Oil and gas reservoir geological modeling and prediction method and system based on machine learning

Abstract

The invention provides an oil and gas reservoir geological modeling and prediction method and system based on machine learning. The precision and efficiency of oil and gas reservoir geological modeling are improved through an intelligent means. Through fusion of multi-source heterogeneous data such as logging, earthquake, geology and production dynamics, data feature extraction and selection are performed by using a deep learning algorithm, and a multi-strategy model construction method of supervised learning, unsupervised learning and semi-supervised learning is combined to realize high-precision prediction of reservoir parameters and fluid distribution of oil and gas reservoirs. The system adopts a modular design, integrates functional modules of data management, model training, geological modeling, prediction analysis, user interaction and the like, supports multi-source data import, automatic feature extraction, model optimization and prediction result visualization, and introduces uncertainty to analyze, evaluate and predict reliability. According to the method, adaptive feature extraction, transfer learning and multi-strategy model construction technologies are innovatively introduced, and the defects of a traditional method in the aspects of data processing, model generalization and prediction precision are effectively overcome.

Description

Technical Field

[0001] The present invention relates to the field of oil and gas exploration and development, and in particular to a method and system for geological modeling and prediction of oil and gas reservoirs using machine learning technology. Background Art

[0002] Reservoir geological modeling and prediction are key components of oil and gas exploration and development. Traditional geological modeling methods rely primarily on the experience of geologists and manual modeling tools, resulting in low efficiency and limited accuracy. In recent years, with the advancement of machine learning technology, its application in reservoir geological modeling has gradually gained attention. However, existing methods still have shortcomings in data processing, model construction, and prediction accuracy. These include insufficient integration of multi-source heterogeneous data, limited model generalization, and insufficient interpretability of prediction results.

[0003] Oil and gas reservoir geological modeling requires processing massive amounts of multi-source data, including well logging, seismic, geological, and production performance data. These data come from a wide range of sources, in diverse formats, and are subject to noise and uncertainty. Effectively fusing this multi-source data and extracting valuable features is a key challenge in improving geological modeling accuracy. Existing data fusion methods mostly rely on manual rules and are difficult to adapt to complex geological conditions and data variations.

[0004] Traditional geological modeling methods typically require extensive manual effort and empirical judgment when constructing models, making automation and efficiency difficult to achieve. In recent years, machine learning has been introduced into geological modeling, but existing methods mostly focus on a single learning strategy, either supervised or unsupervised, which has limited effectiveness in modeling multi-source data under complex geological conditions. Furthermore, existing models lack generalization capabilities and are difficult to adapt to new geological regions and data conditions. Therefore, a method and system for reservoir geological modeling and prediction that integrates multi-strategy learning, adaptive feature extraction, and uncertainty analysis is needed to improve modeling accuracy and efficiency and reduce exploration risks. Summary of the Invention

[0005] The present invention provides a method and system for oil and gas reservoir geological modeling and prediction based on machine learning. Through multi-source data fusion, adaptive feature extraction, multi-strategy model construction, uncertainty analysis and integrated system design, it solves the problems of insufficient data processing, limited model generalization ability and low prediction accuracy in the existing technology, significantly improves the accuracy and efficiency of oil and gas reservoir geological modeling, and reduces the risks of oil and gas exploration.

[0006] Technical points and specific processes

[0007] The multi-source data fusion and preprocessing:

[0008] The present invention first collects various types of geological data, including well logging data (resistivity, acoustic wave time difference, natural gamma, etc.), seismic data (amplitude, phase, frequency, etc.), geological data (lithology, sedimentary facies, stratigraphic interface) and production dynamic data (production, pressure, temperature). These data come from a wide range of sources and in various formats, and are subject to noise and uncertainty. In order to integrate these data, a deep learning algorithm (multimodal data fusion network) is used to align the features of the well logging data and seismic data to ensure the consistency of depth information and spatiotemporal information. The specific steps include:

[0009] Data alignment: Align the depth information of well logging data with the spatiotemporal information of seismic data, extract multi-scale features through deep learning algorithms (convolutional neural networks), and form a unified data representation.

[0010] Data cleaning: Use data cleaning tools to remove outliers and noise, detect and remove outliers in logging data through sliding window algorithms, and remove high-frequency noise in seismic data through wavelet transform.

[0011] Normalization: Normalize the fused data to ensure that the features of different data sources are in the same dimension. Normalize the resistivity of the well logging data to the range [0, 1], and the amplitude of the seismic data to the range [-1, 1].

[0012] Feature alignment: Align features from different data sources through deep learning algorithms (autoencoders) to ensure the continuity of features in space and time.

[0013] These steps provide a high-quality data foundation for subsequent model training and ensure data consistency and availability.

[0014] The adaptive feature extraction and selection:

[0015] After data preprocessing is completed, the present invention uses deep learning algorithms (autoencoders and convolutional neural networks) to perform adaptive feature extraction on the preprocessed data. The specific steps include:

[0016] Feature Extraction: An autoencoder is used to reduce the dimensionality of the data and extract low-dimensional features. The autoencoder compresses the high-dimensional well logging data into a low-dimensional space while retaining key information. Convolutional neural networks are also used to extract local features, with convolutional layers extracting edge and texture features from the seismic data.

[0017] Feature Assessment: Combined with geological domain knowledge, the extracted features are evaluated for importance. Principal component analysis (PCA) is used to assess the variance contribution of the features and select features with high contributions. PCA is used to analyze the features in the well logging data and select features with a variance contribution exceeding 85%.

[0018] Feature selection: Based on the experience of geologists, key features are manually marked to ensure the scientific nature of feature selection. Geologists can mark features related to sedimentary facies based on their experience with sedimentary environments.

[0019] Dynamic update mechanism: During model training, feature selection is dynamically adjusted based on training feedback. When the model performs poorly in certain geological areas, feature weights are automatically adjusted. The Bayesian optimization algorithm dynamically adjusts feature weights to optimize model performance.

[0020] Through these steps, the present invention can adaptively extract and select features related to oil and gas reservoir geological modeling, thereby improving the accuracy and generalization ability of the model.

[0021] The multi-strategy model construction and training:

[0022] Based on the data labels, the present invention selects supervised learning, unsupervised learning or semi-supervised learning strategies to construct geological models. The specific steps include:

[0023] Supervised learning: For labeled data (known reservoir parameters), deep convolutional networks or random forest algorithms are used for training. Deep convolutional networks extract spatial features from the data for classification or regression prediction of reservoir parameters. Convolutional and pooling layers extract spatial features from well logging data for porosity and permeability prediction. Random forest algorithms rank features by importance to optimize model performance. Random forest algorithms assess the importance of different features in well logging data for reservoir parameter prediction.

[0024] Unsupervised Learning: For unlabeled data (unlabeled seismic data), a Generative Adversarial Network (GAN) is used for unsupervised learning. The generator generates a preliminary model of the geological structure, and the discriminator evaluates the authenticity of the generated model. The generator generates geological interfaces in the seismic data, and the discriminator evaluates the authenticity of the generated interfaces. Through adversarial training, the generator and discriminator are optimized to produce high-quality geological models.

[0025] Semi-supervised learning: A self-training algorithm is used for semi-supervised learning with a small amount of labeled data and a large amount of unlabeled data. An initial model is trained using a small amount of labeled data. This initial model is then used to pseudo-label the unlabeled data, gradually expanding the labeled dataset. The self-training algorithm gradually labels the sedimentary facies types in the seismic data, and the model is continuously iterated to improve prediction accuracy.

[0026] Transfer learning: During model training, a transfer learning mechanism is introduced to extract key parameters from existing geological regional models as the initial parameters for the new regional model. These parameters are then fine-tuned on the new regional data to improve the model's generalization and reduce training time. The existing geological model parameters are transferred to the new geological region, and model performance is optimized through fine-tuning.

[0027] Hyperparameter Optimization: Utilizes Bayesian optimization algorithms to automatically tune model hyperparameters. By defining a search space for hyperparameters (such as learning rate and regularization coefficient), these hyperparameters are dynamically adjusted based on model performance feedback to improve training efficiency and prediction accuracy. Bayesian optimization algorithms are used to adjust the learning rate and regularization coefficient of deep convolutional networks to optimize model performance.

[0028] Through these steps, the present invention can select appropriate model building strategies according to different data situations and improve the accuracy and generalization ability of the model.

[0029] The geological modeling and prediction:

[0030] The trained model is applied to oil and gas reservoir geological modeling to generate a 3D geological model. The specific steps include:

[0031] Geological Modeling: The model automatically identifies the boundaries of geological units, sedimentary facies types, and lithologic distribution. The predicted results are used to generate a high-resolution 3D geological model, including reservoir porosity, permeability, and saturation distribution. The porosity and permeability distribution maps of the 3D geological model are generated using the model's predicted porosity and permeability distributions.

[0032] Parameter Prediction: Use the model to predict reservoir parameters (porosity, permeability, saturation) and fluid distribution in oil and gas reservoirs. Interpolate the predicted results to generate a high-resolution parameter distribution map. Use the Kriging interpolation algorithm to interpolate the predicted porosity and permeability to generate a high-resolution parameter distribution map.

[0033] Visualization: Use visualization tools to display geological model details and prediction results. Use 3D visualization tools to display the porosity, permeability, and saturation distribution of the geological model, facilitating analysis by geologists and engineers.

[0034] Uncertainty Analysis: This paper proposes a prediction result evaluation method based on uncertainty analysis, quantifying the uncertainty of prediction results through Monte Carlo simulation. Multiple random samplings are performed on the prediction results, and the standard deviation and confidence interval of the prediction results for each sampling are calculated. The uncertainty assessment report for the prediction results is then generated. Uncertainty analysis of the predicted porosity and permeability is performed through Monte Carlo simulation, and confidence intervals and probability distributions of the prediction results are generated, providing a scientific basis for oil and gas exploration decision-making.

[0035] Through these steps, the present invention can generate a high-precision three-dimensional geological model and perform uncertainty analysis on the prediction results, providing a scientific basis for oil and gas exploration decision-making.

[0036] The system architecture design:

[0037] This paper constructs an integrated oil and gas reservoir geological modeling and prediction system, covering multiple modules such as data management, model training, geological modeling, prediction analysis and user interaction. Specific module functions include:

[0038] Data Management Module: This module supports importing, exporting, querying, and updating data from multiple sources, and provides data preprocessing capabilities. It supports importing and exporting data in various formats (CSV, LAS, and SEG-Y), and offers data cleaning tools (for outlier detection and noise removal), normalization tools (for normalizing the resistivity of well logging data to the [0, 1] range), and feature alignment tools (for aligning features from different data sources using deep learning algorithms).

[0039] Model Training Module: This module integrates multiple machine learning algorithms, supports user-defined algorithms and parameter configuration, and provides model training, validation, and optimization capabilities. It supports training for multiple algorithms, including deep convolutional networks, random forests, and generative adversarial networks. It also supports user-defined feature selection rules and provides transfer learning tools (extracting parameters from existing models and migrating them to new areas), as well as hyperparameter optimization tools (dynamically adjusting hyperparameters using Bayesian optimization algorithms).

[0040] Geological Modeling Module: Generates 3D geological models based on model predictions and supports model visualization and editing. It automatically generates 3D geological models (including reservoir porosity, permeability, and saturation distribution maps) and provides visualization tools (3D visualization tools display geological model details) and model editing tools (users can manually adjust geological model boundaries and parameter distribution).

[0041] Prediction and Analysis Module: This module provides visualization of prediction results and uncertainty analysis capabilities, supporting user analysis and interpretation of prediction results. It provides high-resolution parameter distribution maps (porosity, permeability, and saturation distribution maps), uncertainty analysis tools (outputting confidence intervals and probability distributions of prediction results through Monte Carlo simulation), and prediction result analysis tools (allowing users to perform statistical and comparative analysis of prediction results).

[0042] User Interaction Module: Provides a user-friendly interface that supports operations such as data input, algorithm selection, parameter configuration, and result viewing. It also provides detailed help documentation and online technical support. It offers a graphical user interface (user-defined operations), detailed help documentation (user manual and online tutorials), and online technical support (users can obtain technical support through online customer service).

[0043] Through modular design, the system is easy to expand and maintain, providing users with a one-stop oil and gas reservoir geological modeling and prediction solution.

[0044] The above technical solution can bring the following technical effects:

[0045] 1. Significantly improve modeling accuracy: Through multi-source data fusion and adaptive feature extraction, the present invention can fully utilize multiple data sources such as well logging, seismic, geology and production dynamics to form a more comprehensive and accurate data foundation. Adaptive feature extraction technology combined with deep learning algorithms can automatically identify key features related to oil and gas reservoir geological modeling, and dynamically adjust feature selection to adapt to different geological conditions. This combination of multi-source data fusion and intelligent feature extraction significantly improves the accuracy of oil and gas reservoir geological modeling, enabling the generated three-dimensional geological model to more accurately reflect the actual situation of reservoir parameters (porosity, permeability, saturation) and fluid distribution, providing a more reliable geological basis for oil and gas exploration.

[0046] 2. Significantly improve work efficiency: Through an integrated system design, this invention integrates multiple modules, including data management, model training, geological modeling, predictive analysis, and user interaction, to achieve full-process automation of oil and gas reservoir geological modeling and prediction. The data management module supports the rapid import, export, and preprocessing of multi-source data. The model training module supports the automatic training and optimization of multiple machine learning algorithms. The geological modeling module can automatically generate high-precision three-dimensional geological models. The prediction and analysis module provides visual display and uncertainty analysis of prediction results. This integrated design greatly reduces manual operation and calculation time, improves work efficiency, and enables geologists and engineers to complete oil and gas reservoir geological modeling and prediction tasks more quickly.

[0047] 3. Effectively Reduce Exploration Risk: This invention incorporates uncertainty analysis technology, using Monte Carlo simulation to quantify the uncertainty of prediction results and output confidence intervals and probability distributions. This uncertainty analysis method helps geologists and engineers better understand the reliability of prediction results and assess the risks of oil and gas reservoir development. Based on this scientific uncertainty assessment, decision makers can more accurately formulate exploration strategies and optimize resource allocation, thereby effectively reducing the risks of oil and gas exploration and improving the success rate and economic benefits of exploration. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a step diagram of the oil and gas reservoir geological modeling and prediction method based on machine learning of the present invention DETAILED DESCRIPTION

[0049] In order to better understand the technical solution and advantages of the present invention, the following is a specific embodiment of the present invention. Figure 1 The present invention will be described in detail with respect to the application process of the oil and gas reservoir geological modeling and prediction method and system based on machine learning. This embodiment will be closely combined with the claims and the invention content to highlight the technical points and advantages of the present invention.

[0050] Background of the embodiment:

[0051] Suppose a new round of exploration and development is underway at an oil and gas field, requiring high-precision modeling and prediction of the reservoir's geological structure, reservoir parameters (porosity, permeability, saturation), and fluid distribution. Traditional methods struggle to meet these high-precision exploration requirements due to complex data processing, limited model accuracy, and a lack of uncertainty analysis. Therefore, the method and system proposed in this invention are employed for reservoir geological modeling and prediction.

[0052] Multi-source data fusion and preprocessing:

[0053] Data Collection:

[0054] First, multi-source geological data of the oil and gas field was collected, including:

[0055] Well logging data: resistivity, acoustic transit time, natural gamma, etc.

[0056] Seismic data: amplitude, phase, frequency, etc.

[0057] Geological data: lithology, sedimentary facies, stratigraphic interfaces, etc.

[0058] Production dynamic data: output, pressure, temperature, etc.

[0059] These data come from a wide range of sources, in various formats, and are subject to noise and uncertainty.

[0060] Data fusion and preprocessing:

[0061] A deep learning algorithm (multimodal data fusion network) is used to align the features of well logging data and seismic data to ensure the consistency of depth information and spatiotemporal information. A convolutional neural network is used to extract multi-scale features to form a unified data representation. The specific steps are as follows:

[0062] Data alignment: Align the depth information of well logging data with the spatiotemporal information of seismic data, and extract multi-scale features through deep learning algorithms.

[0063] Data cleaning: Use the sliding window algorithm to detect and remove abnormal points in the logging data, and use wavelet transform to remove high-frequency noise in the seismic data.

[0064] Normalization: Normalize the resistivity of logging data to the range of [0,1], and normalize the amplitude of seismic data to the range of [-1,1].

[0065] Feature alignment: Autoencoders are used to align features from different data sources to ensure the continuity of features in space and time.

[0066] Through these preprocessing steps, the quality and consistency of the data were significantly improved, laying a solid foundation for subsequent model training.

[0067] Adaptive feature extraction and selection:

[0068] Feature extraction:

[0069] Use an autoencoder to reduce the dimensionality of the preprocessed data and extract low-dimensional features. At the same time, use a convolutional neural network to extract local features to ensure the integrity and representativeness of the features. The specific steps are as follows:

[0070] Dimensionality reduction: Compress high-dimensional logging data into a low-dimensional space through an autoencoder to retain key information.

[0071] Local feature extraction: Extract edge features and texture features from seismic data through convolutional neural networks.

[0072] Feature evaluation and selection

[0073] Combined with geological domain knowledge, the importance of the extracted features is evaluated. Principal component analysis (PCA) is used to evaluate the variance contribution of the features and select features with higher contribution rates. The specific steps are as follows:

[0074] Variance contribution rate assessment: PCA was used to analyze the features in the logging data, and features with a variance contribution rate exceeding 85% were selected.

[0075] Manual marking of key features: Geologists manually mark features associated with sedimentary facies based on their experience with the depositional environment.

[0076] Dynamic update mechanism

[0077] During model training, feature selection is dynamically adjusted based on training feedback. When the model performs poorly in certain geological areas, feature weights are automatically adjusted. The Bayesian optimization algorithm dynamically adjusts feature weights to optimize model performance.

[0078] Through adaptive feature extraction and selection, the present invention can automatically identify and select features related to oil and gas reservoir geological modeling, significantly improving the accuracy and generalization ability of the model.

[0079] Multi-strategy model building and training:

[0080] Supervised Learning:

[0081] For labeled data (known reservoir parameters), deep convolutional networks or random forest algorithms are used for training. The specific steps are as follows:

[0082] Deep Convolutional Network: Extracts spatial features from well logging data through convolutional and pooling layers to predict porosity and permeability.

[0083] Random Forest Algorithm: Ranks features by importance to optimize model performance. Uses Random Forest to assess the importance of different features in well logging data for reservoir parameter prediction.

[0084] Unsupervised Learning:

[0085] For unlabeled data (unlabeled earthquake data), we use Generative Adversarial Network (GAN) for unsupervised learning. The specific steps are as follows:

[0086] Generator and Discriminator: The generator generates a preliminary model of the geological structure, and the discriminator evaluates the authenticity of the generated model. Through adversarial training, the generator and discriminator are optimized to produce high-quality geological models.

[0087] Semi-supervised learning:

[0088] For a small amount of labeled data and a large amount of unlabeled data, a self-training algorithm is used for semi-supervised learning. The specific steps are as follows:

[0089] Initial model training: Train the initial model using a small amount of labeled data.

[0090] Pseudo-label annotation: Use the initial model to pseudo-label the unlabeled data and gradually expand the labeled data set.

[0091] Iterative optimization: Continuously iterate and optimize the model to improve prediction accuracy.

[0092] Transfer Learning:

[0093] During the model training process, the transfer learning mechanism is introduced. The specific steps are as follows:

[0094] Parameter migration: Extract key parameters from the existing geological regional model as the initial parameters of the new regional model.

[0095] Fine-tuning optimization: Fine-tune the model on data from new regions to optimize model performance, reduce training time, and improve generalization capabilities.

[0096] Hyperparameter optimization:

[0097] Use the Bayesian optimization algorithm to automatically tune the model hyperparameters. The specific steps are as follows:

[0098] Hyperparameter search: Define the search space for hyperparameters (learning rate, regularization coefficient).

[0099] Dynamic Adjustment: Dynamically adjust hyperparameters based on model performance feedback to improve training efficiency and prediction accuracy.

[0100] Through multi-strategy model construction and training, the present invention can select appropriate model construction strategies according to different data situations, significantly improving the accuracy and generalization ability of the model.

[0101] Geological Modeling and Prediction:

[0102] Geological Modeling:

[0103] The trained model is applied to oil and gas reservoir geological modeling to generate a 3D geological model. The specific steps are as follows:

[0104] Model application: The model can automatically identify the boundaries of geological units, sedimentary facies types and lithologic distribution.

[0105] 3D model generation: Use the prediction results to generate a high-resolution 3D geological model, including the porosity, permeability, and saturation distribution of the reservoir.

[0106] Visualization: 3D visualization tools are used to display the details of the geological model, making it easier for geologists and engineers to analyze.

[0107] Parameter prediction:

[0108] The model is used to predict reservoir parameters (porosity, permeability, saturation) and fluid distribution of oil and gas reservoirs. The specific steps are as follows:

[0109] Interpolation processing: The predicted porosity and permeability are interpolated through the Kriging interpolation algorithm to generate a high-resolution parameter distribution map.

[0110] Visualization: Prediction results are displayed through visualization tools to facilitate analysis by geologists and engineers.

[0111] Uncertainty Analysis:

[0112] A prediction result evaluation method based on uncertainty analysis is proposed, which quantitatively analyzes the uncertainty of the prediction results through Monte Carlo simulation. The specific steps are as follows:

[0113] Monte Carlo simulation: Perform multiple random sampling of the prediction results and calculate the standard deviation and confidence interval of the prediction results for each sampling.

[0114] Uncertainty assessment: Output uncertainty assessment report of prediction results to provide a scientific basis for oil and gas exploration decision-making.

[0115] Through geological modeling and prediction, the present invention can generate a high-precision three-dimensional geological model and perform uncertainty analysis on the prediction results, providing a scientific basis for oil and gas exploration decision-making.

[0116] System architecture design:

[0117] This paper constructs an integrated oil and gas reservoir geological modeling and prediction system, covering multiple modules such as data management, model training, geological modeling, prediction analysis and user interaction. The specific module functions are as follows:

[0118] Data management module:

[0119] It supports importing, exporting, querying, and updating data from multiple sources, and provides data preprocessing capabilities. Specific functions are as follows:

[0120] Data import and export: Supports import and export of multiple data formats (CSV, LAS, SEG-Y).

[0121] Data preprocessing: Provides data cleaning tools (outlier detection and noise removal), normalization tools (normalize the resistivity of logging data to the range of [0, 1]), and feature alignment tools (align features from different data sources using deep learning algorithms).

[0122] Model training module:

[0123] It integrates multiple machine learning algorithms, supports user-defined algorithms and parameter configuration, and provides model training, verification, and optimization functions. Specific functions are as follows:

[0124] Algorithm integration: supports training functions for multiple algorithms such as deep convolutional networks, random forests, and generative adversarial networks.

[0125] User-defined: Supports user-defined feature selection rules.

[0126] Transfer learning: Provides transfer learning tools that support extracting parameters from existing models and migrating them to new regions.

[0127] Hyperparameter optimization: Provides hyperparameter optimization tools to dynamically adjust hyperparameters through Bayesian optimization algorithms.

[0128] Geological Modeling Module:

[0129] Generate a 3D geological model based on the model prediction results, and support model visualization and editing. Specific functions are as follows:

[0130] 3D model generation: Automatically generate 3D geological models, including reservoir porosity, permeability and saturation distribution maps.

[0131] Visualization tools: Provide 3D visualization tools to display the details of geological models.

[0132] Model editing: Provides model editing tools that allow users to manually adjust the boundaries and parameter distribution of geological models.

[0133] Prediction and Analysis Module:

[0134] Provides visualization of forecast results and uncertainty analysis functions, supporting users to analyze and interpret forecast results. Specific functions are as follows:

[0135] Visualization: Provides high-resolution parameter distribution diagrams to display prediction results.

[0136] Uncertainty Analysis: Provides uncertainty analysis tools to output confidence intervals and probability distributions of prediction results through Monte Carlo simulation.

[0137] Result analysis: Provides forecast result analysis tools, allowing users to perform statistical and comparative analysis on forecast results.

[0138] User interaction module:

[0139] It provides a user-friendly interface that supports operations such as inputting data, selecting algorithms, configuring parameters, and viewing results. It also provides detailed help documentation and online technical support. Specific functions are as follows:

[0140] Graphical interface: Provides a graphical user interface where users can customize operations.

[0141] Help Documentation: Provides detailed help documentation, including user manuals and online tutorials.

[0142] Technical support: Provide online technical support, users can obtain technical support through online customer service.

[0143] Through modular design, the system is easy to expand and maintain, providing users with a one-stop oil and gas reservoir geological modeling and prediction solution.

[0144] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for oil and gas reservoir geological modeling and prediction based on machine learning, characterized in that: The following steps are involved: S1. Collect multi-source geological data, including well logging data, seismic data, geological data, and production performance data; S2. fusing and preprocessing the multi-source geological data, including data cleaning, denoising, normalization, and feature alignment, to form a unified data set; S3. Use deep learning algorithms to perform adaptive feature extraction on preprocessed data to automatically identify key features relevant to reservoir geological modeling; S4. Based on the data labels, select supervised learning, unsupervised learning, or semi-supervised learning strategies to construct geological models: For labeled data, supervised learning algorithms such as deep convolutional networks or random forests are used for training; for unlabeled data, generative adversarial networks (GANs) are used for unsupervised learning; for small amounts of labeled data and large amounts of unlabeled data, self-training algorithms are used for semi-supervised learning; S5. During model training, a transfer learning mechanism is introduced to transfer existing geological knowledge and model parameters to new geological areas, reducing training time and improving model generalization capabilities; S6. Use Bayesian optimization algorithms to automatically tune model hyperparameters and optimize model performance. S7. Apply the trained model to oil and gas reservoir geological modeling to generate a three-dimensional geological model and predict reservoir parameters (porosity, permeability, saturation) and fluid distribution; S8. Conduct uncertainty analysis on the prediction results and evaluate the reliability of the prediction results through Monte Carlo simulation to provide a scientific basis for oil and gas exploration decision-making.

2. The method according to claim 1, characterized in that In step S2, a multimodal data fusion network based on deep learning is used to align the features of the well logging data and seismic data to form a unified data representation. Specifically, the process includes: aligning the depth information of the well logging data with the spatiotemporal information of the seismic data; extracting the multi-scale features of the data through a convolutional neural network; and normalizing the extracted features to ensure that the features of different data sources are in the same dimension.

3. The method according to claim 1, characterized in that In the S3 step, an autoencoder or a convolutional neural network is used to automatically extract key features from the data, and the importance of the features is evaluated and selected in combination with domain knowledge. Specifically, the steps include: using an autoencoder to reduce the dimensionality of the data and extract low-dimensional features; using a convolutional neural network to extract local features; and screening the extracted features based on geological domain knowledge to remove redundant features.

4. The method according to claim 1, wherein In the step S4, supervised learning algorithms include but are not limited to deep convolutional networks, support vector machines, and random forests; unsupervised learning algorithms include but are not limited to generative adversarial networks (GANs) and principal component analysis (PCA); semi-supervised learning algorithms include but are not limited to self-training algorithms and graph semi-supervised learning algorithms, specifically including: for labeled data, using deep convolutional networks to extract features and perform classification or regression; for unlabeled data, using generative adversarial networks to generate a preliminary model of geological structure; for a small amount of labeled data and a large amount of unlabeled data, using self-training algorithms to gradually expand the labeled data set.

5. The method according to claim 1, characterized in that In step S5, the existing geological model parameters are used as initial parameters and migrated to the new geological area for fine-tuning to adapt to different geological conditions. Specifically, the steps include: extracting key parameters from the model of the existing geological area; using these parameters as initial values ​​for the model of the new area; and fine-tuning the model on the data of the new area to optimize model performance.

6. The method according to claim 1, characterized in that In step S8, the uncertainty of the prediction results is quantitatively analyzed through Monte Carlo simulation, and the confidence interval and probability distribution of the prediction results are output, which specifically includes: performing multiple random sampling on the prediction results; calculating the standard deviation and confidence interval of the prediction results of each sampling; and outputting an uncertainty assessment report of the prediction results.

7. The method according to claim 1, characterized in that The method support system includes: a data management module for storing and managing multi-source geological data, supporting data import, export, query and update operations, and preprocessing the data, including data cleaning, denoising, normalization and feature alignment; a model training module, integrating multiple machine learning algorithms, supporting user-defined algorithms and parameter configuration, and providing model training, verification and optimization functions, including supervised learning, unsupervised learning and semi-supervised learning strategies; a geological modeling module, generating a three-dimensional geological model based on model prediction results, supporting model visualization and editing, and being able to automatically identify the boundaries of geological units, sedimentary phase types and lithologic distribution; a prediction and analysis module, providing visualization and uncertainty analysis functions for prediction results, supporting users to analyze and interpret prediction results, and evaluating the reliability of prediction results through Monte Carlo simulation; a user interaction module, providing a friendly user interface, supporting users to input data, select algorithms, configure parameters, view results and other operations, and providing detailed help documents and online technical support.

8. The system according to claim 7, characterized in that The data management module supports the fusion and preprocessing of multi-source data, and can integrate logging data, seismic data, geological data and production dynamic data to form a unified data set. Specifically, it supports the import and export of multiple data formats; provides data cleaning tools to remove outliers and noise; and provides data normalization tools to ensure that the characteristics of different data sources are in the same dimension.

9. The system according to claim 7, wherein: The model training module supports adaptive feature extraction, using deep learning algorithms to automatically extract key features and combining domain knowledge for feature selection. It also supports transfer learning mechanisms to migrate existing geological knowledge to new geological areas. Specifically, it provides training functions for autoencoders and convolutional neural networks; supports user-defined feature selection rules; and provides transfer learning tools to support the extraction of parameters from existing models and migration to new areas.

Images