Artificial intelligence three-dimensional attribute modeling method

Abstract

The invention provides an artificial intelligence three-dimensional attribute modeling method. The artificial intelligence three-dimensional attribute modeling method comprises the following steps: step 1, constructing a target sedimentary facies mode geological knowledge base; step 2, constructing a three-dimensional model sample set conforming to a sedimentary mode; 3, establishing a training well point condition data set; 4, establishing a training probability volume data set; 5, geologic model training is carried out on the training probability volume data set based on the generative adversarial network, if the training model evaluation is unqualified, model training continues to be executed, and if the training model evaluation is qualified, the step 6 is executed; and step 6, importing training geologic model data qualified in model evaluation into a geologic model simulator, and realizing display and updating of a three-dimensional attribute model. The method is completely based on data driving, an artificial intelligence modeling technology method is adopted, multi-dimensional information data intelligent fusion is achieved, data nonlinearity is achieved, modeling quality is completely controlled by data, modeling efficiency is high, and real-time model updating is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas field development, and in particular to an artificial intelligence three-dimensional attribute modeling method. Background Art

[0002] Traditional geological modeling technology is to comprehensively utilize geological, seismic, logging and production dynamic data, combine the results of geological experience analysis, use geostatistical methods to predict inter-well reservoir parameters, and form a three-dimensional spatial quantitative representation of the geological model.

[0003] Traditional geological modeling methods and processes are experience-driven, and the modeling steps are relatively complex. They require stratification, zoning, and phase-based multivariate collaborative constraints to fit the variogram and distribution function, and require manual parameter fitting adjustment and characterization. This takes a long time, is inefficient, and the modeling quality varies from person to person. Summary of the invention

[0004] In view of this, the present invention aims to propose an artificial intelligence three-dimensional attribute modeling method, which can establish a set of data-driven artificial intelligence modeling technical method processes, realize nonlinear data representation through intelligent fusion of multi-dimensional information, and has high modeling efficiency. The modeling quality is only controlled by data.

[0005] To achieve the above object, the technical solution of the present invention is implemented as follows: an artificial intelligence three-dimensional attribute modeling method, comprising the following steps:

[0006] Step 1, collecting and arranging relevant sedimentary patterns of target sedimentary facies, and constructing a geological knowledge base of target sedimentary facies patterns;

[0007] Step 2, using a sedimentary process geological sample generation algorithm, extracting morphological features of reservoir characteristic parameters corresponding to each model in the target sedimentary facies model geological knowledge base, and constructing a three-dimensional model sample set that conforms to the sedimentary model;

[0008] Step 3, based on the geological sample scale and generative algorithm design in the three-dimensional model sample set, using the unconditionally constrained generative adversarial network geological modeling algorithm, establish a training well point condition data set;

[0009] Step 4, based on multiple conditional constraints, a generative adversarial network geological modeling algorithm is used to establish a training probability volume data set; the conditional constraints include core data constraints, well logging data constraints, and seismic data constraints;

[0010] Step 5: Perform geological model training on the training probability volume data set based on the deep learning model of the generative adversarial network. If the training model evaluation fails, continue to perform model training. If the training model evaluation passes, proceed to step 6.

[0011] Step 6: Import the trained geological model data that has passed the model evaluation into the geological model simulator to display and update the three-dimensional attribute model.

[0012] Furthermore, in step 1, it is first necessary to clarify the type of target sedimentary phase, then collect relevant pattern data of this type of sedimentary phase, and organize and summarize the patterns to form a pattern map knowledge base of the target sedimentary phase.

[0013] Furthermore, in step 2, reservoir characteristic parameters are extracted from the target sedimentary phase model geological knowledge base, and a scale- and pattern-driven morphological feature extraction method is adopted for the sedimentary morphology of the phase type. By building a network model, a three-dimensional model sample set that conforms to the sedimentary model is constructed.

[0014] Furthermore, in step 3, based on the geological sample scale in the three-dimensional model sample set, the loss function is calculated according to the conditional data to achieve conditional constraints and error reduction, and the training well point conditional data set is established using the unconditionally constrained generative adversarial network geological modeling algorithm.

[0015] Furthermore, in step 5, model training and evaluation are performed on the training probability volume data set based on the generative adversarial network model, and the training results are evaluated according to the experience of geological experts. If the training model evaluation fails, the model training is continued; if the training model evaluation passes, step 6 is executed.

[0016] Furthermore, in step 6, based on the data conversion program, the trained geological model that has passed the model evaluation is seamlessly connected to the geological model simulator. When the well and seismic data change, the geological model can be updated in real time, and the three-dimensional attribute model results can be presented in real time.

[0017] Compared with the prior art, the artificial intelligence three-dimensional attribute modeling method described in the present invention has the following advantages: the present invention is completely data-driven and adopts artificial intelligence modeling technology to realize intelligent geological modeling with sedimentary phase pattern control, conditional data as hard data, and deep learning algorithm as the main factor, eliminating the cumbersome variogram fitting steps in conventional modeling. The whole process from sedimentary phase pattern characteristic parameter extraction to geological, logging, seismic and other conditional data application adopts relevant deep learning algorithms to realize the intelligent fusion of multi-dimensional information data. Since the generative adversarial network structure has nonlinear activation functions and multi-layer network structures, data nonlinearity is realized, which can more accurately describe the three-dimensional spatial model morphology, improve the flexibility and adaptability of the model, improve the accuracy of prediction and classification, reveal hidden patterns and relationships, and deal with complexity and uncertainty. The modeling quality is completely controlled by data, the modeling efficiency is high, and the geological model can be updated in real time, which has great positive significance for promoting the intelligent development of geological modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0019] Figure 1 A flowchart of the steps of an artificial intelligence three-dimensional attribute modeling method;

[0020] Figure 2 Schematic diagram of morphological feature extraction driven by scale and mode;

[0021] Figure 3 It is a display diagram of the three-dimensional attribute model (Figure a is the sandstone probability body model; Figure b is the porosity model). DETAILED DESCRIPTION

[0022] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0023] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0024] like Figure 1-Figure 3 As shown, the present invention is an artificial intelligence three-dimensional attribute modeling method, comprising the following steps:

[0025] Step 1: Collect and organize the relevant sedimentary patterns of the target sedimentary facies and construct a geological knowledge base of the target sedimentary facies pattern. Specifically, first, the type of the target sedimentary facies needs to be clarified, and then the relevant pattern data of this type of sedimentary facies is collected, and the pattern is organized and summarized, such as: the plane and cross-sectional morphology, plane combination relationship, longitudinal superposition relationship, lateral splicing relationship, sand body scale and width-to-thickness ratio of each microfacies of the delta front facies (distributary channel, estuary bar, sheet sand, distant sand bar, inter-channel, etc.), etc. Based on the pattern characteristics, a pattern map knowledge base of the plane and cross-section of the target sedimentary facies is formed, which provides important geological prior knowledge and basic support for the subsequent extraction of sedimentary pattern morphology and the construction of geological models.

[0026] Step 2: Using the sedimentary process geological sample generation algorithm, the reservoir characteristic parameters corresponding to each model in the target sedimentary phase model geological knowledge base (such as the combination relationship of each microphase, distribution morphology, scale, width-to-thickness ratio, etc.) are extracted respectively, and then a three-dimensional model sample set that conforms to the sedimentary model is constructed; wherein, the reservoir characteristic parameters are extracted from the target sedimentary phase model geological knowledge base, and a morphological feature extraction method based on scale and pattern driving is adopted for the sedimentary morphology of the phase type. By building a reasonable generative adversarial network model and fully considering the complexity and diversity of geological data, appropriate network structure and parameters are selected to ensure that the generated model samples are representative and generalizable, and a three-dimensional model sample set that conforms to the sedimentary model is constructed to provide sample data for generative adversarial network geological modeling.

[0027] Step 3: Based on the scale of geological samples in the three-dimensional model sample set and the generative algorithm design, a training well point conditional data set is established using the unconditionally constrained generative adversarial network geological modeling algorithm to provide adjustment and optimization space for the conditional constraints of the training well points. Specifically, based on the scale of geological samples in the three-dimensional model sample set, the loss function is calculated according to the conditional data, and the model parameters are optimized using cross-validation, regularization and other methods to achieve conditional constraints and error reduction, and a training well point conditional data set is established using the unconditionally constrained generative adversarial network geological modeling algorithm.

[0028] Step 4, based on multiple conditional constraints, using the generative adversarial network geological modeling algorithm, establish a training probability volume data set, which is a key link in geological modeling. These conditional constraints not only ensure the accuracy and reliability of the model, but also enable the model to better reflect the actual geological conditions; the conditional constraints include core data constraints, well logging data constraints and seismic data constraints; that is, multiple conditional data such as core, well logging, and seismic data are required for constraints, and the generative adversarial network geological modeling algorithm is used to establish a training probability volume data set; in the generative adversarial network geological modeling, core data is used to train the network to identify and generate similar geological features, and the core data constraints improve the accuracy of the model by ensuring that the geological features output by the model are consistent with the core observation results. The well logging data is used as a longitudinal (vertical) constraint to help the model build an accurate stratigraphic structure in the vertical direction, and the seismic data is used as a lateral (horizontal) constraint to help the model build a geological structure that conforms to the seismic interpretation results on the plane. In the description of the present invention, unless otherwise specified, the meaning of "multiple" is two or more.

[0029] Step 5, geological model training is performed on the training probability volume data set based on the generative adversarial network. The core, well logging and seismic data are used as inputs, and adversarial training is performed through the generator and discriminator of the generative adversarial network. The generator is responsible for generating three-dimensional geological model samples that conform to geological laws, while the discriminator is responsible for evaluating the similarity between these samples and real geological data. If the training model evaluation fails, the model training continues. If the training model evaluation passes, step 6 is performed. Specifically, during the training process, the constraints of the core, well logging and seismic data are incorporated by designing a specific loss function. These loss functions may include losses based on physical properties (such as porosity, permeability, etc.), losses based on structural similarity (such as sand body thickness, morphology, etc.) and losses based on seismic attributes (such as amplitude, frequency, etc.). The training results are evaluated according to the experience of geological experts. Quantitative evaluation indicators, such as accuracy, recall rate, F1 score, etc., may also be used to objectively evaluate the training results. If the training model evaluation fails, the model training continues. If the training model evaluation passes, step 6 is performed.

[0030] Step 6: Import the training geological model data (such as spatial distribution information of geological structure and probability distribution information of geological attributes) that have passed the model evaluation into a mature geological model simulator to realize the display and update of the three-dimensional attribute model. Specifically, based on the data conversion and docking program, the training geological model that has passed the model evaluation is seamlessly connected to the existing and mature geological model simulator. When the well and seismic data change, the model can be updated in real time, and the three-dimensional attribute model results can be presented in real time, which improves the application value of the geological model and provides strong support for production decisions.

[0031] Taking the offshore C oil field as an example, the specific implementation manner of the present invention is further described in conjunction with the accompanying drawings.

[0032] Firstly, according to the geological knowledge base of fluvial facies sedimentary patterns and the sedimentary process geological sample generation algorithm, the morphological characteristics of the sedimentary reservoir characteristic parameters are extracted to construct a three-dimensional model sample set that conforms to the sedimentary pattern, such as Figure 2 shown.

[0033] Secondly, according to the geological sample scale and generative algorithm design in the three-dimensional model sample set, the unconditional generative adversarial network modeling algorithm is used to establish a training well point conditional data set, and based on multiple conditional constraints such as well logging, core, and seismic data, a training probability volume data set is established. Model training is performed on the probability volume data set based on the generative adversarial network model. If the model evaluation fails, training needs to continue. If it passes, the next step is executed.

[0034] Finally, the qualified training geological model data is imported into the mature geological model simulator to update and display the 3D attribute model. Figure 3 shown.

[0035] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An artificial intelligence three-dimensional attribute modeling method, characterized in that: The following steps are involved: Step 1, collecting and arranging relevant sedimentary patterns of target sedimentary facies, and constructing a geological knowledge base of target sedimentary facies patterns; Step 2, using a sedimentary process geological sample generation algorithm, extracting morphological features of reservoir characteristic parameters corresponding to each model in the target sedimentary facies model geological knowledge base, and constructing a three-dimensional model sample set that conforms to the sedimentary model; Step 3, based on the geological sample scale and generative algorithm design in the three-dimensional model sample set, using the unconditionally constrained generative adversarial network geological modeling algorithm, establish a training well point condition data set; Step 4, based on multiple conditional constraints, a generative adversarial network geological modeling algorithm is used to establish a training probability volume data set; the conditional constraints include core data constraints, well logging data constraints, and seismic data constraints; Step 5: Perform geological model training on the training probability volume data set based on the deep learning model of the generative adversarial network. If the training model evaluation fails, continue to perform model training. If the training model evaluation passes, proceed to step 6. Step 6: Import the trained geological model data that has passed the model evaluation into the geological model simulator to display and update the three-dimensional attribute model.

2. The artificial intelligence three-dimensional attribute modeling method according to claim 1, characterized in that: In step 1, it is necessary to first clarify the type of target sedimentary phase, then collect the relevant pattern data of this type of sedimentary phase, and organize and summarize the patterns to form a pattern map knowledge base of the target sedimentary phase.

3. The artificial intelligence three-dimensional attribute modeling method according to claim 1, characterized in that: In step 2, reservoir characteristic parameters are extracted from the target sedimentary phase model geological knowledge base, and a scale- and pattern-driven morphological feature extraction method is adopted for the sedimentary morphology of the phase type. By building a network model, a three-dimensional model sample set that conforms to the sedimentary model is constructed.

4. The artificial intelligence three-dimensional attribute modeling method according to claim 3, characterized in that: In step 3, based on the geological sample scale in the 3D model sample set, the loss function is calculated according to the conditional data to achieve conditional constraints and error reduction, and the training well point conditional data set is established using the unconditionally constrained generative adversarial network geological modeling algorithm.

5. The artificial intelligence three-dimensional attribute modeling method according to claim 4, characterized in that: In step 5, model training and evaluation are performed on the training probability volume data set based on the generative adversarial network model, and the training results are evaluated according to the experience of geological experts. If the training model evaluation fails, the model training continues. If the training model evaluation passes, step 6 is executed.

6. The artificial intelligence three-dimensional attribute modeling method according to claim 4, characterized in that: In step 6, based on the data conversion program, the trained geological model that has passed the model evaluation is seamlessly connected to the geological model simulator. When the well and seismic data change, the geological model can be updated in real time, and the three-dimensional attribute model results can be presented in real time.

Images