A method, device and equipment for intelligent identification of compact gas

Abstract

The application discloses a kind of compact gas intelligent identification method, device and equipment, make full use of professional knowledge, determine data and chart knowledge to build well logging knowledge graph, and according to the atlas of fluid property identification obtained from the graph, the well logging knowledge graph is combined with fluid property identification atlas, and the fluid property intelligent identification model of chart constraint is jointly constructed, and the intelligent identification of compact gas is carried out using it.By the above-mentioned mode, a kind of compact gas intelligent identification method with high calculation precision, good accuracy, strong generalization ability and strong interpretability is successfully constructed, and the timeliness of compact gas identification work is greatly improved.Furthermore, the atlas embedding representation algorithm based on TransH is used in the present application, which maps the relationship between wells and reservoirs in high-dimensional space to a low-dimensional relationship space, so that the representation vector of wells and reservoirs can cover the global geological information of the current block, further ensuring that the compact gas identification effect of the method can reach the level of expert manual analysis.

Description

Technical Field

[0001] This invention relates to the field of geophysical exploration technology, and in particular to a method, apparatus and equipment for intelligent identification of tight gas. Background Technology

[0002] With the continuous expansion and deepening of oil and gas exploration and development, unconventional oil and gas has become an extremely important oil and gas resource in China. Tight gas is one of the key areas of unconventional natural gas exploration and development, and it is also the unconventional natural gas with the largest reserves and production scale now and for many years to come. China has abundant tight gas resources and great potential for increasing reserves and production. Accelerating the development and utilization of tight gas plays an important role in improving China's natural gas self-sufficiency and reducing carbon emissions.

[0003] Tight gas identification technology has always been a key research topic both domestically and internationally. Well logging can reveal a wide range of physical properties of underground strata, including acoustic, electrical, nuclear, pressure, and nuclear magnetic resonance characteristics. Furthermore, its data is characterized by continuous depth, high vertical resolution, and high accuracy, making it a commonly used and dominant technology for tight gas identification. However, due to the tightness of tight gas reservoirs, their pronounced ultra-low porosity and permeability, complex pore structure, large differences in gas saturation, and complex gas-water distribution relationships, well logging calculations of reservoir parameters (mainly porosity, permeability, and saturation) often result in low accuracy and misidentification of gas and water layers, hindering the overall evaluation and efficient development of gas fields.

[0004] In conventional well logging interpretation, porosity is calculated using porosity logging (density, neutron, and acoustic logging), but the accuracy of calculations for reservoirs with complex lithology is difficult to guarantee. Permeability calculations are mainly achieved through empirical formula conversion between porosity and permeability based on core analysis data, but the complex pore structure of tight gas makes this method inaccurate. Gas saturation is mainly calculated using the Archie equation or the Indonesian formula; the key to controlling its accuracy is the correct selection of rock electrical parameters closely related to pore structure. This work is crucial, but the limited availability of core samples leads to poor representativeness of rock electrical parameters, thus affecting the accuracy of their values ​​and reducing the accuracy of saturation calculations. Based on the "four-property relationship" of well logging lithology, electrical properties, physical properties, and gas content, and combined with well logging and oil testing data, a reservoir fluid identification template is established to identify gas and water layers. A key element is accurately selecting well logging parameters sensitive to fluid identification, but the characteristics of tight gas reservoirs make this very difficult.

[0005] In view of this, there is an urgent need to develop a tight gas intelligent identification method with high computational precision, good accuracy, strong generalization ability, and strong interpretability to solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to address the bottleneck problem of tight gas reservoir characteristics and well logging identification by fully utilizing professional knowledge, judgment data, and map knowledge to construct a well logging knowledge graph for intelligent identification of tight gas. It also aims to research targeted machine learning models to establish reservoir parameter prediction models and fluid property intelligent identification models based on the well logging knowledge graph. This solves the problems of poor generalization ability and lack of interpretability of results in existing machine learning models, making tight gas identification technology more capable of learning, thereby improving calculation accuracy and identification accuracy, and significantly improving work efficiency.

[0007] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0008] A method for intelligent identification of tight gas, the method comprising:

[0009] Constructing a well logging knowledge graph;

[0010] Based on the well logging knowledge graph, obtain a fluid property identification chart;

[0011] Based on the well logging knowledge graph and the fluid property recognition chart, a chart-constrained intelligent fluid property recognition model is constructed.

[0012] Intelligent identification of tight gas is performed using the fluid property intelligent identification model constrained by the aforementioned diagram.

[0013] As a further improvement of the present invention, the well logging knowledge graph includes:

[0014] Well logging characteristics, well logging features, the relationship between the four properties of well logging, the distribution pattern of core data, block geology, and judgment data.

[0015] As a further improvement of the present invention, the construction of the well logging knowledge graph includes the following steps:

[0016] Constructing a knowledge ontology for tight gas identification;

[0017] Based on the knowledge ontology of tight gas identification, entities and relationships are extracted from the tight gas identification-related data, which includes: well base information, logging data, well logging data, lithological property analysis data, stratification data, logging interpretation results, oil testing and production data, and research reports.

[0018] The well logging knowledge graph is obtained by constructing triples based on the entities and relationships.

[0019] As a further improvement of the present invention, the step of constructing a fluid property intelligent identification model constrained by the logging knowledge graph and the fluid property identification chart includes the following steps:

[0020] Based on the relationship between wells and reservoirs in the well logging knowledge graph, determine the representation vectors of wells and reservoirs;

[0021] Based on the logging curve data in the logging knowledge graph, determine the logging response characteristics of the reservoir;

[0022] The representation vectors of the well and reservoir are integrated with the logging response characteristics of the reservoir;

[0023] The fusion result is input into the hybrid neural network, and the fluid property recognition map is used as an error constraint for training the hybrid neural network to perform hybrid learning, thereby constructing a map-constrained intelligent fluid property recognition model.

[0024] As a further improvement of the present invention, determining the representation vectors of the well and the reservoir based on the relationship between the well and the reservoir includes:

[0025] The TransH-based graph embedding representation algorithm is used to map the relationship between the well and the reservoir in the high-dimensional space to the low-dimensional relationship space, thereby obtaining the representation vectors of the well and the reservoir.

[0026] As a further improvement of the present invention, determining the well logging response characteristics of the reservoir based on the well logging curve data includes:

[0027] By employing deep convolutional neural networks, calculations of different series and scales are performed on the well logging curve data to obtain the well logging response characteristics of the reservoir.

[0028] As a further improvement of the present invention, the method further includes:

[0029] Using the well logging knowledge graph and rock physics model, a machine learning-based reservoir parameter prediction model is established.

[0030] The reservoir parameters are predicted using the aforementioned machine learning-based reservoir parameter prediction model;

[0031] Using the well logging knowledge graph and reservoir parameter prediction results, a machine learning reservoir identification model is established;

[0032] The reservoir identification model is used to identify reservoirs.

[0033] As a further improvement of the present invention, the method further includes:

[0034] A comprehensive analysis was conducted on the identified gas-bearing reservoirs to screen out those with exploitation potential.

[0035] The present invention also provides a tight gas intelligent identification device, the device comprising:

[0036] The first building unit is used to construct the well logging knowledge graph;

[0037] The chart acquisition unit is used to acquire fluid property identification charts based on the well logging knowledge graph;

[0038] The second construction unit is used to construct a fluid property intelligent recognition model constrained by the logging knowledge graph and the fluid property recognition chart.

[0039] The identification unit is used to intelligently identify tight gas using the fluid property intelligent identification model constrained by the drawing.

[0040] As a further improvement of the present invention, the first construction unit includes a first construction module, an extraction module, and a graph module, wherein...

[0041] The first construction module is used to build the knowledge ontology for tight gas identification;

[0042] The extraction module is connected to the first construction module and is used to extract entities and relationships from the tight gas identification related data based on the knowledge ontology of tight gas identification. The tight gas identification related data includes: well base information, logging data, well logging data, lithological property analysis data, stratification data, logging interpretation results, oil testing and production data, and research reports.

[0043] The graph module is connected to the extraction module and is used to construct triples based on the entities and relationships to obtain the well logging knowledge graph. The well logging knowledge graph includes: well logging features, well logging characteristics, the four properties of well logging, the distribution pattern of core data, block geology, and judgment data.

[0044] As a further improvement of the present invention, the second construction unit includes a first determining module, a second determining module, a fusion module, and a second construction module, wherein...

[0045] The first determining module is used to determine the representation vectors of wells and reservoirs based on the relationship between wells and reservoirs in the well logging knowledge graph.

[0046] The second determining module is used to determine the logging response characteristics of the reservoir based on the logging curve data in the logging knowledge graph.

[0047] The fusion module is connected to the first determination module and the second determination module respectively, and is used to fuse the representation vectors of the well and the reservoir and the logging response characteristics of the reservoir;

[0048] The second construction module is connected to the fusion module and is used to input the fusion result into the hybrid neural network. The fluid property recognition map is used as an error constraint for training the hybrid neural network to perform hybrid learning, thereby constructing a map-constrained intelligent fluid property recognition model.

[0049] As a further improvement of the present invention, the first determining module includes a mapping submodule.

[0050] The mapping submodule uses a TransH-based graph embedding representation algorithm to map the relationship between the well and the reservoir in the high-dimensional space to a low-dimensional relationship space, thereby obtaining the representation vectors of the well and the reservoir.

[0051] As a further improvement of the present invention, the second determining module includes a calculation submodule.

[0052] The calculation submodule uses a deep convolutional neural network to perform calculations on the logging curve data at different series and scales to obtain the logging response characteristics of the reservoir.

[0053] As a further improvement of the present invention, the device further includes:

[0054] The reservoir parameter prediction unit is used to establish a machine learning-based reservoir parameter prediction model using the well logging knowledge graph and rock physics model, and to predict reservoir parameters using the machine learning-based reservoir parameter prediction model.

[0055] The reservoir identification unit is used to establish a machine learning-based reservoir identification model using the well logging knowledge graph and reservoir parameter prediction results, and to identify reservoirs using the machine learning-based reservoir identification model.

[0056] As a further improvement of the present invention, the device further includes:

[0057] The potential reservoir identification unit is used to conduct a comprehensive analysis of the identified gas-bearing reservoirs and screen out reservoirs with exploitation potential.

[0058] The present invention also provides a tight gas intelligent identification device, the device comprising a processor and a memory; wherein,

[0059] The memory is used to store machine-executable instructions;

[0060] The processor is used to read and execute machine-executable instructions stored in the memory to implement the aforementioned tight gas intelligent identification method.

[0061] The beneficial effects of this invention are:

[0062] The present invention provides a method, apparatus, and device for intelligent identification of tight gas. This method constructs a well logging knowledge graph, then obtains a fluid property identification chart based on this graph, and finally constructs a chart-constrained intelligent fluid property identification model using the well logging knowledge graph and the fluid property identification chart. This model is then used for intelligent identification of tight gas. Through this approach, a method for intelligent identification of tight gas with high computational accuracy, good accuracy, strong generalization ability, and strong interpretability is successfully constructed, thereby significantly improving the timeliness of tight gas identification work.

[0063] The intelligent tight gas identification method, apparatus, and equipment provided by this invention employ a TransH-based map embedding representation algorithm to map the relationship between wells and reservoirs in a high-dimensional space to a low-dimensional relationship space, obtaining representation vectors for wells and reservoirs. This allows the representation vectors to encompass the global geological information of the current block, and further enables multi-dimensional representation of the geological information of the study block through modeling. This results in a tight gas well logging identification performance approaching expert interpretation, with a model prediction accuracy exceeding 86%. Test results on actual datasets further demonstrate that the method's identification accuracy has reached the level of expert analysis.

[0064] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0065] Figure 1 This is a flowchart of the tight gas intelligent identification method of the present invention;

[0066] Figure 2 This is a diagram showing the logging interpretation results of well Jinqian 815 in an embodiment of the present invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] This invention provides an intelligent identification method for tight gas that utilizes artificial intelligence technology and cognitive learning methods, integrating well logging knowledge graphs and chart constraints, to construct a method with high computational precision, good accuracy, strong generalization ability, and strong interpretability. Figure 1As shown, by constructing a well logging knowledge graph, obtaining fluid property identification charts from the well logging knowledge graph, and finally using the well logging knowledge graph combined with the fluid property identification charts to construct a chart-constrained intelligent fluid property identification model, a method is achieved that, based on the well logging knowledge graph, well logging response features of tight gas layers are extracted, and a chart-constrained intelligent fluid property identification model is established through a neural network combined with the knowledge-constrained charts (i.e., fluid property identification charts). This model is then used for intelligent identification of tight gas, which significantly improves the timeliness of tight gas identification work, achieves accurate identification of tight gas layers, and makes the intelligent identification of tight gas layers more interpretable.

[0069] The specific implementation steps are described below:

[0070] (a) Constructing a well logging knowledge graph;

[0071] Guided by an ontology construction methodology, this invention constructs a knowledge ontology for tight gas identification. Under the constraints of this ontology, based on tight gas identification-related data such as well basic information, logging data, well logging data, lithological property analysis data, stratification data, well logging interpretation results, oil testing and production data, and research reports, entities and relationships are extracted using entity recognition and relational attribute extraction models. Through knowledge fusion and manual verification, high-quality triplets are obtained. For example, for a triplet with the structure <logging curve, numerical index, interpretation> (where the logging curve and interpretation are entities, and the numerical index is a relation), if expressed as <sonic transit time, [280, 320], gas layer>, it means that when the response characteristics of the sonic logging curve value are between 280 and 320, the well logging interpretation represents a gas layer. This establishes a tight gas logging knowledge graph, which includes information on logging characteristics, well logging "four-property relationships," core data distribution patterns, block geology, and judgment data.

[0072] (ii) Based on the well logging knowledge graph, establish a machine learning reservoir parameter prediction model and a reservoir identification model;

[0073] Effective reservoir identification can be achieved by establishing a machine learning-based intelligent prediction model for reservoir parameters based on rock physics models and well logging knowledge graphs. Then, based on the predicted reservoir parameters and the well logging knowledge graph, a machine learning-based reservoir identification model can be established. This model, combined with the judgment criteria, can then be used to identify reservoirs.

[0074] (iii) Obtain fluid property identification charts based on well logging knowledge graphs;

[0075] Fluid property identification charts refer to charts established using data such as tight gas reservoir logging response characteristics, oil testing, and porosity saturation, combined with chart methods and judgment data, to create different logging responses and interpretation conclusions. Examples include density porosity charts, neutron porosity charts, and resistivity porosity charts.

[0076] (iv) Based on the well logging knowledge graph and combined with the fluid property identification chart, a chart-constrained intelligent fluid property identification model is jointly constructed;

[0077] This invention employs a TransH-based graph embedding representation algorithm to map the relationships between wells and reservoirs in high-dimensional space (e.g., adjacent well relationships, formation structure, well-reservoir inclusion relationships, etc., all derived from well logging knowledge graphs) to a low-dimensional relationship space, obtaining representation vectors for wells and reservoirs. This allows the representation vectors to encompass the global geological information of the current block. Furthermore, this invention uses deep convolutional neural networks to perform calculations on well logging curve data at different series and scales to obtain the well logging response characteristics of tight gas. Specifically, this invention analyzes the well logging response characteristics of tight gas and selects high-quality parameters. Specifically, based on the principles of highlighting gas content and logging contrast while minimizing parameter-introduced errors, this approach combines data from well testing, logging, and core analysis with features such as sedimentary, structural, reservoir, gas reservoir, and production dynamics characteristics to clarify the relationship between logging response characteristics and the four properties (lithology, physical properties, and gas content). It identifies different interference terms and influencing factors, establishes an intelligent preprocessing method, and extracts logging-sensitive parameters, such as lithology index, physical property index, cycle jump characteristics, gas content index, mining effect characteristics, resistivity amplitude difference, and gas logging characteristics. These features are then optimized using correlation analysis and principal component analysis to construct standardized logging characterization parameters, thereby obtaining the logging response characteristics of reservoir tight gas. Subsequently, the logging response characteristics of reservoir tight gas and the representation vectors of the well and reservoir are fused using an attention mechanism. The fused result is then input into a hybrid neural network, using a fluid property recognition chart as the training error constraint for hybrid learning. Finally, a softmax layer is used to obtain the target reservoir identification result, thus constructing a chart-constrained intelligent fluid property recognition model. This model employs a joint learning method to simultaneously train a neural network with fluid property recognition maps as constraints. This approach obtains the embedded representation vectors (representation vectors for wells and reservoirs) from the domain knowledge graph. This significantly enhances the model's ability to understand and utilize domain knowledge, while also providing interpretability to the model's predictions. Therefore, this invention constructs a data- and knowledge-driven intelligent recognition model based on knowledge graph constraints.

[0078] Finally, the intelligent identification model of fluid properties constrained by the chart was used to achieve high-accuracy tight gas identification. Based on the judgment index system, the identified reservoirs containing tight gas were comprehensively analyzed, and the optimal potential gas layer was recommended. This greatly supports the increase of tight gas reserves and production, and is of great significance for exploration and development.

[0079] This invention also provides a tight gas intelligent identification device, which includes a first construction unit for constructing a well logging knowledge graph; a map acquisition unit for acquiring fluid property identification maps based on the well logging knowledge graph; a second construction unit for constructing a map-constrained tight gas intelligent identification model based on the well logging knowledge graph and the fluid property identification maps; and an identification unit for intelligently identifying tight gas using the map-constrained tight gas intelligent identification model. Sometimes, the device may also include a reservoir parameter prediction unit for establishing a machine learning-based reservoir parameter prediction model using the well logging knowledge graph and a rock physics model, and using the machine learning-based reservoir parameter prediction model to predict reservoir parameters; a reservoir identification unit for establishing a machine learning-based reservoir identification model using the well logging knowledge graph and reservoir parameter prediction results, and using the machine learning-based reservoir identification model to identify reservoirs; and a potential layer identification unit for comprehensively analyzing the identified gas-bearing reservoirs and screening out reservoirs with exploitable potential.

[0080] Regarding the aforementioned intelligent identification device for tight gas, the specific methods by which each unit performs its operations have been described in detail in the relevant embodiments of the method, and will not be elaborated upon here.

[0081] The present invention also provides a tight gas intelligent identification device, which includes a processor and a memory; wherein the memory is used to store machine-executable instructions; and the processor is used to read and execute the machine-executable instructions stored in the memory to implement the aforementioned tight gas intelligent identification method.

[0082] The following examples illustrate the effectiveness of the intelligent tight gas identification method provided by the present invention:

[0083] This embodiment provides a method for identifying tight gas in the Wenxingchang-Laoguanmiao, Jinqiu, and Jinqiu-Tianfu research blocks using the intelligent tight gas identification method of this invention. The fluid properties of the Shaximiao Formation in 54 wells were identified, and the overall intelligent identification results showed a consistency rate of 86.4% with the interpretations of well logging experts. Figure 2 The image shown is a logging interpretation result diagram of well Jinqian 815, which meets the needs of operational applications, and the timeliness of intelligent identification is significantly improved. Production efficiency has increased by more than 100 times without reducing, and even improving, the interpretation quality.

[0084] In summary, the tight gas intelligent identification method, apparatus, and equipment provided by this invention construct a well logging knowledge graph, then obtain a fluid property identification chart based on this knowledge graph, and finally construct a chart-constrained fluid property intelligent identification model based on the well logging knowledge graph and the fluid property identification chart. This model is then used for intelligent identification of tight gas. Through this method, a tight gas intelligent identification method with high computational accuracy, good accuracy, strong generalization ability, and strong interpretability is successfully constructed, thereby significantly improving the timeliness of tight gas identification work. Meanwhile, the intelligent tight gas identification method, device, and equipment provided by this invention, further utilize a TransH-based graph embedding representation algorithm to map the relationship between wells and reservoirs in a high-dimensional space to a low-dimensional relationship space, obtaining representation vectors for wells and reservoirs. This allows the representation vectors to encompass the global geological information of the current block, thereby achieving multi-dimensional representation of the geological information of the study block through modeling. This makes the well logging identification effect of this method for tight gas approach that of expert interpretation, with a model prediction accuracy exceeding 86%. Test results on actual datasets further demonstrate that the method's identification accuracy has reached the level of expert analysis, significantly improving the development efficiency and economic benefits of tight gas identification.

[0085] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent identification of tight gas, the method comprising: Constructing a well logging knowledge graph; The well logging knowledge graph includes: well logging characteristics, well logging features, the relationship between the four properties of well logging, the distribution pattern of core data, block geology, and judgment data; Based on the well logging knowledge graph, obtain a fluid property identification chart; Based on the well logging knowledge graph and the fluid property identification chart, a chart-constrained intelligent fluid property identification model is constructed, which includes the following steps: Based on the relationship between wells and reservoirs in the well logging knowledge graph, the representation vectors of wells and reservoirs are determined, which includes: using a graph embedding representation algorithm based on TransH to map the relationship between wells and reservoirs in the high-dimensional space to a low-dimensional relationship space to obtain the representation vectors of wells and reservoirs that cover the global geological information of the current block; Based on the logging curve data in the logging knowledge graph, the logging response characteristics of the reservoir are determined, including: using a deep convolutional neural network to perform calculations on the logging curve data at different series and scales to obtain the logging response characteristics of the reservoir. The representation vectors of the well and reservoir are integrated with the logging response characteristics of the reservoir; The fusion result is input into the hybrid neural network, and the fluid property recognition map is used as the error constraint for training the hybrid neural network to perform hybrid learning, thereby constructing a map-constrained intelligent fluid property recognition model. Intelligent identification of tight gas is performed using the fluid property intelligent identification model constrained by the aforementioned diagram.

2. The tight gas intelligent identification method according to claim 1, wherein, The construction of the well logging knowledge graph includes the following steps: Constructing a knowledge ontology for tight gas identification; Based on the knowledge ontology of tight gas identification, entities and relationships are extracted from the tight gas identification-related data, which includes: well base information, logging data, well logging data, lithological property analysis data, stratification data, logging interpretation results, oil testing and production data, and research reports. The well logging knowledge graph is obtained by constructing triples based on the entities and relationships.

3. The tight gas intelligent identification method according to claim 1, wherein, The method further includes: Using the well logging knowledge graph and rock physics model, a machine learning-based reservoir parameter prediction model is established. The reservoir parameters are predicted using the aforementioned machine learning reservoir parameter prediction model; Using the well logging knowledge graph and reservoir parameter prediction results, a machine learning reservoir identification model is established; The reservoir is identified using the reservoir identification model.

4. The tight gas intelligent identification method according to claim 1, wherein, The method further includes: A comprehensive analysis was conducted on the identified gas-bearing reservoirs to screen out those with exploitation potential.

5. A tight gas intelligent identification device, the device comprising: The first construction unit is used to construct a well logging knowledge graph, which includes: well logging characteristics, well logging features, the relationship between the four properties of well logging, the distribution pattern of core data, block geology, and judgment data. The chart acquisition unit is used to acquire fluid property identification charts based on the well logging knowledge graph; The second construction unit is used to construct a fluid property intelligent recognition model constrained by the logging knowledge graph and the fluid property recognition chart. The identification unit is used to intelligently identify tight gas using the fluid property intelligent identification model constrained by the drawing; The second construction unit includes a first determining module, a second determining module, a fusion module, and a second construction module. The first determining module is used to determine the representation vectors of wells and reservoirs based on the relationship between wells and reservoirs in the well logging knowledge graph. The second determining module is used to determine the logging response characteristics of the reservoir based on the logging curve data in the logging knowledge graph. The fusion module is connected to the first determination module and the second determination module respectively, and is used to fuse the representation vectors of the well and the reservoir and the logging response characteristics of the reservoir; The second construction module is connected to the fusion module and is used to input the fusion result into the hybrid neural network. The fluid property recognition map is used as an error constraint for training the hybrid neural network to perform hybrid learning, thereby constructing a map-constrained intelligent fluid property recognition model. The first determining module includes a mapping submodule, which uses a TransH-based graph embedding representation algorithm to map the relationship between the well and the reservoir in the high-dimensional space to a low-dimensional relationship space, thereby obtaining a representation vector of the well and the reservoir that covers the global geological information of the current block. The second determining module includes a calculation submodule, which uses a deep convolutional neural network to perform calculations on the logging curve data at different series and scales to obtain the logging response characteristics of the reservoir.

6. The tight gas intelligent identification device according to claim 5, wherein, The first construction unit includes a first construction module, an extraction module, and a graph module, wherein, The first construction module is used to build the knowledge ontology for tight gas identification; The extraction module is connected to the first construction module and is used to extract entities and relationships from the tight gas identification related data based on the knowledge ontology of tight gas identification. The tight gas identification related data includes: well base information, logging data, well logging data, lithological property analysis data, stratification data, logging interpretation results, oil testing and production data, and research reports. The graph module is connected to the extraction module and is used to construct triples based on the entities and relationships to obtain the well logging knowledge graph.

7. The tight gas intelligent identification device according to claim 5, wherein, The device further includes: The reservoir parameter prediction unit is used to establish a machine learning-based reservoir parameter prediction model using the well logging knowledge graph and rock physics model, and to predict reservoir parameters using the machine learning-based reservoir parameter prediction model. The reservoir identification unit is used to establish a machine learning-based reservoir identification model using the well logging knowledge graph and reservoir parameter prediction results, and to identify reservoirs using the machine learning-based reservoir identification model.

8. The intelligent identification device for tight gas according to claim 5, wherein, The device further includes: The potential reservoir identification unit is used to conduct a comprehensive analysis of the identified gas-bearing reservoirs and screen out reservoirs with exploitation potential.

9. A tight gas intelligent identification device, the device comprising a processor and a memory; wherein, The memory is used to store machine-executable instructions; The processor is configured to read and execute machine-executable instructions stored in the memory to implement the method as described in any one of claims 1 to 4.

Images