A multi-scale multi-level fault detection method and system based on a skip connection network
By adopting a multi-scale, multi-level fault detection method based on skip connection networks, the problems of long processing time and low accuracy of traditional methods are solved. This method enables efficient and accurate detection of complex geological conditions and micro-faults, thereby improving the accuracy of geological modeling and reservoir prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PETROCHINA CO LTD
- Filing Date
- 2023-03-30
- Publication Date
- 2026-06-12
AI Technical Summary
Traditional tomography methods are time-consuming and have difficulty guaranteeing accuracy, while deep learning methods are not effective in identifying low signal-to-noise ratio and small tomography.
A multi-scale, multi-level fault detection method based on jump connection networks is adopted. By constructing multi-level jump connection and multi-scale, multi-level parallel networks, combined with fault fusion feature network and squeezing model, multi-scale, multi-level feature extraction and fusion are performed.
It improves the accuracy and efficiency of fault detection, especially in complex geological conditions and areas with small faults, and provides more reliable data for subsequent geological modeling and reservoir prediction.
Smart Images
Figure CN116990859B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of seismic signal processing technology, and in particular relates to a multi-scale, multi-level fault detection method and system based on a skip connection network. Background Technology
[0002] Oil and gas are generated by fault control and exist due to fault sealing; faults can cause oil and gas migration, dispersion, or divide reservoirs into several parts. Fault interpretation provides crucial foundational information for geological modeling and subsequent reservoir evaluation. Traditionally, fault identification relied on interpreters manually picking faults from seismic profiles using experience and some seismic attributes. This was often time-consuming and lacked accuracy. In recent years, to improve the efficiency and accuracy of fault interpretation, several automatic fault tracking technologies have been proposed. These typically involve edge detection on bedding slices, combined with attribute information, significantly improving efficiency compared to manual interpretation, but accuracy often falls short of production requirements. To address the problems of manual fault identification, deep learning technology has been introduced for fault detection.
[0003] Conventional deep learning-based tomography detection techniques utilize an end-to-end approach based on the U-Net neural network. This method trains the network model using 3D forward modeling data as training samples, and then directly applies the trained model to predict tomographic data. This approach yields good results with real-world data that has a high signal-to-noise ratio and is relatively simple in construction, but it performs poorly in identifying tomographic data with low signal-to-noise ratios and small tomographic features.
[0004] This invention proposes a novel multi-scale, multi-level fusion tomography detection method and system, which can significantly improve the detection accuracy of low signal-to-noise ratio and small faults. Summary of the Invention
[0005] The purpose of this invention is to provide a multi-scale, multi-level tomographic detection method and system based on skip connection networks in order to solve the above-mentioned problems.
[0006] The present invention achieves the above objectives through the following technical solutions:
[0007] A multi-scale, multi-level tomography detection method based on skip connection networks includes the following steps:
[0008] We construct a deep learning model for 3D fault detection, a fusion model for multi-scale and multi-level fault detection, and a compression model for extracting fault attribute compression features.
[0009] Seismic data is acquired and input into the deep learning model to obtain multiple three-dimensional fault bodies with different high precision.
[0010] The three-dimensional fault body is input into the fusion model to obtain fault fusion body data containing multi-scale and multi-level features;
[0011] The fault fusion body data is input into the extrusion model to obtain the fault data body, thereby realizing fault detection.
[0012] As a further optimization of the present invention, the specific process of constructing a deep learning model for three-dimensional tomographic detection is as follows:
[0013] Construct multi-level jumper connections and multi-scale, multi-level parallel networks;
[0014] Seismic sample data is acquired, and the multi-level jumper connection and multi-scale multi-level parallel network are trained using the seismic sample data to obtain a deep learning model.
[0015] As a further optimization of the present invention, the seismic sample data is obtained by three-dimensional synthesis.
[0016] As a further optimization of the present invention, the specific process of constructing a fusion model for multi-scale and multi-level fault detection is as follows:
[0017] Construct a fault-fusion feature network;
[0018] The earthquake sample data is input into the deep learning model to obtain multiple three-dimensional fault bodies with different high precision.
[0019] The sample three-dimensional fault volume is input into the fault fusion feature network for training to obtain the fusion model.
[0020] As a further optimization of the present invention, the multi-level jumper connection and multi-scale multi-level parallel network are used to fuse features of different levels.
[0021] As a further optimization of the present invention, the multi-level jumper connection and multi-scale multi-level parallel network fuse features of different levels, specifically by fusing large-scale features with small-scale features through a fully connected network, including the expansion of large-scale features to small-scale features and the compatibility of small-scale features to large-scale features.
[0022] As a further optimization of the present invention, the fault fusion feature network consists of an end-to-end convolutional neural network; multiple sample three-dimensional fault volumes are taken as input, and a mapping structure is obtained through the end-to-end network to output the fusion result.
[0023] As a further optimization of the present invention, the extrusion model uses multiple Long Short-Term Memory (LSTM) networks connected in series to obtain the extrusion result through an end-to-end network.
[0024] As a further optimization of the present invention, the specific process of inputting the fault fusion body data into the extrusion model to obtain the fault data body is as follows:
[0025] The fault fusion data is segmented into two-dimensional profile data along the inline direction for processing and prediction to obtain two-dimensional data.
[0026] The two-dimensional data is stitched together to generate three-dimensional data;
[0027] Repeat the above operation along the crossline direction to obtain the fault data volume.
[0028] A multi-scale, multi-level tomographic detection system based on skip connection networks includes:
[0029] The model building module is used to build a deep learning model for 3D fault detection, a fusion model for multi-scale and multi-level fault detection, and a compression model for fault attribute compression feature extraction.
[0030] The data acquisition module is used to acquire earthquake data;
[0031] The first processing module is used to input the seismic data into the deep learning model to obtain multiple three-dimensional fault bodies with different high precision.
[0032] The second processing module is used to input the three-dimensional fault body into the fusion model to obtain fault fusion body data containing multi-scale and multi-level features;
[0033] The third processing module is used to input the fault fusion body data into the extrusion model to obtain the fault data body, so as to realize fault detection.
[0034] An electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;
[0035] Memory, used to store computer programs;
[0036] The processor, when executing a program stored in memory, implements a multi-scale, multi-level tomographic detection method based on a skip connection network.
[0037] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a multi-scale, multi-level tomographic detection method based on a skip connection network.
[0038] The beneficial effects of this invention are as follows:
[0039] This invention considers both the accuracy of fault detection and the efficiency of computation, achieving more accurate fault detection results while maintaining computational efficiency. It is particularly effective in areas with complex geological conditions and well-developed micro-faults. Compared to traditional methods, this approach improves the efficiency and accuracy of fault interpretation, providing reliable data for subsequent geological modeling and reservoir prediction. Attached Figure Description
[0040] Figure 1 This is a flowchart of the method of the present invention;
[0041] Figure 2 This is a schematic diagram of the structure of the multi-level jumper connection and multi-scale multi-level parallel network of the present invention;
[0042] Figure 3 yes Figure 2 Enlarged view of part of the image;
[0043] Figure 4 This is a schematic diagram of the fault fusion feature network structure of the present invention;
[0044] Figure 5 This is a schematic diagram of the extrusion model of the present invention;
[0045] Figure 6 This is a schematic diagram of the fault profile identification results of the present invention;
[0046] Figure 7 This is a schematic diagram of the horizontal slice identification results of the present invention. Detailed Implementation
[0047] The present application will now be described in further detail with reference to the accompanying drawings. It should be noted that the following specific embodiments are only used to further illustrate the present application and should not be construed as limiting the scope of protection of the present application. Those skilled in the art can make some non-essential improvements and adjustments to the present application based on the above application content.
[0048] like Figure 1 As shown, a multi-scale, multi-level tomography detection method based on skip connection networks includes the following steps:
[0049] S1: Construct a deep learning model for 3D fault detection, a fusion model for multi-scale and multi-level fault detection, and a compression model for fault attribute compression feature extraction;
[0050] In this embodiment, the specific process is as follows:
[0051] Construct multi-level jumper connections and multi-scale, multi-level parallel networks;
[0052] Seismic sample data is acquired, which is synthesized using three-dimensional technology. The multi-level jumper connection and multi-scale multi-level parallel network are trained using the seismic sample data to obtain a deep learning model.
[0053] Construct a fault-fusion feature network;
[0054] The earthquake sample data is input into the deep learning model to obtain multiple three-dimensional fault bodies with different high precision.
[0055] The sample three-dimensional fault volume is input into the fault fusion feature network for training to obtain the fusion model.
[0056] The multi-level jumper connection and multi-scale multi-level parallel network can fuse features at different levels. Specifically, it can fuse large-scale features with small-scale features through a fully connected network, including the expansion of large-scale features to small-scale features and the compatibility of small-scale features to large-scale features.
[0057] The fault fusion feature network consists of an end-to-end convolutional neural network; multiple sample three-dimensional fault volumes are taken as input, and a mapping structure is obtained through the end-to-end network to output the fusion result.
[0058] The extrusion model uses multiple Long Short-Term Memory (LSTM) networks connected in series to obtain the extrusion result through an end-to-end network.
[0059] S2: Acquire seismic data, input the seismic data into the deep learning model, and obtain multiple three-dimensional fault bodies with different high precision;
[0060] S3: Input the three-dimensional fault body into the fusion model to obtain fault fusion body data containing multi-scale and multi-level features;
[0061] S4: Input the fault fusion body data into the extrusion model to obtain the fault data body, thereby realizing fault detection; in this embodiment, the specific process is as follows:
[0062] The fault fusion data is segmented into two-dimensional profile data along the inline direction for processing and prediction to obtain two-dimensional data.
[0063] The two-dimensional data is stitched together to generate three-dimensional data;
[0064] Repeat the above operation along the crossline direction to obtain the fault data volume.
[0065] In this case, the results obtained in step S2 are satisfactory for most of the data, and subsequent steps can be omitted. If none of the results in step S2 yield unsatisfactory interpretations, then steps S3-S4 can be used to fuse and extract the results from S2 to obtain satisfactory results. Each step from S2 to S4 can be run independently, or they can be combined to form the novel tomographic detection workflow of this invention.
[0066] 1. High-precision multi-scale, multi-level three-dimensional fault prediction
[0067] Three-dimensional fault identification based on convolutional neural networks often uses multiple convolutional layers to extract features from seismic data. Typical end-to-end networks concatenate different convolutional layers, using concatenation layers to connect high- and low-level features, which leads to significant accuracy loss. To improve prediction accuracy, this invention introduces a high-resolution convolutional neural network based on skip connection features. It replaces the concatenated network with a multi-level skip connection and a multi-scale, multi-level parallel network. Simultaneously, it uses "dilation" and "contraction" feature fusion techniques to fuse features of different levels, ensuring full utilization of features at each level. The network model is as follows: Figure 2 As shown.
[0068] Network structures that link large-scale and small-scale features across levels, such as... Figure 3 As shown, this method fuses large-scale features with small-scale features through a fully connected network. The solid lines in the figure represent the cross-link relationships between different levels, including the expansion of large-scale features to small-scale features and the compatibility of small-scale features to large-scale features. Unlike traditional algorithms that use direct addition, this algorithm better preserves feature information and improves the resolution of prediction results. The multi-scale, multi-level fusion operation is repeated. Multiple networks with different depths and levels of high and low resolution and features are used to improve prediction accuracy. The network depth and size are determined by the user based on actual test data and quality control. This special network structure can accommodate the different responses of large-scale and small faults, significantly improving the prediction accuracy of 3D models. The multi-resolution, multi-scale, multi-level strategy extracts large-scale features near faults while simultaneously extracting small-scale information to train the network. A fault fusion network is used to obtain a fault data volume that conforms to the geological background and expert expectations. Compared with traditional methods, this method improves the efficiency and accuracy of fault interpretation, providing reliable data for subsequent geological modeling and reservoir prediction.
[0069] 2. Extraction and multi-scale fusion of multi-level fault feature information
[0070] Multi-level jumper connections and multi-scale multi-level parallel networks predict multiple 3D fault data volumes. Small-scale network features have high spatial resolution and good response to small fractures, but due to their small sensitivity, they lack the ability to analyze large-scale feature information around the fault. This leads to the prediction results of small-scale 3D networks being affected by features such as lateral discontinuity and longitudinal jitter in the fault results, and often resulting in missing fault predictions due to the small longitudinal span. Large-scale feature extraction networks can bring higher longitudinal continuity, but due to the large scale, the spatial resolution is low. To integrate the advantages of multiple prediction results to obtain the final fault prediction result, this invention introduces a multi-level fault feature information extraction and multi-scale fusion algorithm, namely, constructing a fault fusion feature network to fuse multiple data volumes, expanding the high spatial resolution and continuity features of a single result. The fault fusion feature network is a supervised learning neural network composed of an end-to-end convolutional neural network. First, multiple fault data volumes are used as input, and a mapping structure is obtained through the end-to-end network. Finally, the fused result is output, such as... Figure 4 As shown.
[0071] 3. Fault attribute feature compression and skeletonization extraction
[0072] The compression model is an end-to-end convolutional neural network that takes the fused fault data volume as input. Unlike the previous network models, this step uses multiple Long Short-Term Memory (LSTM) blocks connected sequentially. The final compression result is obtained through this end-to-end network. Because LSTM is slow and requires significant memory, this step segments the seismic data along the inline direction into two-dimensional profile data for processing and prediction. The refined profile data is then stitched together with the output of the high-precision two-dimensional network to form a three-dimensional result. The processed result is then repeated along the crossline direction to obtain the fault data volume. The network structure diagram is shown below. Figure 5 As shown.
[0073] Fault data volume display is generally presented in the form of fault profile data and fault horizontal slice data, as shown below. Figure 6 and Figure 7 As shown, the purpose of fault detection is achieved through fault profile identification results and fault horizontal slice identification results.
[0074] This invention proposes a novel fault detection process based on deep learning. This method utilizes a skip connection neural network to predict faults in 3D seismic data through a multi-scale, multi-level convolutional network, ultimately obtaining multiple fault data volumes. A fault fusion feature network is introduced to extract feature information from faults at different scales near the initial fault, thus obtaining a fused fault volume. Finally, a fault attribute compression feature extraction network is used to obtain a fault data volume that conforms to the geological background and expert expectations.
[0075] This invention considers both the accuracy of fault detection and the efficiency of computation, achieving more accurate fault detection results while maintaining computational efficiency. It is particularly effective in areas with complex geological conditions and well-developed micro-faults. Compared to traditional methods, this approach improves the efficiency and accuracy of fault interpretation, providing reliable data for subsequent geological modeling and reservoir prediction.
[0076] A multi-scale, multi-level tomographic detection system based on skip connection networks includes:
[0077] The model building module is used to build a deep learning model for 3D fault detection, a fusion model for multi-scale and multi-level fault detection, and a compression model for fault attribute compression feature extraction.
[0078] The data acquisition module is used to acquire earthquake data;
[0079] The first processing module is used to input the seismic data into the deep learning model to obtain multiple three-dimensional fault bodies with different high precision.
[0080] The second processing module is used to input the three-dimensional fault body into the fusion model to obtain fault fusion body data containing multi-scale and multi-level features;
[0081] The third processing module is used to input the fault fusion body data into the extrusion model to obtain the fault data body, so as to realize fault detection.
[0082] The implementation process of the functions and roles of each unit in the above system is detailed in the implementation process of the corresponding steps in the above method, and will not be repeated here.
[0083] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the present invention according to actual needs. Those skilled in the art can understand and implement this without creative effort.
[0084] In the above embodiments, any number of the units can be combined into one module, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in one module. At least one of the units can be at least partially implemented as hardware circuitry, such as a Field Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a System-on-Chip, a System-on-Substrate, a System-on-Package, an Application-Specific Integrated Circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in hardware or firmware, or in any one of software, hardware, and firmware implementations, or in a suitable combination of any of these. Alternatively, at least one of the units can be at least partially implemented as a computer program module, which, when run, can perform corresponding functions.
[0085] An electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;
[0086] Memory, used to store computer programs;
[0087] When the processor executes the program stored in the memory, it implements the multi-scale, multi-level tomographic detection method based on a skip connection network.
[0088] The aforementioned communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc.
[0089] The communication interface is used for communication between the aforementioned electronic devices and other devices.
[0090] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.
[0091] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0092] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned multi-scale, multi-level tomographic detection method based on a skip connection network.
[0093] The computer-readable storage medium may be included in the device / apparatus described in the above embodiments; or it may exist independently and not assembled into the device / apparatus. The computer-readable storage medium carries one or more programs, which, when executed, implement a multi-scale, multi-level tomographic detection method based on a skip connection network according to an embodiment of this disclosure.
[0094] According to embodiments of this disclosure, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
[0095] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.
Claims
1. A multi-scale, multi-level fault detection method based on skip connection networks, characterized in that, Includes the following steps: This paper constructs a deep learning model for 3D fault detection, a fusion model for multi-scale, multi-level fault detection, and a compression model for fault attribute compression feature extraction. The specific process for constructing the deep learning model for 3D fault detection includes: constructing a multi-level jumper connection and a multi-scale, multi-level parallel network; acquiring seismic sample data; training the multi-level jumper connection and multi-scale, multi-level parallel network using the seismic sample data to obtain the deep learning model. The specific process for constructing the fusion model for multi-scale, multi-level fault detection includes: constructing a fault fusion feature network; inputting the seismic sample data into the deep learning model to obtain multiple 3D fault bodies of different high precision; inputting the 3D fault bodies into the fault fusion feature network for training to obtain the fusion model. Seismic data is acquired and input into the deep learning model to obtain multiple three-dimensional fault bodies with different high precision. The three-dimensional fault body is input into the fusion model to obtain fault fusion body data containing multi-scale and multi-level features; The fault fusion body data is input into the extrusion model to obtain the fault data body for fault detection. Specifically, the fault fusion body data is divided into two-dimensional profile data along the inline direction for processing and prediction to obtain two-dimensional data; the two-dimensional data is stitched together to generate three-dimensional data; the above operation is repeated along the crossline direction to obtain the fault data body.
2. The multi-scale, multi-level fault detection method based on skip connection networks according to claim 1, characterized in that, The earthquake sample data was synthesized using three-dimensional technology.
3. The multi-scale, multi-level fault detection method based on skip connection networks according to claim 1, characterized in that, The multi-level jumper connection and multi-scale multi-level parallel network are used to fuse features at different levels.
4. The multi-scale, multi-level tomographic detection method based on skip connection networks according to claim 3, characterized in that, The multi-level jumper connection and multi-scale multi-level parallel network fuse features of different levels. Specifically, it fuses large-scale features with small-scale features through a fully connected network, including the expansion of large-scale features to small-scale features and the compatibility of small-scale features to large-scale features.
5. The multi-scale, multi-level fault detection method based on skip connection networks according to claim 1, characterized in that, The fault fusion feature network consists of an end-to-end convolutional neural network; multiple sample three-dimensional fault volumes are taken as input, and a mapping structure is obtained through the end-to-end network to output the fusion result.
6. The multi-scale, multi-level fault detection method based on skip connection networks according to claim 1, characterized in that, The extrusion model uses multiple Long Short-Term Memory (LSTM) networks connected in series to obtain the extrusion result through an end-to-end network.
7. A multi-scale, multi-level tomographic detection system based on a skip connection network, characterized in that, include: The model building module is used to construct a deep learning model for 3D fault detection, a fusion model for multi-scale, multi-level fault detection, and a compression model for fault attribute compression feature extraction. The specific process of constructing the deep learning model for 3D fault detection includes: constructing a multi-level jumper connection and a multi-scale, multi-level parallel network; acquiring seismic sample data; and training the multi-level jumper connection and multi-scale, multi-level parallel network using the seismic sample data to obtain the deep learning model. The specific process of constructing the fusion model for multi-scale, multi-level fault detection includes: constructing a fault fusion feature network; inputting the seismic sample data into the deep learning model to obtain multiple sample 3D fault bodies with different high precision; and inputting the sample 3D fault bodies into the fault fusion feature network for training to obtain the fusion model. The data acquisition module is used to acquire earthquake data; The first processing module is used to input the seismic data into the deep learning model to obtain multiple three-dimensional fault bodies with different high precision. The second processing module is used to input the three-dimensional fault body into the fusion model to obtain fault fusion body data containing multi-scale and multi-level features; The third processing module is used to input the fault fusion body data into the extrusion model to obtain the fault data body for fault detection. Specifically, it includes: cutting the fault fusion body data into two-dimensional profile data along the inline direction for processing and prediction to obtain two-dimensional data; stitching the two-dimensional data to generate three-dimensional data; and repeating the above operations along the crossline direction to obtain the fault data body.
8. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the multi-scale, multi-level tomographic detection method based on a skip connection network as described in any one of claims 1-6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multi-scale, multi-level tomographic detection method based on a skip connection network as described in any one of claims 1-6.