Oil and gas reservoir geological modeling method and system based on deep learning

Abstract

The invention provides an oil and gas reservoir geological modeling method and system based on deep learning, and the method comprises the steps: obtaining multi-source typical geological data of different reservoirs according to the classification of a reservoir geological knowledge system, and inputting the data in a formulated storage data format; a geological knowledge management system is designed and established according to different types, dimensions and scale characteristic functions of data, a geological data knowledge base corresponding to a reservoir is constructed, mining of various geological knowledge information is carried out by using deep learning in combination with the constructed integrated digital outcrop, a training image library, a sample library and a prototype model data set are constructed, and the data are subjected to data processing. Creating a neural network structure, and designing a loss function to train a target neural network model, so as to generate a matched reservoir three-dimensional geologic model based on the earthquake and wellbore data of the required reservoir; the problems that in the prior art, sample data are difficult to obtain, and the accuracy of a geologic model is limited by the sample scale are solved, the sample library is enriched in a multi-level mode, and a basis is provided for high-quality convolutional neural network training.

Description

Technical Field

[0001] The present invention relates to the field of oil and gas reservoir development geological technology, and in particular to an oil and gas reservoir geological modeling method and system based on deep learning. Background Art

[0002] Oil and gas reservoir geological modeling is an important part of oil and gas reservoir description. It is the key to characterize the spatial distribution of underground reservoirs, physical properties and fluids, and to describe the heterogeneity of oil and gas reservoirs by integrating multiple types of data such as geology, well logging, seismic and production dynamics. Traditional geological modeling technology mainly includes three categories of geological modeling methods, namely, geological modeling methods based on seismic attribute truncation, geological modeling methods based on two-point geostatistics of variograms, and geological modeling methods based on multi-point geostatistics of training images.

[0003] The seismic attribute truncation method is a deterministic method. It is mainly based on the strong correlation between seismic attributes and reservoir (sedimentary phase, lithofacies) / reservoir (pores, caves, fractures) development, determines the seismic attribute threshold, truncates the seismic attributes, and then converts them into a deterministic reservoir / reservoir geological model. The reservoir / reservoir model obtained by this method is consistent with the seismic attributes, but it cannot achieve the transformation from seismic anomaly to real geological body, cannot control the shape of the constructed reservoir / reservoir, and when the correlation between seismic attributes and reservoir / reservoir is not strong, the error of the established reservoir / reservoir model is large. In the two-point geostatistical simulation, the distance covariance between two points in the reservoir space, also called the variogram, is used to describe the reservoir space characteristics, which can make the prediction results consistent with the spatial distribution law of the well point observation data in the geological body to a certain extent, but the variogram cannot well describe the complex reservoir / reservoir heterogeneity characteristics and spatial structure. Multi-point geostatistical modeling can better characterize the geometric characteristics and spatial combination patterns of reservoirs through the identification of geological patterns, and realize the integration of conditional data and geological knowledge. However, this method has problems such as the difficulty in obtaining training images and insufficient pattern reproduction ability. Moreover, the prediction process is still not constrained by multiple types of prior geological knowledge.

[0004] With the increase in the acquisition of oil and gas exploration and development data, how to extract as much geological knowledge and knowledge as possible from massive data and use it for geological modeling is a current research difficulty. The current deep learning method based on big data and artificial intelligence algorithms provides a new way for the development and improvement of geological modeling technology. Some existing technologies propose that geological modeling based on deep learning mainly includes large-scale training sample preparation, unconditional adversarial generative network construction and training, and point out that training samples can be obtained by using target-based simulation, geological process simulation and satellite images; among them, the target-based method is used to simulate geological bodies with simple geometric shapes; the geological process-based method can reflect the constraints of geological laws (such as sedimentation simulation, dissolution simulation, etc.), which requires a lot of expert experience and research knowledge as a basis; although the use of satellite images can provide some actual sample data, it is difficult to collect and expensive.

[0005] It can be seen that the existing technology has the following problems when applied: the modeling method based on two-point geostatistics cannot characterize the geometric shape and distribution law of complex geological bodies; the target-based and forward simulation-based modeling methods cannot fully integrate conditional data such as well points; the modeling method based on multi-point geostatistics relies too much on training images, but the training images are difficult to obtain. For complex structural oil and gas reservoirs, it only characterizes the geological pattern characteristics to a certain extent and cannot meet the accuracy requirements; the geological modeling method based on artificial intelligence can better simulate the complex pattern characteristics of space, but there are problems such as difficulty in obtaining real training samples and difficulty in building deep learning modeling simulators.

[0006] The information disclosed in the background technology section of the present invention is only intended to deepen the understanding of the general background technology of the present invention, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art known to those skilled in the art. Summary of the invention

[0007] To solve the above problems, the present invention provides a method for oil and gas reservoir geological modeling based on deep learning. In one embodiment, the method includes:

[0008] Research data acquisition steps: Based on the reservoirs in the set research area, determine the matching geological knowledge system according to the reservoir type of the reservoir, determine the typical geological blocks of the reservoir to be studied based on it, and obtain multi-source typical geological data of different reservoirs as research data;

[0009] Steps for setting the storage format: formulate matching storage data formats for various research data collected according to the data source;

[0010] Knowledge management system design steps: Carry out platform function design according to different types, dimensions and scale characteristics of data, and establish a geological knowledge management system;

[0011] The geological knowledge base construction step is: entering various research data into the geological knowledge management system using matching storage data formats to form a geological data knowledge base corresponding to the reservoir in the study area;

[0012] Digital outcrop construction steps: Use unified coordinate system technology to organically combine the data in the geological knowledge base with the drone photography results and 3D laser scanning results to form a digital outcrop model that integrates the inside and outside of the geological body;

[0013] Pattern and law mining step: Based on the data in the geological data knowledge base, various types of geological knowledge information are mined using artificial intelligence deep learning methods, including geological body information, geological pattern information and laws between geological bodies;

[0014] Sample library construction step: intelligently constructing a training image library based on the data in the geological data knowledge base, and then constructing a training sample library and a prototype model data set;

[0015] Neural network model construction steps: building a neural network structure based on a three-dimensional convolutional neural network structure, outputting a three-dimensional geological model, designing a loss function for training the convolutional neural network based on the prototype model data set and the three-dimensional geological model, and training model parameters based on the loss function using a parallel computing approach to obtain a target neural network model;

[0016] Reservoir geological model determination step: inputting the seismic data and wellbore data of the reservoir of the required block into the target neural network model to obtain the corresponding three-dimensional reservoir geological model.

[0017] Optionally, in one embodiment, in the research data acquisition step, the process of determining a typical geological block of the reservoir to be studied based on the geological knowledge system includes:

[0018] According to the correlation between reservoir elements and structures of the geological knowledge system, a corresponding geological knowledge classification system is further constructed, and according to the knowledge classification system, typical geological blocks with reservoir data acquisition requirements are determined, including various field outcrop geological blocks and typical dense well network areas.

[0019] Further, in one embodiment, in the research data acquisition step, the process of acquiring multi-source typical geological data of different reservoirs includes:

[0020] When obtaining geological data of different typical field outcrops of reservoirs, it includes development information of different geological bodies, scanning information inside geological bodies, scanning information around geological bodies, scanning information of specific details of geological bodies, aerial stereo scanning information and satellite telemetry data;

[0021] When obtaining geological data of reservoir development areas with different typical dense well patterns, it includes single well drilling data, mud logging data, well logging data, dynamic data, seismic data and geological research profile data.

[0022] In an optional embodiment, the data formats for storage matching different research data are determined according to the following logic:

[0023] For the collected outcrop data, the multi-scale and multi-source outcrop geological information is classified and graded according to the quantitative description system of outcrop reservoirs, and the relevant data format standards are formulated;

[0024] Reservoir data are classified according to stratigraphic characteristics, structural characteristics, and reservoir development characteristics, and standardized data format standards are formulated for each type of data entry.

[0025] Preferably, in one embodiment, in the knowledge management system design step, in the process of establishing a geological knowledge management system, the following platform function design is carried out: including data content management function design, platform module design, knowledge content management design, platform framework design, interactive interface design and visual interface design.

[0026] Furthermore, in one embodiment, the pattern and regularity mining step includes:

[0027] Intelligent identification steps of geological bodies: using the pre-built geological body identification model based on the data in the geological data knowledge base to realize the identification of different types of geological bodies;

[0028] Geological model classification and extraction steps: Extract and construct geological models based on geological body combination model samples from multi-source knowledge. The geological body combination model from multi-source knowledge includes geological information model extracted from outcrops, geological model of dense well network anatomy, and classic hand-drawn model.

[0029] Steps for obtaining the laws between geological bodies: By analyzing the correlation between different geological bodies, the statistical laws between different geological bodies are obtained to support the analysis of inter-well fracture characteristics centered on faults.

[0030] In an optional embodiment, in the sample library construction step, the method of intelligently constructing the training image library includes:

[0031] Using the data in the geological data knowledge base to construct training images based on outcrop extraction, dense well pattern dissection, seismic hollowing and hand-drawing; and

[0032] The pre-built multi-source data source training image library is used to realize the automatic vectorization of hand-drawn geological body pattern images and construct training images.

[0033] Furthermore, in one embodiment, in the sample library construction step, the method of intelligently constructing the training sample library includes:

[0034] Based on the fully constructed geological body training data, a geological body knowledge simulator is trained and constructed using neural network technology. Random noise, conditional well data and probability body parameters are input into the geological body knowledge simulator according to needs to generate different geological body models.

[0035] Preferably, in one embodiment, in the neural network model building step, the process of designing a loss function for training a convolutional neural network includes:

[0036] The prototype model data set and the corresponding wellbore data and seismic data are used as the training data set. The three-dimensional seismic data and three-dimensional wellbore data corresponding to each prototype model are used as the input of the convolutional neural network to generate a three-dimensional geological model. The difference between all prototype models and the three-dimensional geological models generated by their corresponding convolutional neural networks is used as the loss function when training the convolutional neural network.

[0037] Based on the application aspects of the method described in any one or more of the above embodiments, the present invention also provides an oil and gas reservoir geological modeling system based on deep learning, which executes the method described in any one or more of the above embodiments.

[0038] Compared with the closest prior art, the present invention also has the following beneficial effects:

[0039] The present invention provides a method and system for oil and gas reservoir geological modeling based on deep learning. The method determines a matching geological knowledge system according to the reservoir type, classifies and obtains multi-source typical geological data of different reservoirs, formulates a matching data entry format, and enters a geological knowledge management system formed by carrying out platform function design for different types, dimensions and scale characteristics of data to build a geological data knowledge base corresponding to the reservoir. Through the construction of the reservoir geological knowledge base, a standardized and comprehensive database foundation is provided for the subsequent intelligent identification of geological body parameters, geological patterns and statistical laws based on a neural network method, and a training sample library is built by combining multiple methods, so as to facilitate the construction of massive reservoir development and structural patterns based on the knowledge base combined with a deep learning algorithm, and the reconstructed reservoir pattern is consistent with the actual geological characteristics.

[0040] Then, we build training image libraries, sample libraries, and prototype model data sets, create neural network structures, and design loss functions to train target neural network models, which are used to generate matching reservoir 3D geological models based on seismic data and wellbore data of the required reservoirs. Based on comprehensive massive sample data and designed loss function training, we obtain high-quality neural network models, which can be applied to the required reservoirs to reconstruct 3D geological models with higher authenticity and accuracy. This model has high requirements on the professionalism of technicians, low probability of operational errors, and is applicable to various types of reservoirs with good practicality.

[0041] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0043] Figure 1 It is a flowchart of a method for oil and gas reservoir geological modeling based on deep learning provided by one embodiment of the present invention;

[0044] Figure 2 This is an example of a knowledge graph of a carbonate fracture-vuggy reservoir geological classification system of an oil and gas reservoir geological modeling method based on deep learning provided in an embodiment of the present invention;

[0045] Figure 3 It is an example of a knowledge graph of the underground river karst reservoir geological classification system of the oil and gas reservoir geological modeling method based on deep learning provided in an embodiment of the present invention;

[0046] Figure 4 It is an example of a knowledge graph of a pipeline system structural geological classification system of a deep learning-based oil and gas reservoir geological modeling method provided in an embodiment of the present invention;

[0047] Figure 5 It is a schematic diagram of outcrop data collection of a method for oil and gas reservoir geological modeling based on deep learning provided by an embodiment of the present invention;

[0048] Figure 6 It is an example of a standard format for storing outcrop parameters of a deep learning-based oil and gas reservoir geological modeling method provided by another embodiment of the present invention;

[0049] Figure 7 It is a schematic diagram of the technical process of constructing an integrated outcrop model inside and outside the cave of a method for geological modeling of oil and gas reservoirs based on deep learning provided by one embodiment of the present invention;

[0050] Figure 8 It is a schematic diagram of the automatic identification process of cracks in outcrop images of the oil and gas reservoir geological modeling method based on deep learning provided by an embodiment of the present invention;

[0051] Fig. 9 It is a schematic diagram of the automatic recognition process of karst caves in outcrop images of the oil and gas reservoir geological modeling method based on deep learning provided by an embodiment of the present invention;

[0052] Fig.10It is a schematic diagram of determining the main controlling factors of fracture-cavity development by using multivariate linear and Lasso regression models in the oil and gas reservoir geological modeling method based on deep learning provided by an embodiment of the present invention;

[0053] Fig.11 It is a schematic diagram of a process for determining a typical karst fracture-cave combination development pattern of a method for oil and gas reservoir geological modeling based on deep learning provided by an embodiment of the present invention;

[0054] Fig.12 It is a schematic diagram of a process for analyzing statistical laws between different fracture-cavity reservoirs in a method for oil and gas reservoir geological modeling based on deep learning provided by an embodiment of the present invention;

[0055] Fig.13 It is a schematic diagram of a process of producing an ancient karst underground river training image based on outcrop data in a deep learning-based oil and gas reservoir geological modeling method provided by an embodiment of the present invention;

[0056] Fig.14 It is a schematic diagram of the automatic vectorization of hand-drawn geological body model images and the training image construction process of the oil and gas reservoir geological modeling method based on deep learning provided by an embodiment of the present invention;

[0057] Fig.15 It is a schematic diagram of a process flow of generating underground dark river-type karst caves based on a simulation process method in a deep learning-based oil and gas reservoir geological modeling method provided by an embodiment of the present invention;

[0058] Fig.16 It is an example diagram of generating multiple different underground dark river type models through a simulator in the oil and gas reservoir geological modeling method based on deep learning provided by an embodiment of the present invention;

[0059] Fig.17 It is an example diagram of a typical underground river training geological model, a three-dimensional well point data set and a seismic probability volume in a deep learning-based oil and gas reservoir geological modeling method provided by an embodiment of the present invention;

[0060] Fig.18 (a) is an example diagram of the main process structure diagram of the generator in the oil and gas reservoir geological modeling method based on deep learning provided in an embodiment of the present invention;

[0061] Fig.18 (b) is an example diagram of the discriminator structure in the oil and gas reservoir geological modeling method based on deep learning provided by an embodiment of the present invention;

[0062] Fig.19 It is a schematic diagram of the simulator construction and training process in the oil and gas reservoir geological modeling method based on deep learning provided by an embodiment of the present invention;

[0063] Fig. 20It is a schematic diagram of comparing well data and probability volume data of a generated model and a theoretical model in a method for oil and gas reservoir geological modeling based on deep learning provided by an embodiment of the present invention;

[0064] Fig.21 It is a comparison diagram between the simulation implementation of different slices and the theoretical model in the oil and gas reservoir geological modeling method based on deep learning provided by an embodiment of the present invention;

[0065] Fig. 22 It is a schematic diagram of the comparison of simulation results and before and after denoising in the oil and gas reservoir geological modeling method based on deep learning provided by an embodiment of the present invention;

[0066] Fig.23 It is a schematic diagram of different types of reservoirs and fusion models of the R4 oil reservoir in the oil and gas reservoir geological modeling method based on deep learning provided in an embodiment of the present invention;

[0067] Fig.24 It is a structural schematic diagram of an oil and gas reservoir geological modeling system based on deep learning provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0068] The following will describe the implementation methods of the present invention in detail in conjunction with the accompanying drawings and embodiments, so that the implementers of the present invention can fully understand how the present invention applies technical means to solve technical problems and achieve the implementation process of technical effects and implement the present invention specifically according to the above implementation process. It should be noted that as long as there is no conflict, the various embodiments and various features of the embodiments in the present invention can be combined with each other, and the technical solutions formed are all within the protection scope of the present invention.

[0069] Although the flowcharts describe the operations as sequential processes, many of the operations may be performed in parallel, concurrently, or simultaneously. The order of the operations may be rearranged. A process may be terminated when its operations are completed, but may also have additional steps not included in the accompanying drawings. A process may correspond to a method, function, procedure, subroutine, subprogram, etc.

[0070] Computer devices include user devices and network devices. Among them, user devices or clients include but are not limited to computers, smart phones, PDAs, etc.; network devices include but are not limited to a single network server, a server group consisting of multiple network servers, or a cloud composed of a large number of computers or network servers based on cloud computing. Computer devices can be operated alone to implement the present invention, or they can be connected to the network and implement the present invention through interactive operations with other computer devices in the network. The network where the computer device is located includes but is not limited to the Internet, wide area network, metropolitan area network, local area network, VPN network, etc.

[0071] The terms "first", "second", etc. may be used herein to describe various units, but these units should not be limited by these terms, and these terms are used only to distinguish one unit from another unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items. When a unit is referred to as being "connected" or "coupled" to another unit, it can be directly connected or coupled to the other unit, or there can be intermediate units.

[0072] The terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments. Unless the context clearly indicates otherwise, the singular forms "one", "one" and "item" used herein are also intended to include plural numbers. It should also be understood that the terms "include" and / or "comprise" used herein specify the existence of stated features, integers, steps, operations, units and / or components, without excluding the existence or addition of one or more other features, integers, steps, operations, units, components and / or combinations thereof.

[0073] Oil and gas reservoir geological modeling is an important part of oil and gas reservoir description. It is the key to characterize the spatial distribution of underground reservoirs, physical properties and fluids, and depict the heterogeneity of oil and gas reservoirs by integrating multiple types of data such as geology, well logging, seismic and production dynamics.

[0074] Traditional geological modeling technology mainly includes three categories of geological modeling methods, namely, geological modeling methods based on seismic attribute truncation, geological modeling methods based on two-point geostatistics of variograms, and geological modeling methods based on multi-point geostatistics of training images.

[0075] Among them, the seismic attribute truncation method is a deterministic method, which is mainly based on the strong correlation between seismic attributes and reservoir (sedimentary phase, lithofacies) / reservoir body (pores, holes, fractures) development, determines the seismic attribute threshold, truncates the seismic attributes, and then converts them into a deterministic reservoir / reservoir body geological model. The reservoir / reservoir body model obtained by this method is consistent with the seismic attributes, but it cannot realize the transformation from seismic anomaly body to real geological body, and cannot control the shape of the constructed reservoir / reservoir body. When the correlation between seismic attributes and reservoir / reservoir body is not strong, the error of the established reservoir / reservoir body model is large.

[0076] In two-point geostatistical simulation, the distance covariance between two points in the reservoir space, also called the variogram, is used to describe the reservoir space characteristics. This can, to a certain extent, make the prediction results consistent with the spatial distribution law of the well point observation data in the geological body. However, the variogram cannot well describe the complex heterogeneity characteristics and spatial structure of the reservoir / reservoir body.

[0077] Multi-point geostatistical modeling can better characterize the geometric characteristics and spatial combination patterns of reservoirs through the identification of geological patterns, and realize the integration of conditional data and geological knowledge. However, according to existing research, there are problems such as difficulty in obtaining training images and insufficient pattern reproduction ability when applying this method, and the prediction process is still not constrained by multiple types of prior geological knowledge; for example, the theory, methods, applications and development status of multi-point geostatistics by Wang Kaiqi et al.

[0078] As more and more data is acquired for oil and gas exploration and development, how to extract as much geological knowledge and understanding as possible from massive data and use it for geological modeling is a current research difficulty. At present, deep learning methods based on big data and artificial intelligence algorithms have achieved rapid development in the field of geology, and have also provided a new way for the development and improvement of geological modeling technology. Through deep neural networks, more and more complex geological patterns and statistical laws can be considered, and geological models that can better represent the actual underground situation can be generated. This is an important development direction for geological modeling of oil and gas reservoirs.

[0079] Dupont E. et al. (Emilien Dupont, Tuanfeng Zhang, Peter Tilke, Lin Liang, William Bailey. Generating Realistic Geology Conditioned on Physical Measurements with Generative Adversarial Networks[J]. Statistics, 2018.) provide a method for generating realistic geological models conditioned on physical measurements using generative adversarial networks. They propose that geological modeling based on deep learning mainly includes large-scale training sample preparation, construction and training of unconditional adversarial generative networks, and secondary optimization of neural network parameters considering conditional data.

[0080] Song Suihong et al. (Song Suihong, Shi Yanqing, Hou Jiagen. Research progress on reservoir geological modeling methods based on generative adversarial networks [J]. Bulletin of Petroleum Science, 2022, 01: 34-49) applied artificial intelligence deep learning methods to the geological modeling of dark river cave reservoirs. The model has a good consistency with the input wellbore interpretation data and is also close to the probability volume data. The generated dark river caves are also consistent with the geological understanding of the study area in terms of morphology and direction.

[0081] Nesvold E. et al. (Nesvold E., Mukerji T. Geomodeling using generative adversarial networks and a database of satellite imagery of modern river deltas [C] / / Anon. Petroleum Geostatistics 2019. European Association of Geoscientists & Engineers, 2019.) proposed a method of using generative adversarial networks and a database of satellite imagery of modern river deltas for geomodeling, and pointed out that training samples can be obtained by using target-based simulation, geological process simulation, and satellite images. Among them, the target-based method is used to simulate geological bodies with simple geometric shapes; the geological process-based method can reflect the constraints of geological laws (such as sedimentation simulation, dissolution simulation, etc.), which requires a lot of expert experience and research knowledge as a basis; although the use of satellite images can provide some actual sample data, it is difficult to collect and expensive.

[0082] In summary, the existing technologies have the following problems when constructing geological models: the modeling method based on two-point geostatistics cannot characterize the geometric shape and distribution law of complex geological bodies; the target-based and forward simulation-based modeling methods cannot completely integrate conditional data such as well points; the modeling method based on multi-point geostatistics relies too much on training images, but the training images are difficult to obtain. For complex structural oil and gas reservoirs, it only characterizes the geological pattern characteristics to a certain extent and cannot meet the accuracy requirements; the geological modeling method based on artificial intelligence can better simulate the complex pattern characteristics of space, but there are problems such as difficulty in obtaining real training samples and difficulty in building deep learning modeling simulators.

[0083] To solve the above problems, the present invention provides a method and system for oil and gas reservoir geological modeling based on deep learning. By constructing a reservoir geological knowledge base, the geological body parameters, geological patterns and statistical laws are intelligently identified based on the neural network method. A training sample library is constructed by combining multiple methods, a deep learning network model is built, and parameter optimization and loss function design are performed to complete the training of the convolutional neural network, establish a model generator, and complete the geological modeling of the actual reservoir.

[0084] Next, the detailed process of the method of the embodiment of the present invention is described in detail based on the accompanying drawings. The steps shown in the flowchart of the accompanying drawings can be executed in a computer system including a set of computer executable instructions. Although the logical order of each step is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0085] Embodiment 1

[0086] Figure 1 The flowchart of the oil and gas reservoir geological modeling method based on deep learning provided by the first embodiment of the present invention is shown. Figure 1 It can be seen that the method includes the following steps.

[0087] Research data acquisition steps: Based on the reservoirs in the set research area, determine the matching geological knowledge system according to the reservoir type of the reservoir, determine the typical geological blocks of the reservoir to be studied based on it, and obtain multi-source typical geological data of different reservoirs as research data;

[0088] Steps for setting the storage format: formulate matching storage data formats for various research data collected according to data sources;

[0089] Knowledge management system design steps: Carry out platform function design according to different types, dimensions and scale characteristics of data, and establish a geological knowledge management system;

[0090] The geological knowledge base construction step is: entering various research data into the geological knowledge management system using matching storage data formats to form a geological data knowledge base corresponding to the reservoir in the study area;

[0091] Digital outcrop construction steps: Use unified coordinate system technology to organically combine the data in the geological knowledge base with the drone photography results and 3D laser scanning results to form a digital outcrop model that integrates the inside and outside of the geological body;

[0092] Pattern and law mining step: Based on the data in the geological data knowledge base, various types of geological knowledge information are mined using artificial intelligence deep learning methods, including geological body information, geological pattern information and laws between geological bodies;

[0093] Sample library construction step: intelligently constructing a training image library based on the data in the geological data knowledge base, and then constructing a training sample library and a prototype model data set;

[0094] Neural network model construction steps: building a neural network structure based on a three-dimensional convolutional neural network structure, outputting a three-dimensional geological model, designing a loss function for training the convolutional neural network based on the prototype model data set and the three-dimensional geological model, and training model parameters based on the loss function using a parallel computing approach to obtain a target neural network model;

[0095] Reservoir geological model determination step: input the seismic data and wellbore data of the reservoir of the required block into the target neural network model to obtain the corresponding reservoir three-dimensional geological model.

[0096] The deep learning-based oil and gas reservoir geological modeling method in the above-mentioned embodiment of the present invention not only covers the construction of deep learning neural network, optimization of neural network parameters, design of loss function, training of convolutional neural network, actual reservoir geological modeling based on trained convolutional neural network, etc., but also focuses on optimizing the early geological knowledge base construction, geological model and law mining, and training sample library construction.

[0097] The embodiment of the present invention analyzes the reservoir type in the study area, determines the classification standard of the geological knowledge system, collects relevant data, enters relevant data, constructs digital outcrops, conducts structured management of data, and forms a reservoir geological knowledge base.

[0098] Specifically, in a preferred embodiment, in the research data acquisition step, the process of determining a typical geological block of the reservoir to be studied based on the geological knowledge system includes:

[0099] According to the correlation between reservoir elements and structures of the geological knowledge system, a corresponding geological knowledge classification system is further constructed, and according to the knowledge classification system, typical geological blocks with reservoir data acquisition requirements are determined, including various field outcrop geological blocks and typical dense well network areas.

[0100] In actual application, the set study area is one or more study areas containing comprehensive types of reservoirs. In the study data acquisition step, it includes: determining the reservoir geological knowledge system type according to the reservoir type of the reservoir as the geological knowledge system. Further subdividing according to the reservoir elements step by step, establishing the correlation between the structures, forming a geological knowledge classification system corresponding to the current reservoir and the current reservoir type, based on which the geological data of typical field outcrops of different reservoirs and dense well network reservoir development areas are obtained.

[0101] Further, in an optional embodiment, in the research data acquisition step, the process of acquiring multi-source typical geological data of different reservoirs includes:

[0102] When obtaining geological data of different typical field outcrops of reservoirs, it includes development information of different geological bodies, scanning information inside geological bodies, scanning information around geological bodies, scanning information of specific details of geological bodies, aerial stereo scanning information and satellite telemetry data;

[0103] When obtaining geological data of reservoir development areas with different typical dense well networks, it includes single well drilling data, mud logging data, well logging data, well production dynamic data, seismic data and geological research profile data.

[0104] For example, when collecting relevant data, typical field outcrops and dense well network reservoir development areas are selected for data collection:

[0105] ① The length, opening, occurrence, etc. of geological bodies (such as fractures, caves, and lithofacies) can be measured in the field by using tape measures and compasses;

[0106] Use 3D laser scanner to scan the internal and surrounding structures of geological bodies;

[0107] Use handheld scanners to scan the entrance, exit and small geological bodies on the surface of geological bodies;

[0108] Use drone aerial photography to conduct three-dimensional scanning from the air;

[0109] Satellite remote sensing is used to scan large-scale macro structures, topography, and landforms.

[0110] ② Single well drilling, logging, well logging, dynamic data, seismic data, geological research profiles and other related data in typical dense well network areas.

[0111] Taking the R1 reservoir as an example, the reservoir is a typical fracture-cavity carbonate reservoir, including weathering crust karst, ancient underground river pipeline system and fault-controlled karst, and a reservoir geological knowledge base is constructed according to the method of the present invention.

[0112] Firstly, a hierarchical classification system of information of such reservoir geological knowledge base is constructed.

[0113] The outcrop data of carbonate fracture-vuggy reservoirs are divided into three categories according to their karst origin, such as Figure 2 It is the knowledge map of geological knowledge classification system of carbonate fracture-vuggy reservoir (level 1); Figure 3 A knowledge map of the geological knowledge classification system for the underground river karst system (secondary); Figure 4 The knowledge map of the geological knowledge classification system of the pipeline system structure (level 3) is defined. Similarly, the knowledge map of reservoir data is also defined. Then, the geological data of fractures, holes, etc. are collected.

[0114] The typical outcrop collection points of ancient karst and modern karst were selected, and the size of the cave, the length, opening, occurrence of the cracks, etc. were measured by tape measure and compass; the structure inside and around the cave was scanned by 3D laser scanner; the entrance, exit and terminal hole were scanned by handheld scanner; and the three-dimensional scanning from the air was performed by drone aerial photography or satellite remote sensing. The relevant data of single well drilling, well logging, well logging, dynamic, seismic, fracture-cavity research profiles, etc. in typical fracture-cavity development and dense well network areas were classified and sorted, such as Figure 5 shown.

[0115] Furthermore, a storage format setting step is performed to formulate matching storage data formats for the various research data collected according to the data sources.

[0116] In a preferred embodiment, the data formats for different research data matches are determined according to the following logic:

[0117] For the collected outcrop data, the multi-scale and multi-source outcrop geological information is classified and graded according to the knowledge map of the geological knowledge classification system corresponding to the outcrop reservoir, and the relevant data format standards are formulated;

[0118] Reservoir data are classified according to stratigraphic characteristics, structural characteristics, and reservoir development characteristics, and standardized data format standards are formulated for each type of data entry.

[0119] In order to standardize the storage of various types of data, for the collected fracture-cavity outcrop data, the multi-scale and multi-source outcrop geological information is classified and graded according to the quantitative description system of outcrop fracture-cavity reservoirs, and the relevant storage data format standards are formulated, such as Figure 6 As shown; similarly, for reservoir data, the data is classified from the aspects of stratigraphic characteristics, structural characteristics, reservoir development characteristics, etc., and then standardized and organized.

[0120] Furthermore, the present invention aims to achieve structured management of various types of data. Therefore, in one embodiment, a knowledge management system is first designed to carry out platform function design for different types, dimensions and scale characteristics of data to establish a geological knowledge management system.

[0121] Then, the geological knowledge base construction step is performed, and various research data are respectively entered into the geological knowledge management system using matching storage data formats to form a geological data knowledge base corresponding to the reservoir in the study area.

[0122] Among them, in a preferred embodiment, in the process of establishing a geological knowledge management system, the following platform function designs are carried out:

[0123] It includes data content management function design, platform module design, knowledge content management design, platform framework design, interactive interface design and visual interface design.

[0124] In actual application, in order to manage data of different types, dimensions and scales separately, a geological knowledge data management platform is designed and established first. The platform function design is carried out, including management of data content, platform module design and knowledge content management standards; platform framework design, interface and interface design are carried out to form database and visualization and other related functions, and also includes platform service function expansion interface design.

[0125] Based on this, a further reservoir geological knowledge base is constructed, and a knowledge base system is developed according to the functional requirements of knowledge collection, management, and application of outcrop and reservoir data. A systematic outcrop knowledge management framework is constructed, structured data tables are designed, the knowledge achievement collection system is optimized, and a knowledge display window for automatic extraction, collection, and automatic updating is established to form a complete reservoir geological knowledge base.

[0126] In a preferred embodiment, the method of the present invention also includes: a digital outcrop construction step, using a unified coordinate system technology to organically combine the data in the geological knowledge base with the drone photography results and the three-dimensional laser scanning results to form a digital outcrop model that integrates the inside and outside of the geological body.

[0127] In actual application, in one embodiment, in the process of constructing the digital outcrop of fracture-cave reservoir, the drone photography and three-dimensional laser scanning technology are organically combined based on the unified coordinate system technology of the control points to establish the digital outcrop model construction technology integrated inside and outside the cave, so as to realize the digital outcrop construction, such as Figure 7 shown.

[0128] Furthermore, based on the collected and managed data, various types of geological knowledge information are mined. In one embodiment, by executing pattern and rule mining steps, artificial intelligence deep learning methods are used based on the data in the geological data knowledge base to mine various types of geological knowledge information, including geological body information, geological model information and rules between geological bodies.

[0129] In a preferred embodiment, the pattern and regularity mining step includes:

[0130] Intelligent identification steps of geological bodies: using the pre-built geological body identification model based on the data in the geological data knowledge base to realize the identification of different types of geological bodies;

[0131] Geological model classification and extraction steps: Extract and construct geological models based on geological body combination model samples from multi-source knowledge. The geological body combination model from multi-source knowledge includes geological information model extracted from outcrops, geological model of dense well network anatomy, and classic hand-drawn model.

[0132] Steps for obtaining the laws between geological bodies: By analyzing the correlation between different geological bodies, the statistical laws between different geological bodies are obtained to support the analysis of inter-well fracture characteristics centered on faults.

[0133] In actual application, various types of geological knowledge information are mined based on the collected and managed data according to the following operations:

[0134] (1) Intelligent identification of geological targets. Based on the artificial intelligence deep learning algorithm, the inherent laws and representation levels of sample data are mined, and a geological body identification model is established to realize the identification of different types of geological bodies in view of the complexity of geological body identification.

[0135] (2) Classification and extraction of geological patterns. Based on the convergence of multi-source knowledge, the geological body combination pattern samples are enhanced, and artificial intelligence machine learning methods are used to classify and extract typical geological body combination patterns. Including: extraction and construction of geological information patterns based on outcrop extraction; extraction and construction of geological patterns based on dense well network anatomy; extraction and construction based on traditional classic hand-drawn patterns.

[0136] (3) Obtaining statistical laws between geological bodies. By analyzing the correlation between different geological bodies, the statistical laws between different geological bodies are displayed, which supports the analysis of fracture characteristics centered on faults.

[0137] Taking the R1 fracture-vuggy reservoir as an example, based on the constructed knowledge base, the extraction and mining of geological patterns and laws are carried out.

[0138] First, the automatic identification of cracks on outcrop images. A crack polygon marking tool was established to complete the annotation of a large number of training data sets.

[0139] The U-net neural network method is used to complete the intelligent identification of crack information on the outcrop. Figure 8 As shown;

[0140] According to the identified crack information, the length, opening, area, perimeter and center point of the crack are calculated and stored in the data table for the extraction of outcrop crack rules.

[0141] Secondly, automatic identification of karst caves on outcrop images. The “YOLO V5 neural network + image processing method” method is used to automatically identify karst caves, such as Fig. 9 shown.

[0142] Third, through multivariate linear fitting quantitative analysis of data, cave width, cave height, cave depth and fracture density are important parameters for distinguishing dark river caves, fault-controlled caves and weathering crust caves. A deep learning training model based on outcrop fracture and cave data is established for the discrimination of karst types, such as Fig.10 shown.

[0143] Fourth, based on a large number of digital outcrop models in the knowledge base, the U-Net deep neural network algorithm was improved to solve the problem of intelligent identification of fracture-cavity combinations, and two typical karst development patterns of fault control and underground rivers were established, laying the foundation for modeling training image production. Fig.11 shown.

[0144] Finally, based on the relevant structured data of the knowledge base, the parameters related to the main controlling factors of different types of reservoirs are determined through multivariate linear regression, and the statistical laws between different types of caves and faults, caves and fractures, faults and fractures are established, which provides a reliable basis for inter-well constraints in geological modeling, such as Fig.12 shown.

[0145] Furthermore, a sample library construction step is performed to intelligently construct a training image library based on the data in the geological data knowledge base, and then construct a training sample library and a prototype model data set.

[0146] In a preferred embodiment, in the sample library construction step, the method of intelligently constructing the training image library includes:

[0147] Using the data in the geological data knowledge base to construct training images based on outcrop extraction, dense well pattern dissection, seismic hollowing and hand-drawing; and

[0148] The pre-built multi-source data source training image library is used to realize the automatic vectorization of hand-drawn geological body pattern images and construct training images.

[0149] Furthermore, in an optional embodiment, in the sample library construction step, the method of intelligently constructing the training sample library includes:

[0150] Based on the fully constructed geological body training data, a geological body knowledge simulator is trained and constructed using neural network technology. Random noise, conditional well data and probability body parameters are input into the geological body knowledge simulator according to needs to generate different geological body models.

[0151] In actual application, the following operations are performed to intelligently construct a training sample library based on the massive data in the knowledge base, including the construction of training images and the construction of the training sample library.

[0152] (1) Training image construction: There are two main methods: ① Training images can be constructed based on traditional methods such as outcrop extraction, dense well pattern dissection, seismic hollowing, and hand-drawing;

[0153] ② The training image library construction idea based on multi-source data sources is adopted to realize the automatic vectorization of hand-drawn geological model images, thereby improving the construction speed of training images.

[0154] (2) Construction of a neural network training sample library to train and learn the pattern knowledge in the fully constructed geological body training data to obtain a simulator.

[0155] After inputting random noise, conditional well data and probability volume data, different geological body models are generated.

[0156] (3) Construct a geological body prototype model dataset. Using the measured data of the study area (such as the probability density distribution of geological body diameter, direction and morphology, etc.) as input data, a target-based method is used to construct a large number of three-dimensional geological body prototype models. The size of each prototype model must be consistent with the size of the geological model output by the convolutional neural network, and the number of grids must also be consistent. In the process of constructing a large number of three-dimensional geological body prototype models, a target-based modeling method is used to simulate and generate a large number of geological body three-dimensional prototype models based on the input measured data; the target-based method is used to simulate the prototype model using the measured data of the study area.

[0157] Taking the R2 paleokarst underground river type oil reservoir as an example, the karst reservoir type is mainly paleokarst underground river system. Based on the massive data in the knowledge base, a training sample library is intelligently constructed to realize the construction of training images and training sample library.

[0158] The training images are constructed by traditional methods such as outcrop data extraction, dense well network dissection, seismic attribute hollowing, and hand-drawing. The process of constructing ancient karst underground river training images from outcrop data is as follows: Fig.13 The training image library construction method based on multi-source data sources is adopted to realize the automatic vectorization of hand-drawn geological model images, which improves the construction speed of training images. Fig.14 shown.

[0159] According to the formation mechanism of dark river caves, a random generation algorithm of dark river caves based on simulation process is established. First, based on the quantitative statistical laws of three parameters, namely, regional crack density distribution map, crack angle and length, the dissolution cracks are simulated one by one to form fluid dissolution channels; then, based on the three parameters of dark river cave height, aspect ratio and profile shape, the channels are expanded vertically outward to form the three-dimensional shape of dark river caves, such as Fig.15 shown.

[0160] Considering that the training images obtained only by the above method are limited and cannot meet the sample size requirements of the deep learning modeling method, the construction of the neural network training sample library is carried out in combination with the dark river generation method of the simulation process. The pattern knowledge in the fully constructed geological body training data is trained and learned to obtain the simulator. After inputting random noise, conditional well data and probability body data, a total of more than 15,000 different dark river models are realized, such as Fig.16 shown.

[0161] 100 underground river cave prototype models were randomly cut along the underground river caves to form a total of 15,000 training geological models (64×64×64 grids). Three-dimensional well point data sets were randomly sampled from each training geological model, and a total of 15,000 sets of training well point data were constructed. The Gaussian smoothing method was used to obtain the training geological model set; a certain proportion of Gaussian noise was added to simulate the real earthquake probability volume; a total of 15,000 sets of probability volume data were obtained, such as Fig.17 shown.

[0162] Further, a neural network model construction step is performed: a neural network structure is built based on a three-dimensional convolutional neural network structure, a three-dimensional geological model is output, a loss function for training the convolutional neural network is designed based on the prototype model data set and the three-dimensional geological model, and a target neural network model is obtained by training model parameters using a parallel computing idea based on the loss function;

[0163] In an optional embodiment, a deep learning neural network model is constructed based on a training sample library, mainly including: building and training a deep learning neural network, designing a loss function, and training a convolutional neural network model.

[0164] (1) Construction and training of deep learning neural network. Based on the "encoder-decoder" structure, a three-dimensional convolutional neural network structure is designed. Its input is a three-dimensional seismic data volume and a wellbore data volume, and its output is a three-dimensional model of the geological volume; all operations of this convolutional network are three-dimensional convolutional neural network layers.

[0165] (2) Design of loss function. The prototype model data set and the corresponding wellbore data and seismic data are used as training data sets. The three-dimensional seismic data and three-dimensional wellbore data corresponding to each prototype model are used as the input of the convolutional neural network to generate a three-dimensional geological model. The difference between all prototype models and the three-dimensional geological models generated by their corresponding convolutional neural networks is used as the loss function Loss when training the convolutional neural network; its specific calculation formula is as follows:

[0166]

[0167] Among them, S is the three-dimensional geological model generated by the convolutional neural network, and M is the three-dimensional prototype model constructed using the target-based method.

[0168] (3) Training the convolutional neural network model. Using the prototype model dataset as the training dataset, applying the aforementioned loss function, and training the parameters to be trained in the constructed convolutional neural network based on the idea of ​​parallel computing on a computer equipped with multiple GPUs.

[0169] Taking the R3 reservoir as an example, the advanced progressive growth GAN method is used. The simulator trains the learning pattern knowledge layer by layer from large scale to small scale. The GANSim of the data is conditioned by the loss function to extract, calculate and condition the data, and design a high-dimensional potential vector body to realize 3D simulation. The construction of deep learning neural network mainly includes generator and discriminator: the generator includes the main process of learning geological pattern knowledge and the conditioned data process, such as Fig.18(a); The discriminator structure is basically symmetrical to the generator main process, except for the two fully connected layers and the minibatch standard layer, as shown in Fig.18 (b) as shown.

[0170] The prototype model data set and the corresponding wellbore data and seismic data are used as training data sets. The three-dimensional seismic data and three-dimensional wellbore data corresponding to each prototype model are used as the input of the convolutional neural network to generate a three-dimensional geological model. The difference between all prototype models and the three-dimensional geological models generated by their corresponding convolutional neural networks is used as the loss function when training the convolutional neural network; the specific calculation formula is as follows:

[0171]

[0172] Among them, S is the three-dimensional geological model generated by the convolutional neural network, and M is the three-dimensional prototype model constructed using the target-based method.

[0173] Using the prototype model dataset as the training dataset, the aforementioned loss function is applied, and the parameters to be trained in the constructed convolutional neural network are trained based on the idea of ​​parallel computing on a computer equipped with multiple GPUs. The algorithm parameter settings include the structural parameters of the simulator and discriminator convolutional neural networks, the interface structural parameters of the conditional data input simulator, etc. The construction of the training dataset includes: training geological model set, training well point dataset, training probability volume dataset; the simulator construction and training process are as follows: Fig.19 shown.

[0174] Through the control of the discriminator, this method simulates and realizes a geological model that is consistent with the dark river cave and has diversity; 100% well point matching rate; the frequency map basically matches the input probability volume data. The model is projected into 2D space using a multi-scale scalar method. From the scatter plot and density map, it can be seen that the generated model is within the distribution range of the test model; at the same time, under the influence of specific well data and probability volume data, the generated model is closer to the theoretical model, such as Fig. 20 As shown in Figure 2. From the model slices, it can be seen that the simulation has a higher prediction accuracy. The probability volume contour, the simulation frequency volume contour and the theoretical model are sliced ​​along three different directions of x, y and z, indicating that the simulation is closer to the cave morphology and has a higher prediction accuracy, as shown in Figure 2. Fig.21 shown.

[0175] Based on this, a reservoir geological model determination step is further performed, and the seismic data and wellbore data of the reservoir in the required block are input into the target neural network model to obtain a corresponding three-dimensional reservoir geological model.

[0176] In a preferred embodiment, in the process of actual reservoir geological modeling based on the trained convolutional neural network, three-dimensional seismic data and wellbore data are obtained for the actual working area of ​​the reservoir, and are input into the trained convolutional neural network to establish a corresponding three-dimensional geological model of the target reservoir.

[0177] Taking carbonate fracture-vuggy reservoir R4 as an example, its block area is 13.4km 2 , 12 wells were drilled, 11 caves and 27 faults were encountered, and the cumulative oil production of the unit was 92.9×10 4 t. Obtain the conditional well data and probability volume data of the dark river development layer, and the modeling process is as follows:

[0178] First, a geological knowledge base similar to the study area is constructed, or geological model knowledge is extracted from a large number of geological model sets in the established knowledge base; then, the training parameters are fixed, and the generator structure and discriminator structure are redesigned according to the actual work area; then, multiple dark river-type cave models are simulated and generated; fourth, after inputting different random noises, well data and probability volume data, multiple different underground dark river-type models are generated through the simulator; fifth, the dark river-type cave realization generated by the simulator has some sporadic noise points, and the simulation realization is denoised to obtain the dark river reservoir model, such as Fig. 22 Finally, other types of reservoir models are established, and all classification models are integrated to obtain the final R4 reservoir geological model, as shown in Fig.23 shown.

[0179] The method of the embodiment of the present invention can construct a comprehensive and flexible massive reservoir development and structural model based on the knowledge base combined with the deep learning algorithm, and the reconstructed reservoir model is consistent with the actual geological characteristics; based on this as a sample, the trained convolutional neural network model can reconstruct the geological model, and the modeling result has good consistency with the well point and seismic data. At the same time, the modeling speed is greatly improved, and the obtained simulator can be used for modeling any reservoir.

[0180] For the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the order of the actions described, because according to the present invention, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0181] It should be pointed out that in other embodiments of the present invention, the method can also obtain a new oil and gas reservoir geological modeling method based on artificial intelligence technology by combining one or several of the above embodiments to achieve high-precision geological modeling research on various oil and gas reservoirs.

[0182] It should be noted that, based on the method in any one or more of the above embodiments of the present invention, the present invention also provides a storage medium, which stores program codes that can implement the method as described in any one or more of the above embodiments, and when the code is executed by the operating system, it can implement the deep learning-based oil and gas reservoir geological modeling method as described above.

[0183] Embodiment 2

[0184] The above embodiments disclosed in the present invention describe the method in detail. The method of the present invention can be implemented by various devices or systems. Therefore, based on other aspects of the method described in any one or more embodiments, the present invention also provides a deep learning-based oil and gas reservoir geological modeling system, which is used to execute the deep learning-based oil and gas reservoir geological modeling method described in any one or more embodiments. Specific embodiments are given below for detailed description.

[0185] Specifically, Fig.24 The structural schematic diagram of the oil and gas reservoir geological modeling system based on deep learning provided in an embodiment of the present invention is shown in FIG. Fig.24 As shown, the system includes:

[0186] A research data acquisition module is configured to determine a matching geological knowledge system based on the reservoir type of the reservoir in the set research area, determine the typical geological block of the reservoir to be studied based on the system, and obtain multi-source typical geological data of different reservoirs as research data;

[0187] The module for setting the storage format is configured to formulate matching storage data formats for various research data collected according to the data source;

[0188] The knowledge management system design module is configured to carry out platform function design for different types, dimensions and scale characteristics of data and establish a geological knowledge management system;

[0189] A geological knowledge base construction module, which is configured to input a variety of research data into the geological knowledge management system using matching storage data formats to form a geological data knowledge base corresponding to the reservoir in the study area;

[0190] A digital outcrop construction module is configured to organically combine the data in the geological knowledge base with the drone photography results and the three-dimensional laser scanning results using a unified coordinate system technology to form a digital outcrop model that integrates the inside and outside of the geological body;

[0191] A pattern and law mining module, which is configured to mine various types of geological knowledge information based on the data in the geological data knowledge base using an artificial intelligence deep learning method, including geological body information, geological pattern information and laws between geological bodies;

[0192] A sample library construction module, which is configured to intelligently construct a training image library based on the data in the geological data knowledge base, and then construct a training sample library and a prototype model data set;

[0193] A neural network model building module is configured to build a neural network structure based on a three-dimensional convolutional neural network structure, output a three-dimensional geological model, design a loss function for training the convolutional neural network based on the prototype model data set and the three-dimensional geological model, and train model parameters using a parallel computing approach based on the loss function to obtain a target neural network model;

[0194] The reservoir geological model determination module is configured to input the seismic data and wellbore data of the reservoir of the required block into the target neural network model to obtain the corresponding three-dimensional reservoir geological model.

[0195] Optionally, in one embodiment, the research data acquisition module is configured to determine a typical geological block of the reservoir to be studied based on the geological knowledge system through the following operations:

[0196] According to the correlation between reservoir elements and structures of the geological knowledge system, a corresponding geological knowledge classification system is further constructed, and according to the knowledge classification system, typical geological blocks with reservoir data acquisition requirements are determined, including various field outcrop geological blocks and typical dense well network areas.

[0197] Furthermore, in one embodiment, the research data acquisition module is configured to acquire multi-source typical geological data of different reservoirs in the following manner:

[0198] When obtaining geological data of different typical field outcrops of reservoirs, it includes development information of different geological bodies, scanning information inside geological bodies, scanning information around geological bodies, scanning information of specific details of geological bodies, aerial stereo scanning information and satellite telemetry data;

[0199] When obtaining geological data of reservoir development areas with different typical dense well patterns, it includes single well drilling data, mud logging data, well logging data, dynamic data, seismic data and geological research profile data.

[0200] In an optional embodiment, the storage format setting module determines the storage data format that matches different research data according to the following logic:

[0201] For the collected outcrop data, the multi-scale and multi-source outcrop geological information is classified and graded according to the quantitative description system of outcrop reservoirs, and the relevant data format standards are formulated;

[0202] Reservoir data are classified according to stratigraphic characteristics, structural characteristics, and reservoir development characteristics, and standardized data format standards are formulated for each type of data entry.

[0203] Preferably, in one embodiment, the knowledge management system design module is configured to carry out the following platform function design during the process of establishing the geological knowledge management system: including data content management function design, platform module design, knowledge content management design, platform framework design, interactive interface design and visual interface design.

[0204] Furthermore, in one embodiment, the pattern and regularity mining module includes:

[0205] A geological body intelligent identification unit configured to realize identification of different types of geological bodies using a pre-built geological body identification model based on data in a geological data knowledge base;

[0206] A geological model classification and extraction unit, which is configured to extract and construct geological models based on geological body combination model samples from multi-source knowledge. The geological body combination model from multi-source knowledge includes geological information models extracted from outcrops, geological models of dense well network anatomy, and classic hand-drawn models;

[0207] The geological body law acquisition unit is configured to acquire the statistical laws between different geological bodies by analyzing the correlation between different geological bodies, so as to support the inter-well fracture characteristic analysis centered on the fault.

[0208] In an optional embodiment, the sample library construction module intelligently constructs a training image library according to the following operations:

[0209] Using the data in the geological data knowledge base to construct training images based on outcrop extraction, dense well pattern dissection, seismic hollowing and hand-drawing; and

[0210] The pre-built multi-source data source training image library is used to realize the automatic vectorization of hand-drawn geological body pattern images and construct training images.

[0211] Furthermore, in one embodiment, the sample library construction module is configured as follows:

[0212] Based on the fully constructed geological body training data, a geological body knowledge simulator is trained and constructed using neural network technology. Random noise, conditional well data and probability body parameters are input into the geological body knowledge simulator according to needs to generate different geological body models.

[0213] Preferably, in one embodiment, the neural network model building module designs a loss function for training a convolutional neural network according to the following operations:

[0214] The prototype model data set and the corresponding wellbore data and seismic data are used as the training data set. The three-dimensional seismic data and three-dimensional wellbore data corresponding to each prototype model are used as the input of the convolutional neural network to generate a three-dimensional geological model. The difference between all prototype models and the three-dimensional geological models generated by their corresponding convolutional neural networks is used as the loss function when training the convolutional neural network.

[0215] In the deep learning-based oil and gas reservoir geological modeling system provided by the embodiment of the present invention, each module or unit structure can operate independently or in combination according to the knowledge data processing requirements and model training requirements to achieve corresponding technical effects.

[0216] It should be understood that the embodiments disclosed in the present invention are not limited to the specific structures, processing steps or materials disclosed herein, but should be extended to equivalent substitutions of these features understood by ordinary technicians in the relevant field. It should also be understood that the terms used herein are only used for the purpose of describing specific embodiments and are not meant to be limiting.

[0217] The "one embodiment" mentioned in the specification means that a particular feature, structure or characteristic described in conjunction with the embodiment is included in at least one embodiment of the present invention. Therefore, the phrase "one embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment.

[0218] Although the embodiments disclosed in the present invention are as above, the above contents are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the patent protection scope of the present invention shall still be subject to the scope defined in the attached claims.

Claims

1. A method for oil and gas reservoir geological modeling based on deep learning, characterized in that: The method comprises: Research data acquisition steps: Based on the reservoirs in the set research area, determine the matching geological knowledge system according to the reservoir type of the reservoir, determine the typical geological blocks of the reservoir to be studied based on it, and obtain multi-source typical geological data of different reservoirs as research data; Steps for setting the storage format: formulate matching storage data formats for various research data collected according to data sources; Knowledge management system design steps: Carry out platform function design according to different types, dimensions and scale characteristics of data, and establish a geological knowledge management system; The geological knowledge base construction step is: entering various research data into the geological knowledge management system using matching storage data formats to form a geological data knowledge base corresponding to the reservoir in the study area; Digital outcrop construction steps: Use unified coordinate system technology to organically combine the data in the geological knowledge base with the drone photography results and 3D laser scanning results to form a digital outcrop model that integrates the inside and outside of the geological body; Pattern and law mining step: Based on the data in the geological data knowledge base, various types of geological knowledge information are mined using artificial intelligence deep learning methods, including geological body information, geological pattern information and laws between geological bodies; Sample library construction step: intelligently constructing a training image library based on the data in the geological data knowledge base, and then constructing a training sample library and a prototype model data set; Neural network model construction steps: building a neural network structure based on a three-dimensional convolutional neural network structure, outputting a three-dimensional geological model, designing a loss function for training the convolutional neural network based on the prototype model data set and the three-dimensional geological model, and training model parameters based on the loss function using a parallel computing approach to obtain a target neural network model; Reservoir geological model determination step: inputting the seismic data and wellbore data of the reservoir of the required block into the target neural network model to obtain the corresponding three-dimensional reservoir geological model.

2. The method according to claim 1, characterized in that In the research data acquisition step, the process of determining the typical geological blocks of the reservoir to be studied based on the geological knowledge system includes: According to the correlation between reservoir elements and structures of the geological knowledge system, a corresponding geological knowledge classification system is further constructed, and according to the knowledge classification system, typical geological blocks with reservoir data acquisition requirements are determined, including various field outcrop geological blocks and typical dense well network areas.

3. The method according to claim 1, characterized in that In the research data acquisition step, the process of obtaining multi-source typical geological data of different reservoirs includes: When obtaining geological data of different typical field outcrops of reservoirs, it includes development information of different geological bodies, scanning information inside geological bodies, scanning information around geological bodies, scanning information of specific details of geological bodies, aerial stereo scanning information and satellite telemetry data; When obtaining geological data of reservoir development areas with different typical dense well patterns, it includes single well drilling data, mud logging data, well logging data, dynamic data, seismic data and geological research profile data.

4. The method according to claim 1, characterized in that: Determine the data format for different research data matching according to the following logic: For the collected outcrop data, the multi-scale and multi-source outcrop geological information is classified and graded according to the quantitative description system of outcrop reservoirs, and the relevant data format standards are formulated; Reservoir data are classified according to stratigraphic characteristics, structural characteristics, and reservoir development characteristics, and standardized data format standards are formulated for each type of data entry.

5. The method according to claim 1, characterized in that In the knowledge management system design step, in the process of establishing the geological knowledge management system, the following platform function designs are carried out: including data content management function design, platform module design, knowledge content management design, platform framework design, interactive interface design and visual interface design.

6. The method according to claim 1, characterized in that The pattern and regularity mining steps include: Intelligent identification steps of geological bodies: using the pre-built geological body identification model based on the data in the geological data knowledge base to realize the identification of different types of geological bodies; Geological model classification and extraction steps: Extract and construct geological models based on geological body combination model samples from multi-source knowledge. The geological body combination model from multi-source knowledge includes geological information model extracted from outcrops, geological model of dense well network anatomy, and classic hand-drawn model. Steps for obtaining the laws between geological bodies: By analyzing the correlation between different geological bodies, the statistical laws between different geological bodies are obtained to support the analysis of inter-well fracture characteristics centered on faults.

7. The method according to claim 1, characterized in that In the sample library construction step, the intelligent construction of the training image library includes: Using the data in the geological data knowledge base to construct training images based on outcrop extraction, dense well pattern dissection, seismic hollowing and hand-drawing; and The pre-built multi-source data source training image library is used to realize the automatic vectorization of hand-drawn geological body pattern images and construct training images.

8. The method according to claim 1, characterized in that In the sample library construction step, the methods of intelligently constructing a training sample library include: Based on the fully constructed geological body training data, a geological body knowledge simulator is trained and constructed using neural network technology. Random noise, conditional well data and probability body parameters are input into the geological body knowledge simulator according to needs to generate different geological body models.

9. The method according to claim 1, characterized in that: In the neural network model construction step, the process of designing a loss function for training a convolutional neural network includes: The prototype model data set and the corresponding wellbore data and seismic data are used as the training data set. The three-dimensional seismic data and three-dimensional wellbore data corresponding to each prototype model are used as the input of the convolutional neural network to generate a three-dimensional geological model. The difference between all prototype models and the three-dimensional geological models generated by their corresponding convolutional neural networks is used as the loss function when training the convolutional neural network.

10. A deep learning-based oil and gas reservoir geological modeling system, characterized in that: The system executes the method according to any one of claims 1 to 9.

Images