WO2024032604A1 - 储层属性建模方法及装置 - Google Patents

储层属性建模方法及装置 Download PDF

Info

Publication number
WO2024032604A1
WO2024032604A1 PCT/CN2023/111727 CN2023111727W WO2024032604A1 WO 2024032604 A1 WO2024032604 A1 WO 2024032604A1 CN 2023111727 W CN2023111727 W CN 2023111727W WO 2024032604 A1 WO2024032604 A1 WO 2024032604A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
reservoir
work area
network
lithology
Prior art date
Application number
PCT/CN2023/111727
Other languages
English (en)
French (fr)
Inventor
张克鑫
张超前
陈和平
徐芳
孟征
郭松伟
武军昌
王玉生
韩彬
田园
Original Assignee
中国石油天然气股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中国石油天然气股份有限公司 filed Critical 中国石油天然气股份有限公司
Publication of WO2024032604A1 publication Critical patent/WO2024032604A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
    • Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping

Definitions

  • the present invention relates to the field of oil reservoir development, and in particular to a reservoir property modeling method and device.
  • geological models include reservoir structure models and reservoir attribute models.
  • the reservoir attribute model is particularly important for reservoir management and development plan optimization because it can directly provide basic input parameters for the reservoir numerical simulation model.
  • reservoir property modeling methods usually include two types: deterministic modeling and stochastic modeling. in:
  • the deterministic modeling method is mainly constructed based on the logging data of the work area and methods such as Kriging interpolation. For each input, there is only one deterministic output.
  • Stochastic modeling methods mainly including sequential Gaussian modeling, multi-point geostatistical modeling (MPS) and other methods, are mainly based on the input logging data to obtain statistical characteristics, including variance, variation, average, etc., and then based on Monte Carlo sampling to generate stochastic models.
  • Monte Carlo sampling when the input is determined, the results of each Monte Carlo sampling modeling have a certain degree of randomness on the premise that they comply with the statistical characteristics, and the modeling results are different each time.
  • reservoir property modeling is mainly based on stochastic modeling methods.
  • existing stochastic modeling of reservoir properties has the problem of inconsistency between dynamic and static data.
  • Embodiments of the present invention provide a reservoir attribute modeling method to improve the dynamic and static data consistency rate of random modeling of reservoir attributes.
  • the method includes:
  • the work area well logging data includes: work area well logging sedimentary facies data, work area well logging sand body interlayer data and work area well logging data. attribute data;
  • an adversarial generation network is a deep convolutional adversarial generation network based on multi-level well-seismic combination, in which the preset type of sedimentary phase training template data selected from the sedimentary phase training template library is used as the initial input data of the network;
  • the second confrontation generation network is a geological attribute generation network based on lithology constraints
  • the reservoir attribute model is modified through the reservoir attribute model of the reservoir numerical simulation agent and the pre-established third confrontation generation network.
  • the third confrontation generation network is based on attributes under dynamic constraints. Optimize your network.
  • Embodiments of the present invention provide a reservoir attribute modeling device to improve the dynamic and static data consistency rate of random modeling of reservoir attributes.
  • the device includes:
  • the data acquisition module is used to obtain the logging data of the work area, the production dynamic data of the work area, the sedimentary facies training template library and the lithology training template library.
  • the logging data of the work area includes: the logging sedimentary facies data of the work area, the logging sand body isolation data of the work area. Interlayer data and work area logging attribute data;
  • the reservoir lithology distribution model generation module is used to generate a network based on well logging sedimentary facies data in the work area, sand body interlayer logging data in the work area, sedimentary facies training template library and lithology training template library, as well as the pre-established first confrontation generation network. Generate a reservoir lithology distribution model.
  • the first adversarial generation network is a deep convolution adversarial generation network based on multi-level well-seismic combination, in which the preset type of sedimentary phase training template data is selected from the sedimentary phase training template library. As the initial input data of the network;
  • the reservoir attribute model generation module is used to generate a reservoir attribute model based on the well logging attribute data of the work area, the reservoir lithology distribution model and a pre-established second confrontation generation network.
  • the second confrontation generation network is based on lithology constraints.
  • the reservoir attribute model optimization module is used to modify the reservoir attribute model based on the production dynamic data of the work area, the reservoir attribute model through the reservoir numerical simulation agent, and the pre-established third confrontation generation network.
  • the third confrontation Generative networks are based on attribute optimization networks under dynamic constraints.
  • An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor.
  • the processor executes the computer program, it implements the above reservoir property modeling method. .
  • Embodiments of the present invention also provide a computer-readable storage medium, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above reservoir property modeling method is implemented.
  • An embodiment of the present invention also provides a computer program product, which includes a computer program that implements the above reservoir property modeling method when executed by a processor.
  • the embodiments of the present invention obtain the well logging data of the work area, the production dynamic data of the work area, the sedimentary facies training template library and the lithology training template library.
  • the data includes: logging sedimentary facies data in the work area, logging sand body interlayer data in the work area and logging attribute data in the work area; based on the logging sedimentary facies data in the work area, logging sand body interlayer data in the work area, sedimentary facies training template library and lithology
  • the training template library and the pre-established first adversarial generation network generate the reservoir lithology distribution model.
  • the first adversarial generation network is a deep convolution adversarial generation network based on multi-level well-seismic combination, in which the sedimentary facies will be generated from
  • the preset type of sedimentary facies training template data selected in the training template library is used as the initial input data of the network;
  • the reservoir attribute model is generated based on the well logging attribute data of the work area, the reservoir lithology distribution model and the pre-established second adversarial generation network.
  • the second adversarial generation network is a geological attribute generation network based on lithological constraints; based on the production dynamic data of the work area, the reservoir attribute model of the reservoir numerical simulation agent and the pre-established third adversarial generation network, the reservoir attributes are The model is modified, and the third adversarial generation network is based on an attribute optimization network under dynamic constraints.
  • the embodiment of the present invention introduces dynamic production content into the reservoir attribute modeling process, combines the sedimentary facies training template library and lithology training template library of well logging data in the work area, and sequentially generates the reservoir lithology distribution model and lithology distribution model based on the artificial intelligence adversarial generation network.
  • the reservoir attribute model is then modified based on the dynamic production content, effectively constructing a reservoir attribute model that is consistent with dynamic and static conditions, and improving the dynamic and static data consistency rate of random modeling of reservoir attributes.
  • Figure 1 is a schematic diagram of a reservoir attribute modeling method in an embodiment of the present invention
  • Figure 2 is a schematic diagram of a method for establishing a template library in an embodiment of the present invention
  • Figure 3 is a flow chart of reservoir attribute modeling in a specific embodiment of the present invention.
  • Figure 4 is a schematic diagram of reservoir attribute model generation in the embodiment of the present invention.
  • Figure 5 is a structural diagram of a reservoir attribute modeling device in an embodiment of the present invention.
  • Figure 6 is a schematic structural diagram of a computer device according to an embodiment of the present invention.
  • Artificial intelligence is based on computer implementation and produces a model with certain intelligence that can respond similarly to human intelligence. It mainly involves robots, language recognition, image recognition, natural language processing and expert systems.
  • Dynamic and static data generally refers to changes in liquid production during the reservoir production process. The dynamic data characteristics are directly related to the static parameter characteristics of the reservoir. Static data generally refers to initial porosity, permeability, saturation, etc. in reservoir analysis. Static data also includes seismic data and other data that do not change with the development process.
  • Work area production dynamic data that is, dynamic data, which is the data on changes in oil production, water production, etc. over time during the oil field production process.
  • This data package comes with corresponding engineering parameter data.
  • Geological modeling Based on the well logging geological data, well logging data, seismic data, etc. in the oil reservoir working area, computer technology is used to generate a three-dimensional static model of the reservoir. Generally, it includes reservoir porosity model, permeability model, and saturation model. This model will be used for reservoir numerical simulation research, predicting changes in future production data, and guiding well location design, etc.
  • Reservoir attribute modeling Based on the well logging data in the work area, a static attribute model of the reservoir is constructed, including a reservoir porosity model, permeability model, and saturation model.
  • Sedimentation dynamics simulation Based on the depositional environment parameters of the study area, establish a depositional model consistent with the depositional facies of the study area, carry out depositional simulation, and obtain a three-dimensional depositional data volume under specific depositional environment parameters.
  • Encoder Also called autoencoder, it is a widely used neural network in the field of artificial intelligence that automatically encodes two-dimensional and three-dimensional image data.
  • the autoencoder consists of two parts: encoder and decoder.
  • Autoencoders have the function of representing learning algorithms in a general sense and are used in dimensionality reduction feature analysis.
  • Hidden variable dimensionality reduction feature array obtained by the encoder.
  • It is a set of combined application networks of deconvolution, depooling and nonlinear mapping, which can generate three-dimensional data images based on the input feature array.
  • Discriminator It is a combined application network of convolution, pooling and non-linear mapping, which can generate three-dimensional data images based on the input feature array.
  • Multi-point geostatistical modeling A method that uses three-dimensional images as a training template, extracts statistical features between data from the template, and then builds a model through Monte Carlo sampling based on the statistical features.
  • Template The basic image used in MPS modeling. The MPS method is based on the statistical characteristics between the statistical attribute data of this image. Templates include: three-dimensional sedimentary facies template and three-dimensional lithology distribution template.
  • Lithology Reservoir sand bodies are classified as interlayers, with sand bodies as 1 and interlayers as 0.
  • inventions of the present invention provide a reservoir attribute modeling method, as shown in Figure 1.
  • the method may include:
  • Step 101 Obtain the work area logging data, the work area production dynamic data, the sedimentary facies training template library and the lithology training template library.
  • the work area logging data includes: the work area logging sedimentary facies data, the work area logging sand body interlayer data and Logging attribute data of the work area;
  • Step 102 Generate a reservoir lithology distribution model based on the well logging sedimentary facies data in the work area, the sand body interlayer logging data in the work area, the sedimentary facies training template library and the lithology training template library, as well as the pre-established first adversarial generation network.
  • the first adversarial generation network is a deep convolutional adversarial generation network based on multi-level well-seismic combination, in which the preset type of sedimentary phase training template data selected from the sedimentary phase training template library is used as the initial input data of the network;
  • Step 103 Generate a reservoir attribute model based on the well logging attribute data of the work area, the reservoir lithology distribution model and the pre-established second confrontation generation network.
  • the second confrontation generation network is a geological attribute generation network based on lithology constraints. ;
  • Step 104 Modify the reservoir attribute model based on the production dynamic data of the work area through the reservoir attribute model of the reservoir numerical simulation agent and the pre-established third confrontation generation network.
  • the third confrontation generation network is based on dynamic constraints. Optimize the network under the properties below.
  • the embodiment of the present invention obtains the well logging data of the work area, the production dynamic data of the work area, the sedimentary facies training template library and the lithology Training template library, the work area logging data includes: work area logging sedimentary facies data, work area logging sand body interlayer data and work area logging attribute data; according to the work area logging sedimentary facies data, work area logging sand body interlayer data , a sedimentary facies training template library and a lithology training template library, as well as a pre-established first confrontation generation network to generate a reservoir lithology distribution model.
  • the first confrontation generation network is a deep convolution confrontation based on multi-level well-seismic combination. Generate a network, in which the preset type of sedimentary phase training template data selected from the sedimentary phase training template library is used as the initial input data of the network; it is generated based on the well logging attribute data of the work area, the reservoir lithology distribution model and the pre-established second confrontation network to generate a reservoir attribute model.
  • the second adversarial generation network is a geological attribute generation network based on lithology constraints; according to the production dynamic data of the work area, the reservoir attribute model through the reservoir numerical simulation agent and the pre-established third
  • the third adversarial generation network is used to modify the reservoir attribute model.
  • the third adversarial generation network is based on an attribute optimization network under dynamic constraints.
  • the embodiment of the present invention introduces dynamic production content into the reservoir attribute modeling process, and combines the well logging data of the work area with the sedimentary facies training template. library and lithology training template library, and sequentially generate the reservoir lithology distribution model and reservoir attribute model based on the artificial intelligence adversarial generation network, and then modify the reservoir attribute model based on dynamic production content, effectively constructing a dynamic and static consistent
  • the reservoir attribute model improves the consistency rate of dynamic and static data in random modeling of reservoir attributes.
  • the reservoir attribute model as the basic input parameter, is input into the reservoir numerical simulation model to predict the well logging dynamic production curve in the work area. By comparing the difference between the well logging production curve obtained by numerical simulation and the actual production curve of the oil field, it can be determined whether there is a dynamic and static inconsistency problem in the reservoir attribute model.
  • the reservoir property model constructed based on the conventional stochastic modeling method basically does not have the problem of dynamic and static inconsistency. Otherwise, reverse adjustment is required, through history matching and a large number of corrections, to adjust the reservoir property model until the dynamic and static data match.
  • the history matching method to solve the problem of dynamic and static inconsistency in existing random modeling of reservoir attributes.
  • the simulation workload is very large.
  • the model needs to be continuously revised in reverse, and each revision requires a reservoir numerical simulation, which consumes huge calculation data.
  • the existing history matching method lacks consideration of the geological laws between wells when making reverse adjustments to the reservoir attribute model.
  • the embodiment of the present invention is based on the artificial intelligence method and combines the existing sedimentary dynamics simulation data, seismic data and well logging sand body interlayer data in the work area, as well as the dynamic production data in the work area, to construct a three-dimensional reservoir attribute model that is consistent with dynamic and static conditions. .
  • sedimentation dynamics data is the preliminary basic data required for the invention.
  • the specific implementation includes three parts: artificial intelligence-based sedimentary facies modeling, artificial intelligence-based lithology modeling and artificial intelligence-based attribute modeling. It adopts a hierarchical control method and is based on the encoder, generator and discriminator network to generate a consistent Attribute model of static data and dynamic production data in the study area.
  • step 101 the work area logging data, the work area production dynamic data, the sedimentary facies training template library and the lithology training template library are obtained.
  • the work area logging data includes: the work area logging sedimentary facies data, the work area logging sand body interlayers data and work area logging attribute data.
  • a sedimentary facies training template library and a lithology training template library are pre-established as follows:
  • Step 201 Obtain sedimentary phase information and sand body interlayer model result data volume.
  • the sedimentary phase information includes: sedimentation dynamics data, existing sedimentary phase diagrams, satellite images, one or any combination thereof;
  • Step 202 Establish a sedimentary phase training template library based on the sedimentary phase information
  • Step 203 Establish a lithology training template library based on the sand body interlayer mode result data volume.
  • a first preset number of sedimentation phase training template data volumes can be extracted from the sedimentation dynamics data; a second preset number of sedimentation phase training template data volumes can be extracted from the water tank experimental data volume. ; Merge the extracted first preset number of sedimentary phase training template data volumes and the second preset number of sedimentary phase training template data volumes to establish a sedimentary phase training template library.
  • FIG 3 is a flow chart of reservoir attribute modeling in a specific embodiment of the present invention.
  • p sedimentary phase training templates are extracted from the three-dimensional data volume of the sedimentary phase simulation result data in the sedimentary dynamics data, that is, ⁇ SDi
  • i 1...p ⁇ , and a sedimentary phase training template library is established.
  • the length, width, height and number of channels of each template data volume SDi are SL, SW, SH and SC.
  • the sedimentation water tank simulation is completed using the scanning water tank device, and the water tank experimental data volume is obtained.
  • the scanning water tank device is shown in Figure 4.
  • i 1...q ⁇ , from the three-dimensional data volume of the flume experimental data, and establish a sedimentary phase training template library.
  • the length, width, height and number of channels of each template data volume SCi are SL, SW, SH and SC.
  • i 1...n ⁇ , and establish a sedimentary phase training template library.
  • the length, width, height and number of channels of each template data volume Si are SL, SW, SH and SC.
  • h lithology training template data volumes are extracted for each sedimentary phase area, ⁇ Lis
  • the possible value range of Lis is ⁇ 0,1 ⁇ , where 0 is sandstone and 1 is mudstone.
  • the length, width, height and number of channels of each template data volume Lis are LL, LW, LH and LC.
  • the sink device is a device that simulates the deposition of particulate matter in a liquid. Devices for transport and deposition phenomena.
  • the water tank device can be used to reproduce the sedimentation process and analyze the structural characteristics of the internal sedimentary body.
  • the scanning water tank device is a combination of a large CT scanning device and a water tank device.
  • the large CT scanning device is used to three-dimensionally scan the water tank sedimentation results to obtain a three-dimensional model of the water tank results, which can clearly display the internal configuration and structure of different sedimentary bodies, and the scanning results Stored in the form of a three-dimensional data volume, artificial intelligence analysis can be applied.
  • the reservoir lithology distribution is generated based on the well logging sedimentary facies data of the work area, the sand body interlayer logging data of the work area, the sedimentary facies training template library and the lithology training template library, and the pre-established first adversarial generation network.
  • the first adversarial generation network is a deep convolutional adversarial generation network based on multi-level well-seismic combination, in which the preset type of sedimentary phase training template data selected from the sedimentary phase training template library is used as the initial input data of the network.
  • the first adversarial generation network includes: a first adversarial generation sub-network based on sedimentation dynamics simulating sedimentary facies patterns, a second adversarial generation sub-network based on well log seismic sedimentary facies patterns, under sedimentation facies constraints
  • the third adversarial generation sub-network based on sedimentary dynamics simulation lithology model, and the fourth adversarial generation sub-network based on well log seismic lithology model under the constraints of sedimentary facies;
  • the reservoir lithology distribution model is generated, including:
  • the well logging sedimentary facies data of the work area is input into the second adversarial generation sub-network for training.
  • multi-point geostatistical modeling is performed to generate a reservoir sedimentary facies model.
  • the second adversarial generation sub-network is based on the trained
  • the first adversarial generator sub-network is established;
  • the logging sand body interlayer data of the work area is input into the fourth adversarial generation sub-network for training. Based on the training results and the reservoir sedimentary facies model, multi-point geological statistical modeling is performed to generate a reservoir lithology distribution model.
  • the fourth adversarial generator sub-network is established based on the trained third adversarial generator sub-network.
  • multiple sedimentary phase training template data volumes in the sedimentary phase training template library are input into the first adversarial generation subnetwork for training, including:
  • the preset type sedimentary phase training template data selected from the sedimentary phase training template library is used as the initial input data of the network, and is input into the first adversarial generation sub-network for training;
  • the well logging sedimentary facies data of the work area is input into the second adversarial generation sub-network for training, and multi-point geostatistical modeling is performed based on the training results to generate a reservoir sedimentary facies model, including: based on the trained first adversarial generation sub-network Establish a second adversarial generation sub-network, input the logging sedimentary facies data of the work area into the second adversarial generation sub-network for training, and output a sedimentary facies template that represents the characteristics of the logging sedimentary facies data of the work area. Multi-point geostatistical modeling is performed using facies templates and actual well data to generate a reservoir depositional facies model.
  • a corresponding encoder E1 and generation G1 are established based on the length, width, height and number of channels of each template data volume Si.
  • the input layer size of the encoder module EE1 of E1 is SL ⁇ SW ⁇ SH ⁇ SC, and the batch training number is SM.
  • the output layer size of the decoder module ED1 of E0 is SL ⁇ SW ⁇ SH ⁇ SC, and the batch training number is SM.
  • For Sedimentary phase SC 1.
  • G,1 is a copy of E1's decoder ED1.
  • the encoder module EE1 of the encoder E1 is input, and the decoder module ED1 of E1 outputs the result ⁇ SOj
  • j k...k+SM-1 ⁇ , use the input discriminator D1 to compare ⁇ Sj
  • j k...k+SM-1 ⁇ and ⁇ SOj
  • j k...k+SM-1 ⁇
  • the error of E1 that is, the first error term
  • the error of E1 is obtained by reversely correcting the encoder module and decoder module of E1, so that ⁇ Sj
  • j k...k+SM-1 ⁇ and ⁇ SOj
  • j k...k+SM-1 ⁇ The error is the smallest.
  • the output of E1's encoder EE1 is a hidden variable SZ that can effectively characterize the three-dimensional data volume characteristics of the sedimentary phase simulation result data. Its length, width, height and number of channels are SZL, SZW, SZH and SZC, and the design size is SZL ⁇ SZW ⁇ SZH ⁇ SZC.
  • the decoder module ED1 of E1 can generate a sedimentation phase template ED_S that represents the characteristics of the sedimentation phase simulation result data volume based on SZ and random noise.
  • WS [WSD1, WSS1, WSD2, WSS2,..., WSDK, WSSK]
  • WSD is the coordinate data
  • WSS is the sedimentary facies number data
  • WS is the 2K ⁇ D image
  • D is the number of data points of sedimentary facies logged in each work area.
  • the hidden layer WZ output by EE2 is The length, width, height and number of channels are SZL, SZW, SZH and SZC, and the size is SZL ⁇ SZW ⁇ SZH ⁇ SZC, which is consistent with the size of SZ.
  • the hidden layer WZ of E2 after training can be used as a hidden variable to characterize the characteristics of the well logging sedimentary facies data in the work area.
  • the generator G2 is established, the size of its input layer is SZL ⁇ SZW ⁇ SZH ⁇ SZC, and the length, width, height and number of channels of the output layer are SL, SW, SH and SC.
  • G2 initial settings copy the parameters of G1 to use the learned deposition mode.
  • G2S as a template and actual well data as constraints, multi-point geostatistical modeling was used to obtain the reservoir sedimentary phase model FS.
  • the modeling is first carried out based on the use of well data and sedimentary facies training network.
  • the sedimentary facies training network will adjust the sedimentation on its own based on error feedback.
  • the facies type output is to correct people's understanding to achieve the optimal sedimentary facies model; then, based on the lithology data, he will also adjust the output results of the interlayer model accordingly.
  • the principle is the same as above.
  • the first step is to enable the artificial intelligence network to have the ability to generate the different deposition phases required, but which deposition phase the artificial intelligence network actually does not know. Therefore, we will tell the artificial intelligence network to generate sedimentary phase type A based on basic research knowledge to start modeling, and automatically correct the sedimentary phase type during continuous modeling. The final result may be sedimentary phase B. However, our modeling results usually cannot withstand dynamic data verification.
  • the last step is to add dynamic data to correct the geological model to unify static and dynamic conditions. Since it is carried out in hierarchical classification, this process is more intelligent and convenient than letting the artificial intelligence network slowly explore on its own, and the calculation will be more stable and efficient. It also makes the entire modeling method expandable. If there are completely new sedimentary facies types, just update the initial template library.
  • A For example, if you initially think it is A, but it is actually B, and then use A to train the network, the calculated error is large, and the model will automatically find B data from the template library. And due to the characteristics of deep learning, a deposition type between A and B may be generated.
  • the core idea is to use artificial intelligence networks to help correct deposition understanding and increase the accuracy of modeling.
  • the reservoir property modeling method further includes:
  • the sedimentary facies template that characterizes the well logging sedimentary facies data characteristics in the work area and the actual well data are used for multi-point geostatistical modeling to generate a reservoir sedimentary facies model.
  • the discriminator D2 is used to calculate the comprehensive error between FS, seismic attribute data volume SEGY and work area logging data, that is, the second error term.
  • the calculation error of D2 is used to correct G2 until the sedimentary phase model generated based on the sedimentary phase template generated by G2 has the smallest error with the well logging and seismic in the work area, and the optimal reservoir sedimentary phase model OFS is output.
  • multiple lithology training template data volumes in the lithology training template library are input into a third adversarial generation subnetwork for training, including:
  • the logging sand body interlayer data in the work area is input into the fourth adversarial generation sub-network for training, and multi-point geostatistical modeling is performed based on the training results and the reservoir sedimentary facies model to generate a reservoir lithology distribution model, including : Establish a fourth adversarial generation sub-network based on the trained third adversarial generation sub-network, input the logging sand body interlayer data in the work area into the fourth adversarial generation sub-network for training, and output the lithology characterizing the characteristics of the logging lithology data in the work area Distribution template: perform multi-point geostatistical modeling based on the lithology distribution template that characterizes the logging lithology data characteristics of the work area and actual well data to generate a reservoir lithology distribution model.
  • the corresponding encoder E3 and generator G3 are established according to the length, width, height and number of channels of each template data volume Lis.
  • the input layer size of E3's encoder module EE3 is LL ⁇ LW ⁇ LH ⁇ LC, and the batch training number is LM.
  • the output layer size of E3's decoder module ED3 is LL ⁇ LW ⁇ LH ⁇ LC, and the batch training number is LM.
  • For Sedimentary phase LC 1.
  • G3 is a copy of E3's decoder module ED3.
  • the encoder module EE3 of the encoder E3 is input, and the decoder module ED3 of E3 outputs the result ⁇ LOj
  • j k...k+LM-1 ⁇ , use the compliance module to compare ⁇ Lj
  • j k..k+LM-1 ⁇ and ⁇ LOj
  • j k...k+LM-1 ⁇ , and get E3 Error, by reversely correcting the encoder module and decoder module of E3, let ⁇ Lj
  • j k...k+LM-1 ⁇ and ⁇ LOj
  • j k...k+LM-1 ⁇ errors Minimum.
  • the output of the encoder module EE3 of E3 is the hidden variable LZ that can effectively characterize the three-dimensional data volume characteristics of the sand body interlayer model result data. Its length, width, height and number of channels are LZL, LZW, LZH and LZC. The design size It is LZL ⁇ LZW ⁇ LZH ⁇ LZC.
  • the decoder module ED3 of E3 can generate a lithology distribution template ED_L that represents the characteristics of the sand body interlayer model result data volume based on LZ and random noise.
  • the output of EE4 The length, width, height and number of channels of the hidden layer LZ are LZL, LZW, LZH and LZC, and the size is LZL ⁇ LZW ⁇ LZH ⁇ LZC, which is consistent with the size of LZ.
  • the hidden layer WZ of E4 after training can be used as a parameter to characterize the logging core characteristics of the upper work area in the entire area.
  • the generator G4 is established, the size of its input layer is LZL ⁇ LZW ⁇ LZH ⁇ LZC, and the length, width, height and number of channels of the output layer are LL, LW, LH and LC.
  • G4 initial settings copy the parameters of G3 to use the learned deposition pattern.
  • G1L as a template, taking actual well data as constraints, and using multi-point geostatistical modeling, the reservoir lithology distribution model FLs of the s sedimentary facies area was obtained.
  • the reservoir property modeling method further includes:
  • discriminator D4 is used to calculate the comprehensive error between FLs, seismic attribute data volume SEGY and work area logging data, that is, the fourth error term. D4 calculates the error and is used to correct G4 until the sedimentary facies model generated based on the three-dimensional sand body interlayer template generated by G4 has the smallest error with the well logging and seismic in the work area, and the optimal reservoir lithology distribution model of the sth sedimentary facies area is output.
  • OFLs Traverse all n sedimentary areas, perform the above modeling process for each sedimentary facies area, complete the intelligent lithology modeling of all n sedimentary facies areas, and obtain the optimal reservoir lithology distribution model OFL.
  • a reservoir attribute model is generated based on the well logging attribute data of the work area, the reservoir lithology distribution model and a pre-established second confrontation generation network.
  • the second confrontation generation network is based on the geological attributes under lithology constraints. Generate network.
  • the logging attribute data of the work area includes: logging porosity attribute data of the work area, logging permeability attribute data of the work area, and logging saturation attribute data of the work area;
  • the reservoir attribute model is generated, including:
  • WPq [WPD1q, WPP1q, WPD2q, WPP2q,..., WPDKq, WPPKq]
  • WPD is the coordinate Data
  • WPP is the porosity data
  • WP is the 2K ⁇ D image, where D is the number of data points of the logging attribute data of each working area, and q is the attribute number to be analyzed.
  • the input layer size of the encoder module EE5 of E5 is 2K ⁇ D, and the batch training number is PW.
  • the output layer size of the decoder module ED5 of E5 is 2K ⁇ D, and the batch training number is PW.
  • the output of EE5 The length, width, height and number of channels of the hidden layer PZ are PZL, PZW, PZH and PZC, and the size is PZL ⁇ PZW ⁇ PZH ⁇ PZC, which is consistent with the size of PZ.
  • the hidden layer PZq of E5 after training can be used as a parameter to characterize the qth attribute data characteristics of the well logging in the entire area, that is, a hidden variable that represents the characteristics of the logging saturation attribute data in the work area.
  • Build generator G5 the size of its input layer 1 is PZL ⁇ PZW ⁇ PZH ⁇ (2 ⁇ PZC), and the size of the output layer is PZL ⁇ PZW ⁇ PZH.
  • the first PZC channels in the input layer are placed with PZ data, and the following PZC channels are used to extract PZL ⁇ PZW ⁇ PZH lithological data blocks from OFL in a translational traversal manner.
  • G5 completes traversing OFL, as shown in Figure 5, and generates the reservoir attribute model OFPp with the pth attribute. After G4 traversal is completed, OFL generates a three-dimensional reservoir attribute model OFPp. Compared with inputting all OFL into G4 at once, through mobile traversal, the size of the G4 input layer can be reduced and the demand for hardware is reduced.
  • the reservoir attribute models OPFS ⁇ OPF1, OFP2, OPF3 ⁇ for porosity, permeability and saturation attributes are generated in sequence, where OPF1 is the optimal reservoir porosity attribute model generated, and OPF2 is the optimal generated reservoir porosity attribute model.
  • the optimal reservoir permeability attribute model, OPF3 is the optimal reservoir saturation attribute model generated.
  • step 104 the reservoir attribute model is modified according to the production dynamic data of the work area through the reservoir attribute model of the reservoir numerical simulation agent and the pre-established third confrontation generation network.
  • the third confrontation generation network is based on Attribute optimization networks under dynamic constraints.
  • the reservoir property model is modified through the reservoir property model of the reservoir numerical simulation agent and the pre-established third confrontation generation network, including:
  • the fifth error term is obtained by comparing the consistency between the production dynamic data of the work area and the reservoir attribute model through the reservoir numerical simulation agent;
  • the reservoir attribute model output by the third adversarial generation network when the fifth error term reaches the minimum value is used as the modified reservoir attribute model.
  • the reservoir numerical simulation agent RS is used based on OPFS to predict the production parameters of the work area
  • the discriminator D5 is used to calculate the error Err1 between the predicted data and the actual work area production parameters FO, as well as the generated porosity and permeability.
  • the generator G5 is reversely corrected until the error obtained by D5 is the smallest, and the optimal reservoir property model OFP is output.
  • Block X in South America is used as an application example.
  • a sedimentary dynamics model that conforms to the sedimentary environment of Block X was constructed, and 2,000 three-dimensional sedimentary facies simulation result data volumes and 4,000 three-dimensional sand body-interlayer model result data volumes were obtained as training data volumes.
  • the work area logging sedimentary facies data of 400 wells in the study area, the sand body-interlayer data of 400 wells, and the porosity, permeability, and saturation data of 400 wells were collected.
  • the dynamic production data of 400 wells in the study area were compiled.
  • Seismic data volumes in the study area were collected to achieve full coverage of the work area.
  • the generated three-dimensional reservoir attribute model has a coincidence rate of 98.2% with the well logging porosity data of the work area, a coincidence rate with the well log permeability of the work area is 97.3%, and a coincidence rate with the well log saturation data of the work area is 98.5 %, and the coincidence rate with the well logging fluid production data in the work area is 94.1%.
  • the obtained three-dimensional reservoir attribute model is consistent with the well logging porosity data of the work area at a rate of 82.4%, consistent with the well log permeability of the work area at a rate of 83.2%, and consistent with the well log saturation of the work area.
  • the rate is 81.1%
  • the coincidence rate with the well logging fluid production data in the work area is 79.2%.
  • the dynamic and static data consistency rate of the reservoir attribute model obtained by the present invention is higher than that of the existing random modeling method.
  • the embodiment of the present invention is based on a multi-point geostatistical three-dimensional template generation method based on artificial intelligence and a combination of dynamic and static methods. Effectively solves the problem of existing MPS method templates that do not consider dynamic production data.
  • reservoir property modeling under multi-level control based on artificial intelligence.
  • a sedimentary facies model is established based on artificial intelligence methods.
  • a lithology model based on artificial intelligence is established.
  • an attribute model is established based on artificial intelligence.
  • embodiments of the present invention also provide a reservoir property modeling device, as described in the following embodiments. Since these problem-solving principles are similar to the reservoir property modeling method, the implementation of the reservoir property modeling device can be referred to the implementation of the method, and repeated details will not be repeated.
  • FIG. 5 is a structural diagram of a reservoir property modeling device in an embodiment of the present invention. As shown in Figure 5, the reservoir property modeling device includes:
  • the data acquisition module 501 is used to obtain work area logging data, work area production dynamic data, sedimentary facies training template library and lithology training template library.
  • the work area logging data includes: work area logging sedimentary facies data, work area logging sand bodies Interlayer data and work area logging attribute data;
  • the reservoir lithology distribution model generation module 502 is used to generate a first adversarial generation network based on well logging sedimentary facies data in the work area, sand body interlayer logging data in the work area, sedimentary facies training template library and lithology training template library, as well as a pre-established first confrontation generation network , generate a reservoir lithology distribution model.
  • the first adversarial generation network is a deep convolution adversarial generation network based on multi-level well and seismic combination, in which a preset type of sedimentary phase training template will be selected from the sedimentary phase training template library.
  • the data serves as the initial input data of the network;
  • the reservoir attribute model generation module 503 is used to generate a reservoir attribute model based on the well logging attribute data of the work area, the reservoir lithology distribution model, and a pre-established second confrontation generation network.
  • the second confrontation generation network is based on lithology. Geological attribute generation network under constraints;
  • the reservoir attribute model optimization module 504 is used to modify the reservoir attribute model based on the production dynamic data of the work area, the reservoir attribute model of the reservoir numerical simulation agent, and the pre-established third confrontation generation network.
  • the third confrontation generation network is Adversarial generative networks are based on attribute optimization networks under dynamic constraints.
  • an embodiment of the present invention also provides a computer device 600, which includes a memory 610, a processor 620, and a computer program 630 stored on the memory 610 and executable on the processor 620.
  • a computer program 630 stored on the memory 610 and executable on the processor 620.
  • embodiments of the present invention also provide a computer-readable storage medium, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above reservoir property modeling method is implemented.
  • An embodiment of the present invention also provides a computer program product, which includes a computer program that implements the above reservoir property modeling method when executed by a processor.
  • the embodiments of the present invention obtain the well logging data of the work area, the production dynamic data of the work area, the sedimentary facies training template library and the lithology training template library.
  • the data includes: logging sedimentary facies data in the work area, logging sand body interlayer data in the work area and logging attribute data in the work area; based on the logging sedimentary facies data in the work area, logging sand body interlayer data in the work area, sedimentary facies training template library and lithology
  • the training template library and the pre-established first adversarial generation network generate the reservoir lithology distribution model.
  • the first adversarial generation network is a deep convolution adversarial generation network based on multi-level well-seismic combination, in which the sedimentary facies will be generated from
  • the preset type of sedimentary facies training template data selected in the training template library is used as the initial input data of the network;
  • the reservoir attribute model is generated based on the well logging attribute data of the work area, the reservoir lithology distribution model and the pre-established second adversarial generation network.
  • the second adversarial generation network is a geological attribute generation network based on lithological constraints; based on the production dynamic data of the work area, the reservoir attribute model of the reservoir numerical simulation agent and the pre-established third adversarial generation network, the reservoir attributes are The model is modified, and the third adversarial generation network is based on an attribute optimization network under dynamic constraints.
  • dynamic production content is introduced, combined with the sedimentary facies training template library and lithology training template library of well logging data in the work area, and the reservoir lithology distribution model and reservoir attribute model based on the artificial intelligence adversarial generation network are sequentially generated, and then combined with dynamic production
  • the content corrects the reservoir attribute model, effectively constructs a reservoir attribute model that is consistent with dynamic and static conditions, and improves the consistency rate of dynamic and static data in random modeling of reservoir attributes.
  • embodiments of the present invention may be provided as methods, systems, or computer program products.
  • the invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects.
  • the invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • These computer program instructions may also be stored in a computer-readable memory that causes a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction means, the instructions
  • the device implements the functions specified in a process or processes of the flowchart and/or a block or blocks of the block diagram.
  • These computer program instructions may also be loaded onto a computer or other programmable data processing device, causing a series of operating steps to be performed on the computer or other programmable device to produce computer-implemented processing, thereby executing on the computer or other programmable device.
  • Instructions provide steps for implementing the functions specified in a process or processes of a flowchart diagram and/or a block or blocks of a block diagram.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Geometry (AREA)
  • Computer Hardware Design (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明公开了一种储层属性建模方法及装置，包括：根据沉积相训练模板库和岩性训练模板库以及第一对抗生成网络生成储层岩性分布模型，第一对抗生成网络是基于多级井震联合的深度卷积对抗生成网络，将从沉积相训练模板库中选取的预设类型沉积相训练模板数据作为网络初始输入数据；根据工区测井属性数据，储层岩性分布模型以及第二对抗生成网络生成储层属性模型，第二对抗生成网络是基于岩性约束下的地质属性生成网络；根据工区生产动态数据，经过油藏数值模拟代理器的储层属性模型以及第三对抗生成网络对储层属性模型进行修正，第三对抗生成网络是基于动态约束下的属性优化网络。本发明可以有效提升储层属性随机建模动静态数据符合率。

Description

储层属性建模方法及装置 技术领域
本发明涉及油藏开发领域，尤其涉及储层属性建模方法及装置。
背景技术
本部分旨在为权利要求书中陈述的本发明实施例提供背景或上下文。此处的描述不因为包括在本部分中就承认是现有技术。
随着储层开发的深化，油藏精细开发对地质模型的要求越来越高。通常地质模型包括储层构造模型和储层属性模型等。其中储层属性模型，因其可直接为油藏数值模拟模型提供基础输入参数，对油藏管理和开发方案优化尤为重要。
目前储层属性建模方法通常包括两种：确定性建模和随机建模。其中：
①确定性建模方法，主要根据工区测井数据，基于克里金插值等方法构建，对于每次输入，只有1个确定性的输出。
②随机建模方法，主要包括序贯高斯建模、多点地质统计建模(MPS)等方法，主要基于输入的测井数据，获取统计特征，包括方差、变差、平均值等，然后基于蒙特卡洛抽样，生成随机模型。随机建模方法，在输入确定情况下，每次蒙特卡洛抽样建模的结果在符合统计特征的前提下，具有一定的随机性，每次建模结果不尽相同。
相比于随机建模，确定性建模的结果，采用空间插值，结果缺乏对实际地质复杂性的反应。因此目前，储层属性建模主要还是以随机建模方法为主，但是现有的储层属性随机建模存在动静态数据不符的问题。
因此，亟需一种可以克服上述问题的储层属性建模方案。
发明内容
本发明实施例提供一种储层属性建模方法，用以提升储层属性随机建模动静态数据符合率，该方法包括：
获得工区测井数据，工区生产动态数据，沉积相训练模板库和岩性训练模板库，所述工区测井数据包括：工区测井沉积相数据，工区测井砂体隔夹层数据和工区测井属性数据；
根据工区测井沉积相数据，工区测井砂体隔夹层数据，沉积相训练模板库和岩性训练模板库，以及预先建立的第一对抗生成网络，生成储层岩性分布模型，所述第一对抗生成网络是基于多级井震联合的深度卷积对抗生成网络，其中，将从沉积相训练模板库中选取的预设类型沉积相训练模板数据作为网络初始输入数据；
根据工区测井属性数据，储层岩性分布模型以及预先建立的第二对抗生成网络，生成储层属性模型，所述第二对抗生成网络是基于岩性约束下的地质属性生成网络；
根据工区生产动态数据，经过油藏数值模拟代理器的储层属性模型以及预先建立的第三对抗生成网络，对储层属性模型进行修正，所述第三对抗生成网络是基于动态约束下的属性优化网络。
本发明实施例提供一种储层属性建模装置，用以提升储层属性随机建模动静态数据符合率，该装置包括：
数据获得模块，用于获得工区测井数据，工区生产动态数据，沉积相训练模板库和岩性训练模板库，所述工区测井数据包括：工区测井沉积相数据，工区测井砂体隔夹层数据和工区测井属性数据；
储层岩性分布模型生成模块，用于根据工区测井沉积相数据，工区测井砂体隔夹层数据，沉积相训练模板库和岩性训练模板库，以及预先建立的第一对抗生成网络，生成储层岩性分布模型，所述第一对抗生成网络是基于多级井震联合的深度卷积对抗生成网络，其中，将从沉积相训练模板库中选取的预设类型沉积相训练模板数据作为网络初始输入数据；
储层属性模型生成模块，用于根据工区测井属性数据，储层岩性分布模型以及预先建立的第二对抗生成网络，生成储层属性模型，所述第二对抗生成网络是基于岩性约束下的地质属性生成网络；
储层属性模型优化模块，用于根据工区生产动态数据，经过油藏数值模拟代理器的储层属性模型以及预先建立的第三对抗生成网络，对储层属性模型进行修正，所述第三对抗生成网络是基于动态约束下的属性优化网络。
本发明实施例还提供一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现上述储层属性建模方法。
本发明实施例还提供一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序被处理器执行时实现上述储层属性建模方法。
本发明实施例还提供一种计算机程序产品，所述计算机程序产品包括计算机程序，所述计算机程序被处理器执行时实现上述储层属性建模方法。
本发明实施例与现有技术中储层属性随机建模的技术方案相比，通过获得工区测井数据，工区生产动态数据，沉积相训练模板库和岩性训练模板库，所述工区测井数据包括：工区测井沉积相数据，工区测井砂体隔夹层数据和工区测井属性数据；根据工区测井沉积相数据，工区测井砂体隔夹层数据，沉积相训练模板库和岩性训练模板库，以及预先建立的第一对抗生成网络，生成储层岩性分布模型，所述第一对抗生成网络是基于多级井震联合的深度卷积对抗生成网络，其中，将从沉积相训练模板库中选取的预设类型沉积相训练模板数据作为网络初始输入数据；根据工区测井属性数据，储层岩性分布模型以及预先建立的第二对抗生成网络，生成储层属性模型，所述第二对抗生成网络是基于岩性约束下的地质属性生成网络；根据工区生产动态数据，经过油藏数值模拟代理器的储层属性模型以及预先建立的第三对抗生成网络，对储层属性模型进行修正，所述第三对抗生成网络是基于动态约束下的属性优化网络。本发明实施例在储层属性建模过程中引入动态生产内容，结合工区测井数据沉积相训练模板库和岩性训练模板库，依次生成基于人工智能对抗生成网络的储层岩性分布模型和储层属性模型，进而再结合动态生产内容对储层属性模型进行修正，有效构建了动静态相符的储层属性模型，提升了储层属性随机建模动静态数据符合率。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。在附图中：
图1为本发明实施例中储层属性建模方法示意图；
图2为本发明实施例中模板库建立方法示意图；
图3为本发明具体实施例中储层属性建模流程图；
图4为本发明实施例中储层属性模型生成示意图；
图5为本发明实施例中储层属性建模装置结构图；
图6是本发明实施例的计算机设备结构示意图。
具体实施方式
为使本发明实施例的目的、技术方案和优点更加清楚明白，下面结合附图对本发明实施例做进一步详细说明。在此，本发明的示意性实施例及其说明用于解释本发明，但并不作为对本发明的限定。
首先，对本发明实施例中的技术名词进行介绍：
人工智能：人工智能是基于计算机实现，产生具有一定智能的模型，该模型可以做出与人类智能相似的反应，主要涉及机器人、语言识别、图像识别、自然语言处理和专家系统等。
动静态数据：动态数据一般指在油藏生产过程中，产液变化情况。而动态数据特征直接与油藏的静态参数特征关联。静态数据一般指在储层分析中的初始孔隙度、渗透率和饱和度等，静态数据也包括地震数据，等不随开发过程而改变的数据。
工区生产动态数据：也就是动态数据，是油田生产过程中的产油、产水等随时间的变化数据，该数据包附带相应的工程参数数据。
地质建模：基于油藏工区测井地质数据、测井数据、地震数据等，利用计算机技术，生成储层三维静态模型。一般包括储层孔隙度模型、渗透率模型、饱和度模型，该模型将用于油藏数值模拟研究，预测未来生产数据的变化，指导井位设计等。
储层属性建模：基于工区测井数据，构建油藏静态属性模型，包括储层孔隙度模型、渗透率模型、饱和度模型。
沉积动力学模拟：基于研究区沉积环境参数，建立符合研究区沉积相的沉积模型，开展沉积模拟，得到特定沉积环境参数下的三维沉积数据体。
编码器：也叫自编码器，是人工智能领域对二维和三维图像数据进行自动编码的一种广泛使用的一种神经网络。自编码器包含编码器(encoder)和解码器(decoder)两部分。自编码器具有一般意义上表征学习算法的功能，被应用于降维特征分析。
隐变量：编码器得到的降维特征数组。
生成器：是一组反卷积、反池化和非线性映射的组合申请网络，可以根据输入的特征数组，生成三维数据图像。
判别器：是一组卷积、池化和非线性映射的组合申请网络，可以根据输入的特征数组，生成三维数据图像。
多点地质统计建模(MPS建模)：以三维图像作为训练模板，从模板中提取数据之间的统计特征，然后基于统计特征，通过蒙特卡洛抽样构建模型的方法。
模板：MPS建模中使用的基础图像，MPS方法基于该图像统计属性数据之间的统计特征。模板包括：三维沉积相模板、三维岩性分布模板。
岩性：储层砂体隔夹层分类，砂体为1，隔夹层为0。
为了进行储层属性建模，提升储层属性随机建模动静态数据符合率，本发明实施例提供一种储层属性建模方法，如图1所示，该方法可以包括：
步骤101、获得工区测井数据，工区生产动态数据，沉积相训练模板库和岩性训练模板库，所述工区测井数据包括：工区测井沉积相数据，工区测井砂体隔夹层数据和工区测井属性数据；
步骤102、根据工区测井沉积相数据，工区测井砂体隔夹层数据，沉积相训练模板库和岩性训练模板库，以及预先建立的第一对抗生成网络，生成储层岩性分布模型，所述第一对抗生成网络是基于多级井震联合的深度卷积对抗生成网络，其中，将从沉积相训练模板库中选取的预设类型沉积相训练模板数据作为网络初始输入数据；
步骤103、根据工区测井属性数据，储层岩性分布模型以及预先建立的第二对抗生成网络，生成储层属性模型，所述第二对抗生成网络是基于岩性约束下的地质属性生成网络；
步骤104、根据工区生产动态数据，经过油藏数值模拟代理器的储层属性模型以及预先建立的第三对抗生成网络，对储层属性模型进行修正，所述第三对抗生成网络是基于动态约束下的属性优化网络。
由图1所示可以得知，本发明实施例与现有技术中储层属性随机建模的技术方案相比，通过获得工区测井数据，工区生产动态数据，沉积相训练模板库和岩性训练模板库，所述工区测井数据包括：工区测井沉积相数据，工区测井砂体隔夹层数据和工区测井属性数据；根据工区测井沉积相数据，工区测井砂体隔夹层数据，沉积相训练模板库和岩性训练模板库，以及预先建立的第一对抗生成网络，生成储层岩性分布模型，所述第一对抗生成网络是基于多级井震联合的深度卷积对抗生成网络，其中，将从沉积相训练模板库中选取的预设类型沉积相训练模板数据作为网络初始输入数据；根据工区测井属性数据，储层岩性分布模型以及预先建立的第二对抗生成网络，生成储层属性模型，所述第二对抗生成网络是基于岩性约束下的地质属性生成网络；根据工区生产动态数据，经过油藏数值模拟代理器的储层属性模型以及预先建立的第三对抗生成网络，对储层属性模型进行修正，所述第三对抗生成网络是基于动态约束下的属性优化网络。本发明实施例在储层属性建模过程中引入动态生产内容，结合工区测井数据沉积相训练模板 库和岩性训练模板库，依次生成基于人工智能对抗生成网络的储层岩性分布模型和储层属性模型，进而再结合动态生产内容对储层属性模型进行修正，有效构建了动静态相符的储层属性模型，提升了储层属性随机建模动静态数据符合率。
发明人发现，目前储层属性随机建模，仍然存在动静态数据不符的问题。虽然目前建模中充分考虑了工区测井的静态数据，但完全未考虑工区测井的动态数据。而动态数据，对于一些测井解释不确信较大的属性具有很高的参考价值。比如：目前测井方法，基于AC声波测井曲线，可以较为精确得解释孔隙度特征，但对于渗透率和饱和度特征，虽然有多个模型，但在实践中，这两个属性的不确性远远大于孔隙度。因为目前渗透率没有直接的测井解释方案，主要依靠孔渗关系模型进行解释，但目前复杂储层中孔渗关系模型难以用简单的线性模型表示，对于致密储层，孔渗相关性也非常差，因此目前基于测井解释渗透率模型，难度较大，不确定性也较大。而基于测井解释获得的渗透率属性模型，不确定也远远高于孔隙度。同样，储层饱和度的测井解释难度也较大，主要是因为储层中泥质和煤质成分会干扰解释模型，导致计算出的含水饱和度偏离实际较大。常规随机建模方法，仅依赖工区测井数据，当工区测井测井解释数据存在较大偏差时，必然导致模型无法准确预测动态数据，产生动静不符问题。历史拟合是现在解决地质模型动静不符的主要手段。储层属性模型，作为基础输入参数，输入进油藏数值模拟模型后，预测工区测井动态生产曲线。通过对比数值模拟得到的工区测井生产曲线与油田实际的生产曲线的差别，可以确定储层属性模型是否存在动静不符问题。如果预测动态生产区域与实际生产曲线差别较小，则一般认为基于常规随机建模方法，所构建的储层属性模型基本不存在动静不符问题。否则，就需要进行反向调整，通过历史拟合，通过大量认为修正，调整储层属性模型，直到动静数据相符为止。然而，历史拟合方法解决现有储层属性随机建模动静不符问题，仍然存在较多问题。首先是，模拟工作量非常大，历史拟合阶段，需要不断反向修正模型，而每次修正，都需要开展一次油藏数值模拟，计算资料消耗巨大。其次，现有历史拟合方法，在对储层属性模型进行反向调整的时候，缺乏对井间地质规律的考虑，主要是分区分块，对存在动静不符问题的井周属性进行硬性调整，缺乏对邻井数据的考虑。因此，本发明实施例基于人工智能方法，结合工区已有的沉积动力学模拟数据、地震数据和工区测井砂体隔夹层数据，以及工区动态生产数据，构建动静态相符的储层三维属性模型。其中沉积动力学数据、地震数据、工区测井沉积相数据、砂体隔夹层数据、工区测井孔隙度数据、工区测井渗透率数据、工区测井饱和度数据、工区测井生产数据(包含工程数据)为该发明需要的前期基础数据。 具体实现包括基于人工智能的沉积相建模、基于人工智能的岩性建模和基于人工智能的属性建模三个部分，采用分级控制方法，基于编码器、生成器和判别器网络，生成符合研究区静态数据和动态生产数据的属性模型。
下面对每个步骤进行详细分析。
在步骤101中，获得工区测井数据，工区生产动态数据，沉积相训练模板库和岩性训练模板库，所述工区测井数据包括：工区测井沉积相数据，工区测井砂体隔夹层数据和工区测井属性数据。
在一个实施例中，如图2所示，按如下方式预先建立沉积相训练模板库和岩性训练模板库：
步骤201、获得沉积相信息和砂体隔夹层模式结果数据体，所述沉积相信息包括：沉积动力学数据，已有沉积相图，卫星图像其中之一或任意组合；
步骤202、根据所述沉积相信息，建立沉积相训练模板库；
步骤203、根据所述砂体隔夹层模式结果数据体，建立岩性训练模板库。
具体实施时，以水槽数据为例，可以从沉积动力学数据中抽取第一预设数量的沉积相训练模板数据体；从水槽实验数据体中抽取第二预设数量的沉积相训练模板数据体；将抽取的第一预设数量的沉积相训练模板数据体和第二预设数量的沉积相训练模板数据体合并，建立沉积相训练模板库。
图3为本发明具体实施例中储层属性建模流程图。具体实施时，从沉积动力学数据中的沉积相模拟结果数据三维数据体中抽取p个沉积相训练模板，也即{SDi|i＝1...p}，建立沉积相训练模板库。沉积相训练模板库中，每个模板数据体SDi的长、宽、高和通道数为SL、SW、SH和SC。利用扫描水槽装置完成沉积水槽模拟，并得到水槽实验数据体，其中，扫描水槽装置如图4所示。从水槽实验数据三维数据体中抽取q个沉积相训练模板，{SCi|i＝1...q}，建立沉积相训练模板库。沉积相训练模板库中，每个模板数据体SCi的长、宽、高和通道数为SL、SW、SH和SC。将两部分沉积相训练模板合并为一起，形成{Si|i＝1...n}，建立沉积相训练模板库。沉积相训练模板库中，每个模板数据体Si的长、宽、高和通道数为SL、SW、SH和SC。从沉积动力学数据中的砂体隔夹层模式结果数据三维数据体中，为每个沉积相区域抽取h个岩性训练模板数据体，{Lis|i＝1...h,s＝1...n}，其中s为沉积相编号，建立岩性训练模板库。Lis的可取值范围为{0,1}，其中0为砂岩、1为泥岩。岩性训练模板库中，每个模板数据体Lis的长、宽、高和通道数为LL、LW、LH和LC。其中，水槽装置是一种模拟沉积颗粒物质在液体中 输运和沉积现象的装置。可以利用水槽装置，重现沉积过程，分析沉积体内部的构型结构特征。扫描水槽装置即将大型CT扫描装置和水槽装置合并在一起，利用大型CT扫描装置对水槽沉积结果进行三维扫描，得到水槽结果的三维模型，可以清晰展示不同沉积体内部的构型结构，并且扫描结果以三维数据体形式存储，可以应用人工智能分析。
在步骤102中，根据工区测井沉积相数据，工区测井砂体隔夹层数据，沉积相训练模板库和岩性训练模板库，以及预先建立的第一对抗生成网络，生成储层岩性分布模型，所述第一对抗生成网络是基于多级井震联合的深度卷积对抗生成网络，其中，将从沉积相训练模板库中选取的预设类型沉积相训练模板数据作为网络初始输入数据。
在一个实施例中，所述第一对抗生成网络包括：基于沉积动力学模拟沉积相模式的第一对抗生成子网络，基于测井地震沉积相模式的第二对抗生成子网络，沉积相约束下的基于沉积动力学模拟岩性模式的第三对抗生成子网络，以及沉积相约束下的基于测井地震岩性模式的第四对抗生成子网络；
根据工区测井沉积相数据，工区测井砂体隔夹层数据，沉积相训练模板库和岩性训练模板库，以及预先建立的第一对抗生成网络，生成储层岩性分布模型，包括：
将所述沉积相训练模板库中的多个沉积相训练模板数据体输入第一对抗生成子网络进行训练；
将所述工区测井沉积相数据输入第二对抗生成子网络进行训练，根据训练的结果进行多点地质统计建模，生成储层沉积相模型，所述第二对抗生成子网络根据训练好的第一对抗生成子网络建立；
将所述岩性训练模板库中的多个岩性训练模板数据体输入第三对抗生成子网络进行训练；
将所述工区测井砂体隔夹层数据输入第四对抗生成子网络进行训练，根据训练的结果以及所述储层沉积相模型进行多点地质统计建模，生成储层岩性分布模型，所述第四对抗生成子网络根据训练好的第三对抗生成子网络建立。
在一个实施例中，将所述沉积相训练模板库中的多个沉积相训练模板数据体输入第一对抗生成子网络进行训练，包括：
将从沉积相训练模板库中选取的预设类型沉积相训练模板数据作为网络初始输入数据，输入第一对抗生成子网络进行训练；
对选取的沉积相训练模板数据与第一对抗生成子网络的输出结果进行符合度比对，得到第一误差项；
根据第一误差项和多个沉积相训练模板数据体对第一对抗生成子网络进行修正直至第一误差项达到最小值；
将所述工区测井沉积相数据输入第二对抗生成子网络进行训练，根据训练的结果进行多点地质统计建模，生成储层沉积相模型，包括：根据训练好的第一对抗生成子网络建立第二对抗生成子网络，将工区测井沉积相数据输入第二对抗生成子网络进行训练，输出表征工区测井沉积相数据特征的沉积相模板，根据表征工区测井沉积相数据特征的沉积相模板和实际井数据进行多点地质统计建模，生成储层沉积相模型。
具体实施时，根据每个模板数据体Si的长、宽、高和通道数，建立相应的编码器E1和生成G1。E1的编码机模块EE1输入层的大小为SL×SW×SH×SC，批训练数为SM，E0的解码机模块ED1输出层大小为SL×SW×SH×SC，批训练数为SM，对于沉积相SC＝1。G,1为E1的解码机ED1的拷贝。每次输入SM个沉积相训练模板数据体{Sj|j＝k...k+SM-1}，输入编码器E1的编码机模块EE1，E1的解码机模块ED1输出的结果{SOj|j＝k...k+SM-1},利用输入判别器D1对比{Sj|j＝k...k+SM-1}和{SOj|j＝k...k+SM-1}的符合度，得到E1的误差，也即第一误差项，通过反向修正E1的编码机模块和解码机模块，令{Sj|j＝k...k+SM-1}和{SOj|j＝k...k+SM-1}误差最小。此时E1的编码机EE1的输出为可有效表征沉积相模拟结果数据三维数据体特征的隐变量SZ，其长、宽、高和通道数为SZL、SZW、SZH和SZC，设计大小为SZL×SZW×SZH×SZC。E1的解码机模块ED1可以根据SZ和随机噪声，生成表征沉积相模拟结果数据体特征的沉积相模板ED_S。
然后，将所有K口工区测井的每个点的坐标和工区测井沉积相数据整理为全区井数据WS＝[WSD1,WSS1,WSD2,WSS2,…,WSDK,WSSK]，WSD为坐标数据，WSS为沉积相编号数据，WS为2K×D的图像，其中D为每口工区测井沉积相的数据点数。建立编码器E2，E2编码机EE2输入层的大小为2K×D，批训练数为SW，E2解码机ED1输出层大小为2K×D，批训练数为SW，其中EE2输出的隐含层WZ的长、宽、高和通道数为SZL、SZW、SZH和SZC，大小为SZL×SZW×SZH×SZC，与SZ的大小一致。训练完毕后的E2的隐含层WZ，可以作为表征工区测井沉积相数据特征的隐变量。
进而，建立生成器G2，其输入层的大小为SZL×SZW×SZH×SZC，输出层长、宽、高和通道数为SL、SW、SH和SC。G2初始设置拷贝G1的参数，以使用学习到的沉积模式。将E2输出的隐变量WZ输入到G2，生成新的三维沉积相模板G2S，其长、宽、高和通道数为SL、SW、SH和SC。以G2S为模板，以实际井数据为约束，使用多点地质统计建模，得到储层沉积相模型FS。
需要说明的是，首先根据利用井数据和沉积相训练网络开展建模，这里可以首先人为假设研究区是某种沉积相，但如果人为的认识不对，沉积相训练网络会根据误差反馈自行调整沉积相类型输出，即校正人的认识，达到沉积相模型的最优；然后，针对岩性数据，他也会相应的调整隔夹层模式输出结果，原理同上。
具体的，第一步是让人工智能网络可以有能力生成需要的不同沉积相，但是哪种沉积相实际上人工智能网络是不知道的。因此我们会根据基础研究认识告诉人工智能网络生成沉积相类型A开始建模，并在不断建模中自动校正沉积相类型，可能最终的结果是沉积相B。但我们建模结果通常经不起动态数据验证，最后一步加入动态数据来校正地质模型，就使静态与动态统一起来。由于分级分类进行，因此这个过程就比让人工智能网络自己慢慢摸索要更加智能便捷，计算会更加稳定，效率也高一些。并且也使整建模方法的拓展性，如果再有完全新的沉积相类型，更新开头的模板库即可。举一例，比如人为最开始认为是A，实际上是B，然后用A去训练网络，算出来误差大，模型会自动从模板库里找B数据。并且由于深度学习的特性可能还会生成介于A、B之间的一种沉积类型。核心思想是利用人工智能网络帮助校正沉积认识，增加建模的准确度。
在一个实施例中，储层属性建模方法还包括：
获得工区地震数据体；
对工区地震数据体和储层沉积相模型进行符合度比对，得到第二误差项；
根据第二误差项对第二对抗生成子网络进行修正直至第二误差项达到最小值；
根据表征工区测井沉积相数据特征的沉积相模板和实际井数据进行多点地质统计建模，生成储层沉积相模型，包括：根据第二误差项达到最小值时第二对抗生成子网络输出的表征工区测井沉积相数据特征的沉积相模板和实际井数据进行多点地质统计建模，生成储层沉积相模型。
具体实施时，使用判别器D2，计算FS与地震属性数据体SEGY和工区测井数据之间的综合误差，也即第二误差项。D2计算误差用于修正G2，直到基于G2生成的沉积相模板产生的沉积相模型与工区测井和地震误差最小，输出最优的储层沉积相模型OFS。
在一个实施例中，将所述岩性训练模板库中的多个岩性训练模板数据体输入第三对抗生成子网络进行训练，包括：
将多个岩性训练模板数据体输入第三对抗生成子网络进行训练；
对岩性训练模板数据体与第三对抗生成子网络的输出结果进行符合度比对，得到第三误差项；
根据第三误差项对第三对抗生成子网络进行修正直至第三误差项达到最小值；
将所述工区测井砂体隔夹层数据输入第四对抗生成子网络进行训练，根据训练的结果以及所述储层沉积相模型进行多点地质统计建模，生成储层岩性分布模型，包括：根据训练好的第三对抗生成子网络建立第四对抗生成子网络，将工区测井砂体隔夹层数据输入第四对抗生成子网络进行训练，输出表征工区测井岩性数据特征的岩性分布模板，根据表征工区测井岩性数据特征的岩性分布模板和实际井数据进行多点地质统计建模，生成储层岩性分布模型。
具体实施时，根据每个模板数据体Lis的长、宽、高和通道数，建立相应的编码器E3和生成器G3。E3的编码机模块EE3输入层的大小为LL×LW×LH×LC，批训练数为LM，E3的解码机模块ED3输出层大小为LL×LW×LH×LC，批训练数为LM，对于沉积相LC＝1。G3为E3的解码机模块ED3的拷贝。每次输入LM个岩性训练模板数据体{Lj|j＝k...k+LM-1}，输入编码器E3的编码机模块EE3，E3的解码机模块ED3输出的结果{LOj|j＝k...k+LM-1}，利用符合度模块对比{Lj|j＝k..k+LM-1}和{LOj|j＝k...k+LM-1}，得到E3的误差，通过反向修正E3的编码机模块和解码机模块，令{Lj|j＝k...k+LM-1}和{LOj|j＝k...k+LM-1}误差最小。此时E3的编码机模块EE3的输出为可有效表征砂体隔夹层模式结果数据三维数据体特征的隐变量LZ，其长、宽、高和通道数为LZL、LZW、LZH和LZC，设计大小为LZL×LZW×LZH×LZC。E3的解码机模块ED3可以根据LZ和随机噪声，生成表征砂体隔夹层模式结果数据体特征的岩性分布模板ED_L。
然后，将所有K口工区测井的每个点的坐标和工区测井砂体隔夹层数据整理为全区井数据WL＝[WLD1,WLL1,WLD2,WLL2,…,WLDK,WLLK]，WLD为坐标数据，WLL为砂体隔夹层编号数据，WL为2K×D的图像，其中D为每口工区测井砂体隔夹层的数据点数。建立编码器E4，E4的编码机模块EE4输入层的大小为2K×D，批训练数为LW，E4的解码机模块ED4输出层大小为2K×D，批训练数为LW，其中EE4输出的隐含层LZ的长、宽、高和通道数为LZL、LZW、LZH和LZC，大小为LZL×LZW×LZH×LZC，与LZ的大小一致。训练完毕后的E4的隐含层WZ，可以作为表征全区上工区测井岩心性特征的参量。
进而，建立生成器G4，其输入层的大小为LZL×LZW×LZH×LZC，输出层长、宽、高和通道数为LL、LW、LH和LC。G4初始设置拷贝G3的参数，以使用学习到的 沉积模式。将E4输出的隐变量LZ输入到G4，生成新的表征工区测井岩性数据特征的岩性分布模板G1L，其长、宽、高和通道数为LL、LW、LH和LC。以G1L为模板，以实际井数据为约束，使用多点地质统计建模，得到第s个沉积相区域的储层岩性分布模型FLs。
在一个实施例中，储层属性建模方法还包括：
获得工区地震数据体；
对工区地震数据体和储层岩性分布模型进行符合度比对，得到第四误差项；
根据第四误差项对第四对抗生成子网络进行修正直至第四误差项达到最小值；
根据表征工区测井岩性数据特征的岩性分布模板和实际井数据进行多点地质统计建模，生成储层岩性分布模型，包括：根据第四误差项达到最小值时第四对抗生成子网络输出的表征工区测井岩性数据特征的岩性分布模板和实际井数据进行多点地质统计建模，生成储层岩性分布模型。
具体实施时，使用判别器D4，计算FLs与地震属性数据体SEGY和工区测井数据之间的综合误差，即第四误差项。D4计算误差，用于修正G4，直到基于G4生成的三维砂体隔夹层模板产生的沉积相模型与工区测井和地震误差最小，输出第s个沉积相区最优的储层岩性分布模型OFLs。遍历全部n个沉积区，为每个沉积相区执行上述建模过程，完成全部n个沉积相区的岩性智能建模，得到最优的储层岩性分布模型OFL。
在步骤103中，根据工区测井属性数据，储层岩性分布模型以及预先建立的第二对抗生成网络，生成储层属性模型，所述第二对抗生成网络是基于岩性约束下的地质属性生成网络。
在一个实施例中，所述工区测井属性数据包括：工区测井孔隙度属性数据，工区测井渗透率属性数据，工区测井饱和度属性数据；
根据工区测井属性数据，储层岩性分布模型以及预先建立的第二对抗生成网络，生成储层属性模型，包括：
将所述工区测井孔隙度属性数据和储层岩性分布模型输入第二对抗生成网络，输出对应的储层孔隙度属性模型；
将所述工区测井渗透率属性数据和储层岩性分布模型输入第二对抗生成网络，输出对应的储层渗透率属性模型；
将所述工区测井饱和度属性数据和储层岩性分布模型输入第二对抗生成网络，输出对应的储层饱和度属性模型；
具体实施时，将所有K口工区测井的每个点的坐标和工区测井属性数据整理为全区井数据WPq＝[WPD1q,WPP1q,WPD2q,WPP2q,…,WPDKq,WPPKq]，WPD为坐标数据，WPP为孔隙度数据，WP为2K×D的图像，其中D为每口工区测井属性数据的数据点数，q为要分析的属性编号。孔隙度的属性编号q＝1，渗透率的属性编号q＝2，饱和度的属性编号q＝3。建立编码器E5，E5的编码机模块EE5输入层的大小为2K×D，批训练数为PW，E5的解码机模块ED5输出层大小为2K×D，批训练数为PW，其中EE5输出的隐含层PZ的长、宽、高和通道数为PZL、PZW、PZH和PZC，大小为PZL×PZW×PZH×PZC，与PZ的大小一致。训练完毕后的E5的隐含层PZq，可以作为表征全区上工区测井第q个属性数据特征的参量，也即表征工区测井饱和度属性数据特征的隐变量。建立生成器G5，其输入层1的大小为PZL×PZW×PZH×(2×PZC)，输出层大小为PZL×PZW×PZH。输入层中前PZC个通道，放置PZ数据，后面PZC个通道，以平移遍历方式从OFL中提取PZL×PZW×PZH大小的岩性数据块。G5遍历完毕OFL，如图5所示，生成第p个属性的储层属性模型OFPp。G4遍历完毕OFL生成三维储层属性模型OFPp，相比于一次把OFL全部输入G4,通过移动遍历，可以降低G4输入层的大小，减少对硬件的需求。按照上述流程依次分别生成孔隙度、渗透率和饱和度属性的储层属性模型OPFS＝{OPF1,OFP2,OPF3}，其中OPF1为生成的最优的储层孔隙度属性模型，OPF2为生成的最优的储层渗透率属性模型，OPF3为生成的最优的储层饱和度属性模型。
在步骤104中，根据工区生产动态数据，经过油藏数值模拟代理器的储层属性模型以及预先建立的第三对抗生成网络，对储层属性模型进行修正，所述第三对抗生成网络是基于动态约束下的属性优化网络。
在一个实施例中，根据工区生产动态数据，经过油藏数值模拟代理器的储层属性模型以及预先建立的第三对抗生成网络，对储层属性模型进行修正，包括：
将工区生产动态数据和经过油藏数值模拟代理器的储层属性模型进行符合度比对，得到第五误差项；
根据第五误差项对第三对抗生成网络进行修正直至第五误差项达到最小值；
将第五误差项达到最小值时第三对抗生成网络输出的储层属性模型作为修正后的储层属性模型。
具体实施时，基于OPFS使用油藏数值模拟代理器RS，预测工区生产参数，并利用判别器D5，计算预测数据与实际工区生产参数FO的误差Err1，以及生成的孔隙度、渗 透率、饱和度属性与工区测井属性数据之间的误差Err2，进而得到动态和静态数据的总误差Err＝Err1+Err2。根据这个误差Err，反向修正生成器G5，直到D5得到的误差最小，输出最优的储层属性模型OFP。
下面给出一个具体实施例，说明本发明实施例中储层属性建模的具体应用。在本具体实施例中以南美X区块为应用实例。首先构建了符合X区块沉积环境的沉积动力学模型，得到了2000个三维沉积相模拟结果数据体和4000个三维砂体-隔夹层模式结果数据数据体，作为训练数据体。同时搜集研究区400口井的工区测井沉积相数据和、400口井的砂体-隔夹层数据和400口井的孔、渗、饱数据。整理了研究区400口井的动态生产数据。收集研究区地震数据体，实现了工区全覆盖。基于AMAX深度学习环境，使用8块48GB显存的RTX 8000显卡环境，搭建了编码器E0、生成器G0、生成器G1和编码器E1、生成器G2和编码器E2、生成器G3和编码器E3、生成器G4，以及判别器D1、D2、D3和D4。通过21000次反向训练，生成的三维储层属性模型，与工区测井孔隙度数据符合率为98.2％，与工区测井渗透率符合度为97.3％，与工区测井饱和度符合率为98.5％,与工区测井产液数据符合率为94.1％。采用已有的MPS非人工智能方法，得到的三维储层属性模型，与工区测井孔隙度数据符合率为82.4％，与工区测井渗透率符合度为83.2％，与工区测井饱和度符合率为81.1％,与工区测井产液数据符合率为79.2％。本发明得到的储层属性模型的动静数据符合率高于已有的随机建模方法。
本发明实施例基于人工智能、动静态结合的多点地质统计学三维模板生成方法。有效得解决现有MPS方法模板，没有考虑动态生产数据的问题。并且基于人工智能的多级控制下的储层属性建模。首先基于人工智能方法建立沉积相模型，然后在沉积相模型控制下，建立基于人工智能的岩性模型；最后在岩性控制下，基于人工智能建立属性模型。通过多级控制，不仅降低了训练难度，也提高了模型精度。
基于同一发明构思，本发明实施例还提供了一种储层属性建模装置，如下面的实施例所述。由于这些解决问题的原理与储层属性建模方法相似，因此储层属性建模装置的实施可以参见方法的实施，重复之处不再赘述。
图5为本发明实施例中储层属性建模装置的结构图，如图5所示，该储层属性建模装置包括：
数据获得模块501，用于获得工区测井数据，工区生产动态数据，沉积相训练模板库和岩性训练模板库，所述工区测井数据包括：工区测井沉积相数据，工区测井砂体隔夹层数据和工区测井属性数据；
储层岩性分布模型生成模块502，用于根据工区测井沉积相数据，工区测井砂体隔夹层数据，沉积相训练模板库和岩性训练模板库，以及预先建立的第一对抗生成网络，生成储层岩性分布模型，所述第一对抗生成网络是基于多级井震联合的深度卷积对抗生成网络，其中，将从沉积相训练模板库中选取的预设类型沉积相训练模板数据作为网络初始输入数据；
储层属性模型生成模块503，用于根据工区测井属性数据，储层岩性分布模型以及预先建立的第二对抗生成网络，生成储层属性模型，所述第二对抗生成网络是基于岩性约束下的地质属性生成网络；
储层属性模型优化模块504，用于根据工区生产动态数据，经过油藏数值模拟代理器的储层属性模型以及预先建立的第三对抗生成网络，对储层属性模型进行修正，所述第三对抗生成网络是基于动态约束下的属性优化网络。
基于前述发明构思，如图6所示，本发明实施例还提供一种计算机设备600，包括存储器610、处理器620及存储在存储器610上并可在处理器620上运行的计算机程序630，所述处理器620执行所述计算机程序630时实现上述储层属性建模方法。
基于前述发明构思，本发明实施例还提供一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序被处理器执行时实现上述储层属性建模方法。
本发明实施例还提供一种计算机程序产品，所述计算机程序产品包括计算机程序，所述计算机程序被处理器执行时实现上述储层属性建模方法。
本发明实施例与现有技术中储层属性随机建模的技术方案相比，通过获得工区测井数据，工区生产动态数据，沉积相训练模板库和岩性训练模板库，所述工区测井数据包括：工区测井沉积相数据，工区测井砂体隔夹层数据和工区测井属性数据；根据工区测井沉积相数据，工区测井砂体隔夹层数据，沉积相训练模板库和岩性训练模板库，以及预先建立的第一对抗生成网络，生成储层岩性分布模型，所述第一对抗生成网络是基于多级井震联合的深度卷积对抗生成网络，其中，将从沉积相训练模板库中选取的预设类型沉积相训练模板数据作为网络初始输入数据；根据工区测井属性数据，储层岩性分布模型以及预先建立的第二对抗生成网络，生成储层属性模型，所述第二对抗生成网络是基于岩性约束下的地质属性生成网络；根据工区生产动态数据，经过油藏数值模拟代理器的储层属性模型以及预先建立的第三对抗生成网络，对储层属性模型进行修正，所述第三对抗生成网络是基于动态约束下的属性优化网络。本发明实施例在储层属性建模过 程中引入动态生产内容，结合工区测井数据沉积相训练模板库和岩性训练模板库，依次生成基于人工智能对抗生成网络的储层岩性分布模型和储层属性模型，进而再结合动态生产内容对储层属性模型进行修正，有效构建了动静态相符的储层属性模型，提升了储层属性随机建模动静态数据符合率。
本领域内的技术人员应明白，本发明的实施例可提供为方法、系统、或计算机程序产品。因此，本发明可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本发明可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本发明是参照根据本发明实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
以上所述的具体实施例，对本发明的目的、技术方案和有益效果进行了进一步详细说明，所应理解的是，以上所述仅为本发明的具体实施例而已，并不用于限定本发明的保护范围，凡在本发明的精神和原则之内，所做的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。

Claims (11)

  1. 一种储层属性建模方法，其特征在于，包括：
    获得工区测井数据，工区生产动态数据，沉积相训练模板库和岩性训练模板库，所述工区测井数据包括：工区测井沉积相数据，工区测井砂体隔夹层数据和工区测井属性数据；
    根据工区测井沉积相数据，工区测井砂体隔夹层数据，沉积相训练模板库和岩性训练模板库，以及预先建立的第一对抗生成网络，生成储层岩性分布模型，所述第一对抗生成网络是基于多级井震联合的深度卷积对抗生成网络，其中，将从沉积相训练模板库中选取的预设类型沉积相训练模板数据作为网络初始输入数据；
    根据工区测井属性数据，储层岩性分布模型以及预先建立的第二对抗生成网络，生成储层属性模型，所述第二对抗生成网络是基于岩性约束下的地质属性生成网络；
    根据工区生产动态数据，经过油藏数值模拟代理器的储层属性模型以及预先建立的第三对抗生成网络，对储层属性模型进行修正，所述第三对抗生成网络是基于动态约束下的属性优化网络。
  2. 如权利要求1所述的储层属性建模方法，其特征在于，按如下方式预先建立沉积相训练模板库和岩性训练模板库：
    获得沉积相信息和砂体隔夹层模式结果数据体，所述沉积相信息包括：沉积动力学数据，已有沉积相图，卫星图像其中之一或任意组合；
    根据所述沉积相信息，建立沉积相训练模板库；
    根据所述砂体隔夹层模式结果数据体，建立岩性训练模板库。
  3. 如权利要求1所述的储层属性建模方法，其特征在于，所述第一对抗生成网络包括：基于沉积动力学模拟沉积相模式的第一对抗生成子网络，基于测井地震沉积相模式的第二对抗生成子网络，沉积相约束下的基于沉积动力学模拟岩性模式的第三对抗生成子网络，以及沉积相约束下的基于测井地震岩性模式的第四对抗生成子网络；
    根据工区测井沉积相数据，工区测井砂体隔夹层数据，沉积相训练模板库和岩性训练模板库，以及预先建立的第一对抗生成网络，生成储层岩性分布模型，包括：
    将所述沉积相训练模板库中的多个沉积相训练模板数据体输入第一对抗生成子网络进行训练；
    将所述工区测井沉积相数据输入第二对抗生成子网络进行训练，根据训练的结果进行多点地质统计建模，生成储层沉积相模型，所述第二对抗生成子网络根据训练好的第一对抗生成子网络建立；
    将所述岩性训练模板库中的多个岩性训练模板数据体输入第三对抗生成子网络进行训练；
    将所述工区测井砂体隔夹层数据输入第四对抗生成子网络进行训练，根据训练的结果以及所述储层沉积相模型进行多点地质统计建模，生成储层岩性分布模型，所述第四对抗生成子网络根据训练好的第三对抗生成子网络建立。
  4. 如权利要求3所述的储层属性建模方法，其特征在于，将所述沉积相训练模板库中的多个沉积相训练模板数据体输入第一对抗生成子网络进行训练，包括：
    将从沉积相训练模板库中选取的预设类型沉积相训练模板数据作为网络初始输入数据，输入第一对抗生成子网络进行训练；
    对选取的沉积相训练模板数据与第一对抗生成子网络的输出结果进行符合度比对，得到第一误差项；
    根据第一误差项和多个沉积相训练模板数据体对第一对抗生成子网络进行修正直至第一误差项达到最小值；
    将所述工区测井沉积相数据输入第二对抗生成子网络进行训练，根据训练的结果进行多点地质统计建模，生成储层沉积相模型，包括：根据训练好的第一对抗生成子网络建立第二对抗生成子网络，将工区测井沉积相数据输入第二对抗生成子网络进行训练，输出表征工区测井沉积相数据特征的沉积相模板，根据表征工区测井沉积相数据特征的沉积相模板和实际井数据进行多点地质统计建模，生成储层沉积相模型。
  5. 如权利要求4所述的储层属性建模方法，其特征在于，还包括：
    获得工区地震数据体；
    对工区地震数据体和储层沉积相模型进行符合度比对，得到第二误差项；
    根据第二误差项对第二对抗生成子网络进行修正直至第二误差项达到最小值；
    根据表征工区测井沉积相数据特征的沉积相模板和实际井数据进行多点地质统计建模，生成储层沉积相模型，包括：根据第二误差项达到最小值时第二对抗生成子网络输出的表征工区测井沉积相数据特征的沉积相模板和实际井数据进行多点地质统计建模，生成储层沉积相模型。
  6. 如权利要求3所述的储层属性建模方法，其特征在于，将所述岩性训练模板库中的多个岩性训练模板数据体输入第三对抗生成子网络进行训练，包括：
    将多个岩性训练模板数据体输入第三对抗生成子网络进行训练；
    对岩性训练模板数据体与第三对抗生成子网络的输出结果进行符合度比对，得到第三误差项；
    根据第三误差项对第三对抗生成子网络进行修正直至第三误差项达到最小值；
    将所述工区测井砂体隔夹层数据输入第四对抗生成子网络进行训练，根据训练的结果以及所述储层沉积相模型进行多点地质统计建模，生成储层岩性分布模型，包括：根据训练好的第三对抗生成子网络建立第四对抗生成子网络，将工区测井砂体隔夹层数据输入第四对抗生成子网络进行训练，输出表征工区测井岩性数据特征的岩性分布模板，根据表征工区测井岩性数据特征的岩性分布模板和实际井数据进行多点地质统计建模，生成储层岩性分布模型。
  7. 如权利要求6所述的储层属性建模方法，其特征在于，还包括：
    获得工区地震数据体；
    对工区地震数据体和储层岩性分布模型进行符合度比对，得到第四误差项；
    根据第四误差项对第四对抗生成子网络进行修正直至第四误差项达到最小值；
    根据表征工区测井岩性数据特征的岩性分布模板和实际井数据进行多点地质统计建模，生成储层岩性分布模型，包括：根据第四误差项达到最小值时第四对抗生成子网络输出的表征工区测井岩性数据特征的岩性分布模板和实际井数据进行多点地质统计建模，生成储层岩性分布模型。
  8. 如权利要求1所述的储层属性建模方法，其特征在于，所述工区测井属性数据包括：工区测井孔隙度属性数据，工区测井渗透率属性数据，工区测井饱和度属性数据；
    根据工区测井属性数据，储层岩性分布模型以及预先建立的第二对抗生成网络，生成储层属性模型，包括：
    将所述工区测井孔隙度属性数据和储层岩性分布模型输入第二对抗生成网络，输出对应的储层孔隙度属性模型；
    将所述工区测井渗透率属性数据和储层岩性分布模型输入第二对抗生成网络，输出对应的储层渗透率属性模型；
    将所述工区测井饱和度属性数据和储层岩性分布模型输入第二对抗生成网络，输出对应的储层饱和度属性模型。
  9. 如权利要求8所述的储层属性建模方法，其特征在于，根据工区生产动态数据，经过油藏数值模拟代理器的储层属性模型以及预先建立的第三对抗生成网络，对储层属性模型进行修正，包括：
    将工区生产动态数据和经过油藏数值模拟代理器的储层属性模型进行符合度比对，得到第五误差项；
    根据第五误差项对第三对抗生成网络进行修正直至第五误差项达到最小值；
    将第五误差项达到最小值时第三对抗生成网络输出的储层属性模型作为修正后的储层属性模型。
  10. 一种储层属性建模装置，其特征在于，包括：
    数据获得模块，用于获得工区测井数据，工区生产动态数据，沉积相训练模板库和岩性训练模板库，所述工区测井数据包括：工区测井沉积相数据，工区测井砂体隔夹层数据和工区测井属性数据；
    储层岩性分布模型生成模块，用于根据工区测井沉积相数据，工区测井砂体隔夹层数据，沉积相训练模板库和岩性训练模板库，以及预先建立的第一对抗生成网络，生成储层岩性分布模型，所述第一对抗生成网络是基于多级井震联合的深度卷积对抗生成网络，其中，将从沉积相训练模板库中选取的预设类型沉积相训练模板数据作为网络初始输入数据；
    储层属性模型生成模块，用于根据工区测井属性数据，储层岩性分布模型以及预先建立的第二对抗生成网络，生成储层属性模型，所述第二对抗生成网络是基于岩性约束下的地质属性生成网络；
    储层属性模型优化模块，用于根据工区生产动态数据，经过油藏数值模拟代理器的储层属性模型以及预先建立的第三对抗生成网络，对储层属性模型进行修正，所述第三对抗生成网络是基于动态约束下的属性优化网络。
  11. 一种计算机程序产品，其特征在于，所述计算机程序产品包括计算机程序，所述计算机程序被处理器执行时实现权利要求1所述方法。
PCT/CN2023/111727 2022-08-08 2023-08-08 储层属性建模方法及装置 WO2024032604A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210943817.7A CN117592350A (zh) 2022-08-08 2022-08-08 储层属性建模方法及装置
CN202210943817.7 2022-08-08

Publications (1)

Publication Number Publication Date
WO2024032604A1 true WO2024032604A1 (zh) 2024-02-15

Family

ID=89850935

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/111727 WO2024032604A1 (zh) 2022-08-08 2023-08-08 储层属性建模方法及装置

Country Status (2)

Country Link
CN (1) CN117592350A (zh)
WO (1) WO2024032604A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100161300A1 (en) * 2006-09-01 2010-06-24 Chevron U.S.A. Inc. System and method for forecasting production from a hydrocarbon reservoir
CN112419493A (zh) * 2019-08-21 2021-02-26 中国石油天然气股份有限公司 页岩储层三维属性模型建立方法及装置
CN112507618A (zh) * 2020-12-03 2021-03-16 中国石油大学(华东) 一种基于生成对抗网络的油藏自动历史拟合方法
US20210224682A1 (en) * 2020-01-16 2021-07-22 Saudi Arabian Oil Company Training of machine learning algorithms for generating a reservoir digital twin

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100161300A1 (en) * 2006-09-01 2010-06-24 Chevron U.S.A. Inc. System and method for forecasting production from a hydrocarbon reservoir
CN112419493A (zh) * 2019-08-21 2021-02-26 中国石油天然气股份有限公司 页岩储层三维属性模型建立方法及装置
US20210224682A1 (en) * 2020-01-16 2021-07-22 Saudi Arabian Oil Company Training of machine learning algorithms for generating a reservoir digital twin
CN112507618A (zh) * 2020-12-03 2021-03-16 中国石油大学(华东) 一种基于生成对抗网络的油藏自动历史拟合方法

Also Published As

Publication number Publication date
CN117592350A (zh) 2024-02-23

Similar Documents

Publication Publication Date Title
EP3293552B1 (en) System and method for editing geological models by switching between volume-based models and surface-based structural models augmented with stratigraphic fiber bundles
US10671775B2 (en) Sediment transport simulation with parameterized templates for depth profiling
NO309548B1 (no) Fremgangsmåte for generering av et tredimensjonalt gittermönster som samsvarer med et legemes geometri, for å oppnå en modell som representerer legemet
CN112394404B (zh) 一种渐进式储层精细表征方法
CN106803124A (zh) 基于流形正则和范数正则的领域迁移极限学习机方法
CN104504754A (zh) 一种多点统计建模的方法及装置
CN110765665B (zh) 一种地学动态建模方法及系统
Lee et al. Regeneration of channelized reservoirs using history-matched facies-probability map without inverse scheme
CN105738952A (zh) 一种水平井区储层岩石相建模方法
CN111766635A (zh) 砂体连通程度分析方法和系统
Razak et al. Physics-guided deep learning for improved production forecasting in unconventional reservoirs
WO2024032604A1 (zh) 储层属性建模方法及装置
CN103543478A (zh) 地质形态插值的km方法
Hauge et al. Object-based modeling with dense well data
CN115421181A (zh) 一种基于深度学习的三维地质模型相控属性建模方法
Pershin et al. Approximation Models for the Hydrolitospheric Processes
Milliff A modified capacitance matrix method to implement coastal boundaries in the Harvard Open Ocean Model
CN115469361B (zh) 一种碎屑岩地层三维地质建模方法
Jo et al. Conditioning stratigraphic, rule-Based models with generative adversarial networks: A deepwater lobe, deep learning example
CN113219534B (zh) 一种叠前深度偏移速度质控方法、装置、介质及电子设备
CN113985493B (zh) 地下多信息约束的等时地层格架智能建模方法
Cai et al. 3D conditional simulation of porosity for a heterogeneous core
Dzidolikas et al. Neural solution of elliptic partial differential equation problem for single phase flow in porous media
Chen et al. Optimal model-based reservoir management with model parameter uncertainty updates
Curtis et al. A New Workflow to Upscale and Propagate Saturation-dependent Petrophysical Properties from Wireline Logs to 3D Geocellular Models

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23851825

Country of ref document: EP

Kind code of ref document: A1