A prediction method and related device for the development of effective hydrocarbon source rocks

Abstract

The present application discloses a prediction method and related device for the development of effective source rocks, which relates to the field of geophysical exploration and comprehensive technology. The method uses a total organic carbon content prediction model to capture spatial feature processing sequence data, trains a strong and complex mapping relationship between total organic carbon content and seismic attributes, and uses a support vector machine classification model to analyze the mapping relationship between seismic attributes and lithology data of virtual wells in the study area, respectively completing the intelligent prediction of total organic carbon content and intelligent identification of lithology, improving the generalization ability of the model, and being able to more effectively predict the regional distribution of total organic carbon content and lithology, thereby maintaining a high accuracy rate in a complex geological environment. Finally, the thickness and position of the effective source rock in the target layer are determined based on the lithology identification results and the total organic carbon content prediction results, realizing an accurate assessment of the development of effective source rocks.

Description

Technical Field

[0001] This application relates to the fields of geophysical exploration and integrated technology, and in particular to a method and related apparatus for predicting the development of effective source rocks. Background Technology

[0002] The presence of effective source rocks is a crucial prerequisite for hydrocarbon accumulation and formation, and a fundamental basis for evaluating the oil and gas resource potential of an exploration area. Effective source rocks play a vital role in supplying hydrocarbons, and their size and distribution range control the size and distribution of oil and gas fields. Therefore, it is necessary to evaluate the source rocks in the exploration area at the initial stage of exploration to provide theoretical and methodological support for subsequent oil and gas exploration work.

[0003] Existing technologies utilize seismic data to predict the planar distribution of source rocks using traditional machine learning models or simple neural network models combined with the characteristics of source rock development. This can solve the problem of predicting the planar distribution of source rocks in areas with limited well logging and analytical data. Although this method can predict the vertical development of source rocks in a single well to a certain extent, it still faces some challenges in practical applications, such as limited generalization ability and poor noise resistance in complex geological environments, which affects the accuracy of predicting the development of source rocks. Summary of the Invention

[0004] The purpose of this application is to provide an effective method and related apparatus for predicting the development of source rocks, which can improve the accuracy of predicting the development of source rocks.

[0005] To achieve the above objectives, this application provides the following solution:

[0006] Firstly, this application provides an effective method for predicting the development of source rocks, including:

[0007] Acquire post-stack seismic data for the study area;

[0008] Seismic attributes of the virtual wells inserted in the study area are extracted based on the post-stack seismic data.

[0009] The seismic attributes of the virtual well are input into the trained lithology prediction model, and the lithology data of each segment of the virtual well are output. The lithology prediction model is a support vector machine classification model, and the lithology data includes lithology type and development thickness of different lithologies.

[0010] The seismic attributes of the virtual well are input into the trained total organic carbon content prediction model, and the total organic carbon content of each segment of the virtual well is output. The total organic carbon content prediction model is a CNN-GRU architecture model based on the combination of convolutional neural network and gated recurrent unit.

[0011] Based on the lithological data and total organic carbon content of each layer of the virtual well, the thickness and location of the effective source rock are determined, wherein the effective source rock is a mudstone layer with a total organic carbon content greater than a threshold.

[0012] The development of effective source rocks is assessed based on their thickness and location.

[0013] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for predicting the effective source rock development described in the first aspect above.

[0014] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for predicting the effective source rock development described in the first aspect above.

[0015] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the method for predicting the effective development of source rocks as described in the first aspect above.

[0016] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0017] This application provides a method and related apparatus for predicting the development of effective source rocks. The method involves inserting virtual wells into the study area and inputting the seismic attributes of these virtual wells into a trained lithology prediction model and a trained total organic carbon (TOC) prediction model. This yields the lithology type and TOC content of each segment of the virtual well, allowing for the identification of effective source rocks. Based on the thickness and location of these effective source rocks, the development of the source rocks is assessed. Accurate lithology identification helps determine the presence and quality of source rocks, while accurate prediction of TOC content directly relates to the hydrocarbon generation capacity of source rocks. Therefore, accurate prediction of lithology and TOC content is a key factor in evaluating the quality and hydrocarbon generation potential of source rocks, thus improving the accuracy of source rock development prediction. Furthermore, this application employs a Support Vector Machine (SVC) classification model and a CNN-GRU architecture model for intelligent lithology identification and TOC prediction, respectively. SVC, with its superior classification performance, demonstrates significant advantages in intelligent lithological identification. By maximizing edge strategies, it enhances the model's generalization ability, thus maintaining high accuracy in complex geological environments. On the other hand, the CNN-GRU model combines the advantages of convolutional neural networks (CNN) and gated recurrent units (GRU), enabling it not only to capture spatial features but also to process sequential data. Training a CNN-GRU model with strong complex mapping capabilities allows for the establishment of nonlinear relationships between TOC data and seismic attributes, more effectively predicting the regional distribution of TOC content. Through multi-attribute prediction using neural networks, this application ultimately applies the trained mathematical model to a three-dimensional seismic data volume, achieving a three-dimensional quantitative evaluation of source rocks. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating an effective method for predicting the development of source rocks, as provided in Embodiment 1 of this application;

[0020] Figure 2 This is a flowchart illustrating the training methods for the SVC model and the CNN-GRU architecture model in Embodiment 1 of this application.

[0021] Figure 3 This is a schematic diagram of the deployment of a virtual well site in the first specific area of ​​Embodiment 1 of this application;

[0022] Figure 4 This is a schematic diagram of the deployment of two virtual well locations in the second specific area in Embodiment 1 of this application;

[0023] Figure 5 This is a thermogram of lithological correlation in Example 1 of this application;

[0024] Figure 6 This is a heatmap of the TOC correlation in Embodiment 1 of this application;

[0025] Figure 7 This is a comparison chart of the loss functions of the lithology prediction models in Example 1 of this application;

[0026] Figure 8 This is a schematic diagram of a segment of a confusion matrix in the first specific region of Embodiment 1 of this application;

[0027] Figure 9 This is a schematic diagram of the two-segment confusion matrix of the second specific region in Embodiment 1 of this application;

[0028] Figure 10 This is a schematic diagram of the ROC curve of a section of mudstone in the first specific region of Embodiment 1 of this application;

[0029] Figure 11 This is a schematic diagram of the ROC curves of the two mudstone sections in the second specific region in Embodiment 1 of this application;

[0030] Figure 12 This is a comparison chart of the loss functions of the TOC prediction model in Embodiment 1 of this application;

[0031] Figure 13 This is a graph showing the loss function of the TOC prediction model in Embodiment 1 of this application;

[0032] Figure 14 This is a schematic diagram of the predicted planar development thickness of a segment of source rock in the first specific region of this application in Embodiment 1;

[0033] Figure 15 This is a schematic diagram of the predicted planar development thickness of the second specific region's second-section source rock in Embodiment 1 of this application. Detailed Implementation

[0034] Research has revealed several methods for assessing the development of effective source rocks, including: seismic prediction methods combining BP and SOM neural networks, which preprocess raw seismic data through clustering using a self-organizing feature map neural network, followed by BP neural network learning and prediction of sample data; or, using TOC curves as constraints and employing seismic multi-attribute inversion techniques, involving multivariate stepwise regression and artificial neural network training to select seismic attributes strongly correlated with TOC, fitting the relationship between seismic attributes and TOC, and obtaining TOC data volumes for assessing the development of effective source rocks; or, establishing a nonlinear relationship between TOC content and seismic elastic parameters by training a ConvLSTM neural network with strong complex mapping capabilities to effectively predict the regional distribution of TOC content; or, using a data-driven machine learning framework to quantitatively characterize source rocks using geological constraints, drilling logging, well logging, geochemical, and pre-stack seismic data. Random forest algorithms perform well on small sample data and can effectively predict the distribution of mudstone and total organic carbon content. However, they face challenges in practical applications, such as limited generalization ability to existing methods, poor noise resistance in complex geological environments, and the inability to effectively establish nonlinear relationships between sequential data such as TOC data and seismic attributes, which in turn affects the accurate assessment of the development of effective source rocks.

[0035] In response, this application provides an effective method and related apparatus for predicting the development of source rocks, in order to overcome the above-mentioned technical deficiencies.

[0036] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0037] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0038] Example 1

[0039] The method for predicting the development of effective source rocks provided in this application can be applied to an environment including terminals and servers. The terminals communicate with the server via a network. A data storage system can store the data that the server needs to process. The data storage system can be set up independently, integrated into the server, or located in the cloud or on other servers. The terminal can send post-stack seismic data of the study area to be processed to the server. After receiving the post-stack seismic data of the study area, the server extracts the seismic attributes of the virtual wells inserted in the study area based on the post-stack seismic data. The seismic attributes of the virtual wells are input into a trained lithology prediction model, which outputs the lithology data of each segment of the virtual well. The lithology prediction model is a support vector machine classification model. The seismic attributes of the virtual wells are input into a trained total organic carbon (TOC) prediction model, which outputs the TOC of each segment of the virtual well. The TOC prediction model is a CNN-GRU architecture model based on a combination of convolutional neural networks and gated recurrent units. Based on the lithology data and TOC of each segment of the virtual well, the thickness and location of the effective source rocks are determined. The effective source rocks are mudstone layers with a TOC greater than a threshold. Based on the thickness and location of the effective source rocks, the development of the effective source rocks is evaluated. The server can feed back the evaluation results of the development of the effective source rocks to the terminal. In addition, in some embodiments, the method for predicting the development of effective source rocks can also be implemented by the server or the terminal alone. For example, the terminal can directly process the post-stack seismic data of the study area to be processed using the method for predicting the development of effective source rocks, or the server can obtain the post-stack seismic data of the study area to be processed from the data storage system and process it using the method for predicting the development of effective source rocks.

[0040] The terminals can be, but are not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle systems. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. Servers can be implemented using independent servers, server clusters composed of multiple servers, or cloud servers.

[0041] In one exemplary embodiment, such as Figure 1 The method provided is an effective method for predicting the development of source rocks. This method is executed by a computer device, specifically a terminal or server, or both. In this embodiment, the method is described using a server as an example, and includes the following steps 201 to 206. Wherein:

[0042] Step 201: Obtain post-stack seismic data for the study area.

[0043] Step 202: Extract the seismic attributes of the virtual wells inserted in the study area based on the post-stack seismic data.

[0044] Based on the obtained post-stack seismic data, this embodiment addresses the issue that the number of actual wells in the study area is small and cannot meet the prediction accuracy required for planar analysis. Virtual wells are inserted in the study block and a fixed interval is set to ensure their uniform distribution.

[0045] Step 203: Input the seismic attributes of the virtual well into the trained lithology prediction model and output the lithology data of each layer of the virtual well. The lithology prediction model is a support vector machine classification model, and the lithology data includes lithology type and development thickness of different lithologies.

[0046] Step 204: Input the seismic attributes of the virtual well into the trained total organic carbon content prediction model, and output the total organic carbon content of each segment of the virtual well. The total organic carbon content prediction model is a CNN-GRU architecture model based on the combination of convolutional neural network and gated recurrent unit.

[0047] Step 205: Based on the lithological data and total organic carbon content of each layer of the virtual well, determine the thickness and location of the effective source rock, wherein the effective source rock is a mudstone layer with a total organic carbon content greater than a threshold.

[0048] By setting a TOC threshold based on existing research experience, mudstone strata with TOC values ​​greater than the threshold are identified.

[0049] Step 206: Assess the development of the effective source rock based on its thickness and location.

[0050] The effective source rocks were statistically analyzed and accumulated to obtain the effective source rock thickness of each layer in each virtual well. The effective source rock thickness values ​​and coordinates at each virtual well point were imported into the "DF-GVision" software to draw a plane contour map of the effective source rock thickness within the study area, thereby evaluating the development of effective source rocks.

[0051] As an optional implementation method, such as Figure 2 As shown, the specific training process for the lithology prediction model and the total organic carbon content prediction model includes:

[0052] (1) Construct a sample training set, wherein the sample training set includes a lithological sample dataset and a total organic carbon content sample dataset of real wells in the research area, the lithological sample dataset includes lithologically sensitive seismic attributes and sample lithological data, and the total organic carbon content sample dataset includes total organic carbon content sensitive seismic attributes and sample total organic carbon content interpretation data.

[0053] The process of constructing the sample training set specifically includes:

[0054] (1-1) Obtain sample geological exploration data of the study area, wherein the sample geological exploration data includes sample well location data, sample well trajectory data, sample back-stack seismic data, sample well logging curve data, sample total organic carbon content interpretation data, sample stratigraphic interpretation data and sample core logging data.

[0055] The SMI interpretation software was used to import sample well location data, sample well trajectory data, sample post-stack seismic data, sample well logging curve data, sample TOC interpretation data, sample stratigraphic interpretation data, and sample core logging data. The sample TOC interpretation data was read to obtain the true well location coordinates with TOC interpretation data within the effective source rocks. Based on the sample core logging data, the lithological data of individual wells was determined, including lithological type and the development thickness of different lithologies. The sample post-stack seismic data was preprocessed, including noise and interference removal, and directional filtering based on structural dip angles.

[0056] The "SMI" software described in this embodiment, also known as seismic waveform indication inversion software, is a high-precision seismic phasic reservoir prediction software independently developed by Beijing Zhongheng Lihua Petroleum Technology Research Institute. The software features a Chinese interface and 3D visualization capabilities, and is suitable for high-precision reservoir prediction of thin interbedded layers, tight reservoirs, and unconventional reservoirs.

[0057] The post-stack seismic data described in this embodiment refers to the information about underground rock layers and strata obtained during seismic exploration by exciting vibrations on the ground and then recording the waves that propagate back to the surface. This type of recorded data is called three-dimensional seismic data. Post-stack refers to the three-dimensional seismic data obtained after superposition processing, which helps to enhance the effective signal, reduce noise, and improve the signal-to-noise ratio of the data.

[0058] (1-2) Determine the lithological data of the actual well based on the core logging data of the sample.

[0059] (1-3) Construct a velocity field based on the sample well location data and the sample logging curve data.

[0060] Based on sample well location data and sample logging curve data, and using velocity and density data measured at the same depth points in actual wells, the acoustic impedance of the formation is calculated. Then, the reflection coefficient sequence R(t) of the formation is calculated. Subsequently, the reflection coefficient sequence is convolved with the seismic wavelet to obtain the synthetic seismic record S(t), as shown in the formula:

[0061] (1);

[0062] In the formula, S(t) is the synthetic seismic record; R(t) is the reflection coefficient; and W(t) is the seismic wavelet.

[0063] A time-depth relationship is constructed using synthetic seismic records, thereby constructing a velocity field.

[0064] The seismic wavelet described in this embodiment refers to the waveform of a seismic wave during its propagation underground. It has finite energy, a definite start time, and a certain duration, and is the basic unit in seismic records.

[0065] The synthetic seismic record described in this embodiment refers to seismic record data obtained by combining actual geological models with the principles of seismic exploration and development.

[0066] The time-depth relationship described in this embodiment refers to the conversion of the propagation time of seismic waves in the underground medium into a corresponding depth or height relationship in the interpretation of seismic data.

[0067] The velocity field described in this embodiment is the seismic velocity field, which reflects the spatial variation and distribution of the sound wave propagation velocity in underground rocks. It can be used to convert time-domain data from earthquakes into depth-domain data.

[0068] (1-4) Based on the sample well trajectory data, the sample post-stack seismic data, the sample layer interpretation data, and the velocity field, obtain the seismic attributes along the well trajectory in the true well depth domain.

[0069] Based on sample post-stack seismic data, sample well trajectory data, and sample stratigraphic interpretation data, time-domain seismic volume attributes are extracted. The time-domain seismic volume attributes are then converted into depth-domain seismic volume attributes based on the velocity field. Subsequently, the required seismic attributes are extracted based on the depth-domain seismic volume attributes, and the depth-domain seismic attributes along the well trajectory of the actual well are extracted based on the stratigraphic interpretation constraints.

[0070] The seismic volume attributes mentioned in this embodiment refer to different three-dimensional attribute volumes calculated based on different algorithms on the basis of three-dimensional seismic data volumes.

[0071] (1-5) The correlation coefficients of the seismic attributes along the well trajectory in the actual well depth domain with the total organic carbon content and lithological attributes were calculated by using the correlation analysis method, and the first correlation coefficient and the second correlation coefficient were obtained accordingly.

[0072] Based on the seismic attributes along the actual well trajectory, preprocessing was performed, including outlier removal, normalization, and smoothing. Then, Pearson correlation coefficients were used to analyze the correlation between TOC and the lithology-sensitive seismic attributes along the actual well trajectory. The correlations were ranked by numerical value, and the top 80% of the seismic attributes along the actual well depth domain with the highest correlation to the label data, and the bottom 80% of the seismic attributes along the actual well depth domain with the lowest correlation among themselves, were selected. The formula is as follows:

[0073] (2);

[0074] In the formula, m is the sample size; r xy The degree of linear correlation between two variables x and y; The average value of variable x; Let y be the average value of the variable y.

[0075] (1-6) Calculate the correlation coefficient between the seismic attributes along the well trajectory in the actual well depth domain to obtain the third correlation coefficient.

[0076] (1-7) Select the seismic attribute along the well trajectory in the real well depth domain where the first correlation coefficient is greater than the first threshold and the third correlation coefficient is less than the third threshold as the lithology-sensitive seismic attribute.

[0077] (1-8) Select the seismic attribute along the well trajectory in the true well depth domain where the second correlation coefficient is greater than the second threshold and the third correlation coefficient is less than the third value, as the seismic attribute sensitive to the total organic carbon content.

[0078] (2) Use the lithology-sensitive seismic attributes as feature data and the sample lithology data as label data to train the lithology prediction model.

[0079] The seismic attributes along the well trajectory in the true well depth domain with the highest correlation value with the lithological data (top 80%) and the seismic attributes along the well trajectory in the true well depth domain with the lowest correlation value (bottom 80%) are selected as feature data. The lithological data of each single well (i.e., sample lithological data) is encoded using a one-hot encoding method. The encoded lithological column is used as label data and together with the feature data as a sample dataset. The dataset is divided into a training set and a validation set. The support vector machine classification (SVC) method is used to identify lithology and predict lithology.

[0080] The accuracy score is used as the evaluation metric to assess the training effect. Based on the binary confusion matrix, the ROC curve is generated using the data to evaluate the quality of the model's prediction. The area under the curve is calculated as the AUC. The closer the AUC is to 1, the better the model performance; the farther away from the 45-degree diagonal, the higher the accuracy.

[0081] The Support Vector Machine (SVM) described in this embodiment is a commonly used supervised learning algorithm, primarily used for classification and regression analysis. Its basic idea is to construct an optimal hyperplane in the feature space to separate sample instances of different categories. Lithological classification prediction typically involves multiple geological parameters and features, which can be high-dimensional and complex. The SVM model performs excellently when handling high-dimensional data and complex features, effectively performing feature mapping and classification.

[0082] (3) Use the total organic carbon content sensitive seismic attributes as feature data and the total organic carbon content interpretation data of the sample as label data to train the total organic carbon content prediction model.

[0083] Seismic attributes along the well trajectory in the true well depth domain with the highest correlation to TOC data (top 80%) and seismic attributes along the well trajectory in each true well depth domain with the lowest correlation to TOC data (bottom 80%) were selected as feature data. TOC interpretation data was used as label data, and together with the feature data, they formed the TOC prediction sample dataset. The dataset was divided into training and validation sets. A CNN-GRU architecture model, combining a convolutional neural network (CNN) and a gated recurrent unit (GRU), was used for TOC regression prediction. In training the neural network model, a loss function based on mean squared error (MSE) was used as the performance metric. By calculating the loss on the training set and the loss on the validation set, a diagnostic model was established, creating a complex mapping relationship between input and output to achieve TOC prediction for unknown samples in virtual wells.

[0084] The Convolutional Neural Network (CNN) described in this embodiment is a deep learning model specifically designed for processing and analyzing grid-like data. Its core idea is to extract local features from the input data through convolutional layers and reduce the spatial dimensionality of the data through pooling layers, thereby effectively capturing patterns and structures in images.

[0085] The Gated Recurrent Unit (GRU) described in this embodiment is a type of recurrent neural network proposed to address issues such as long-term memory and gradients in backpropagation. It controls the flow of information through learnable gates, aiming to better capture dependencies with large time step distances in time series data.

[0086] The mean squared error-based loss function described in this embodiment is an operational function used to measure the difference between the model's predicted value f(dx) and the true value Y. It is a non-negative real-valued function, usually represented by L(Y,f(x)). The smaller the loss function, the better the model's generalization ability. The loss function is mainly used in the model's training phase. After each batch of training data is fed into the model, it outputs predicted values ​​through forward propagation. Then, the loss function calculates the difference between the predicted value and the true value, which is the loss value. After obtaining the loss value, the model updates its parameters through backpropagation to reduce the loss between the true and predicted values, making the model's generated predicted values ​​move closer to the true values, thereby achieving the learning objective.

[0087] Mean square error represents the difference between the actual and predicted values ​​of reservoir parameters, and the formula is:

[0088] (3);

[0089] In the formula, This represents the number of original sample data. These are actual values; The model predicts numerical values.

[0090] This embodiment provides an effective intelligent prediction method for source rocks based on SVC and CNN-GRU. During the modeling process, lithological data and TOC data, using nominal and sequential data respectively, serve as key input parameters for model training and prediction. The relationship between lithological and TOC data and seismic attributes is often non-linear. Traditional neural network models require large amounts of data and complex network structures to effectively model this relationship. This embodiment employs Support Vector Machine (SVC) classification and CNN-GRU architecture models for intelligent lithological identification and TOC prediction, respectively. SVC, with its superior classification performance, demonstrates significant advantages in intelligent lithological identification. It utilizes one-hot encoding to convert nominal data into numerical data and improves the model's generalization ability by maximizing margin strategies, thus maintaining high accuracy in complex geological environments. On the other hand, the CNN-GRU model combines the advantages of convolutional neural networks and gated recurrent units, enabling it not only to capture spatial features but also to process sequential data. Training a CNN-GRU model with strong complex mapping capabilities establishes the non-linear relationship between TOC data and seismic parameters, allowing for more effective prediction of the regional distribution of TOC content. Through multi-attribute prediction using neural networks, this embodiment ultimately applies the trained mathematical model to the three-dimensional seismic data volume, achieving a three-dimensional quantitative evaluation of source rocks.

[0091] This embodiment takes a certain depression study area as an example to describe in detail the specific implementation process of the above-mentioned method for predicting the development of effective source rocks.

[0092] a. Import well location data, well trajectory data, conventional logging curves, TOC interpretation data, post-stack seismic data, stratigraphic interpretation data, and core logging data from a specific depression area using interpretation software. Read the conventional logging curves and core logging data to obtain the actual well data information for the first and second specific segments of the area to be predicted. Based on the core logging data and TOC interpretation data, statistically analyze the actual well coordinates with lithological and TOC data within the two segments. The lithological information includes lithological type and thickness of different lithologies. Preprocess the post-stack seismic data, including noise and interference removal, and directional filtering based on structural dip angles.

[0093] b. Based on the post-stack seismic data and well location data obtained in step a, and considering the limited number of actual wells in the study block, which cannot meet the prediction accuracy requirements for planar analysis, virtual wells were inserted within the study block. According to the seismic horizon interpretation results, 1271 virtual well locations were deployed in the first specific area segment. The deployment results of the virtual well locations in the first specific area segment are as follows: Figure 3 As shown, 715 virtual well locations were deployed in the second specific area, section two. The deployment results of the virtual well locations in the second specific area, section two are as follows. Figure 4 As shown, the virtual wells are uniformly distributed with a spacing of approximately 2700 meters between them.

[0094] c. Based on the well location data and conventional logging curves obtained in step a, the acoustic impedance is calculated using the velocity and density data measured at the same depth points in 31 drilled wells (real wells). Then, the reflection coefficient R(t) sequence is calculated. Subsequently, the convolution of the formation reflection coefficient sequence with the seismic wavelet is calculated to obtain the synthetic seismic record S(t). The time-depth relationship is constructed using the synthetic seismic record, thereby constructing the velocity field.

[0095] d. Based on the work area data obtained in step a, extract the time-domain seismic volume attributes; based on the velocity field established in step c, convert the time-domain seismic volume attributes into depth-domain seismic volume attributes, and then extract nine attributes on the basis of the depth-domain seismic volume attributes: instantaneous amplitude, instantaneous frequency, sweet spot, gradient mode, variance, root mean square amplitude, instantaneous phase, trace integral, and amplitude squared difference. Based on the layer interpretation constraints, extract the depth-domain seismic attributes along the well trajectory of the actual well.

[0096] e. Based on the actual well-trajectory seismic attributes obtained in step d, preprocess them, including outlier removal, smoothing, and normalization. Then, use the Pearson correlation coefficient to perform correlation analysis on TOC and lithology-sensitive seismic attributes, and select characteristic attributes that have high correlation with the label data and low correlation among the attributes.

[0097] Seven attributes most sensitive to lithology were selected: root mean square amplitude, instantaneous amplitude, instantaneous frequency, sweet spot, gradient mode, variance, and trace integral. After calculation, a correlation heatmap between seismic attributes and lithology was obtained, as shown below. Figure 5 As shown, the trace integral attribute is most correlated with the TOC data in this study, with a correlation coefficient of 0.1616. The instantaneous phase attribute has the weakest correlation, at only 0.02765, placing it at the very bottom of the correlation values ​​between all attributes and the TOC data. Furthermore, the trace integral attribute has a high correlation with the instantaneous phase attribute, reaching 0.616. Therefore, the feature selection process should prioritize attributes with high correlation to the label data and low correlation among all attributes. Thus, root mean square amplitude, trace integral, and amplitude squared difference were selected as lithology prediction variables.

[0098] Four attributes most sensitive to TOC were selected: root mean square amplitude, instantaneous phase, trace integral, and amplitude squared difference. After calculation, a heatmap showing the correlation between seismic attributes and TOC was obtained, as shown below. Figure 6 As shown, the trace integral attribute is most correlated with the TOC data in this study, with a correlation coefficient of 0.1616. The instantaneous phase attribute has the weakest correlation, at only 0.02765, placing it at the very bottom of the correlation values ​​between all attributes and the TOC data. Furthermore, the trace integral attribute has a high correlation with the instantaneous phase attribute, reaching 0.616. Therefore, the feature selection process should prioritize features with high correlation to the label data and low correlation among all attributes. Thus, root mean square amplitude, trace integral, and amplitude squared difference were selected as the TOC predictor variables.

[0099] f. Using the optimized and preprocessed lithology-sensitive well trajectory attributes obtained in step e as feature data, the lithology data read in step a is encoded using the One-hot encoding method (see Table 1). The encoded lithology column is used as label data and together with the feature data, it forms a sample dataset.

[0100] Table 1 Lithological Coding Data

[0101]

[0102] Model performance evaluation typically relies on the optimization of the loss function. The decreasing trend of the loss function is a direct indicator of model training convergence. Ideally, the loss value should decrease monotonically with each training epoch, reflecting the model's gradual adaptation to the data distribution and the gradual reduction of error. If the loss function decreases abnormally or oscillates, it may indicate an inappropriate learning rate selection or overfitting or underfitting of the model.

[0103] Support Vector Machine (SVM) models perform particularly well in lithology prediction tasks. By comparing and analyzing traditional neural network models such as Feedforward Neural Networks (FNN), Convolutional Neural Networks (CNN), Long Short-Term Memory Networks (LSTM), and Self-Organizing Maps (SOM), we can understand the performance differences by observing the changes in the loss functions of the models. We selected the Cross-Entropy Loss Function (CELF) as the loss function for this multi-class data analysis.

[0104] In the initial 50 iterations of training, the SVC model showed significantly lower loss on the test set compared to other models, such as... Figure 7 As shown, this reveals that SVC can more accurately capture data features when dealing with lithology prediction problems, thus having a significant advantage in prediction accuracy. This characteristic of the SVC model mainly benefits from its optimization of the classification boundary through the selection of support vectors while maximizing the boundary margin.

[0105] In contrast, while FNN models are simple in structure and easy to implement, their performance may be limited when handling complex lithological data due to their lack of deep understanding of data features. CNN models have achieved great success in image processing, but their application in processing sequence data may not be as good as models specifically designed for such data. LSTM models, by introducing gating mechanisms to process time series data, can effectively capture long-term dependencies, but in some cases, they may face the risk of overfitting due to excessive model complexity. SOM models, as an unsupervised learning method, have unique advantages in data visualization and cluster analysis, but in tasks requiring accurate classification, such as lithological prediction, their performance may be inferior to supervised learning SVC models.

[0106] Therefore, the optimized support vector machine classification model was selected to predict lithology. The accuracy score was used as the evaluation metric to assess the training effectiveness, based on a binary confusion matrix (e.g., ...). Figure 8 and Figure 9 As shown), use its data to create an ROC curve (such as...). Figure 10 and Figure 11 As shown in the figure, the horizontal axis represents the Free Rate of Return (FPR), which is the proportion of samples predicted as positive but actually negative out of all negative samples; the vertical axis represents the Total Rate of Return (TPR), which is the proportion of samples predicted as positive and actually positive out of all positive samples. The area under the ROC curve is calculated as the Average Value Under the Curve (AUC). The larger the AUC, the better the prediction performance. By combining the ROC curve and AUC data, the model's prediction performance can be improved by adjusting model parameters and adding features.

[0107] Depend on Figures 8-11It can be seen that the average TPR of each lithology in the first specific region group (segment 1) is 97.3%, and the AUC reaches 0.97; the average TPR of the second specific region group (segment 2) is 96.9%, and the AUC reaches 0.99. This shows that the support vector machine model has high accuracy and good training effect in this study, and can make a relatively accurate prediction of the overall lithology classification based on existing lithology sample data.

[0108] g. Using the preprocessed TOC-sensitive well trajectory attributes obtained in step e as feature data, the TOC interpretation data as label data, together with the feature data, as the TOC prediction sample dataset, and the dataset is divided into training set and validation set in an 8:2 ratio.

[0109] The CNN-GRU model performs particularly well in lithology prediction tasks. Comparative analysis with traditional neural network models such as Multilayer Perceptron (MLP), Radial Basis Function (RBF), Convolutional Neural Networks (CNN), and a combination of Long Short-Term Memory (LSTM) and CNN models reveals performance differences through variations in their loss functions. Since the prediction data is sequential, the Root Mean Square Error (RMSE) is selected as the loss function. A comparison of the loss functions of the five models is shown in the figure below. Figure 12 As shown, the CNN-GRU model demonstrates significantly higher prediction accuracy and training efficiency compared to traditional architectures such as MLP, RBF, CNN, and CNN-LSTM.

[0110] During the first 50 training rounds, the CNN-GRU model exhibited the most robust downward trend in its loss curve, indicating its superior performance in time series prediction tasks. The CNN-GRU model combines the efficient spatial feature extraction capability of convolutional neural networks with the ability of gated recurrent units to capture temporal dependencies, enabling it to achieve more accurate predictions when processing complex spatiotemporal data. MLP models, due to their simple structure, are often limited in handling nonlinear problems, while RBF networks face challenges of overfitting and computational efficiency due to their large parameter size. CNNs effectively extract spatial features through convolutional and pooling layers, making them suitable for image processing tasks. While CNN-LSTM models offer improvements in handling spatiotemporal data, their parameter count and computational complexity also increase. In contrast, the CNN-GRU model uses GRU units, reducing model parameters while maintaining the ability to capture long-term dependencies, making CNN-GRU more efficient and accurate in sequence data analysis.

[0111] Therefore, a CNN-GRU architecture model combining a convolutional neural network (CNN) and a gated recurrent unit (GRU) was selected for TOC regression prediction, and prediction was performed on the sample dataset.

[0112] In training the neural network model, a loss function based on mean squared error was used as the performance metric. The loss on the training set and the loss on the validation set were calculated to diagnose the model. Through hyperparameter search, the optimal parameters of the neural network model were found, establishing a complex mapping relationship between input and output. After 20 rounds of learning, both loss curves began to converge, and the mean squared error gradually decreased until it reached its minimum value, resulting in a near-perfect fit of the loss curve (e.g., ...). Figure 13 As shown in the figure, the MSE value of the model training set is 0.0285, and the MSE value of the model test set is 0.0275, indicating that the model can predict the TOC distribution in the study area well and can achieve TOC prediction for unknown samples of virtual wells.

[0113] Based on the lithological prediction results obtained in step f and the TOC prediction results obtained in step g, and drawing on existing research experience, the TOC threshold for the first segment of the first specific region group was set to 1%, and the TOC threshold for the second segment of the second specific region group was set to 2%. The thicknesses of mudstone with TOC values ​​greater than the thresholds in both segments were statistically analyzed and accumulated to obtain the effective source rock thickness for each segment of each virtual well. The effective source rock thickness values ​​and coordinates at each virtual well point were imported into the "DF-GVision" software to create a contour map of the effective source rock thickness within the study area. The actual wells within the study area were then projected onto the predicted plane map for observation. Figures 14-15 The virtual well prediction of effective source rock thickness is shown in the planar development map. The actual statistical results on the well are basically consistent with the results of the virtual well prediction, which shows the generalization ability and prediction accuracy of this prediction, and provides theoretical and practical guidance for the next step of oil and gas exploration.

[0114] This application also provides an application scenario in which the above-mentioned method for predicting the development of effective source rocks is applied. Specifically, the method for predicting the development of effective source rocks provided in this embodiment can be applied to the exploration scenario of oil and gas field development. This scenario includes the geological model establishment stage, data analysis stage, and resource assessment stage. The method for predicting the development of effective source rocks provided in this embodiment belongs to the key technology in the data analysis stage. This method can help exploration teams identify potential high-yield areas, improve the success rate and economic benefits of resource development, and thus optimize the development strategy of oil and gas fields.

[0115] Example 2

[0116] This embodiment provides a computer device, which can be a server or a terminal. The computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores data from a method for predicting the development of effective source rocks. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with an external terminal via a network connection. When the computer program is executed by the processor, it implements a method for predicting the development of effective source rocks as described in Embodiment 1.

[0117] Example 3

[0118] This embodiment provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the method for predicting the development of effective source rocks as described in Embodiment 1 above.

[0119] Example 4

[0120] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for predicting the development of effective source rocks as described in Embodiment 1 above.

[0121] Example 5

[0122] This embodiment provides a computer program product, including a computer program that, when executed by a processor, implements a method for predicting the development of effective source rocks as described in Embodiment 1 above.

[0123] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0124] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0125] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0126] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0127] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting the development of effective source rocks, characterized in that: The method for predicting the development of effective source rocks includes: Obtain post-stack seismic data in the study area; Extracting seismic attributes of a virtual well inserted in the study area based on the post-stack seismic data; Inputting the seismic attributes of the virtual well into a trained lithology prediction model, and outputting lithology data of each layer of the virtual well, wherein the lithology prediction model is a support vector machine classification model, and the lithology data includes lithology types and development thicknesses of different lithologies; Inputting the seismic attributes of the virtual well into the trained total organic carbon content prediction model, and outputting the total organic carbon content of each layer of the virtual well, wherein the total organic carbon content prediction model is a CNN-GRU architecture model based on a combination of a convolutional neural network and a gated recurrent unit; Determine the thickness and position of effective source rock according to the lithology data and total organic carbon content of each layer section of the virtual well, wherein the effective source rock is a mudstone layer with a total organic carbon content greater than a threshold value; According to the thickness and location of the effective source rock, evaluating the development of the effective source rock; The training process of the lithology prediction model and the total organic carbon content prediction model specifically includes: Constructing a sample training set, wherein the sample training set includes a lithology sample data set and a total organic carbon content sample data set of real wells in the research area, the lithology sample data set includes lithology-sensitive seismic attributes and sample lithology data, and the total organic carbon content sample data set includes total organic carbon content-sensitive seismic attributes and sample total organic carbon content interpretation data; Using the lithology-sensitive seismic attributes as feature data and the sample lithology data as label data to train a lithology prediction model; Using the total organic carbon content sensitive seismic attributes as feature data and the sample total organic carbon content interpretation data as label data to train a total organic carbon content prediction model; The construction process of the sample training set specifically includes: Acquire sample geological exploration data of the study area, wherein the sample geological exploration data includes sample well location data, sample well trajectory data, sample post-stack seismic data, sample well logging curve data, sample total organic carbon content interpretation data, sample stratigraphic interpretation data and sample core logging data; Determine the sample lithology data of the real well according to the sample core logging data; Constructing a velocity field according to the sample well location data and the sample logging curve data; According to the sample well trajectory data, the sample post-stack seismic data, the sample horizon interpretation data and the velocity field, seismic attributes along the well trajectory in the real well depth domain are obtained; The correlation coefficients of the seismic attributes along the well trajectory in the real well depth domain with the lithological attributes and the total organic carbon content are calculated by using a correlation analysis method, and a first correlation coefficient and a second correlation coefficient are obtained correspondingly; Calculating the correlation coefficient between the seismic attributes along the well trajectory in the real well depth domain to obtain a third correlation coefficient; Selecting the real well depth domain along-well trajectory seismic attribute for which the first correlation coefficient is greater than a first threshold value and the third correlation coefficient is less than a third threshold value as the lithology-sensitive seismic attribute; The seismic attribute along the well trajectory in the real well depth domain, for which the second correlation coefficient is greater than a second threshold value and the third correlation coefficient is less than a third threshold value, is selected as the total organic carbon content sensitive seismic attribute.

2. The method for predicting the development of effective source rocks according to claim 1, characterized in that: Constructing a velocity field according to the sample well location data and the sample logging curve data specifically includes: According to the sample well location data and the sample logging curve data, velocity data and density data of a real well at the same depth point are obtained; Calculating the acoustic impedance of the formation based on the velocity data and density data; Calculating the reflection coefficient of the formation according to the acoustic impedance; Convolution operation is performed on the reflection coefficient and the seismic wavelet to generate a synthetic seismic record; The velocity field is constructed from the synthetic seismic record.

3. The method for predicting the development of effective source rocks according to claim 1, characterized in that: According to the sample well trajectory data, the sample post-stack seismic data, the sample horizon interpretation data and the velocity field, seismic attributes along the well trajectory in the real well depth domain are obtained, specifically including: Extracting real well time domain seismic attributes according to the sample well trajectory data, the sample post-stack seismic data and the sample horizon interpretation data; According to the velocity field, converting the real well time domain seismic attributes into real well depth domain seismic attributes; According to the seismic attributes in the real well depth domain and the sample horizon interpretation data, the seismic attributes along the well trajectory in the real well depth domain are obtained.

4. The method for predicting the development of effective source rocks according to claim 1, characterized in that: According to the thickness and location of the effective source rock, the development of the effective source rock is evaluated, including: Accumulating the thickness of the effective source rock to obtain the thickness of the effective source rock in each layer of each virtual well; According to the thickness and position of each effective source rock, a numerical plane contour map of the thickness of the effective source rock within the research area is drawn; The development of the effective source rock is evaluated based on the numerical plane contour map of the effective source rock thickness.

5. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for predicting the development of effective source rocks as described in any one of claims 1 to 4.

6. 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 for predicting the development of effective source rocks described in any one of claims 1 to 4 is implemented.

7. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for predicting the development of effective source rocks described in any one of claims 1 to 4 is implemented.

Images