AI large model reservoir parameter prediction method based on three-dimensional seismic sub-volume data driving

Abstract

The invention discloses an AI large model reservoir parameter prediction method based on three-dimensional seismic sub-volume data driving, and the method comprises the steps: constructing an initial reservoir porosity prediction model, the reservoir porosity prediction model comprises a three-order tensor input layer, a three-dimensional convolution layer, a maximum pooling layer, a first full connection layer, a second full connection layer and a regression layer which are connected in sequence; obtaining a three-dimensional seismic sub-volume data set, and training the initial reservoir porosity prediction model by using the three-dimensional seismic sub-volume data set to obtain a completely trained reservoir porosity prediction model; and obtaining actually measured three-dimensional seismic sub-volume data, and inputting the actually measured three-dimensional seismic sub-volume data into the well-trained reservoir porosity prediction model to obtain a porosity prediction result. According to the reservoir porosity prediction method and system, through the well-designed six-layer neural network architecture, the accurate porosity prediction value is output, more efficient and accurate reservoir porosity prediction is achieved, and the reservoir porosity prediction method and system are particularly suitable for areas where drilling operation is difficult to conduct or scenes where a large amount of drilling needs to be avoided for cost-benefit consideration.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas exploration, and in particular to an AI large model reservoir parameter prediction method driven by three-dimensional seismic subvolume data. Background Art

[0002] In the field of oil and gas exploration and development, accurate prediction of reservoir parameters is crucial to assessing the commercial value of oil and gas reservoirs and formulating reasonable development plans. The most important reservoir parameter is porosity, which directly affects the recoverable amount of oil and gas.

[0003] With the emergence and successful application of artificial intelligence big models, it is a general trend to use artificial intelligence big model calculations to mine reservoir porosity from three-dimensional data volumes. Although artificial intelligence technology has made some progress in reservoir prediction in recent years, existing AI methods mostly rely on the analysis of two-dimensional or one-dimensional data and fail to fully utilize the spatial structural characteristics of three-dimensional seismic data. In addition, traditional deep learning models have limited processing capabilities for large-scale three-dimensional data, and the training process requires a large amount of labeled data, and still face problems such as insufficient data and insufficient generalization ability.

[0004] Therefore, the present invention proposes an AI large-model reservoir parameter prediction method driven by three-dimensional seismic sub-volume data, which can fully explore the spatial characteristics and deep relationships in the data to achieve more accurate reservoir porosity prediction. It is particularly suitable for application in the absence of direct logging data and relying only on seismic data. Summary of the invention

[0005] In view of this, the present invention provides an AI large model reservoir parameter prediction method driven by three-dimensional seismic sub-volume data to solve the technical problems of low prediction accuracy and low prediction efficiency in the existing reservoir porosity prediction technology under the condition of lack of direct logging data and relying only on seismic data.

[0006] In order to achieve the above technical objectives, the present invention adopts the following technical solutions:

[0007] The present invention provides an AI large model reservoir parameter prediction method based on 3D seismic subvolume data drive, comprising:

[0008] Constructing an initial reservoir porosity prediction model, the reservoir porosity prediction model comprising a third-order tensor input layer, a three-dimensional convolutional layer, a maximum pooling layer, a first fully connected layer, a second fully connected layer and a regression layer connected in sequence;

[0009] Among them, the third-order tensor input layer is used to receive 3D seismic sub-volume data and pre-process the input data; the 3D convolution layer is used to capture the feature map of the data through convolution operation; the maximum pooling layer is used to reduce the spatial dimension of the feature map; the first fully connected layer is used to map high-dimensional data to low-dimensional space for feature integration; the second fully connected layer is used to map the integrated features to the output space; the regression layer is used to generate the final prediction results;

[0010] Acquire a three-dimensional seismic sub-volume data set, and use the three-dimensional seismic sub-volume data set to train an initial reservoir porosity prediction model to obtain a fully trained reservoir porosity prediction model;

[0011] The measured 3D seismic sub-volume data are obtained and input into the well-trained reservoir porosity prediction model to obtain the porosity prediction results.

[0012] Furthermore, the three-dimensional convolution layer includes multiple convolution kernels, each of which has the same size, and the output of the convolution kernel is expressed as follows:

[0013]

[0014] Where X[i,j,k] is the element of the input 3D seismic subvolume data, W pqr,m is the weight of the convolution kernel, b m is the bias of the mth convolution kernel, Z[i,j,k,m] is the element of the output feature map, N×N is the operation window size of the convolution kernel in the spatial dimension, and M is the convolution length in the depth or time dimension.

[0015] Furthermore, the pooling operation of the maximum pooling layer is expressed as follows:

[0016]

[0017] Where P[i,j,k,m] is the output after pooling, A[i,j,k,m] is the input feature map, and w p × q × r Indicates the size of the pooling window.

[0018] Furthermore, the output formula of the first fully connected layer is:

[0019] z fc1 =W fc1 ·x flat +b fc1 ;

[0020] a fc1 =ReLU(z fc1 );

[0021] Where Wfc1 is the first weight matrix, x flat b is the input after flattening the output of the maximum pooling layer into a one-dimensional vector, fc1 is the bias vector, z fc1 is the calculated output, a fc1 is the output after activation by the ReLU activation function.

[0022] Furthermore, the second fully connected layer contains three neurons, corresponding to the predicted values ​​of porosity, permeability and saturation, respectively. The output formula of the second fully connected layer is:

[0023] z fc2 =W fc2 ·a fc1 +b fc2 ;

[0024] Where W fc2 is the second weight matrix, a fc1 is the activation output of the Fc1 fully connected layer, b fc2 is the bias vector, z fc2 is the output feature.

[0025] Furthermore, the regression layer optimizes the model output by using the mean square error as the loss function, and the loss function is expressed as follows:

[0026]

[0027] In the formula, y i Represents the real geological parameters, represents the predicted value of the model, and n is the number of samples.

[0028] Furthermore, a 3D seismic subvolume data set is obtained, including:

[0029] Collecting three-dimensional seismic sub-volume sample data and storing the sample data in the form of a third-order tensor;

[0030] Preprocess the 3D seismic subvolume sample data;

[0031] The porosity, permeability and saturation corresponding to each 3D seismic sub-volume sample data are stored using a label vector set;

[0032] The labels are matched with the corresponding 3D seismic sub-volume sample data to form a 3D seismic sub-volume dataset, and the dataset is divided into a training set, a validation set, and a test set.

[0033] Furthermore, the initial reservoir porosity prediction model is trained using the three-dimensional seismic sub-volume data set to obtain a fully trained reservoir porosity prediction model, including:

[0034] The parameters of the reservoir porosity prediction model are initialized using standard normal distribution or uniform distribution;

[0035] The training set is input into the model for forward propagation, the output prediction value is calculated layer by layer, and the model parameters are optimized using the Adam optimizer;

[0036] Use the validation set to evaluate the generalization ability of the model and determine whether there is overfitting or underfitting, and adjust the model's hyperparameters based on the evaluation results;

[0037] The performance of the adjusted model is verified through the test set. When the model reaches the expected performance standard, a fully trained reservoir porosity prediction model is obtained.

[0038] Furthermore, the measured 3D seismic sub-volume data is obtained and input into a well-trained reservoir porosity prediction model, including:

[0039] Seismic sub-volumes are extracted one by one from the study area according to the measured 3D seismic sub-volume data with the longitudinal line numbers from small to large and the transverse line numbers from small to large, and used as the input of the reservoir porosity prediction model;

[0040] After the model calculation is completed, the porosity parameters are output, and the porosity results corresponding to different longitudinal and transverse line positions are saved in sequence to form a complete porosity prediction result set.

[0041] Furthermore, the method further comprises:

[0042] The spatial distribution of the porosity parameters is analyzed to identify reservoir characteristics, which are combined with permeability and saturation to form a comprehensive geological model that is visualized.

[0043] Compared with the prior art, the AI ​​large-model reservoir parameter prediction method and device based on three-dimensional seismic sub-volume data driven proposed in the present invention receives three-dimensional seismic data through a three-order tensor input layer through a carefully designed six-layer neural network architecture, and gradually refines key features through deep convolution and pooling processing, and finally accurately outputs porosity prediction values ​​through a fully connected layer and a regression layer. This method not only ensures high accuracy of prediction, but also takes into account the effective use of computing resources, so that the porosity of oil and gas reservoirs can be efficiently and accurately predicted in the absence of direct logging data and relying only on seismic data. The present invention greatly broadens the scope of application of reservoir assessment, making it more in line with the actual needs of current oil and gas exploration and development, and is particularly suitable for use in areas where drilling operations are difficult or in scenarios where large-scale drilling needs to be avoided for cost-effectiveness considerations. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1A schematic flow chart of the AI ​​large model reservoir parameter prediction method based on 3D seismic subvolume data drive provided by the present invention;

[0045] Figure 2 A schematic diagram of the structure of the reservoir porosity prediction model provided by the present invention;

[0046] Figure 3 Schematic diagram of the three-dimensional seismic data sub-volume provided by the present invention. DETAILED DESCRIPTION

[0047] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0048] See also Figure 1 This embodiment provides an AI large model reservoir parameter prediction method based on 3D seismic sub-volume data drive, including:

[0049] Step S101: constructing an initial reservoir porosity prediction model, wherein the reservoir porosity prediction model comprises a third-order tensor input layer, a three-dimensional convolution layer, a maximum pooling layer, a first fully connected layer, a second fully connected layer and a regression layer connected in sequence;

[0050] Among them, the third-order tensor input layer is used to receive 3D seismic sub-volume data and pre-process the input data; the 3D convolution layer is used to capture the feature map of the data through convolution operation; the maximum pooling layer is used to reduce the spatial dimension of the feature map; the first fully connected layer is used to map high-dimensional data to low-dimensional space for feature integration; the second fully connected layer is used to map the integrated features to the output space; the regression layer is used to generate the final prediction results;

[0051] Step S102: acquiring a three-dimensional seismic sub-volume data set, and using the three-dimensional seismic sub-volume data set to train an initial reservoir porosity prediction model to obtain a fully trained reservoir porosity prediction model;

[0052] Step S103: Obtain measured 3D seismic sub-volume data and input it into a well-trained reservoir porosity prediction model to obtain a porosity prediction result.

[0053] The AI ​​large-model reservoir parameter prediction method based on three-dimensional seismic sub-volume data driven proposed in this embodiment can achieve more accurate and efficient reservoir porosity prediction by combining the advantages of deep learning, especially the capabilities of three-dimensional convolutional neural networks. This method takes advantage of automatic feature extraction, spatial data processing, and efficient training. Even in the absence of direct logging data and relying only on seismic data, it can efficiently and accurately predict the porosity of oil and gas reservoirs. It is more in line with the actual needs of current oil and gas exploration and development, and is particularly suitable for areas where drilling operations are difficult or where large-scale drilling needs to be avoided for cost-effectiveness considerations.

[0054] The structure of the reservoir porosity prediction model and the data processing process are described in detail below through a specific embodiment.

[0055] This embodiment constructs a reservoir porosity prediction model consisting of six layers, such as Figure 2 As shown, Figure 2 The specific structure of the reservoir porosity prediction model is shown, including Segy3D input layer, Conv3d three-dimensional convolution layer, MaxPool pooling layer, Fc1 fully connected layer, Fc2 fully connected layer, and RegressionOut regression layer.

[0056] The characteristics and functions of each layer of the model are described in detail below.

[0057] The Segy3D layer is an input layer for inputting the 3D seismic subvolume data layer. In 3D seismic exploration, the inline direction, crossline direction, and time or depth together constitute the three dimensions of the 3D seismic data cube. Specifically: Inline (I) means along the survey line, Crossline (X) means perpendicular to the survey line, and time / depth (Z) is the direction perpendicular to the ground. Therefore, we designed the Segy3D input layer to represent the received 3D seismic data in the form of a third-order tensor. In actual processing, the size of the tensor is 7×7×30, where 7×7 represents the 2D plane area of ​​the seismic subvolume on the surface, that is, there are 7 data points in the inline direction and 7 data points in the crossline direction. The setting of 7×7 in the 2D plane area on the surface balances the resolution and computing resources; 30 means that there are 30 data points in the time dimension. The number of data points in the time dimension ensures the integrity of seismic wave propagation information and reduces redundancy. The data quality and format consistency are ensured through input layer preprocessing, which avoids outliers and format errors from interfering with subsequent feature extraction and provides stable and reliable input for accurate porosity prediction.

[0058] After the data input is completed, the data is passed to the Conv3D layer, which is a three-dimensional convolution layer. The three-dimensional convolution layer mainly extracts the local spatial features of the input data through multiple convolution kernels. In some embodiments, this layer uses 10 convolution kernels, and the size of each convolution kernel is 3×3×5. Among them, 3×3 represents the operation window size of the convolution kernel in the spatial dimension, and 5 represents the convolution length in the depth or time dimension. Through the convolution operation, the characteristics of the input data in the local area can be extracted. Each convolution kernel generates a feature map when passing through the data, and finally outputs 10 feature maps. By extracting multiple features through multiple convolution kernels, the ability to characterize complex geological structures and porosity change patterns is enhanced, providing key feature information for porosity prediction. The convolution operation can extract features that are directly or indirectly related to porosity. These features will appear in a specific pattern in the seismic data, and the convolution kernel can identify them and convert them into feature maps that can be processed by the model. Areas with higher porosity may exhibit specific patterns of amplitude and frequency changes in seismic data. The convolution kernel can capture these patterns and reflect them in the feature map, providing key information for the subsequent fully connected layer to accurately predict the porosity.

[0059] The output tensor size is determined by the size of the convolution kernel and the size of the input data. The output tensor is represented as:

[0060]

[0061] Where X[i,j,k] is the element of the input tensor, W pqr,m is the weight of the convolution kernel, b m is the bias of the mth convolution kernel, and Z[i,j,k,m] is the element of the output feature map.

[0062] The feature map after the convolution operation is downsampled through the maximum pooling layer (MaxPool layer). The main purpose of the pooling operation is to reduce the size of the feature map and retain the most important features, thereby reducing the amount of calculation and improving the efficiency of the model. The window size of the pooling layer is 2×2×2, which means that the maximum pooling operation is performed in units of 2×2×2 areas in the spatial and temporal dimensions. Through the pooling process, the spatial and depth dimensions of the feature map are reduced by half. The main features retained after the pooling layer are often more directly related to the porosity, so that the prominent larger amplitude area may correspond to the dense formation with lower reservoir porosity, while the relatively small amplitude area may be related to the loose formation with higher porosity. By reducing the data dimension, the model can process data faster and focus on these key features related to porosity, improving the accuracy and efficiency of prediction.

[0063] For example, the size of the input feature map is 5×5×26×10, and the size of the output feature map after pooling is 2×2×13×10, which effectively reduces the dimension of the feature and retains the key feature information.

[0064] The pooling operation formula is:

[0065]

[0066] Among them, P[i,j,k,m] is the output after pooling, A[i,j,k,m] is the input feature map, and 2×2×2 represents the size of the pooling window.

[0067] The pooled feature map is flattened, that is, the multi-dimensional data is converted into a one-dimensional vector to prepare for the input of the fully connected layer. The flattened data size is 2×2×13×10, that is, 520 data points. The flattened data is passed to the first fully connected layer (Fc1 layer).

[0068] The Fc1 fully connected layer balances feature integration and computing resource utilization, fully combines and nonlinearly transforms the pooling layer features, captures feature interactions, and provides more comprehensive features for porosity prediction. In this embodiment, the Fc1 fully connected layer has 32 neurons and is calculated using the following formula:

[0069] z fc1 =W fc1 ·x flat +b fc1 ;

[0070] Among them, W fc1 is the weight matrix, x flat is the flattened input, b fc1 is the bias vector, z fc1 is the calculated output.

[0071] Then the corresponding data is output after the ReLU activation function:

[0072] a fc1 =ReLU(z fc1 );

[0073] Among them, a fc1 is the output after activation.

[0074] After the Fc1 layer is processed, the features are passed to the second fully connected layer (Fc2 layer). The Fc2 layer contains 3 neurons, corresponding to the three target parameters of porosity, permeability and saturation. The main function of this layer is to map the 32-dimensional features output by the Fc1 layer into 3 output values, which are used to represent the target geological parameters, accurately map the comprehensive features into porosity prediction values, and provide a complete parameter set for reservoir evaluation. After linear transformation calculation, a three-dimensional vector is output, corresponding to the predicted values ​​of the three target parameters. The output formula of the Fc2 layer is:

[0075] z fc2 =W fc2·a fc1 +b fc2 ;

[0076] Among them, W fc2 is the weight matrix, a fc1 is the activation output of the previous layer, b fc2 is the bias vector, z fc2 is the output feature.

[0077] The output of the Fc2 fully connected layer is processed by the regression layer (RegressionOut layer). The main purpose of the regression layer is to minimize the error between the model prediction value and the true value. In this embodiment, the regression layer uses the mean square error (MSE) as the loss function, and the loss function is expressed as:

[0078]

[0079] Among them, y i Represents the real geological parameters, Represents the predicted value of the model, and n is the number of samples. By optimizing the loss function, the model gradually improves the prediction accuracy of the target parameter.

[0080] During the training process, the parameters of each layer are continuously adjusted according to the feedback of the loss function, so that the gap between the predicted porosity value and the actual measured or known porosity value gradually narrows. When the predicted porosity is higher than the true value, the loss function value increases, and the model will adjust the parameters to reduce the predicted value; otherwise, it will increase the predicted value. Through continuous iterative optimization, accurate porosity prediction is finally achieved.

[0081] After the framework of the reservoir porosity prediction model is constructed, it is necessary to prepare training sample data to optimize the model. As a preferred embodiment, in step S102, a 3D seismic sub-volume data set is obtained, including:

[0082] Collecting three-dimensional seismic sub-volume sample data and storing the sample data in the form of a third-order tensor;

[0083] Preprocess the 3D seismic subvolume sample data;

[0084] The porosity, permeability and saturation corresponding to each 3D seismic sub-volume sample data are stored using a label vector set;

[0085] The labels are matched with the corresponding 3D seismic sub-volume sample data to form a 3D seismic sub-volume dataset, and the dataset is divided into a training set, a validation set, and a test set.

[0086] The following is a detailed description of how to construct the dataset.

[0087] First, collect 3D seismic subvolume data. During data collection, porosity data is obtained from core tests of all wells sampled within the 3D seismic coverage area, and seismic subvolumes are extracted from the 3D seismic data volume using their position coordinates. This data collection method ensures the correspondence between seismic subvolume data and porosity data, allowing subsequent models to learn the true relationship between seismic data features and porosity. For example, assuming that the plane coordinates of a well are (x, y), a core sample is obtained at a depth of z, and its porosity, permeability, and saturation parameters are measured. Then, on this basis, the depth coordinate z of the core sample is converted into the two-way time t in the corresponding seismic data volume to ensure accurate data matching.

[0088] The collected 3D seismic sub-volume data is recorded as X∈R n×7×7×30 , where X={X1,X2,…,X n},X i ∈R 7 ×7×30 , X i is the i-th 3D seismic subvolume, representing the amplitude information in space and depth, and n is the number of samples. Figure 3 As shown, Figure 3 The schematic diagram of 3D seismic subvolume data is shown. For a specific target study area, seismic subvolumes are extracted one by one from the 3D seismic data volume. When extracting seismic subvolumes, it is necessary to target a specific stratum. The ordinate of the stratum represents the depth in meters (m), while the ordinate of the 3D seismic is time in milliseconds (ms). Depth is converted into time based on the velocity parameter.

[0089] Secondly, after the data collection is completed, the original data set X raw Preprocessing is performed to ensure data quality and availability, including data cleaning and standardization. After cleaning, the data generates X clean =clean(X raw ), the clean() function is used to perform outlier detection, deduplication and missing value filling. Data cleaning operations (outlier detection, deduplication and missing value filling) in data preprocessing are critical for porosity prediction. Outliers may be caused by measurement errors or geological anomalies. If not processed, they may interfere with the model's learning of normal porosity-related features. Missing value filling ensures the integrity of the data, allowing the model to be trained on complete data and avoiding incomplete or inaccurate feature extraction due to missing data.

[0090] Next, the data is standardized to improve the efficiency and stability of model training. The data is scaled to a uniform range, such as [0,1] or [-1,1]. The standardization formula is:

[0091] In addition, the corresponding reservoir parameter label data, including porosity, permeability and saturation, are collected and recorded as label vector The dataset can be represented as:

[0092] Y={Y1,Y2,…,Y n},Y i ∈R 3 .

[0093] Finally, the preprocessed 3D seismic sub-volume data is paired with the corresponding labels to form a training sample data set Training_Sample = {(X i ,Y i )},i=1,2,…,n.

[0094] To effectively evaluate the model performance, we divide the dataset into training set, validation set and test set, Training_Set,Validation_Set,Test_Set=split(X,Y).

[0095] Through the above steps, high-quality pairing of input data and label data is ensured, laying a data foundation for subsequent model training and prediction.

[0096] In order to improve the prediction accuracy of the model, it is necessary to select a suitable loss function and gradient optimization algorithm, and train and verify the model through a sample data set. As a preferred embodiment, in step S103, the initial reservoir porosity prediction model is trained using the 3D seismic subvolume data set to obtain a fully trained reservoir porosity prediction model, including:

[0097] The parameters of the reservoir porosity prediction model are initialized using standard normal distribution or uniform distribution;

[0098] The training set is input into the model for forward propagation, the output prediction value is calculated layer by layer, and the model parameters are optimized using the Adam optimizer;

[0099] Use the validation set to evaluate the generalization ability of the model and determine whether there is overfitting or underfitting, and adjust the model's hyperparameters based on the evaluation results;

[0100] The performance of the adjusted model is verified through the test set. When the model reaches the expected performance standard, a fully trained reservoir porosity prediction model is obtained.

[0101] The model training process is introduced in detail below.

[0102] Before training begins, all parameters of the model (including weights and biases) are set by random initialization. For the weights W in the convolutional layer and the fully connected layer, standard normal distribution or uniform distribution is used for initialization. Xavier initialization or He initialization is used, which can be expressed as:

[0103] or

[0104] Among them, n in and n out are the dimensions of input and output respectively, and σ is the standard deviation.

[0105] Initialize the bias term to 0, that is: b=0.

[0106] The training data set X train ∈R n×7×7×30 The model is passed in for forward propagation, and the output is calculated layer by layer until the final prediction value of the model is obtained. In the calculation process of each layer, the input features pass through the convolution layer, pooling layer, fully connected layer and activation function, and finally the prediction value of the model is obtained. The Adam optimizer is preferred. This is because the Adam optimizer combines the advantages of Momentum and RMSProp, can adaptively adjust the learning rate during the gradient descent process, and has good convergence speed and robustness.

[0107] During the model training process, the validation set data is used for forward propagation, the prediction results are calculated and the loss value on the validation set is evaluated to determine the generalization ability of the model and identify whether there is overfitting or underfitting. According to the performance of the validation set, hyperparameters such as learning rate α, batch size, regularization coefficient λ, etc. are adjusted. The change in the validation set loss value will determine the direction of hyperparameter adjustment. When the loss value on the validation set reaches the predetermined standard or stops decreasing, the training process will terminate. The usual standard is that the validation set loss reaches a certain threshold or the model performance no longer improves within a number of epochs. Finally, when the model reaches the expected performance standard, the trained model parameters will be saved for subsequent predictions and applications.

[0108] Specifically, the training process includes selecting appropriate loss functions and gradient optimization algorithms to improve the prediction accuracy of the model. The main function of the loss function is to measure the difference between the model's predicted value and the actual value, while the gradient optimization algorithm optimizes the model performance through repeated iterations until the model's performance on the validation set reaches the predetermined standard. The loss function is used to measure the difference between the model's predicted value and the actual value, and the gradient optimization algorithm optimizes the model performance through repeated iterations until the model's performance on the validation set reaches the predetermined standard.

[0109] After obtaining the final fully trained model, we input the measured 3D seismic sub-volume data into the trained model to predict the reservoir porosity.

[0110] As a preferred embodiment, in step S104, the trained model parameter file is read, and the measured 3D seismic sub-volume data are extracted one by one from the study area according to the inline line number from small to large and the crossline line number from small to large, as the input of the model. After the model calculation is completed, the porosity parameters are output. According to different inline and crossline positions, the porosity results of the corresponding positions are saved or output in sequence to form a complete porosity prediction result set:

[0111]

[0112] Among them, X is the extracted seismic sub-volume data, f() is the forward propagation function of the model, W and b are the weight and bias of the model, is the porosity predicted by the model.

[0113] Furthermore, after the porosity results are output, the results can be analyzed and visualized. By analyzing the spatial distribution of the porosity parameters, interpreters can identify reservoir characteristics and evaluate their economic value. In addition, in combination with other reservoir parameters (such as permeability and saturation), a more comprehensive geological model can be formed, and the geological model can be visualized to support decision making. For example, if the predicted porosity shows a specific distribution pattern in a certain area, combined with geological knowledge, the connectivity and oil and gas content of the reservoir in the area can be determined, providing an important basis for oil and gas exploration and development decisions.

[0114] In some embodiments, the prediction results of the model can be compared with the measured data to evaluate the accuracy of the model. If it is found that there is a significant deviation between the model prediction and the actual situation, a feedback mechanism is required to input new data into the model for retraining or fine-tuning. This iterative process ensures that the model is continuously optimized and the accuracy of the prediction is improved. This method ensures that the model can continuously adapt to new data and geological conditions, continuously improve the accuracy of porosity prediction, and enable the model to always maintain good prediction performance in practical applications.

[0115] The AI ​​large-model reservoir parameter prediction method based on 3D seismic subvolume data drive provided by the present invention utilizes deep learning technology, combined with advanced data preprocessing and feature extraction algorithms, and innovative model training strategies to achieve high-precision prediction of reservoir parameters. In addition, the present invention is also a new method for predicting the porosity of oil and gas reservoirs in the absence of well logging data but only seismic data, making it suitable for application in actual oil and gas exploration and development scenarios.

[0116] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. An AI large model reservoir parameter prediction method based on 3D seismic subvolume data drive, characterized in that: include: Constructing an initial reservoir porosity prediction model, the reservoir porosity prediction model comprising a third-order tensor input layer, a three-dimensional convolutional layer, a maximum pooling layer, a first fully connected layer, a second fully connected layer and a regression layer connected in sequence; Among them, the third-order tensor input layer is used to receive 3D seismic sub-volume data and pre-process the input data; the 3D convolution layer is used to capture the feature map of the data through convolution operation; the maximum pooling layer is used to reduce the spatial dimension of the feature map; the first fully connected layer is used to map high-dimensional data to low-dimensional space for feature integration; the second fully connected layer is used to map the integrated features to the output space; the regression layer is used to generate the final prediction results; Acquire a three-dimensional seismic sub-volume data set, and use the three-dimensional seismic sub-volume data set to train an initial reservoir porosity prediction model to obtain a fully trained reservoir porosity prediction model; The measured 3D seismic sub-volume data are obtained and input into the well-trained reservoir porosity prediction model to obtain the porosity prediction results.

2. The AI ​​large model reservoir parameter prediction method based on 3D seismic subvolume data drive according to claim 1 is characterized in that: The three-dimensional convolution layer includes multiple convolution kernels, each of which has the same size, and the output of the convolution kernel is expressed as follows: Where X[i,j,k] is the element of the input 3D seismic subvolume data, W pqr,m is the weight of the convolution kernel, b m is the bias of the mth convolution kernel, Z[i,j,k,m] is the element of the output feature map, N×N is the operation window size of the convolution kernel in the spatial dimension, and M is the convolution length in the depth or time dimension.

3. The AI ​​large model reservoir parameter prediction method based on 3D seismic subvolume data drive according to claim 1 is characterized in that: The pooling operation of the maximum pooling layer is expressed by the formula: Where P[i,j,k,m] is the output after pooling, A[i,j,k,m] is the input feature map, and w p × q × r Indicates the size of the pooling window.

4. The AI ​​large model reservoir parameter prediction method based on 3D seismic subvolume data drive according to claim 1 is characterized in that: The output formula of the first fully connected layer is: z fc1 =W fc1 ·x flat +b fc1 and fc1 =ReLU(from fc1 ) Where W fc1 is the first weight matrix, x flat b is the input after flattening the output of the maximum pooling layer into a one-dimensional vector, fc1 is the bias vector, z fc1 is the calculated output, a fc1 is the output after activation by the ReLU activation function.

5. The AI ​​large model reservoir parameter prediction method based on 3D seismic sub-volume data drive according to claim 1 is characterized in that: The second fully connected layer contains three neurons, corresponding to the predicted values ​​of porosity, permeability and saturation, respectively. The output formula of the second fully connected layer is: z fc2 =W fc2 ·a fc1 +b fc2 Where W fc2 is the second weight matrix, a fc1 is the activation output of the Fc1 fully connected layer, b fc2 is the bias vector, z fc2 is the output feature.

6. The AI ​​large model reservoir parameter prediction method based on 3D seismic sub-volume data drive according to claim 1 is characterized in that: The regression layer optimizes the model output by using the mean square error as the loss function, and the loss function is expressed as: In the formula, y i Represents the real geological parameters, represents the predicted value of the model, and n is the number of samples.

7. The AI ​​large model reservoir parameter prediction method based on 3D seismic subvolume data drive according to claim 1 is characterized in that: Acquire 3D seismic subvolume data sets, including: Collecting three-dimensional seismic sub-volume sample data and storing the sample data in the form of a third-order tensor; Preprocess the 3D seismic subvolume sample data; The porosity, permeability and saturation corresponding to each 3D seismic sub-volume sample data are stored using a label vector set; The labels are matched with the corresponding 3D seismic sub-volume sample data to form a 3D seismic sub-volume dataset, and the dataset is divided into a training set, a validation set, and a test set.

8. The AI ​​large model reservoir parameter prediction method based on 3D seismic sub-volume data drive according to claim 7 is characterized in that: The initial reservoir porosity prediction model is trained using the three-dimensional seismic subvolume data set to obtain a fully trained reservoir porosity prediction model, including: The parameters of the reservoir porosity prediction model are initialized using standard normal distribution or uniform distribution; The training set is input into the model for forward propagation, the output prediction value is calculated layer by layer, and the model parameters are optimized using the Adam optimizer; Use the validation set to evaluate the generalization ability of the model and determine whether there is overfitting or underfitting, and adjust the model's hyperparameters based on the evaluation results; The performance of the adjusted model is verified through the test set. When the model reaches the expected performance standard, a fully trained reservoir porosity prediction model is obtained.

9. The AI ​​large model reservoir parameter prediction method based on 3D seismic subvolume data drive according to claim 1 is characterized in that: Obtain measured 3D seismic subvolume data and input it into a well-trained reservoir porosity prediction model, including: Seismic sub-volumes are extracted one by one from the study area according to the measured 3D seismic sub-volume data with the longitudinal line numbers from small to large and the transverse line numbers from small to large, and used as the input of the reservoir porosity prediction model; After the model calculation is completed, the porosity parameters are output, and the porosity results corresponding to different longitudinal and transverse line positions are saved in sequence to form a complete porosity prediction result set.

10. The AI ​​large model reservoir parameter prediction method based on 3D seismic sub-volume data drive according to claim 9, characterized in that: Also includes: The spatial distribution of the porosity parameters is analyzed to identify reservoir characteristics, which are combined with permeability and saturation to form a comprehensive geological model that is visualized.

Images