Fault vertical sealing performance prediction method

Abstract

The invention discloses a fault vertical sealing performance prediction method, and belongs to the field of oil-gas exploration. Multi-source data acquisition comprises data source expansion: remote sensing data, micro-seismic monitoring data and underground fluid monitoring; dynamic geological historical data: adding geological historical evolution information in a region when geological data is collected; according to the method, information dimensions for prediction are enriched through multi-source data acquisition, remote sensing, micro-seismic monitoring, underground fluid monitoring and dynamic geological historical data, so that the model can analyze the fault vertical closure from a more comprehensive angle, and the prediction precision is improved. An advanced data fusion algorithm is adopted, different types of data are effectively integrated, data contradictions and errors are avoided, and a foundation is laid for accurate prediction. When the comprehensive prediction model is constructed, the processing capability and prediction accuracy of the model on complex geological data are improved by applying a hybrid model, integrated learning and deep reinforcement learning.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas exploration, and in particular to a method for predicting the vertical sealing property of faults. Background Art

[0002] In the process of oil and gas exploration and development, the vertical sealing property of faults plays a key role in the migration, accumulation, and preservation of oil and gas. Accurately predicting the vertical sealing property of faults can effectively guide the deployment of oil and gas exploration, improve the exploration success rate, and reduce the exploration cost. However, traditional methods for predicting the vertical sealing property of faults often only consider one or a few factors, and it is difficult to comprehensively and accurately reflect the true situation of the vertical sealing property of faults under complex geological conditions. For example, some methods only judge based on the geometric shape of the fault, ignoring the influence of factors such as formation pressure and fluid properties on the sealing property; there are also some methods that consider some factors, but lack in-depth analysis of the interaction between various factors, resulting in a large deviation between the prediction results and the actual situation.

[0003] Therefore, it is of great practical significance to develop a method for predicting the vertical sealing property of faults that can comprehensively consider various factors and is more accurate and reliable. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a method for predicting the vertical sealing property of faults, which solves the problems raised in the above background art.

[0005] Technical Solution: To solve the above technical problems, according to one aspect of the present invention, more specifically, a method for predicting the vertical sealing property of faults includes the following steps:

[0006] S1. Multi-source data acquisition

[0007] a. Expand data sources: remote sensing data, microseismic monitoring data, underground fluid monitoring;

[0008] b. Dynamic geological history data: When collecting geological data, add the geological history evolution information in the region;

[0009] S2. Data fusion and preprocessing

[0010] a. Advanced data fusion algorithms: deep learning algorithms and deep learning-based multimodal data fusion algorithms;

[0011] b. Data cleaning and standardization:

[0012] Adopt data standardization and normalization techniques to eliminate the scale differences of different data sources;

[0013] Introduce data denoising techniques, and use wavelet transform and principal component analysis to remove noise and enhance the effectiveness of data;

[0014] c. Feature parameter extraction:

[0015] Extract geological feature parameters closely related to the vertical sealing of faults, including but not limited to: formation pressure coefficient, rock permeability, anisotropy, fault geometry, etc.;

[0016] Apply an automatic feature extraction algorithm, adopt feature selection based on gradient boosting and deep learning models, and automatically extract key parameters from a large amount of data to reduce human intervention;

[0017] S3. Build a comprehensive prediction model

[0018] a. Use hybrid models: convolutional neural network, recurrent neural network, graph neural network;

[0019] b. Ensemble learning:

[0020] Combine multiple machine learning models, including random forest, support vector machine, XGBoost, fuse the advantages of different models, and use ensemble learning to improve the prediction accuracy and stability;

[0021] c. Deep reinforcement learning:

[0022] Use deep reinforcement learning to optimize the parameter adjustment and iterative training of the prediction model, so that the model can continuously adjust the prediction strategy in a complex environment to improve the prediction accuracy;

[0023] S4. Model validation and correction

[0024] a. Cross-validation and bootstrap method:

[0025] Use a combination of K-fold cross-validation and the bootstrap method for validation, and ensure the stability and universality of the model through different dataset and model parameter settings;

[0026] Use the leave-one-out method for validation to enhance the performance of the model in small sample datasets and avoid overfitting problems;

[0027] b. Model performance evaluation:

[0028] Use multiple evaluation metrics for model evaluation, including accuracy, recall rate, F1 score;

[0029] Analyze the error types of the model through the confusion matrix, and find the blind spots and improvement space of the model;

[0030] c. Dynamic feedback and model correction:

[0031] According to the newly obtained test data and actual monitoring data, continuously optimize the model and make real-time corrections;

[0032] S5. Visualization of prediction results

[0033] a. 3D Visualization: Visualize the prediction results through 3D geological modeling tools to generate a three-dimensional fault structure diagram, showing the fault sealing conditions at different depths, and helping geologists better understand the dynamic changes of faults;

[0034] b. Dynamic Update and Real-time Monitoring: Combine real-time monitoring data to dynamically correlate the prediction results with the actual monitoring results and update the prediction model in a timely manner;

[0035] c. Multi-scale Analysis: Combine data at different scales to show the changes in sealing characteristics from local to overall during the visualization process, providing multi-level analysis perspectives.

[0036] Furthermore, the extended data in the S1 step includes the following:

[0037] The remote sensing data: Obtain large-scale geological data through satellite remote sensing and aerial geological exploration technologies, such as regional topography, sedimentary environment changes, etc., providing richer background information for the prediction model;

[0038] The microseismic monitoring data: Collect microseismic monitoring data to capture information on fault activities; these data can reveal the potential activity and fluid movement of faults, providing a reference for vertical sealing evaluation;

[0039] The underground fluid monitoring: Collect fluid monitoring data such as hydrology and oil and gas, and analyze the control effect of faults on fluid flow, which is very important for predicting vertical sealing.

[0040] Furthermore, the geological historical evolution information in the region includes but is not limited to factors such as crustal movement, sedimentary environment, and tectonic stress field, in order to dynamically evaluate the influence of these factors on the formation of fault vertical sealing.

[0041] Furthermore, the advanced data fusion algorithms in the S2 step include the following:

[0042] The deep learning algorithms, including convolutional neural networks and recurrent neural networks, to automatically learn the complex relationships between different data sources; convolutional neural networks can extract spatial features in seismic reflection features, and recurrent neural networks can capture dynamic changes in time series data;

[0043] The deep learning-based multi-modal data fusion algorithm effectively fuses data from different sources, thereby avoiding contradictions and errors between data and improving the accuracy of data integration.

[0044] Furthermore, the hybrid model in the S2 step includes the following:

[0045] The convolutional neural network: For spatial data such as seismic reflections and gravity anomalies, a convolutional neural network is used to extract local spatial features;

[0046] The recurrent neural network: Used to process time series data, such as formation pressure changes, fluid flow, etc., to capture dynamic features;

[0047] The graph neural network: Using the graph neural network to process fault data with complex topological structures, identify the relationships and interactions between different faults, thereby improving the model's adaptability to fault geometries.

[0048] Furthermore, in step S5, virtual reality technology is applied in the three-dimensional visualization, enabling users to view the vertical sealing data of faults immersively and enhancing the user experience.

[0049] The beneficial effects of a method for predicting the vertical sealing of faults according to the present invention are as follows:

[0050] (1) By setting up multi-source data collection, incorporating remote sensing, microseismic monitoring, underground fluid monitoring, and dynamic geological history data, the present invention enriches the information dimension for prediction, enabling the model to analyze the vertical sealing of faults from a more comprehensive perspective, thereby improving the prediction accuracy. By adopting an advanced data fusion algorithm, different types of data are effectively integrated to avoid data contradictions and errors, further improving the data quality and laying a foundation for accurate prediction. When constructing a comprehensive prediction model, a hybrid model, ensemble learning, and deep reinforcement learning are used to give full play to the advantages of different models and learning methods, improving the model's processing ability for complex geological data and prediction accuracy.

[0051] (2) Through three-dimensional visualization, using three-dimensional geological modeling tools and virtual reality technology, the prediction results are presented in a three-dimensional and intuitive manner, helping geologists better understand the dynamic changes of faults. Dynamic update, real-time monitoring, and multi-scale analysis enable the prediction results to be dynamically associated with actual monitoring, showing the changes in sealing characteristics at different scales and providing a multi-level analysis perspective for a more in-depth study of the vertical sealing of faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The present invention will be further described in detail below with reference to the drawings and specific implementation methods.

[0053] Figure 1 It is a schematic structural diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] The present invention will be described in detail below with reference to the drawings and embodiments. It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0055] To make the technical solution of the present invention clearer, the following further describes the present invention in detail with reference to the accompanying drawings and specific embodiments.

[0056] Referring to Figure 1 , a method for predicting the vertical sealing property of a fault includes the following steps:

[0057] S1. Multi-source data collection

[0058] a. Expand data sources: remote sensing data, microseismic monitoring data, underground fluid monitoring;

[0059] b. Dynamic geological history data: When collecting geological data, add the geological history evolution information within the region;

[0060] S2. Data fusion and preprocessing

[0061] a. Advanced data fusion algorithms: deep learning algorithms and deep learning-based multimodal data fusion algorithms;

[0062] b. Data cleaning and standardization:

[0063] Adopt data standardization and normalization techniques to eliminate the scale differences of different data sources;

[0064] Introduce data denoising techniques, such as wavelet transform or principal component analysis (PCA), to remove noise and enhance the effectiveness of data;

[0065] c. Feature parameter extraction:

[0066] Extract geological feature parameters closely related to the vertical sealing property of the fault, including but not limited to: formation pressure coefficient, rock permeability, anisotropy, fault geometry, etc.;

[0067] Apply automatic feature extraction algorithms, such as feature selection based on gradient boosting (GBDT) and deep learning models, to automatically extract key parameters from a large amount of data and reduce human intervention;

[0068] S3. Construct an integrated prediction model

[0069] a. Use hybrid models: convolutional neural network (CNN), recurrent neural network (RNN), graph neural network (GNN);

[0070] b. Ensemble learning:

[0071] Combine multiple machine learning models, including random forest, support vector machine (SVM), XGBoost, fuse the advantages of different models, and use ensemble learning to improve the prediction accuracy and stability;

[0072] c. Deep reinforcement learning:

[0073] Use deep reinforcement learning (DRL) to optimize the parameter adjustment and iterative training of the prediction model, enabling the model to continuously adjust the prediction strategy in complex environments and improve the prediction accuracy;

[0074] S4. Model Validation and Correction

[0075] a. Cross-validation and Bootstrap method:

[0076] Use a combination of K-fold cross-validation and the Bootstrap method for validation. Through different dataset and model parameter settings, ensure the stability and universality of the model;

[0077] Utilize the Leave-One-Out validation method to enhance the model's performance in small-sample datasets and avoid overfitting problems;

[0078] b. Model performance evaluation:

[0079] Use multiple evaluation metrics for model evaluation, including precision, recall rate, and F1 score;

[0080] Analyze the error types of the model through the confusion matrix to identify the blind spots and improvement areas of the model;

[0081] c. Dynamic feedback and model correction:

[0082] Continuously optimize the model and perform real-time correction based on newly acquired test data and actual monitoring data;

[0083] S5. Visualization of Prediction Results

[0084] a. 3D visualization: Through 3D geological modeling tools, visualize the prediction results, generate a three-dimensional fault structure diagram, and display the fault sealing conditions at different depths to help geologists better understand the dynamic changes of faults;

[0085] b. Dynamic update and real-time monitoring: Combine real-time monitoring data, dynamically associate the prediction results with the actual monitoring results, and update the prediction model in a timely manner;

[0086] c. Multi-scale analysis: Combine data at different scales (such as micro-scale core data and macro-scale seismic data), and display the changes in sealing characteristics from local to overall during the visualization process to provide a multi-level analysis perspective;

[0087] Continuously optimize the model and perform real-time correction according to newly acquired test data and actual monitoring data. At the same time, adopt online learning and transfer learning methods to enable the model to adapt to new geological regions or conditions and maintain the ability to continuously track and predict the vertical sealing changes of faults.

[0088] Preferably, the extended data in the S1 step includes the following:

[0089] The remote sensing data: Obtain large-scale geological data through satellite remote sensing and aerial geological exploration technologies, such as the topographic and geomorphic features and sedimentary environment changes in the region, to provide richer background information for the prediction model;

[0090] The microseismic monitoring data: Collect microseismic monitoring data to capture information on fault activities; these data can reveal the potential activity of faults and fluid movement, providing a reference for the evaluation of vertical sealing;

[0091] The underground fluid monitoring: Collect fluid monitoring data such as hydrology and oil and gas, and analyze the control effect of faults on fluid flow, which is very important for predicting vertical sealing.

[0092] Preferably, the geological historical evolution information in the region includes but is not limited to factors such as crustal movement, sedimentary environment, and tectonic stress field, so as to dynamically evaluate the influence of these factors on the formation of fault vertical sealing.

[0093] Preferably, the advanced data fusion algorithms in the S2 step include the following:

[0094] The deep learning algorithms, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), to automatically learn the complex relationships between different data sources; CNNs can extract spatial features in seismic reflection features, while RNNs can capture dynamic changes in time series data;

[0095] The deep learning-based multimodal data fusion algorithm, which effectively fuses data from different sources (such as geological, geophysical, and geochemical data), thus avoiding contradictions and errors between data and improving the accuracy of data integration.

[0096] Preferably, the hybrid model in the S2 step includes the following:

[0097] The convolutional neural network (CNN): For spatial data such as seismic reflection and gravity anomalies, use the convolutional neural network to extract local spatial features;

[0098] The recurrent neural network (RNN): Used to process time series data, such as formation pressure changes and fluid flow, to capture dynamic features;

[0099] The graph neural network (GNN): Use the graph neural network to process fault data with complex topological structures, identify the mutual relationships and interactions between different faults, and thus improve the adaptability of the model to fault geometries.

[0100] Preferably, in the step S5, the three-dimensional visualization applies virtual reality (VR) technology, enabling users to immerse themselves in viewing the vertical sealing data of the fault, enhancing the user experience.

[0101] The above embodiments merely represent several implementation manners of the present invention, and the description thereof is relatively specific and detailed. However, it should not be construed as a limitation on the scope of the patent for the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention shall be subject to the appended claims.

Claims

1. A method for predicting the vertical sealing of a fault, characterized in that: The steps include: S1. Multi-source data collection a. Extended data sources: remote sensing data, microseismic monitoring data, underground fluid monitoring; b. Dynamic geological historical data: When collecting geological data, add the geological historical evolution information in the region; S2. Data fusion and preprocessing a. Advanced data fusion algorithm: deep learning algorithm and multimodal data fusion algorithm based on deep learning; b. Data cleaning and standardization: Use data standardization and normalization techniques to eliminate scale differences among different data sources; Introducing data noise reduction technology, using wavelet transform and principal component analysis to remove noise and enhance data effectiveness; c. Feature parameter extraction: Extract geological characteristic parameters closely related to the vertical sealing of faults, including but not limited to: formation pressure coefficient, rock permeability, anisotropy, fault geometry, etc.; Apply automatic feature extraction algorithms and feature selection based on gradient boosting and deep learning models to automatically extract key parameters from large amounts of data and reduce human intervention; S3. Building a comprehensive prediction model a. Use hybrid models: convolutional neural networks, recurrent neural networks, and graph neural networks; b. Ensemble learning: Combine multiple machine learning models, including random forest, support vector machine, and XGBoost, to integrate the advantages of different models and use ensemble learning to improve prediction accuracy and stability; c. Deep reinforcement learning: Use deep reinforcement learning to optimize the parameter adjustment and iterative training of the prediction model, so that the model can improve the prediction accuracy by continuously adjusting the prediction strategy in complex environments; S4. Model verification and correction a. Cross-validation and bootstrap method: K-fold cross validation and bootstrap method were used for validation, and different data sets and model parameter settings were used to ensure the stability and universality of the model. Use the leave-one-out validation method to enhance the performance of the model in small sample data sets and avoid overfitting problems; b. Model performance evaluation: Use multiple evaluation indicators to evaluate the model, including precision, recall, and F1 score; Analyze the error types of the model through the confusion matrix to find the blind spots and room for improvement of the model; c. Dynamic feedback and model correction: Continuously optimize the model and make real-time corrections based on newly acquired test data and actual monitoring data; S5. Visualization of prediction results a. 3D visualization: Through 3D geological modeling tools, the prediction results are visualized to generate a three-dimensional fault structure map, showing the closure of faults at different depths, helping geologists better understand the dynamic changes of faults; b. Dynamic update and real-time monitoring: Combine real-time monitoring data to dynamically associate the prediction results with the actual monitoring results and update the prediction model in a timely manner; c. Multi-scale analysis: Combine data at different scales to display the changes in closure characteristics from local to overall in the visualization process, providing a multi-level analysis perspective.

2. A method for predicting vertical sealing of a fault according to claim 1, characterized in that: The extended data in step S1 includes the following: Remote sensing data: obtain a wide range of geological data through satellite remote sensing and aerial geological exploration technology, such as regional topography, sedimentary environment changes, etc., to provide richer background information for the prediction model; The microseismic monitoring data: collects microseismic monitoring data to capture information on fault activity; These data can reveal the potential activity and fluid movement of the fault, providing a reference for vertical sealing assessment; The underground fluid monitoring: collecting hydrological, oil and gas and other fluid monitoring data, analyzing the control effect of faults on fluid flow, which is very important for predicting vertical sealing.

3. A method for predicting vertical sealing of a fault according to claim 1, characterized in that: The geological history evolution information in the region includes but is not limited to factors such as crustal movement, sedimentary environment, tectonic stress field, etc., so as to dynamically evaluate the impact of these factors on the formation of vertical closure of faults.

4. A method for predicting vertical sealing of a fault according to claim 1, characterized in that: The advanced data fusion algorithm in step S2 includes the following: The deep learning algorithm includes convolutional neural networks and recurrent neural networks to automatically learn complex relationships between different data sources; the convolutional neural network can extract spatial features in seismic reflection features, and the recurrent neural network can capture dynamic changes in time series data; The multimodal data fusion algorithm based on deep learning effectively fuses data from different sources, thereby avoiding contradictions and errors between data and improving the accuracy of data integration.

5. A method for predicting vertical sealing of a fault according to claim 1, characterized in that: The mixed model in step S2 includes the following: The convolutional neural network: for spatial data such as seismic reflection and gravity anomaly, a convolutional neural network is used to extract local spatial features; The recurrent neural network is used to process time series data, such as formation pressure changes, fluid flow, etc., to capture dynamic characteristics; The graph neural network: uses the graph neural network to process fault data with complex topological structures, identify the relationships and interactions between different faults, and thus improve the adaptability of the model to fault geometry.

6. A method for predicting vertical sealing of a fault according to claim 1, characterized in that: The three-dimensional visualization in step S5 uses virtual reality technology to allow users to immersively view the vertical closure data of the fault, thereby enhancing the user experience.

Images