Intelligent identification method and system for strike-slip fault

Abstract

The invention discloses a strike-slip fault intelligent identification method and system, and belongs to the field of strike-slip fault intelligent identification, and the method comprises the steps: obtaining historical geological data, carrying out the distributed storage of the geological data according to the characteristics of the historical geological data, and obtaining a storage data set; preprocessing the stored data set to obtain a fault feature data set; constructing a multi-modal neural network model, and inputting the fault feature data set into the multi-modal neural network model for training to obtain a strike-slip fault recognition model; acquiring real-time geological data, inputting the real-time geological data into the strike-slip fault recognition model for calculation, and acquiring a strike-slip fault recognition result; and carrying out spatio-temporal data analysis mining on the strike-slip fault identification result, generating an evolution law and a correlation mode of the fault in time and space, and combining the evolution law and the correlation mode of the fault in time and space with the strike-slip fault identification result to obtain an intelligent identification result.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent identification of strike-slip faults, and in particular relates to an intelligent identification method and system for strike-slip faults. Background Art

[0002] There are many types of strike-slip faults on the earth, and they are widely distributed. For example, the San Andreas Fault and the Tanlu Fault are all famous strike-slip faults. Bates et al. formally proposed the strike-slip fault in 1987, which is a type of fault in which the movement direction of the two plates of the fault is mostly parallel or nearly parallel to the fault strike. Now geologists define strike-slip faults as faults of any scale. As long as the movement direction of the two plates of the fault is nearly parallel to the fault strike, they are all called strike-slip faults. However, there are very few strike-slip faults with only strike activity or normal faults with only dip activity in nature. There are also many faults with relative movement between the two plates between pure strike-slip and pure dip-slip. Strike-slip faults are usually large in scale, deep in fracture, and poor in sealing. They are a double-edged sword. In terms of geological storage, they are not conducive to the safe storage of carbon dioxide, but in the field of oil and gas exploration, they are conducive to oil and gas migration. Due to the difficulty in accurately identifying the type of strike-slip faults, it has brought great challenges to the basic geological understanding of carbon dioxide geological storage, oil exploration, tunnel and bridge engineering, and geological site selection for large engineering projects.

[0003] In actual geology, the breakpoints of strike-slip faults are mostly difficult to identify, and the structural style of strike-slip faults is relatively variable. It is difficult to apply the original structural style, and the application method is largely influenced by the subjective experience of researchers. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides an intelligent identification method for strike-slip faults, comprising:

[0005] Acquire historical geological data, and perform distributed storage on the geological data according to the characteristics of the historical geological data to obtain a storage data set;

[0006] Preprocessing and feature engineering the stored data set to obtain a fault feature data set;

[0007] Constructing a multimodal neural network model based on a support vector machine and a convolutional neural network, inputting the fault feature data set into the multimodal neural network model for training, and obtaining a strike-slip fault identification model;

[0008] Acquiring real-time geological data, inputting the real-time geological data into the strike-slip fault identification model for calculation, and obtaining an identification result of the strike-slip fault;

[0009] The identification result of the strike-slip fault is subjected to spatiotemporal data analysis and mining to generate the evolution law and correlation pattern of the fault in time and space, and the evolution law and correlation pattern of the fault in time and space are combined with the identification result of the strike-slip fault to obtain an intelligent identification result.

[0010] Preferably, the process of obtaining the storage data set further includes: uniformly storing and managing the historical geological data by combining a distributed file system and a NoSQL database to obtain the storage data set.

[0011] Preferably, the process of preprocessing and feature engineering the stored data set to obtain the fault feature data set includes:

[0012] Performing denoising and normalization processing on the stored data set to obtain a preprocessed data set;

[0013] Extracting key information reflecting fault characteristics from the preprocessed data set to generate a feature vector; the key information includes seismic waveform and stratum structure;

[0014] A feature extraction sub-network based on deep learning converts the feature vector into a residual mapping feature;

[0015] Extracting boundary structure information of the residual mapping feature to obtain a fault interpretation result;

[0016] The fault interpretation results at different levels are fused based on a multi-layer output fusion model to generate the fault feature data set.

[0017] Preferably, the process of converting the feature vector into a residual mapping feature by the deep learning-based feature extraction subnetwork includes: processing the feature vector through multiple convolutional layers and fully connected layers based on the feature extraction subnetwork, and then comparing the input feature vector with the feature obtained by network transformation, adjusting the input based on the comparison result, mapping the adjusted feature to the target feature space, and obtaining the residual mapping feature.

[0018] Preferably, the process of obtaining the strike-slip fault identification model includes:

[0019] Constructing a support vector machine model and a convolutional neural network model, training the support vector machine model using the fault feature data set, and obtaining a support vector machine fault recognition model;

[0020] Inputting the fault feature data set into the support vector machine fault recognition model for calculation to obtain a preliminary recognition data set;

[0021] Inputting the preliminary recognition data set into the convolutional neural network model for training to obtain a convolutional fault recognition model;

[0022] The output end of the support vector machine fault identification model is connected to the input end of the convolution fault identification model to obtain the strike-slip fault identification model.

[0023] Preferably, the process of acquiring real-time geological data, inputting the real-time geological data into the strike-slip fault identification model for calculation, and obtaining the identification result of the strike-slip fault includes:

[0024] Preprocessing the real-time data and inputting it into the strike-slip fault identification model for forward propagation, and identifying the existence and activity characteristics of the strike-slip fault based on the calculation of the strike-slip fault identification model;

[0025] The existence and activity characteristics of the strike-slip fault are transferred to the three-dimensional map data to obtain a final strike-slip fault identification result, and a three-dimensional visualization model of the strike-slip fault is constructed based on the final strike-slip fault identification result.

[0026] Preferably, the process of obtaining the intelligent recognition result includes:

[0027] Acquiring the spatiotemporal evolution characteristics of the fault based on the three-dimensional visualization model of the strike-slip fault, and constructing a spatiotemporal data analysis model based on the spatiotemporal evolution characteristics of the fault;

[0028] Inputting the identification results of the strike-slip fault into the spatiotemporal data analysis model will generate fault interpretation results of different scales;

[0029] Based on the multi-layer output fusion technology, the fault interpretation results at different scales are integrated to obtain the spatiotemporal data analysis results;

[0030] The development trend of the fault is predicted based on the spatiotemporal data analysis results to generate the intelligent identification results.

[0031] On the other hand, the present invention also provides an intelligent identification system for strike-slip faults, comprising:

[0032] A data acquisition module is used to acquire historical geological data, and store the geological data in a distributed manner according to the characteristics of the historical geological data to obtain a storage data set;

[0033] A data processing module, used for preprocessing and feature engineering the stored data set to obtain a fault feature data set;

[0034] A model building module, used to build a multimodal neural network model based on a support vector machine and a convolutional neural network, input the fault feature data set into the multimodal neural network model for training, and obtain a strike-slip fault identification model;

[0035] An identification result calculation module is used to obtain real-time geological data, input the real-time geological data into the strike-slip fault identification model for calculation, and obtain the identification result of the strike-slip fault;

[0036] The prediction result calculation module is used to perform spatiotemporal data analysis and mining on the identification results of the strike-slip fault, generate the evolution law and correlation pattern of the fault in time and space, combine the evolution law and correlation pattern of the fault in time and space with the identification results of the strike-slip fault, and obtain an intelligent identification result.

[0037] On the other hand, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the intelligent identification method of the strike-slip fault when executing the computer program.

[0038] On the other hand, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program implements the intelligent identification method of the strike-slip fault when executed by a processor.

[0039] Compared with the prior art, the present invention has the following advantages and technical effects:

[0040] The present invention constructs a multimodal neural network model by combining support vector machine (SVM) and convolutional neural network (CNN), which can better fuse information from different modes (such as historical data and real-time data). This multimodal fusion can effectively improve the robustness and accuracy of the model, especially when facing complex geological data, the model can integrate multiple feature sources, thereby reducing the limitations of a single mode. SVM is good at processing high-dimensional feature data (such as fault features), while CNN is good at extracting features from images or spatial data. Combining the two can give full play to their respective advantages and obtain more accurate model predictions. By inputting real-time geological data into a trained strike-slip fault identification model, the intelligent identification results of strike-slip faults can be obtained instantly. This is very critical for fields such as disaster warning and environmental monitoring, and can respond to geological changes in real time and provide timely feedback to decision makers. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0042] Figure 1 A schematic diagram of a method flow chart of an embodiment of the present invention;

[0043] Figure 2 Schematic diagram of the system structure of an embodiment of the present invention. DETAILED DESCRIPTION

[0044] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0045] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0046] Embodiment 1

[0047] like Figure 1-2 As shown, this embodiment provides an intelligent identification method for strike-slip faults, including:

[0048] Step S101, in view of the characteristics of massive geological data, a distributed file system and a NoSQL database are combined to perform unified storage and management, laying the foundation for subsequent data processing and analysis.

[0049] According to the characteristics of massive geological data, a distributed file system is used to store raw seismic data, and metadata is managed through a NoSQL database to build a unified data storage and management platform. Raw seismic data is read from the distributed file system, and quality control and standardization of the data are performed through the data preprocessing module to obtain a standardized seismic data set.

[0050] Furthermore, in modern seismic data processing, distributed file systems are used to store raw seismic data because these data are usually high in volume and high in dimension, and it is difficult for a single-machine storage system to meet their storage and processing requirements. For example, using the Hadoop Distributed File System (HDFS) can effectively store and manage seismic data because HDFS supports data sharding storage across multiple servers, which not only improves the data read and write speed, but also enhances the system's fault tolerance. For metadata management, using NoSQL databases such as MongoDB can provide more flexible data models and faster query performance to process unstructured or semi-structured metadata. For example, the metadata of seismic data may include information such as acquisition time, location, depth, etc., which can be flexibly stored and queried through MongoDB's document storage model. The role of the data preprocessing module is to ensure the quality and consistency of the data input into the analysis model. For example, pre-training on earthquake data in California and then fine-tuning applied to the interpretation of earthquake data in Sichuan, China. For example, using spatial database extensions in geographic information systems (GIS), such as PostGIS, can efficiently perform spatial queries and support complex geological data analysis tasks. Storing the comprehensive geological interpretation results in a NoSQL database and performing parallel processing and visualization through a distributed computing framework such as Apache Spark can effectively support geological experts in fault analysis.

[0051] Step S102, by preprocessing and feature engineering the original geological data, key information that can reflect the fault characteristics, such as seismic waveforms, stratigraphic structures, and historical fault activity records, is extracted to prepare for building a machine learning model.

[0052] The data quality and availability are improved by preprocessing the raw seismic data, including denoising and normalization. Based on the preprocessed seismic data, key information reflecting the fault characteristics, such as seismic waveforms and stratigraphic structures, is extracted to construct feature vectors. A feature extraction subnetwork based on deep learning is used to further extract residual mapping features related to strike-slip faults from the constructed feature vectors. The boundary structure information in the residual map is extracted through the structure extraction subnetwork to obtain preliminary results of fault interpretation. If there is seismic reflection discontinuity caused by noise or non-fault structure in the preliminary results, a denoising convolution subnetwork is used for processing to ensure the accuracy and clarity of the prediction results.

[0053] Furthermore, in seismic data processing, the first thing to do is data preprocessing, including denoising and normalization operations. For example, a bandpass filter is used to remove high-frequency noise and low-frequency interference in the seismic signal, making the effective signal in the seismic data more prominent. Normalization processing is to adjust the data to the same numerical range, such as scaling the amplitude values ​​of all data points to between 0 and 1. This can avoid errors caused by different data dimensions in subsequent analysis. Next, key information is extracted from the preprocessed seismic data to construct a feature vector. In this step, features such as the velocity, amplitude, and frequency of seismic waves can be extracted. For example, by analyzing the propagation speed of seismic waves in different strata, changes in underground structures can be inferred. Deep learning methods are used to further extract features, especially residual mapping features related to strike-slip faults.

[0054] Step S103, constructing a suitable machine learning model, such as a support vector machine and a convolutional neural network, based on the extracted key features, and realizing intelligent identification of strike-slip faults by training and optimizing model parameters.

[0055] Through the feature extraction sub-network, the residual mapping features related to the strike-slip fault are extracted from the 3D seismic data to obtain the fault feature data. According to the extracted fault feature data, the support vector machine model is used to train the fault feature data to obtain the support vector machine fault recognition model. The structure extraction sub-network is used to extract the boundary structure residual mapping of the 3D seismic data to obtain the fault interpretation target data. The support vector machine fault recognition model is applied to the fault interpretation target data to perform preliminary identification of the fault and obtain the preliminary identification results. If there is noise or seismic reflection discontinuity caused by non-fault structure in the preliminary identification results, the preliminary identification results are denoised by the denoising convolution sub-network, otherwise the next step is directly entered. According to the identification results after denoising, the convolutional neural network model is used to train the network with a large number of training samples and their labels to obtain the fault intelligent detection network model. Through the multi-layer output fusion technology, the output results of the support vector machine fault recognition model and the convolutional neural network fault intelligent detection model are fused to improve the robustness and generalization of the classification interpretation of faults of different scales. By adopting the transfer learning method, the trained fault intelligent detection network model is applied to the new 3D seismic data, and the strike-slip faults in the new data are intelligently identified to obtain the final strike-slip fault identification results. Based on the final strike-slip fault identification results, a 3D visualization model of the strike-slip fault is generated to provide an intuitive reference for subsequent geological analysis and interpretation.

[0056] Furthermore, by using the feature extraction subnetwork, residual mapping features related to strike-slip faults can be extracted from 3D seismic data. For example, assuming that the 3D seismic data of a certain area contains multiple seismic event records, the feature extraction subnetwork gradually extracts the high-frequency and low-frequency components in the seismic waveform through a combination of convolutional layers and pooling layers, and then generates residual mapping feature maps. These feature maps can highlight abnormal signals in the fault area, such as the reflection, refraction and attenuation characteristics of seismic waves at the fault. Based on the extracted fault feature data, a support vector machine (SVM) model is used for training. Assume that there is a set of fault feature data sets containing 1,000 samples, of which 500 samples are marked as fault areas and 500 samples are marked as non-fault areas. Through SVM model training, a decision boundary that can distinguish between fault and non-fault areas can be obtained. For example, the SVM model may find a hyperplane that effectively separates fault samples and non-fault samples in the feature space, thereby constructing a support vector machine fault recognition model. Using the structure extraction subnetwork, information can be further extracted from the boundary structure residual map of 3D seismic data. Assuming that the seismic data of a certain area shows obvious seismic wave reflection discontinuities at the fault boundary, the structure extraction subnetwork can highlight these discontinuities through specific convolution kernels and activation functions, and generate boundary structure residual maps. These maps can be used as target data for fault interpretation and provide a basis for subsequent fault identification. The support vector machine fault identification model is applied to the fault interpretation target data for preliminary identification. For example, assuming that there are multiple potential fault areas in the boundary structure residual map of a certain area, the SVM model classifies the features of these areas and identifies which areas are more likely to be faults. The preliminary identification results may show multiple fault candidate areas, but they may contain seismic reflection discontinuities caused by noise or non-fault structures. If there are seismic reflection discontinuities caused by noise or non-fault structures in the preliminary identification results, they are processed by the denoising convolution subnetwork. Assuming that in the preliminary identification results of a certain area, some areas are mistakenly identified as faults due to noise interference, the denoising convolution subnetwork can effectively remove these noises through specific denoising algorithms, such as denoising autoencoders based on convolutional neural networks, and improve the accuracy of the identification results. Based on the recognition results after denoising, a convolutional neural network (CNN) model is used for training. Assume that there is a training data set containing 2,000 samples, of which 1,000 samples are labeled as fault areas and 1,000 samples are labeled as non-fault areas. By training the CNN network with a large number of training samples and their labels, a network model that can intelligently detect faults in three-dimensional seismic data can be obtained. For example, the CNN model may gradually extract the features of the fault area through multiple layers of convolution and pooling operations, and finally output a probability map indicating the probability that each pixel belongs to a fault.The output results of the support vector machine fault recognition model and the convolutional neural network fault intelligent detection model are fused through multi-layer output fusion technology. For example, suppose the SVM model identifies multiple potential fault areas in the recognition results of a certain area, and the CNN model also identifies similar fault areas in the recognition results of the same area. Through the multi-layer output fusion technology, the output results of the two models can be weighted and fused to improve the robustness and generalization of the classification interpretation of faults of different scales. The trained fault intelligent detection network model is applied to new 3D seismic data by transfer learning. For example, suppose there is a CNN model trained in a certain area, and now the model needs to be applied to 3D seismic data in another area. Through transfer learning, the knowledge of the existing model can be used to intelligently identify the strike-slip faults in the new data, reduce training time and computing resources, and improve the generalization ability of the model. According to the final strike-slip fault recognition results, a 3D visualization model of the strike-slip fault is generated. For example, suppose the final recognition results show that there are multiple strike-slip faults in a certain area. Through 3D visualization technology, these faults can be displayed in 3D space to generate a 3D model of the fault. These models can provide intuitive references for geologists, helping them to better understand the spatial distribution and geological structure of faults, and provide strong support for subsequent geological analysis and interpretation. Through the above steps and technical means, not only can the accuracy and robustness of fault identification be improved, but also more intuitive and reliable data support can be provided for geological analysis and interpretation, thus playing an important role in practical applications.

[0057] Step S104, based on the machine learning model, introduce spatiotemporal data analysis and mining technology, make full use of the spatiotemporal characteristics of geological data, and discover the evolution laws and association patterns of faults in time and space through methods such as spatiotemporal clustering and association rule mining, so as to improve the accuracy and interpretability of fault identification.

[0058] A convolutional neural network model based on deep learning is used to preprocess and extract features of 3D seismic data to obtain residual mapping features related to strike-slip faults. The extracted residual mapping features are clustered by spatiotemporal clustering algorithm, and potential fault areas are identified based on the clustering results. The association rule mining technology is used to analyze the evolution law and association pattern of the fault area in time and space to obtain the spatiotemporal evolution characteristics of the fault. According to the spatiotemporal evolution characteristics of the fault, a spatiotemporal data analysis model is constructed to model and predict the development process of the fault. The structure extraction subnetwork is used to extract the boundary structure of the potential fault area and generate the residual mapping of the fault interpretation. The generated residual mapping is denoised by the denoising convolution subnetwork to improve the accuracy and clarity of the fault interpretation results. The multi-layer output fusion technology is used to fuse the fault interpretation results at different scales to improve the robustness of the classification interpretation of faults of different scales. Through transfer learning, the existing fault identification model is used to optimize and adjust the current model to improve the generalization ability of the model. The optimized fault identification model is applied to new seismic data to automatically identify and interpret strike-slip faults, and the development trend of faults is predicted based on the results of spatiotemporal data analysis.

[0059] Furthermore, a convolutional neural network model based on deep learning is used to preprocess and extract features of three-dimensional seismic data to obtain residual mapping features related to strike-slip faults. For example, in the preprocessing stage, the original seismic data can be converted into a format suitable for convolutional neural network processing through normalization and denoising techniques. The convolutional neural network extracts local features in the data layer by layer through multi-layer convolution and pooling operations, and finally generates residual mapping features related to strike-slip faults. These features can reflect the changes in different frequencies and amplitudes in seismic data, thereby revealing potential fault information. The extracted residual mapping features are clustered by spatiotemporal clustering algorithms, and potential fault areas are identified based on the clustering results. Specifically, the residual mapping features can be divided into several clusters using clustering algorithms such as K-means or DBSCAN. Each cluster represents a potential fault area. For example, assuming that 10 clusters are obtained through cluster analysis, the feature values ​​of some clusters are significantly higher than those of other clusters. These clusters may correspond to the main active areas of strike-slip faults. Association rule mining technology is used to analyze the evolution law and association pattern of fault areas in time and space, and obtain the spatiotemporal evolution characteristics of faults. For example, the Apriori algorithm can be used to mine frequent item sets and association rules between different fault areas. Assuming that a certain fault area frequently appears in the time series with several other areas, this may indicate that there is a certain geological connection between these areas, thereby revealing the spatiotemporal evolution law of faults. According to the spatiotemporal evolution characteristics of faults, a spatiotemporal data analysis model is constructed to model and predict the development process of faults. For example, a hidden Markov model (HMM) or a dynamic Bayesian network (DBN) can be used to model the spatiotemporal evolution process of faults. By training the model, the possible development of faults in a certain period of time in the future can be predicted. Assuming that the model predicts that the activity of a certain fault area will increase significantly in the next five years, this can provide an important reference for geological exploration and disaster prevention. The structure extraction subnetwork is used to extract the boundary structure of the potential fault area and generate a residual map for fault interpretation. For example, the structure extraction subnetwork can extract the boundary information of the potential fault area through edge detection algorithms, such as the Canny operator. The generated residual map can clearly show the geometric morphology and spatial distribution of the fault. The generated residual map is denoised by the denoising convolution subnetwork to improve the accuracy and clarity of the fault interpretation results. For example, the denoising convolution subnetwork can use an autoencoder or a generative adversarial network (GAN) to remove noise interference in the residual map. Assuming that there is a large amount of random noise in the original residual map, after denoising, the noise is significantly reduced and the fault boundary is clearer. The multi-layer output fusion technology is used to fuse the fault interpretation results at different scales to improve the robustness of the classification interpretation of faults of different scales. The structure extraction subnetwork is used to extract the boundary structure of the potential fault area and generate a residual map for fault interpretation.

[0060] Acquire three-dimensional seismic data, the three-dimensional seismic data including seismic reflection information related to the strike-slip fault; use a feature extraction subnetwork to process the three-dimensional seismic data, extract residual mapping features related to the strike-slip fault, and obtain residual mapping feature data; according to the residual mapping feature data, use a structure extraction subnetwork to extract boundary structures of potential fault areas to generate residual mappings for fault interpretation; if there is noise or seismic reflection discontinuity caused by non-fault structures in the residual mapping, use a denoising convolution subnetwork to denoise the residual mapping to obtain a denoised residual mapping; according to the denoised residual mapping, use a multi-layer output fusion technology to fuse the classification interpretation results of faults of different scales to obtain a fused fault classification interpretation result; use a transfer learning method to optimize the fused fault classification interpretation result using a pre-established fault classification interpretation model to obtain an optimized fault classification interpretation result;

[0061] Further, three-dimensional seismic data are obtained, which contain seismic reflection information related to strike-slip faults. In actual operation, seismic data are usually collected by seismic exploration equipment, which can send sound waves and receive sound waves reflected by underground structures to obtain information about underground structures. The feature extraction subnetwork processes these three-dimensional seismic data and extracts residual mapping features related to strike-slip faults. For example, low-frequency features may indicate large-scale faults, while high-frequency features may indicate small-scale faults. By fusing these features, fault data can be interpreted more comprehensively. Using the transfer learning method, the pre-established fault classification interpretation model is used to optimize the fused fault classification interpretation results. In this step, the model trained in other regions or conditions can be adjusted and applied to the current data set, and the parameters of the model can be fine-tuned to adapt to the new geological data environment. The optimized fault classification interpretation results will be used to construct a convolutional neural network based on image segmentation, and the model will be further trained to improve its accuracy and generalization ability. The three-dimensional seismic data to be identified is input into the fault intelligent identification model to obtain preliminary fault prediction results. If there is noise or seismic reflection discontinuity caused by non-fault structures in the prediction results, the model's adaptive learning ability will automatically adjust the network parameters to determine whether it is a real fault. During this adaptive adjustment process, the model may analyze the statistical characteristics of the seismic data and automatically adjust the recognition threshold and parameters to distinguish between real geological faults and data noise. Based on the accurate fault prediction results, 3D visualization technology is used to intuitively display the spatial distribution and geometric morphology of the fault. This can be achieved through 3D modeling software, such as using VTK (Visualization Toolkit) to create 3D models of faults, which can clearly show the location, shape and relationship of the fault relative to the surrounding strata. Finally, based on the 3D visualization model of the fault, combined with the experience and knowledge of geological experts, a comprehensive evaluation of the cause, activity and control of the fault on the oil and gas reservoir is carried out. Geological experts may analyze the type of fault, its activity history and its interaction with the surrounding strata, and evaluate its possible impact on the formation and distribution of oil and gas reservoirs, so as to obtain the final fault identification and interpretation results.

[0062] The generated residual map is denoised through a denoising convolutional subnetwork to improve the accuracy and clarity of the fault interpretation results.

[0063] The feature extraction subnetwork is used to process seismic data and extract residual mapping features related to strike-slip faults. Based on the extracted residual mapping features, the structure extraction subnetwork is used to extract the boundary structure and obtain the residual mapping for fault interpretation. The residual mapping generated by the feature extraction subnetwork and the structure extraction subnetwork is input into the denoising convolution subnetwork. The denoising convolution subnetwork denoises the input residual mapping to remove the noise generated by the network. The denoised residual mapping is used to obtain a fault interpretation result with higher accuracy and clarity. The multi-layer output fusion technology is used to fuse the fault interpretation results at different levels to improve the robustness of the classification interpretation of faults of different scales. Through transfer learning, the existing fault interpretation model is used to fine-tune the current network to improve the generalization ability of the network. A convolutional neural network based on image segmentation is constructed, and a large number of training samples and their labels are used to train the network to continuously optimize the network parameters. The trained and optimized network model is obtained for intelligent detection and interpretation of faults in three-dimensional seismic data, avoiding the problem of sensitivity to seismic reflection discontinuity caused by noise or non-fault structures in traditional methods.

[0064] Step S105, in order to support the rapid processing and analysis of massive geological data, a distributed computing framework such as Hadoop and Spark is used to optimize the data retrieval and filtering performance by designing an efficient data partitioning and indexing mechanism to meet the real-time query and analysis requirements during the fault identification process.

[0065] According to the characteristics of massive geological data, the distributed computing framework Hadoop and Spark are used to design efficient data partitioning and indexing mechanisms to optimize data retrieval and filtering performance. By constructing a convolutional neural network based on image segmentation and training the network with a large number of training samples and their labels, a network model that can intelligently detect faults in three-dimensional seismic data is obtained. If it is necessary to extract residual mapping features related to strike-slip faults, a feature extraction subnetwork is used to extract these features from seismic data. If the residual mapping of boundary structures is to be extracted to achieve the goal of fault interpretation, a structure extraction subnetwork is used to complete this task. In order to remove the noise generated by the network cumulatively and ensure the accuracy and clarity of the prediction results, a denoising convolution subnetwork is used for processing. Through multi-layer output fusion technology and transfer learning, the robustness and generalization of the classification interpretation of faults of different scales are enhanced, and the accuracy and adaptability of fault identification are improved. In the prediction process, the network model can effectively avoid the problem of sensitivity to seismic reflection discontinuity caused by noise or non-fault structures in traditional methods, and improve the reliability of fault identification. According to the characteristics of the distributed computing framework, by designing a reasonable data partitioning and indexing strategy, the data distribution and access efficiency in the cluster are optimized, and the data processing and analysis in the fault identification process are accelerated. By adopting the distributed computing framework and the optimized data partitioning indexing mechanism, combined with the fault identification method based on deep learning, an efficient, accurate and robust strike-slip fault identification system is finally obtained to meet the needs of massive geological data processing and real-time analysis.

[0066] Furthermore, according to the characteristics of massive geological data, the distributed computing frameworks Hadoop and Spark are used to design efficient data partitioning and indexing mechanisms to optimize data retrieval and filtering performance. For example, in Hadoop, geological data can be partitioned horizontally and vertically to divide the data according to the time and space dimensions, so that the size of each data block is moderate and convenient for parallel processing. Assuming that there is 1PB of seismic data, it can be divided into 1000 1TB data blocks through horizontal partitioning, each of which contains seismic records within a specific time period. Then, each data block is further subdivided into multiple sub-blocks through vertical partitioning, each of which contains data in a specific geographical area. In this way, when seismic data of a specific time and area needs to be retrieved, the system only needs to access a few data blocks, which significantly improves the retrieval efficiency. By constructing a convolutional neural network based on image segmentation and training the network with a large number of training samples and their labels, a network model that can intelligently detect faults in three-dimensional seismic data is obtained. For example, a convolutional neural network with a U-Net structure can be designed, which performs well in image segmentation tasks. Suppose there are 10,000 annotated 3D seismic data samples, each containing labels for faults and non-fault regions. Through training, the network model can learn the features of the fault region, so that it can automatically identify faults in new seismic data. If it is necessary to extract residual map features related to strike-slip faults, a feature extraction subnetwork is used to extract these features from the seismic data. For example, the feature extraction subnetwork can contain multiple convolutional layers and pooling layers to extract high-frequency and low-frequency features from the seismic data layer by layer. Suppose that through the feature extraction subnetwork, a vector containing 100 features is obtained, which can reflect the geometry and spatial distribution of the strike-slip fault. If the residual map of the boundary structure is to be extracted to achieve the goal of fault interpretation, the structure extraction subnetwork is used to complete this task. For example, the multi-layer output fusion technique can integrate low-frequency and high-frequency features, where low-frequency features mainly reflect large-scale faults and high-frequency features mainly reflect small-scale faults. Suppose that through the multi-layer fusion network, a comprehensive feature vector is obtained, which can simultaneously identify faults of different scales. Through transfer learning, the weights of existing models trained in other regions can be applied to the seismic data in the current region through fine-tuning technology, reducing training time and data requirements. In the prediction process, the network model can effectively avoid the problem of traditional methods being sensitive to seismic reflection discontinuities caused by noise or non-fault structures, and improve the reliability of fault identification. For example, traditional methods may mistakenly identify noise as faults, while deep learning-based models can learn the difference between noise and faults through training, thereby reducing misidentification.For example, suppose the system processes 1PB of seismic data. Through a distributed computing framework, the data processing time is shortened from days to hours. Through a deep learning model, the accuracy of fault identification is increased from 80% to 95%, significantly improving the efficiency and accuracy of geological exploration.

[0067] Step S106, comprehensively utilize big data technology, machine learning algorithms and spatiotemporal databases to build an overall solution for intelligent identification of strike-slip faults. Through seamless integration of data management, feature extraction, model training, spatiotemporal analysis and other links, an end-to-end process from massive geological data to precise fault identification is realized, providing strong support for geological disaster warning and prevention.

[0068] Big data technology is used to collect, store and manage massive geological data, and a spatiotemporal database is constructed to provide a data basis for subsequent fault identification. Through the feature extraction subnetwork, residual mapping features related to strike-slip faults are extracted from seismic data to obtain key information for fault identification. According to the structure extraction subnetwork, the residual mapping of the boundary structure is extracted to achieve the goal of fault interpretation and obtain preliminary fault structure information. The denoising convolution subnetwork is used to remove the noise generated by the network cumulatively, ensure the accuracy and clarity of the prediction results, and obtain high-quality fault identification results. Through multi-layer output fusion technology and transfer learning, the robustness and generalization of the classification interpretation of faults of different scales are enhanced, and the adaptability and reliability of fault identification are improved. A convolutional neural network based on image segmentation is constructed, and the network is trained using a large number of training samples and their labels to obtain a network model that can intelligently detect faults in three-dimensional seismic data. In the prediction process, the network model can effectively avoid the problem of sensitivity to seismic reflection discontinuity caused by noise or non-fault structures in traditional methods, and improve the accuracy of fault identification. By comprehensively applying machine learning algorithms, the extracted fault features are classified and identified, and the end-to-end process from massive geological data to accurate fault identification is realized through the seamless integration of time-space analysis and other links. The identification results are combined with geological disaster warning and prevention, and through visualization and data analysis technology, strong support is provided for the prediction, assessment and decision-making of geological disasters, which improves the scientificity and effectiveness of geological disaster prevention and control.

[0069] Furthermore, big data technology is used to collect, store and manage massive geological data, build a spatiotemporal database, and provide a data basis for subsequent fault identification. For example, a distributed file system such as HDFS is used to store TB-level seismic data generated in geological exploration, and the spatiotemporal index of these data is managed through NoSQL databases such as HBase. In this way, researchers can efficiently query geological data within a specific time and space range to provide accurate data support for fault identification. Through the feature extraction subnetwork, residual mapping features related to strike-slip faults are extracted from seismic data to obtain key information for fault identification. Specifically, assuming a set of three-dimensional seismic data, the feature extraction subnetwork will gradually extract features such as edges and textures of the fault area through convolutional layers and pooling layers to generate a residual map. For example, in the seismic data of a certain area, the feature extraction subnetwork may identify an obvious linear feature, which indicates that there may be a strike-slip fault in the area. According to the structure extraction subnetwork, the residual map of the boundary structure is extracted to achieve the goal of fault interpretation and obtain preliminary fault structure information. The structure extraction subnetwork further refines the output of the feature extraction subnetwork and focuses on the boundary structure of the fault. For example, by using more sophisticated convolution kernels and deconvolution operations, the network can clearly outline the specific location and direction of the fault and generate a high-precision fault boundary map. The denoising convolution subnetwork is used to remove the noise generated by the network, ensure the accuracy and clarity of the prediction results, and obtain high-quality fault identification results. In practical applications, seismic data often contains a lot of noise, which affects the recognition effect. The denoising convolution subnetwork effectively filters out noise and improves image quality through a specific convolution structure and activation function. For example, there may be random noise points in the original seismic image. After denoising, the image becomes smoother and the fault features are more prominent. Through multi-layer output fusion technology and transfer learning, the classification interpretation of faults of different scales is enhanced. Robustness and generalization power are enhanced, and the adaptability and reliability of fault identification are improved. Multi-layer output fusion technology fuses feature maps at different levels, comprehensively utilizes global and local information, and improves recognition results. For example, the fusion of detailed information in shallow feature maps and global information in deep feature maps can more comprehensively describe fault characteristics. Transfer learning uses pre-trained models to fine-tune new tasks to improve the generalization ability of the model. For example, using a convolutional neural network model pre-trained on a large amount of image data and fine-tuning it for fault identification can quickly converge and achieve good results. Construct a convolutional neural network based on image segmentation, and use a large number of training samples and their labels to train the network to obtain a network model that can intelligently detect faults in three-dimensional seismic data. For example, thousands of seismic images with fault locations marked are collected as training data. Through network training, the model gradually learns to identify fault features. During the training process, the cross entropy loss function is used to evaluate the difference between the model's prediction results and the true labels, and the model parameters are continuously optimized through the back-propagation algorithm.In the prediction process, the network model can effectively avoid the problem of traditional methods being sensitive to seismic reflection discontinuities caused by noise or non-fault structures, and improve the accuracy of fault identification. Traditional methods such as manual interpretation are often affected by subjective factors and are sensitive to noise. The deep learning-based model can automatically distinguish between noise and real fault signals through large amounts of data training. For example, in a complex geological area, traditional methods may misjudge noise as a fault, while the deep learning model can accurately identify the real fault. The machine learning algorithm is used in combination to classify and identify the extracted fault features, and through the seamless integration of time-space analysis and other links, an end-to-end process from massive geological data to accurate fault identification is achieved.

[0070] On the other hand, this embodiment also provides an intelligent identification system for strike-slip faults, including:

[0071] A data acquisition module is used to acquire historical geological data, and store the geological data in a distributed manner according to the characteristics of the historical geological data to obtain a storage data set;

[0072] A data processing module, used for preprocessing and feature engineering the stored data set to obtain a fault feature data set;

[0073] A model building module, used to build a multimodal neural network model based on a support vector machine and a convolutional neural network, input the fault feature data set into the multimodal neural network model for training, and obtain a strike-slip fault identification model;

[0074] An identification result calculation module is used to obtain real-time geological data, input the real-time geological data into the strike-slip fault identification model for calculation, and obtain the identification result of the strike-slip fault;

[0075] The prediction result calculation module is used to perform spatiotemporal data analysis and mining on the identification results of the strike-slip fault, generate the evolution law and correlation pattern of the fault in time and space, combine the evolution law and correlation pattern of the fault in time and space with the identification results of the strike-slip fault, and obtain an intelligent identification result.

[0076] On the other hand, this embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the intelligent identification method of the strike-slip fault when executing the computer program.

[0077] On the other hand, this embodiment further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and is characterized in that when the computer program is executed by a processor, the intelligent identification method of the strike-slip fault is implemented.

[0078] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An intelligent identification method for strike-slip faults, characterized in that: include: Acquire historical geological data, and perform distributed storage on the geological data according to the characteristics of the historical geological data to obtain a storage data set; Preprocessing and feature engineering the stored data set to obtain a fault feature data set; Constructing a multimodal neural network model based on a support vector machine and a convolutional neural network, inputting the fault feature data set into the multimodal neural network model for training, and obtaining a strike-slip fault identification model; Acquiring real-time geological data, inputting the real-time geological data into the strike-slip fault identification model for calculation, and obtaining an identification result of the strike-slip fault; The identification result of the strike-slip fault is subjected to spatiotemporal data analysis and mining to generate the evolution law and correlation pattern of the fault in time and space, and the evolution law and correlation pattern of the fault in time and space are combined with the identification result of the strike-slip fault to obtain an intelligent identification result.

2. The intelligent identification method of a strike-slip fault according to claim 1, characterized in that: The process of obtaining the storage data set also includes: uniformly storing and managing the historical geological data by combining a distributed file system and a NoSQL database to obtain the storage data set.

3. The intelligent identification method of a strike-slip fault according to claim 1, characterized in that: The process of preprocessing and feature engineering the stored data set to obtain the fault feature data set includes: Performing denoising and normalization processing on the stored data set to obtain a preprocessed data set; Extracting key information reflecting fault characteristics from the preprocessed data set to generate a feature vector; the key information includes seismic waveform and stratum structure; A feature extraction sub-network based on deep learning converts the feature vector into a residual mapping feature; Extracting boundary structure information of the residual mapping feature to obtain a fault interpretation result; The fault interpretation results at different levels are fused based on a multi-layer output fusion model to generate the fault feature data set.

4. The intelligent identification method of a strike-slip fault according to claim 3, characterized in that: The process of converting the feature vector into residual mapping features by the deep learning-based feature extraction subnetwork includes: processing the feature vector through multiple convolutional layers and fully connected layers based on the feature extraction subnetwork, and then comparing the input feature vector with the features obtained through network transformation, adjusting the input based on the comparison result, mapping the adjusted features to the target feature space, and obtaining the residual mapping features.

5. The intelligent identification method of a strike-slip fault according to claim 1, characterized in that: The process of obtaining the strike-slip fault identification model includes: Constructing a support vector machine model and a convolutional neural network model, training the support vector machine model using the fault feature data set, and obtaining a support vector machine fault recognition model; Inputting the fault feature data set into the support vector machine fault recognition model for calculation to obtain a preliminary recognition data set; Inputting the preliminary recognition data set into the convolutional neural network model for training to obtain a convolutional fault recognition model; The output end of the support vector machine fault identification model is connected to the input end of the convolution fault identification model to obtain the strike-slip fault identification model.

6. The intelligent identification method of a strike-slip fault according to claim 1, characterized in that: The process of acquiring real-time geological data, inputting the real-time geological data into the strike-slip fault identification model for calculation, and obtaining the identification result of the strike-slip fault includes: Preprocessing the real-time data and inputting it into the strike-slip fault identification model for forward propagation, and identifying the existence and activity characteristics of the strike-slip fault based on the calculation of the strike-slip fault identification model; The existence and activity characteristics of the strike-slip fault are transferred to the three-dimensional map data to obtain a final strike-slip fault identification result, and a three-dimensional visualization model of the strike-slip fault is constructed based on the final strike-slip fault identification result.

7. The intelligent identification method of strike-slip fault according to claim 6, characterized in that: The process of obtaining the intelligent recognition result includes: Acquiring the spatiotemporal evolution characteristics of the fault based on the three-dimensional visualization model of the strike-slip fault, and constructing a spatiotemporal data analysis model based on the spatiotemporal evolution characteristics of the fault; Inputting the identification results of the strike-slip fault into the spatiotemporal data analysis model will generate fault interpretation results of different scales; Based on the multi-layer output fusion technology, the fault interpretation results at different scales are integrated to obtain the spatiotemporal data analysis results; The development trend of the fault is predicted based on the spatiotemporal data analysis results to generate the intelligent identification results.

8. An intelligent identification system for strike-slip faults, characterized in that: include: A data acquisition module is used to acquire historical geological data, and store the geological data in a distributed manner according to the characteristics of the historical geological data to obtain a storage data set; A data processing module, used for preprocessing and feature engineering the stored data set to obtain a fault feature data set; A model building module, used to build a multimodal neural network model based on a support vector machine and a convolutional neural network, input the fault feature data set into the multimodal neural network model for training, and obtain a strike-slip fault identification model; An identification result calculation module is used to obtain real-time geological data, input the real-time geological data into the strike-slip fault identification model for calculation, and obtain the identification result of the strike-slip fault; The prediction result calculation module is used to perform spatiotemporal data analysis and mining on the identification results of the strike-slip fault, generate the evolution law and correlation pattern of the fault in time and space, combine the evolution law and correlation pattern of the fault in time and space with the identification results of the strike-slip fault, and obtain an intelligent identification result.

9. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method described in any one of claims 1 to 7 is implemented.

Images