A method, apparatus, device, and medium for predicting a carbonate reservoir geodesy sweet spot

Abstract

The application discloses a carbonate reservoir geological dessert prediction method, device, equipment and medium, and belongs to the field of unconventional oil and gas reservoir evaluation. The method comprises the following steps: obtaining logging data and seismic data of a target layer section of a single well and performing pretreatment to obtain a plurality of input feature sequences; inputting the plurality of input feature sequences into a pre-trained geological dessert evaluation parameter prediction model to obtain a plurality of geological dessert evaluation parameter prediction values; constructing a physical constraint condition according to the single-well geological dessert evaluation parameter prediction value and the gas production and embedding the physical constraint condition into a loss function of the prediction model to solve optimal weight coefficients of each geological dessert evaluation parameter; and performing weighted summation on the plurality of geological dessert evaluation parameter prediction values and the corresponding optimal weight coefficients to realize quantitative evaluation of the dessert of the target layer section of the single well. The application adopts a carbonate reservoir geological dessert evaluation parameter prediction model based on a deep multi-task learning neural network, and realizes efficient and automatic geological dessert identification.

Description

Technical Field

[0001] This application belongs to the field of unconventional oil and gas reservoir evaluation and mineral geological exploration services, specifically involving a method, apparatus, equipment and medium for predicting geological sweet spots in carbonate rock reservoirs. Background Technology

[0002] Compared to conventional sandstone oil and gas reservoirs, carbonate oil and gas reservoirs exhibit more significant heterogeneity geologically, with frequent lithofacies changes, thin reservoir thickness, and characteristics of tightness and low permeability. To achieve efficient development of carbonate reservoirs, refined reservoir evaluation is necessary to locate oil and gas-rich geological sweet spots. Traditional research methods (such as core data analysis and conventional well logging interpretation) suffer from low accuracy, poor efficiency, and high dependence on specialized personnel in reservoir geological sweet spot evaluation, making them unsuitable for current demands for efficient and large-scale production. With the rapid development of artificial intelligence technology, data-driven intelligent algorithms are widely used in various stages of oil and gas reservoir sweet spot evaluation. However, existing intelligent algorithms often focus on independent modeling of single geological parameters, severing the inherent correlation between geological parameters and failing to fully utilize multi-source data or consider the synergistic effects between multiple tasks, thus limiting the predictive accuracy and adaptability of intelligent algorithms in reservoir sweet spot evaluation. Summary of the Invention

[0003] To address the problems existing in the prior art, this application proposes a method, device, equipment, and medium for predicting geological sweet spots in carbonate reservoirs. It employs a multi-task deep learning algorithm to deeply explore the synergistic relationship among evaluation parameters of geological sweet spots in reservoirs, thereby establishing a high-precision prediction model for geological sweet spot evaluation parameters. On this basis, an adaptive dynamic weighted summation strategy is further introduced to achieve intelligent quantitative evaluation of geological sweet spots.

[0004] This application is achieved through the following technical solution:

[0005] A method for predicting geological sweet spots in carbonate reservoirs includes:

[0006] Obtain well logging data and seismic data of the target layer in a single well and perform preprocessing to obtain a multi-input feature sequence;

[0007] The multi-input feature sequence is input into the pre-trained geological sweet spot evaluation parameter prediction model to obtain the predicted values ​​of multiple geological sweet spot evaluation parameters for the target layer of the single well.

[0008] Based on the predicted values ​​of single-well geological sweet spot evaluation parameters and gas production, physical constraints are constructed and embedded into the loss function of the geological sweet spot evaluation parameter prediction model to obtain the physically constrained geological sweet spot evaluation parameter prediction model in order to solve the optimal weight coefficients of each geological sweet spot evaluation parameter.

[0009] The predicted values ​​of multiple geological sweet spot evaluation parameters are weighted and summed with their corresponding optimal weight coefficients to achieve quantitative evaluation of the sweet spot of the target layer in the single well.

[0010] In some implementations, the training process of the geological sweet spot evaluation parameter prediction model includes:

[0011] Integrate well logging data, seismic data, and core test data to establish a sample dataset and divide it into training and test sets;

[0012] A deep multi-task learning neural network is established. This network includes a shared feature extraction module, a feature fusion module, and a multi-task module. The shared feature extraction module consists of a low-level feature extraction layer and a shared feature mapping layer. The low-level feature extraction layer extracts low-level feature data from multi-source input data, and the shared feature mapping layer maps the low-level feature data to generate shared features serving multiple tasks and task-specific features. The feature fusion module first calculates the fusion weights of the shared features for each task using an attention mechanism, and then uses a gating network to apply these weights to the shared features of multiple tasks and the task-specific features. The weighted summation yields the final dedicated fusion features for each task. The multi-task module consists of multiple task heads, each of which takes into account the dedicated fusion features of its corresponding prediction task and outputs a predicted value for a geological sweet spot evaluation parameter. To meet the needs of different tasks, a one-dimensional convolutional neural network is combined with a fully connected layer, and dedicated hyperparameters are configured for each task head to achieve independent learning and training of tasks. By introducing an attention mechanism and a dynamic gating strategy, the weights of the loss functions of each task are adjusted in real time, and the loss of each task is weighted to obtain a weighted loss. The weighted losses of all tasks are summed to construct the total loss function of the deep multi-task learning neural network.

[0013] The deep multi-task learning neural network is trained using the training set to establish a prediction model for the geological sweet spot evaluation parameters.

[0014] In some implementations, during the training process, a ten-fold cross-validation method is used to randomly divide the training set into 10 independent and identically distributed subsets. One of these subsets is used as the validation subset, and the remaining nine subsets are used as the training subsets. This process is repeated multiple times for training and validation.

[0015] During training, the Aurora Optimizer is used to optimize the model's hyperparameter combination.

[0016] In some implementations, the integrated well logging data, seismic data, and core test data include:

[0017] Core sample test data is used as the model output label. The core sample test data includes total organic carbon content, porosity, gas saturation, gas content, siliceous minerals, and carbonate minerals. At the same time, well logging data and seismic data corresponding to the core samples are selected as input features to form a basic dataset. The well logging data includes curves of natural gamma, sonic transit time, compensated neutron, density, natural gamma spectrum, potassium content, and resistivity. The seismic data includes root mean square amplitude, wave impedance, elastic modulus, and Poisson's ratio.

[0018] The data in the basic dataset are preprocessed and feature-selected to obtain the sample dataset. The preprocessing includes missing value imputation, outlier handling, and data standardization. Feature selection is carried out using Pearson correlation analysis.

[0019] In some implementations, the step of constructing physical constraints based on the predicted values ​​of single-well geological sweet spot evaluation parameters and gas production, and embedding them into the loss function of the geological sweet spot evaluation parameter prediction model, includes:

[0020] A sweetness index is defined to quantitatively characterize sweetness. By finding a set of weight coefficient combinations that minimizes the mean square error between the sweetness index and the daily gas production of a single well, a physical constraint loss function is constructed and embedded into the loss function of the geological sweetness evaluation parameter prediction model.

[0021] The sweetness index is obtained by weighted summation of predicted values ​​from multiple geological sweetness evaluation parameters.

[0022] In some implementations, the step of weighting and summing the predicted values ​​of multiple geological sweet spot evaluation parameters with their corresponding optimal weight coefficients to achieve quantitative evaluation of the sweet spot in the target formation of a single well includes:

[0023] The sweetness index of the target layer of the single well is obtained by weighting and summing the predicted values ​​of multiple geological sweetness evaluation parameters with their corresponding optimal weight coefficients.

[0024] The process of quantitatively evaluating the sweetness of the target layer in a single well was repeated to obtain the sweetness index of different target layers in the single well. Based on the calculation results of the sweetness index, a columnar section of the geological sweetness profile of the single well was generated.

[0025] Secondly, this application proposes a device for predicting geological sweet spots in carbonate reservoirs, comprising:

[0026] The preprocessing unit is used to acquire and preprocess well logging data and seismic data of the target layer in a single well to obtain a multi-input feature sequence;

[0027] The prediction unit is used to input the multi-input feature sequence into a pre-trained geological sweet spot evaluation parameter prediction model to obtain multiple geological sweet spot evaluation parameter prediction values ​​for the target layer of the single well.

[0028] The optimization unit is used to construct physical constraints based on the predicted values ​​of the geological sweet spot evaluation parameters of a single well and the gas production, and embed them into the loss function of the geological sweet spot evaluation parameter prediction model to obtain the geological sweet spot evaluation parameter prediction model with physical constraints in order to solve the optimal weight coefficients of the geological sweet spot evaluation parameters of each well.

[0029] And an evaluation unit, used to perform a weighted summation of the predicted values ​​of the multiple geological sweet spot evaluation parameters and their corresponding optimal weight coefficients to achieve a quantitative evaluation of the sweet spot of the target layer in the single well.

[0030] In some embodiments, the device further includes:

[0031] A visualization unit is used to display a bar chart of geological sweet spot profiles of a single well, which is generated based on the calculated sweetness index of different target layers in the single well.

[0032] Thirdly, this application proposes an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the above-described methods for predicting geological sweet spots in carbonate reservoirs.

[0033] Fourthly, this application proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for predicting geological sweet spots in carbonate reservoirs.

[0034] This application proposes a method for predicting geological sweet spots in carbonate reservoirs. It adopts a prediction model for evaluation parameters of geological sweet spots in carbonate reservoirs based on a deep multi-task learning neural network, which realizes efficient and automated identification of geological sweet spots. The process is simple and fast, and the identification results are accurate and reliable. By introducing a multi-source data fusion and multi-task collaborative learning mechanism, this method considers all factors, enhances the model's comprehensive representation ability of complex geological conditions, and significantly improves its generalization and prediction robustness.

[0035] Accordingly, the carbonate reservoir geological sweet spot prediction device, computer equipment, and computer-readable storage medium proposed in this application also possess the same technical effects as described above. Attached Figure Description

[0036] The accompanying drawings, which are included to provide a further understanding of the embodiments of this application and form part of this application, do not constitute a limitation on the embodiments of this application. In the drawings:

[0037] Figure 1 This is a schematic diagram of the geological sweet spot prediction method proposed in the embodiments of this application;

[0038] Figure 2 A schematic diagram of the deep multi-task learning neural network architecture established for the embodiments of this application;

[0039] Figure 3 This is a schematic diagram of the gating network and attention mechanism structure used in the embodiments of this application;

[0040] Figure 4 This is a longitudinal quantitative distribution profile of a gas well's geological sweet spot, predicted using the geological sweet spot prediction method proposed in this application.

[0041] Figure 5 This is a block diagram illustrating the principle of the geological sweet spot prediction device proposed in the embodiments of this application;

[0042] Figure 6 This is a schematic diagram of the geological sweet spot prediction system architecture proposed in the embodiments of this application;

[0043] Figure 7 This is a schematic diagram of the electronic device proposed in the embodiments of this application;

[0044] Figure 8 This is a schematic diagram of a computer-readable storage medium proposed in an embodiment of this application;

[0045] Figure reference numerals and corresponding component names:

[0046] 200-Geological sweet spot prediction device, 201-Preprocessing unit, 202-Prediction unit, 203-Optimization unit, 204-Evaluation unit, 205-Visualization unit, 300-Geological sweet spot prediction system, 301-Input device, 302-Output device, 303-Processor A, 304-Memory A, 400-Electronic device, 410-Memory B, 420-Processor B, 411-Computer program A, 500-Computer readable storage medium, 511-Computer program B. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this application are only for explaining this application and are not intended to limit this application.

[0048] Currently, traditional methods for evaluating geological sweet spots in carbonate reservoirs are labor-intensive, highly empirical, and have limited accuracy. Intelligent evaluation methods often focus on independent modeling of single geological parameters, severing the inherent relationships between these parameters and failing to fully utilize multi-source data or consider the synergistic effects between data, thus limiting prediction accuracy and adaptability. To address this, this application proposes a method for predicting geological sweet spots in carbonate reservoirs. This method comprehensively utilizes multi-source data, including regional geology, well logging, seismic data, and core testing. Leveraging the complementary stratigraphic information of multi-source data, and while considering model optimization efficiency, it integrates gating networks and attention mechanisms to construct a Deep Multi-Task Learning Network (DMTL-Net). This fully explores the synergistic relationships between reservoir geological sweet spot evaluation parameters, thereby establishing a high-precision intelligent prediction model for geological sweet spot evaluation parameters. Furthermore, an adaptive dynamic weighted summation strategy based on sweet spot evaluation parameters is introduced to achieve intelligent quantitative evaluation of geological sweet spots. This method has advantages such as high efficiency, convenience, high accuracy, low cost, and strong adaptability.

[0049] like Figure 1 As shown, the geological sweet spot prediction method proposed in this application includes the following steps:

[0050] Step 1: Obtain well logging data and seismic data of the target layer in a single well and perform preprocessing to obtain a multi-input feature sequence;

[0051] Step 2: Input the multi-input feature sequence into the pre-trained geological sweet spot evaluation parameter prediction model to obtain the predicted values ​​of multiple geological sweet spot evaluation parameters for the target layer of the single well.

[0052] Step 3: Based on the predicted values ​​of the geological sweet spot evaluation parameters and the gas production of a single well, construct physical constraints and embed them into the loss function of the geological sweet spot evaluation parameter prediction model to obtain the physical constraint geological sweet spot evaluation parameter prediction model in order to solve the optimal weight coefficients of the geological sweet spot evaluation parameters of each well.

[0053] Step 4: The predicted values ​​of multiple geological sweet spot evaluation parameters are weighted and summed with their corresponding optimal weight coefficients to achieve quantitative evaluation of the sweet spot in the target layer of the single well.

[0054] Furthermore, in this embodiment of the application, the training process of the geological sweet spot evaluation parameter prediction model includes:

[0055] Step 2.1: Integrate multi-source data such as well logging, seismic data, and core testing data to establish a sample dataset and divide it into training and testing sets to provide a data foundation for model training and testing.

[0056] A carbonate reservoir geological sweet spot refers to an area rich in oil and gas resources and with superior physical properties. Its quality is mainly controlled by four categories: source rock quality (with total organic carbon (TOC) as the core, determining hydrocarbon generation capacity), reservoir physical properties (including porosity (Por) and gas saturation (Sg), affecting reservoir space and gas potential), mineral quality (mainly carbonate minerals Car and siliceous minerals Si, determining reservoir type and brittleness index), and gas content quality (represented by gas content (Qg), reflecting resource abundance). Therefore, this application's embodiments preferentially use a series of inherent reservoir geological properties, such as total organic carbon (TOC), porosity (Por), gas saturation (Sg), gas content (Qg), siliceous minerals (Si), and carbonate minerals (Car), as evaluation parameters for geological sweet spots, in order to systematically and scientifically evaluate carbonate reservoir geological sweet spots.

[0057] Based on this, the sample dataset creation process in this step includes:

[0058] First, core sample test data (total organic carbon content TOC, porosity Por, gas saturation Sg, gas content Qg, silica minerals Si, and carbonate minerals Car) are used as output labels, and corresponding well logging and seismic data are selected as input features to form a basic dataset. Well logging data mainly includes curves for natural gamma (GR), sonic transit time (AC), compensated neutron (CNL), density (DEN), natural gamma spectrum (thorium TH, uranium U content and U / TH ratio), potassium (K) content, and resistivity (RT), which together form the foundation for high-resolution vertical data. Seismic data extracts various properties including root mean square amplitude (RMS), acoustic impedance (AI), elastic modulus (E), and Poisson's ratio (v) to macroscopically reflect the reservoir's physical and fluid characteristics.

[0059] Then, the data in the basic dataset is preprocessed to obtain the sample dataset. The preprocessing process includes: addressing issues such as missing data, inconsistent dimensionality, and differences in units of measurement in various types of data by performing missing value imputation, outlier handling, and data standardization to comprehensively improve the quality of the sample data; and employing Pearson correlation analysis to perform feature selection, thereby reducing data dimensionality, avoiding multicollinearity, and improving the training efficiency and generalization ability of the model. It should be noted that existing technologies can be used for the preprocessing process, which will not be elaborated here.

[0060] The sample data in the sample dataset is randomly divided into training set and test set according to a preset ratio (e.g., 7:3, 8:2, etc.), which are used for model training and performance testing respectively, to ensure the randomness in the model building process and the credibility of the results.

[0061] Step 2.2: Establish a deep multi-task learning neural network. This deep multi-task learning neural network uses a convolutional neural network as its underlying architecture and improves the sharing strategy by integrating a gating network and an attention mechanism to achieve multi-task adaptive weight collaborative prediction.

[0062] like Figure 2 As shown, the deep multi-task learning neural network constructed in this embodiment mainly includes a shared feature extraction module, a feature fusion module, and a multi-task module.

[0063] The shared feature extraction module consists of a low-level feature extraction layer and a shared feature mapping layer. Its core objective is to extract shared features from multi-source input data that can efficiently serve multiple tasks, while preserving task-specific features, thus improving overall learning ability while addressing specific task requirements. Specifically, the low-level feature extraction layer extracts multi-scale features through a series of convolutional layers, using weighted filters to perform sliding convolutions on the input data to capture local feature patterns. Subsequently, activation functions such as ReLU are used to introduce non-linear transformations, enhancing the model's ability to express complex relationships. Simultaneously, pooling layers are used to reduce the dimensionality of the convolutional features, preserving key information while filtering redundant details, thereby compressing features and reducing computational and memory overhead. Finally, a flattening layer maps the multi-dimensional features into one-dimensional vectors, which are then input to the shared feature mapping layer. The shared feature mapping layer consists of fully connected layers with multiple neurons. Each fully connected layer is configured with a set of weight matrices and bias terms, transforming the features extracted by the previous layer into a higher-level abstract representation through weighting and biasing, thereby extracting more discriminative features to adapt to the needs of different tasks. This application describes the shared feature mapping layer using a dual-label prediction task (task 1 and task 2) as an example. The shared feature mapping layer generates three types of features: task 1 specific feature F. A Task 2 Specific Feature F B and the shared feature F of the two tasks S The feature mapping process can be represented as:

[0064]

[0065]

[0066]

[0067] in, This represents the underlying feature data (i.e., the features extracted by the underlying feature extraction layer). For activation function, , , Features F A F B F SThe weight matrix, , , Features F A F B F S The weight matrix.

[0068] The core function of the feature fusion module is to dynamically integrate shared features and task-specific features in multi-task learning, achieving a balance between knowledge sharing and task personalization through a collaborative attention mechanism and a gating network. This feature fusion module first uses an attention mechanism to calculate feature importance weights, focusing on key information in the input; simultaneously, it dynamically adjusts the fusion ratio of shared features and task-specific features through a gating network, controlling feature flow with a gating value, ultimately optimizing feature representation and efficiently allocating computational resources. This application describes the feature fusion module using a dual-label prediction task (Task 1 and Task 2) as an example. For Task 1 and Task 2, an attention mechanism is used to calculate the fusion weights. The formula is as follows:

[0069]

[0070]

[0071] in, , Shared features F S The fusion weights for Task 1 and Task 2; , These are task-specific features F A Task-specific features F B The representation after concatenation with shared features; , These are the query vectors for Task 1 and Task 2, respectively; Softmax is the activation function.

[0072] Based on a gated network, the feature F is processed using fusion weights. A F B F S By performing a weighted summation, we obtain the final exclusive fusion features for Task 1 and Task 2. and This information is then fed into the prediction layers of Task 1 and Task 2 to complete the regression prediction task. The calculation formula for the specific fusion feature is as follows:

[0073]

[0074]

[0075] The multi-task module consists of multiple task heads. To meet the needs of different tasks, a one-dimensional convolutional neural network is combined with fully connected layers, and each task head is configured with dedicated hyperparameters (such as kernel size, number of layers, number of neurons, etc.) to achieve independent learning and training for each task. By introducing an attention mechanism and dynamic gating strategy, the weights of the loss functions for each task are adjusted in real time. The loss of each task is weighted to obtain a weighted loss, thereby balancing the optimization objectives of each task and ensuring mutual promotion between tasks. Finally, the weighted losses of all tasks are summed to form the total loss function of the deep multi-task learning neural network, calculated as follows:

[0076]

[0077]

[0078] in, For the number of tasks, For the task The threshold value, For the task The weighted loss weight, For the task The threshold value, For the task loss function, This is the total loss function.

[0079] The attention mechanism and gating network structure used in the embodiments of this application are as follows: Figure 3 As shown, Figure 3 As shown, the gating network uses a learnable gating mechanism to automatically control the flow of information in the network, thereby enhancing the model's information filtering ability. Multiple information vectors (Vector 1, ..., Vector n) are fed into the gating network. The Selector receives these vectors and generates a set of corresponding attention weights through fully connected (FC) and SoftMax (activation function) layers. The gating network uses this set of weights to perform a weighted sum of all input vectors to generate the final output X. The attention mechanism simulates the human brain's attention allocation, used to filter key content from a large amount of information. The similarity between the query q (representing the content that needs attention) and the key kᵢ (i=1,2, ...,n) corresponding to each input is calculated to obtain a similarity score Sᵢ (i=1,2, ...,n). The similarity score is normalized by SoftMax (activation function) to obtain the attention weight aᵢ (i=1,2, ...,n), representing the importance of each value. The attention weights are weighted and summed with the weight pair values ​​vᵢ (i=1,2, ...,n) to generate the output a. in, This indicates element-wise multiplication. This indicates the concatenation of elements.

[0080] Step 2.3: Train the deep multi-task learning neural network using the training set, and finally establish a prediction model for evaluation parameters of the geological sweet spot of carbonate reservoirs.

[0081] During training, a 10-fold cross-validation method is used, which randomly divides the training set into 10 independent and identically distributed subsets. One subset is used as the validation subset, and the remaining 9 subsets are used as the training subsets. This process is repeated multiple times to obtain the model's performance metrics on different data subsets. This method effectively reduces the randomness in the evaluation process by simulating the model's performance on unknown data, thereby more accurately evaluating the model's generalization ability and stability.

[0082] During training, the Aurora Optimizer (PLO) is used to optimize the model's hyperparameter combination. The PLO's mathematical model incorporates mechanisms such as Lorentz force, Newton's second law, damping factors, and adaptive weights. By simulating the motion of charged particles in the Earth's magnetic field and atmosphere, and combining strategies such as rotational motion, aurora elliptical step, dynamic step size, and particle collisions, it achieves a dynamic balance between global exploration and local exploitation, significantly improving the efficiency and accuracy of the optimization algorithm. This optimizer helps the model quickly find the optimal hyperparameter combination, thereby improving overall performance.

[0083] After model training is complete, relevant regression evaluation metrics are introduced, and a test set is used to comprehensively evaluate model performance and determine its feasibility. The embodiments of this application use mean absolute difference (MAE), root mean square error (RMSE), and coefficient of determination (R²). 2 The model performance is evaluated using metrics such as MAE and RMSE. Lower MAE and RMSE values ​​indicate better predictive performance. 2 The higher the value, the better the model's predictive performance.

[0084] Furthermore, in practical applications, the preprocessing of the acquired well logging data and seismic data in step 1 is the same as the preprocessing process in step 2.1 above, including operations such as missing value imputation, outlier handling, and data standardization.

[0085] Furthermore, in step 3 of this embodiment, a sweetness index F is defined to quantitatively characterize the sweetness level. By finding a set of weight coefficients that minimizes the mean square error between the sweetness index and the daily gas production of a single well, physical constraints are constructed based on this principle and embedded into the loss function of the geological sweetness evaluation parameter prediction model, forming a physically constrained geological sweetness evaluation parameter prediction model. A gating network and attention mechanism are used to adaptively and dynamically adjust the weights of each input sweetness evaluation parameter, ultimately obtaining the weight coefficient combination that optimizes the model's prediction performance. Here, the sweetness index is the value obtained by weighted summation of multiple geological sweetness evaluation parameters.

[0086] The physical constraint loss function is:

[0087]

[0088] in, For the first The sweetness index of Kou-i, For the first The daily gas production of the well. This is the physical constraint loss function.

[0089] Furthermore, step 4 of this embodiment of the application performs a weighted summation of the predicted values ​​of multiple geological sweet spot evaluation parameters and their corresponding optimal weight coefficients to obtain the sweetness index of the target interval in a single well. Through the above steps 1 to 4, the sweetness index of different target intervals in a single well can be obtained. Based on the sweetness index calculation results, a single-well geological sweet spot profile columnar diagram is generated to intuitively display the distribution characteristics of geological sweet spots, thus completing the prediction and quantitative evaluation of geological sweet spots in carbonate reservoirs.

[0090] Furthermore, the geological sweet spot prediction method in this application embodiment also includes:

[0091] The importance of each feature parameter is analyzed using SHAP value charts and ranked to determine the contribution of different parameters to the model's prediction results. At the same time, a feature parameter importance heatmap is constructed to visually display the strength and direction of the influence of each feature parameter on the model output, thereby enhancing the interpretability of the model and making the prediction decision-making process more transparent.

[0092] The geological sweet spot prediction method proposed in this application combines a deep multi-task learning neural network with carbonate reservoir sweet spot evaluation parameters. It deeply explores the potential nonlinear relationship between well logging data, seismic data, core test data and sweet spot evaluation parameters, establishes a prediction model for carbonate reservoir geological sweet spot evaluation parameters, and obtains the sweetness index by weighted summation of the optimal weights obtained by combining the prediction results of geological sweet spot evaluation parameters with the model with physical constraints. This method can quickly and accurately complete sweet spot prediction and quantitative evaluation. In addition, this method also has high generalization ability and interpretability.

[0093] To verify the feasibility of the geological sweet spot prediction method proposed in this application embodiment, this application embodiment takes a carbonate reservoir in a certain area as an example. First, the well logging, seismic and core test data and geological sweet spot evaluation parameters of a certain carbonate gas reservoir are determined, as shown in Table 1.

[0094] Table 1. Normalized display of a sample dataset from a carbonate gas reservoir.

[0095] GR AC CNL DEN TH U U / TH RT RMS 0.12 0.56 0.33 0.74 0.28 0.61 0.42 0.15 0.68 0.45 0.22 0.67 0.19 0.84 0.09 0.76 0.53 0.04 0.78 0.89 0.11 0.95 0.47 0.72 0.25 0.87 0.39 0.23 0.41 0.85 0.38 0.13 0.35 0.59 0.31 0.77 0.91 0.07 0.52 0.62 0.69 0.48 0.03 0.64 0.26 0.34 0.63 0.29 0.05 0.56 0.17 0.88 0.22 0.93 ... ... ... ... ... ... ... ... ... AI E v TOC Por Sg Qg Si Car 0.37 0.82 0.23 0.59 0.34 0.71 0.46 0.17 0.63 0.81 0.29 0.67 0.13 0.78 0.18 0.83 0.52 0.08 0.14 0.66 0.05 0.84 0.27 0.95 0.19 0.88 0.74 0.58 0.11 0.79 0.47 0.62 0.36 0.57 0.33 0.41 0.45 0.54 0.42 0.21 0.09 0.64 0.24 0.61 0.55 0.72 0.38 0.16 0.75 0.53 0.02 0.68 0.25 0.12 ... ... ... ... ... ... ... ... ...

[0096] Based on well logging, seismic, and experimental test data, sample data was constructed. A vertical quantitative distribution profile of the geological sweet spot in the gas well was predicted using a sweet spot evaluation parameter prediction model and a sweetness index calculation model. This profile was then compared with actual core interpretation results. Figure 4 As shown, from Figure 4 It can be observed that the geological sweet spot prediction method proposed in this application has high overall accuracy and reliability in predicting and identifying geological sweet spots. Furthermore, by using a physically constrained model to quantitatively calculate the geological sweet spot of a single well in a carbonate gas reservoir, the accuracy of the gas well geological sweet spot prediction is 92.7%, and the single-well prediction process takes approximately 3 minutes. This demonstrates high prediction accuracy and efficiency, meeting engineering requirements.

[0097] Based on the same technical concept described above, this application also proposes a device for predicting the geological sweet spot of carbonate reservoirs, such as... Figure 5 As shown, the geological sweet spot prediction device 200 includes:

[0098] The preprocessing unit 201 is used to acquire and preprocess well logging data and seismic data of the target layer in a single well to obtain a multi-input feature sequence. The specific preprocessing method is as described in the geological sweet spot prediction method above, and will not be repeated here.

[0099] Prediction unit 202 is used to input multiple input feature sequences into a pre-trained geological sweet spot evaluation parameter prediction model to obtain predicted values ​​of multiple geological sweet spot evaluation parameters for the target interval of the single well. The specific model construction process is as described in the geological sweet spot prediction method above, and will not be repeated here.

[0100] Optimization unit 203 is used to construct physical constraints based on the predicted values ​​of single-well geological sweet spot evaluation parameters and gas production, and embed these constraints into the loss function of the geological sweet spot evaluation parameter prediction model to obtain a physically constrained geological sweet spot evaluation parameter prediction model for solving the optimal weight coefficients of each geological sweet spot evaluation parameter. The method of constructing physical constraints is as described in the geological sweet spot prediction method above, and will not be repeated here.

[0101] Furthermore, evaluation unit 204 is used to perform a weighted summation of the predicted values ​​of multiple geological sweet spot evaluation parameters and their corresponding optimal weight coefficients to achieve a quantitative evaluation of the sweet spot in the target layer of the single well. The specific evaluation process is as described in the geological sweet spot prediction method above, and will not be repeated here.

[0102] Furthermore, the geological sweet spot prediction device 200 proposed in this application embodiment also includes:

[0103] Visualization unit 205 is used to display a bar chart of geological sweet spot profile of a single well, which can be generated based on the calculated sweetness index of different target layers of the single well.

[0104] Furthermore, the visualization unit 205 in this embodiment can also be used to display a heatmap of the importance of each feature parameter, wherein the heatmap is generated by analyzing the contribution of each feature parameter to the model prediction results through a SHAP value chart.

[0105] Based on the same technical concept described above, this application also proposes a system for predicting geological sweet spots in carbonate reservoirs, such as... Figure 6 As shown, the geological sweet spot prediction system 300 proposed in this application includes:

[0106] The system comprises an input device 301, an output device 302, a processor A303, and a memory A304; wherein the number of processors A303 and memory A304 can be one or more. Figure 6 The following description uses a processor A303 and a memory A304 as an example. The input device 301, output device 302, processor A303, and memory A304 can be connected via a bus or other means. Figure 6 Taking the example of a connection between China and Israel via a bus.

[0107] Specifically, by calling the operation instructions stored in memory A304, processor A303 executes the following steps:

[0108] Obtain well logging data and seismic data of the target layer in a single well and perform preprocessing to obtain a multi-input feature sequence;

[0109] The multi-input feature sequence is input into the pre-trained geological sweet spot evaluation parameter prediction model to obtain the predicted values ​​of multiple geological sweet spot evaluation parameters for the target layer of the single well.

[0110] Based on the predicted values ​​of geological sweet spot evaluation parameters and gas production of a single well, physical constraints are constructed and embedded into the loss function of the geological sweet spot evaluation parameter prediction model to obtain the physical constraint geological sweet spot evaluation parameter prediction model in order to solve the optimal weight coefficients of the geological sweet spot evaluation parameters of each well.

[0111] The predicted values ​​of multiple geological sweet spot evaluation parameters are weighted and summed with their corresponding optimal weight coefficients to achieve quantitative evaluation of the sweet spot in the target layer of this single well.

[0112] Optionally, by calling the operation instructions stored in memory A304, processor A303 is also used to execute any of the embodiments in the corresponding examples of the above-described geological sweet spot prediction method.

[0113] Based on the same technical concept described above, this application also proposes an electronic device, such as... Figure 7As shown, the electronic device 400 includes: a memory 410B, a processor 420B, and a computer program A411 stored on the memory 410B and executable on the processor 420B. When the processor 420B executes the computer program A411, it performs the following steps:

[0114] Obtain well logging data and seismic data of the target layer in a single well and perform preprocessing to obtain a multi-input feature sequence;

[0115] The multi-input feature sequence is input into the pre-trained geological sweet spot evaluation parameter prediction model to obtain the predicted values ​​of multiple geological sweet spot evaluation parameters for the target layer of the single well.

[0116] Based on the predicted values ​​of geological sweet spot evaluation parameters and gas production of a single well, physical constraints are constructed and embedded into the loss function of the geological sweet spot evaluation parameter prediction model to obtain the physical constraint geological sweet spot evaluation parameter prediction model in order to solve the optimal weight coefficients of the geological sweet spot evaluation parameters of each well.

[0117] The predicted values ​​of multiple geological sweet spot evaluation parameters are weighted and summed with their corresponding optimal weight coefficients to achieve quantitative evaluation of the sweet spot in the target layer of this single well.

[0118] Optionally, when processor B420 executes computer program A411, it can implement any of the embodiments in the corresponding examples of the above-described geological sweet spot prediction method.

[0119] It should be noted that the electronic device proposed in this application embodiment is a device used to implement the above-mentioned geological sweet spot prediction method. Therefore, based on the above-mentioned geological sweet spot prediction method proposed in this application embodiment, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this application embodiment. Therefore, how the electronic device specifically implements the above-mentioned geological sweet spot prediction method will not be described in detail here. Any electronic device used by those skilled in the art to implement the above-mentioned geological sweet spot prediction method falls within the scope of protection of this application.

[0120] Based on the same technical concept described above, embodiments of this application also propose a computer-readable storage medium, such as... Figure 8 As shown, the computer-readable storage medium 500 stores a computer program B511, which, when executed by a processor, performs the following steps:

[0121] Obtain well logging data and seismic data of the target layer in a single well and perform preprocessing to obtain a multi-input feature sequence;

[0122] The multi-input feature sequence is input into the pre-trained geological sweet spot evaluation parameter prediction model to obtain the predicted values ​​of multiple geological sweet spot evaluation parameters for the target layer of the single well.

[0123] Based on the predicted values ​​of geological sweet spot evaluation parameters and gas production of a single well, physical constraints are constructed and embedded into the loss function of the geological sweet spot evaluation parameter prediction model to obtain the physical constraint geological sweet spot evaluation parameter prediction model in order to solve the optimal weight coefficients of the geological sweet spot evaluation parameters of each well.

[0124] The predicted values ​​of multiple geological sweet spot evaluation parameters are weighted and summed with their corresponding optimal weight coefficients to achieve quantitative evaluation of the sweet spot in the target layer of this single well.

[0125] Optionally, when the computer program B511 is executed by the processor, it can implement any of the embodiments corresponding to the above-described geological sweet spot prediction method.

[0126] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0127] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0128] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0129] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0130] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0131] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above description is only a specific embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for predicting geological sweet spots in carbonate reservoirs, characterized in that, include: Obtain well logging data and seismic data of the target layer in a single well and perform preprocessing to obtain a multi-input feature sequence; The multi-input feature sequence is input into the pre-trained geological sweet spot evaluation parameter prediction model to obtain the predicted values ​​of multiple geological sweet spot evaluation parameters for the target layer of the single well. Based on the relationship between the predicted values ​​of single-well geological sweet spot evaluation parameters and gas production, physical constraints are constructed and embedded into the loss function of the geological sweet spot evaluation parameter prediction model to obtain a physically constrained geological sweet spot prediction model in order to solve for the optimal weight coefficients of each geological sweet spot evaluation parameter. The predicted values ​​of multiple geological sweet spot evaluation parameters are weighted and summed with their corresponding optimal weight coefficients to achieve quantitative evaluation of the sweet spot of the target layer in the single well. The training process of the geological sweet spot evaluation parameter prediction model includes: Integrate well logging data, seismic data, and core test data to establish a sample dataset and divide it into training and test sets; A deep multi-task learning neural network is established. This network includes a shared feature extraction module, a feature fusion module, and a multi-task module. The shared feature extraction module consists of a low-level feature extraction layer and a shared feature mapping layer. The low-level feature extraction layer extracts low-level feature data from multi-source input data, and the shared feature mapping layer maps the low-level feature data to generate shared features serving multiple tasks and task-specific features. The feature fusion module first calculates the fusion weights of the shared features for each task using an attention mechanism, and then uses a gating network to apply these weights to the shared features of multiple tasks and the task-specific features. The weighted summation yields the final dedicated fusion features for each task. The multi-task module consists of multiple task heads, each of which takes into account the dedicated fusion features of its corresponding prediction task and outputs a predicted value for a geological sweet spot evaluation parameter. To meet the needs of different tasks, a one-dimensional convolutional neural network is combined with a fully connected layer, and dedicated hyperparameters are configured for each task head to achieve independent learning and training of tasks. By introducing an attention mechanism and a dynamic gating strategy, the weights of the loss functions of each task are adjusted in real time, and the loss of each task is weighted to obtain a weighted loss. The weighted losses of all tasks are summed to construct the total loss function of the deep multi-task learning neural network. The deep multi-task learning neural network is trained using the training set to establish a prediction model for the geological sweet spot evaluation parameters.

2. The method for predicting geological sweet spots in carbonate reservoirs according to claim 1, characterized in that, During the training process, the ten-fold cross-validation method is used to randomly divide the training set into 10 independent and identically distributed subsets. One of these subsets is used as the validation subset, and the remaining nine subsets are used as the training subsets. This process is repeated multiple times for training and validation. During training, the Aurora Optimizer is used to optimize the model's hyperparameter combination.

3. The method for predicting geological sweet spots in carbonate reservoirs according to claim 1, characterized in that, The integrated well logging data, seismic data, and core test data include: Core sample test data is used as the model output label. The core sample test data includes total organic carbon content, porosity, gas saturation, gas content, siliceous minerals, and carbonate minerals. At the same time, well logging data and seismic data corresponding to the core samples are selected as input features to form a basic dataset. The well logging data includes curves of natural gamma, sonic transit time, compensated neutron, density, natural gamma spectrum, potassium content, and resistivity. The seismic data includes root mean square amplitude, wave impedance, elastic modulus, and Poisson's ratio. The data in the basic dataset are preprocessed and feature-selected to obtain the sample dataset. The preprocessing includes missing value imputation, outlier handling, and data standardization. Feature selection is carried out using Pearson correlation analysis.

4. A method for predicting geological sweet spots in carbonate reservoirs according to any one of claims 1-3, characterized in that, The method of constructing physical constraints based on the predicted values ​​of single-well geological sweet spot evaluation parameters and gas production, and embedding them into the loss function of the geological sweet spot evaluation parameter prediction model, includes: A sweetness index is defined to quantitatively characterize sweetness. By finding a set of weight coefficient combinations that minimizes the mean square error between the sweetness index and the daily gas production of a single well, a physical constraint loss function is constructed and embedded into the loss function of the geological sweetness evaluation parameter prediction model. The sweetness index is obtained by weighted summation of predicted values ​​from multiple geological sweetness evaluation parameters.

5. The method for predicting geological sweet spots in carbonate reservoirs according to claim 4, characterized in that, The method of weighted summing of predicted values ​​of multiple geological sweet spot evaluation parameters with their corresponding optimal weight coefficients to achieve quantitative evaluation of the sweet spot in the target formation of a single well includes: The sweetness index of the target layer of the single well is obtained by weighting and summing the predicted values ​​of multiple geological sweetness evaluation parameters with their corresponding optimal weight coefficients. The process of quantitatively evaluating the sweetness of the target layer in a single well was repeated to obtain the sweetness index of different target layers in the single well. Based on the calculation results of the sweetness index, a columnar section of the geological sweetness profile of the single well was generated.

6. A device for predicting geological sweet spots in carbonate reservoirs, characterized in that, include: The preprocessing unit is used to acquire and preprocess well logging data and seismic data of the target layer in a single well to obtain a multi-input feature sequence; The prediction unit is used to input the multi-input feature sequence into a pre-trained geological sweet spot evaluation parameter prediction model to obtain multiple geological sweet spot evaluation parameter prediction values ​​for the target layer of the single well. The optimization unit is used to construct physical constraints based on the predicted values ​​of the geological sweet spot evaluation parameters of a single well and the gas production, and embed them into the loss function of the geological sweet spot evaluation parameter prediction model to obtain the geological sweet spot evaluation parameter prediction model with physical constraints in order to solve the optimal weight coefficients of the geological sweet spot evaluation parameters of each well. And an evaluation unit, used to perform a weighted summation of the predicted values ​​of the multiple geological sweet spot evaluation parameters and their corresponding optimal weight coefficients to achieve a quantitative evaluation of the sweet spot of the target layer in the single well; The training process of the geological sweet spot evaluation parameter prediction model includes: Integrate well logging data, seismic data, and core test data to establish a sample dataset and divide it into training and test sets; A deep multi-task learning neural network is established. This network includes a shared feature extraction module, a feature fusion module, and a multi-task module. The shared feature extraction module consists of a low-level feature extraction layer and a shared feature mapping layer. The low-level feature extraction layer extracts low-level feature data from multi-source input data, and the shared feature mapping layer maps the low-level feature data to generate shared features serving multiple tasks and task-specific features. The feature fusion module first calculates the fusion weights of the shared features for each task using an attention mechanism, and then uses a gating network to apply these weights to the shared features of multiple tasks and the task-specific features. The weighted summation yields the final dedicated fusion features for each task. The multi-task module consists of multiple task heads, each of which takes into account the dedicated fusion features of its corresponding prediction task and outputs a predicted value for a geological sweet spot evaluation parameter. To meet the needs of different tasks, a one-dimensional convolutional neural network is combined with a fully connected layer, and dedicated hyperparameters are configured for each task head to achieve independent learning and training of tasks. By introducing an attention mechanism and a dynamic gating strategy, the weights of the loss functions of each task are adjusted in real time, and the loss of each task is weighted to obtain a weighted loss. The weighted losses of all tasks are summed to construct the total loss function of the deep multi-task learning neural network. The deep multi-task learning neural network is trained using the training set to establish a prediction model for the geological sweet spot evaluation parameters.

7. The device for predicting geological sweet spots in carbonate reservoirs according to claim 6, characterized in that, Also includes: A visualization unit is used to display a bar chart of geological sweet spot profiles of a single well, which is generated based on the calculated sweetness index of different target layers in the single well.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method for predicting geological sweet spots in carbonate reservoirs as described in any one of claims 1-5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for predicting geological sweet spots in carbonate reservoirs as described in any one of claims 1-5.

Images