechanism to achieve cross-scale feature evolution. Pooling layers are placed in different convolutional branches to reduce dimensionality and enhance translation invariance. The fully connected layer contains a hidden layer and an output layer, outputting the predicted TOC or S1 content, and Dropout layers are placed between the fully connected layers to mitigate overfitting risk.
[0033] S3. Reference diagram of the cross-scale evolutionary regression network model constructed in this invention. Figure 2 The model is trained using the training set, and then optimized by combining a loss function and a geological prior constraint term. The loss function is:
[0034]
[0035] in, For mean square error loss, For regularization terms based on geological prior constraints, Weighting factors are used to control The magnitude of its contribution to the overall loss function.
[0036] In a preferred embodiment of the present invention, the geological prior constraints include the following three types: (1) Based on the empirical upper and lower limits of total organic carbon and free hydrocarbon content; For pseudo-label samples If it exceeds the range of experience, then truncate:
[0037]
[0038] in, and The values for total organic carbon and free hydrocarbons predicted by the pseudo-label. and These represent the empirical lower and upper limits for total organic carbon content, respectively. and These represent the empirical lower and upper limits for free hydrocarbon content, respectively. (2) Based on the correlation constraints between well logging curves; When generating pseudo-labels, the correlation coefficient between the pseudo-labels and the key logging curves must meet the following requirements:
[0039] in, The Pearson correlation coefficient is used. These are the predicted values for pseudo-labels. For the first well logging curves, The threshold value set for the lower limit of correlation; (3) Smoothness constraints based on the variation law of stratum thickness; For adjacent depth points , The pseudo-label prediction is required to satisfy the smoothness constraint:
[0040] in, For depth point The predicted value, For depth point The predicted value, This indicates the maximum permissible rate of change of formation properties in the thickness direction.
[0041] The unlabeled dataset is input into the trained model, and dynamic pseudo-label selection is performed by combining the prediction confidence distribution and geological prior constraints. The selected pseudo-label samples are then expanded into the training set.
[0042] The specific steps for dynamic pseudo-tag filtering are as follows: (1) Calculate the prediction results and confidence levels for unlabeled samples; For unlabeled samples The model outputs predicted values. Its confidence level is defined as:
[0043] in, for The confidence level value, This represents the i-th unlabeled sample. Indicates the prediction variance. The model input is The predicted value at that time; (2) Determine the dynamic threshold by combining the confidence distribution and geological prior constraints; In the In the next iteration, the dynamic threshold The calculation formula is:
[0044] in, For the current iteration round, For the confidence set quantiles, As a geological a priori constraint correction factor, This is a dynamic threshold used to determine which pseudo-labeled samples are included in the training set.
[0045] (3) Gradually increase the proportion of pseudo-labeled samples with each iteration; Pseudo-label ratio Increment by the following formula:
[0046] in, For the first The proportion of pseudo-labels selected during each iteration. This represents the initial ratio of pseudo-tags. The maximum proportion of pseudo-tags, For growth rate control parameters, For iteration rounds; (4) Add confidence samples that meet the screening criteria as pseudo-labels to the training set; like > Then pseudo-tags are generated:
[0047] And expand the training set:
[0048] in, For the first Pseudo-labels generated from unlabeled samples For the first The training set for each round of iterations, For the expanded training set, only when the confidence level is greater than the dynamic threshold... Only then will the predicted value be adopted as a pseudo-label.
[0049] S4. Self-train the model based on the expanded training set, test the self-trained model using the test set, and repeat steps S3 and S4 until the final trained model meets the requirements.
[0050] The final optimal model is the one with the highest overall score across training set accuracy, test set accuracy, and pseudo-label sample consistency metrics. The overall score is calculated using the following evaluation function:
[0051] in, The coefficient of determination for the training set. The coefficient of determination for the test set. This is a consistency metric for predictions of pseudo-labeled samples across multiple iterations. , , As a weighting factor, satisfying + + =1, preferably =0.2, =0.5, =0.3.
[0052] S5. Input the well logging curve data to be predicted into the model that meets the requirements to obtain the predicted values of total organic carbon and free hydrocarbon content.
[0053] To verify the effectiveness of the method of this invention, the predicted values of total organic carbon (TOC) and free hydrocarbon content of the cross-scale evolutionary regression network model (SCARNet) integrating dynamic pseudo-labels and geological prior constraints, as well as the predicted values of RNN (recurrent neural network), SSRNN (semi-supervised recurrent neural network), and CNN (convolutional neural network) models, are compared with the actual values. A comparison chart of the predicted and actual TOC values of the model (SCARNet) of this invention with those of the RNN, SSRNN, and CNN models is provided for reference. Figure 3 The comparison chart shows the predicted and actual values of free hydrocarbon content from the model (SCARNet) of this invention, along with those from RNN, SSRNN, and CNN models. Figure 4 In the figure, GD12 and GD14 represent two wells with logging and measured data, respectively, used as comparison objects for model training and validation.
[0054] Both the TOC and S1 validation sets contain 25 samples. The results show that, under the same 25-sample condition, the proposed cross-scale evolutionary regression network model, which integrates dynamic pseudo-labels and geological prior constraints, exhibits higher fitting accuracy and stability in both TOC and S1 predictions. Compared to other models, it better captures the nonlinear relationship between well logging data and oil content parameters, demonstrating strong small-sample learning ability and adaptability to highly heterogeneous formations.
[0055] The above results demonstrate that the cross-scale evolutionary regression network model method integrating dynamic pseudo-labels and geological prior constraints provided in this invention can achieve high prediction accuracy and stability in TOC and S1 predictions under small sample conditions. Under the same validation set conditions, this method shows significant advantages over traditional RNN, SSRNN, and CNN models in terms of fitting accuracy, generalization ability, and adaptability to strongly heterogeneous formations. This also confirms that the method of this application can make full use of limited labeled data and a large amount of unlabeled data, improving the reliability and efficiency of well logging data-driven shale oil resource evaluation.
[0056] In one exemplary embodiment, a computer-readable storage medium is included, which stores a computer program that, when executed by a processor, implements the above-described method for predicting total organic carbon and free hydrocarbon content.
[0057] Please see Figure 5 In one exemplary embodiment, the device further includes an electronic device including at least one processor, at least one memory, and at least one communication bus.
[0058] The memory stores a computer program, which includes computer-readable instructions. The processor calls the computer-readable instructions stored in the memory through the communication bus to execute the above-mentioned method for predicting the total organic carbon and free hydrocarbon content.
[0059] In one exemplary embodiment, a computer program product is proposed, comprising a computer program / instructions that, when executed by a processor, implement the steps of the above-described method for predicting total organic carbon and free hydrocarbon content.
[0060] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for predicting total organic carbon and free hydrocarbon content, characterized in that, Includes the following steps: S1. Acquire well logging curve data and preprocess it. Use the preprocessed well logging curve data and the corresponding labels of measured total organic carbon and free hydrocarbon content as training set and test set, and use the unlabeled preprocessed well logging curve data as unlabeled dataset. S2. Construct a cross-scale evolutionary regression network model, which includes a multi-scale convolutional module and a fully connected layer; the multi-scale convolutional module includes multiple parallel convolutional branches, and each convolutional branch includes multiple consecutive convolutional layers and a max pooling layer; S3. Train the model using the training set, and optimize the model by combining the loss function and geological prior constraints; input the unlabeled dataset into the trained model, and perform dynamic pseudo-label screening by combining the prediction confidence distribution and geological prior constraints, and expand the screened pseudo-label samples into the training set. S4. Self-train the model based on the expanded training set, test the self-trained model using the test set, and repeat steps S3 and S4 until the final trained model meets the requirements. S5. Input the well logging curve data to be predicted into the model that meets the requirements to obtain the predicted values of total organic carbon and free hydrocarbon content.
2. The method for predicting total organic carbon and free hydrocarbon content according to claim 1, characterized in that, The kernel size and number of channels of the multi-scale convolutional layer are dynamically adjusted through a genetic algorithm update mechanism.
3. The method for predicting total organic carbon and free hydrocarbon content according to claim 1, characterized in that, In step S3, during model training, the loss function is: Where L is the total loss, For mean square error loss, For regularization terms based on geological prior constraints, This is the weighting factor.
4. The method for predicting total organic carbon and free hydrocarbon content according to claim 1, characterized in that, Geological a priori constraints include the following three types: (1) Based on the empirical upper and lower limits of total organic carbon and free hydrocarbon content; For pseudo-label samples If it exceeds the range of experience, then truncate: in, and The values for total organic carbon and free hydrocarbons predicted by the pseudo-label. and These are the empirical lower and upper limits for total organic carbon content, respectively. and These are the empirical lower and upper limits for free hydrocarbon content, respectively; (2) Based on the correlation constraints between well logging curves; When generating pseudo-labels, the correlation coefficient between the pseudo-labels and the key logging curves must meet the following requirements: in, The Pearson correlation coefficient is used. These are the predicted values for pseudo-labels. For the first well logging curves, The threshold value set for the lower limit of correlation; (3) Smoothness constraints based on the variation law of stratum thickness; For adjacent depth points , The pseudo-label prediction is required to satisfy the smoothness constraint: in, For depth point The predicted value, For depth point The predicted value, This indicates the maximum permissible rate of change of formation properties in the thickness direction.
5. The method for predicting total organic carbon and free hydrocarbon content according to claim 1, characterized in that, Dynamic pseudo-label selection is performed by combining the predicted confidence distribution and geological prior constraints, as detailed below: (1) Calculate the prediction results and confidence levels for unlabeled samples; For unlabeled samples The model outputs predicted values. Its confidence level is defined as: in, for The confidence level value, This represents the i-th unlabeled sample. Indicates the prediction variance. The model input is The predicted value at that time; (2) Determine the dynamic threshold by combining the confidence distribution and geological prior constraints; In the In the next iteration, the dynamic threshold The calculation formula is: in, For the current iteration round, For the confidence set quantiles, As a geological a priori constraint correction factor, For dynamic thresholds; (3) Gradually increase the proportion of pseudo-labeled samples with each iteration; Pseudo-label ratio Increment by the following formula: in, For the first The proportion of pseudo-labels selected during each iteration. This represents the initial ratio of pseudo-tags. The maximum proportion of pseudo-tags, For growth rate control parameters, For iteration rounds; (4) Add confidence samples that meet the screening criteria as pseudo-labels to the training set; like > Then pseudo-tags are generated: And expand the training set: in, For the first Pseudo-labels generated from unlabeled samples For the first The training set for each round of iterations, For the expanded training set, only when the confidence level is greater than the dynamic threshold... Only then will the predicted value be adopted as a pseudo-label.
6. The method for predicting total organic carbon and free hydrocarbon content according to claim 1, characterized in that, Determine whether the final trained model meets the requirements using the following formula: in, This represents the overall score. , , As a weighting factor, satisfying + + =1, The coefficient of determination for the training set. The coefficient of determination for the test set. It serves as a predictive consistency metric for pseudo-labeled samples across multiple iterations.
7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.
8. An electronic device, characterized in that, The device includes a processor and a memory, the processor being interconnected with the memory, wherein the memory is used to store a computer program, the computer program including computer-readable instructions, and the processor is configured to invoke the computer-readable instructions to perform the method as described in any one of claims 1-6.
9. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1-6.