Stratigraphic flow unit fine division method, computer device and readable storage medium

Abstract

The application provides a formation flow unit fine division method, a computer device and a readable storage medium. The fine division method comprises the following steps: calculating FZI according to core physical property experiment data; obtaining pore throat radius parameter R35 according to mercury injection data; dividing flow unit types by FZI and R35 intersection maps, and marking flow unit classification labels on core data; returning the core; constructing high-dimensional characteristic parameters and reducing dimensions, and extracting characteristic parameters representing flow unit types; constructing machine learning sample data sets; adopting a random gradient boosting decision tree algorithm to establish a flow unit fine division model; and testing the fine division model, and promoting application after meeting the accuracy. The computer device comprises a processor and a memory with program instructions, and the program instructions comprise instructions for executing the above method. The computer readable storage medium comprises computer program instructions, which are executed by the processor to realize the above method. The division method of the application has small errors.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and oil and gas reservoir exploration and development technology, specifically to a method for fine division of stratigraphic flow units, a computer device, and a readable storage medium. Background Technology

[0002] A flow unit refers to a reservoir body with similar lithological characteristics that are continuous both laterally and vertically, influencing fluid flow. Flow unit classification is an effective method for deeply understanding and describing the heterogeneity of oil reservoirs, and is of great significance for finely dividing reservoirs, improving the accuracy of permeability calculations, and tapping remaining oil potential. Since the concept of flow units was proposed in 1984, geologists have proposed many methods for classifying flow units based on basic data and research practice, which can be roughly summarized into the following four categories:

[0003] (1) Flow unit division method based on geological theory

[0004] In the early stages of research, based on geological theories such as sedimentation and diagenesis, geological bodies were roughly divided into several genetic units vertically according to sedimentary interfaces and diagenetic barriers. Then, flow unit types were further subdivided based on the physical properties of each unit. The patent "Method for Dividing Reservoir Flow Units" (application number: CN201610684447.4) discloses a flow unit division method based on high-resolution sequence stratigraphy. This method, after completing sedimentary microfacies analysis within an isochronous stratigraphic framework, establishes a sand body structure model and divides flow units by analyzing their connectivity and seepage characteristics. Both of these methods are primarily qualitative and fail to achieve quantitative characterization of flow units. Subsequently, some geologists, based on core descriptions and lithological property variations, performed stratification and further subdivided the rock into several secondary flow units by calculating parameters such as the rock's conductivity and storage coefficient. The advantage of this method is that it achieves quantification, but the disadvantage is that it can only roughly divide the flow units on a macroscopic level. The division accuracy of flow units in strongly heterogeneous formations is not high, and it is difficult to meet the needs of fine reservoir description. The above method is different from the division method of this application.

[0005] (2) Method for dividing flow units based on core analysis data

[0006] The flow zone index (FZI) method is currently the most popular method for dividing flow units. This method establishes the relationship between FZI and reservoir quality index (RQI) and porosity index (Φ) based on the Kozeny-Carman equation. z The relationship between RQI and Φ is calculated first using the porosity (Φ) and permeability (K) data from core physical property experiments. z Then through RQI and Φ zLog-log plots are used to delineate flow units. Some scholars, based on mercury intrusion porosimetry (MIP) experiments, delineate flow units by analyzing the pore geometry reflected in the core capillary pressure curves, known as the R35 method. This method assumes that pore throat size reflects the fluid flow capacity and development status in the rock, with larger pore throats playing a major controlling role in fluid flow conditions. Typically, the pore throat radius at which mercury intrusion reaches 35% on the MIP curve is extracted as the index parameter (R35), and the reservoir is divided into several flow units based on the R35 value. In the absence of core MIP data, the empirical formula (logR35 = 0.732 + 0.588logk - 0.864log(Φ)) is used. It should be noted that this method works well for porous reservoirs, but its flow unit identification effect is poor when secondary porosity such as dissolution cavities and fractures are developed. Patent application number "CN202011538416.0" discloses a method for delineating reservoir flow units based on nuclear magnetic resonance logging data. This method first converts the transverse relaxation time T2 spectrum into a pseudo-capillary pressure curve, then extracts relevant mercury injection parameters, and establishes deviation coefficients (N index and M index) to quantitatively divide the flow units. Patent application number "CN201910154632.6" discloses a method for dividing dynamic flow units in water-injected reservoirs during the ultra-high water-cut period. This method is based on core relative permeability curves, calculates the water phase permeability coefficient of each grid, and then divides the flow units. All of the above methods are based on core analysis and testing data. Their advantage is the reliability of the division results; their disadvantage is that they rely entirely on core samples, and in practice, they are constrained by drilling engineering technology, coring conditions, and high economic costs, making it inconvenient to widely core at the site, thus limiting their application scope. Such methods also differ from the division method in this application.

[0007] (3) Applying production dynamic data to conduct flow unit research

[0008] Canas et al. divided flow units based on interwell fluid flow velocity and flow capacity data during the production process of the La Cira oilfield. This method first calculates the actual flow rate ratio between two wells within a production unit to obtain the Interwell Flow Capacity Index (IFCI), a parameter characterizing the flow unit, and then divides the flow units. For example, Chinese patent application number "CN201610890913.4" discloses a method and apparatus for dividing vertical flow units in a reservoir, as well as an interwell comparison method and apparatus. This method first subdivides the reservoir into layers, then statistically analyzes and plots histograms of the cumulative production capacity percentage and cumulative storage percentage of each layer, and combines this with the flowability index to complete the fine division of flow units. However, this method differs from the division method in this application. The advantage of this type of method is that the divided flow units are combined with the actual production of the oilfield, which can effectively guide oilfield development and tapping remaining oil potential. The disadvantage is that this type of method requires abundant dynamic data as support. For newly explored areas with scarce dynamic data, if data or empirical parameters from neighboring areas are used for calculation, the calculation results often do not conform to reality.

[0009] (4) Multi-parameter flow unit partitioning method based on simple mathematical models

[0010] This method is currently widely used in major oilfields. The specific implementation involves first intensive sampling from single wells, performing various analyses and tests, and selecting macroscopic and microscopic characteristic parameters reflecting rock structure, physical properties, and fluid properties, such as sedimentary microfacies, clay content, median grain size, porosity, and permeability. Then, combined with well logging curves, mathematical methods such as cluster analysis, discriminant analysis, or BP neural networks are used to classify fluid flow unit types. For example, patent application number "CN201811580931.8" discloses a method for calculating the permeability of sandstone-type uranium deposit sand bodies based on flow unit classification. This method first calculates the flow unit index of core sample points, then applies a probabilistic graphical method to classify core flow units, and finally establishes a quantitative relationship between flow units and parameters such as natural gamma, density, apparent resistivity, and sonic transit time to complete the application of flow unit classification. However, this flow unit classification method differs from the classification method in this application. In recent years, the advantages of machine learning in solving complex classification problems have been increasingly recognized. For example, Chinese patent application number "CN201910889252.7" entitled "A Method for Classifying and Identifying Flow Unit Information Based on Support Vector Machine Algorithm" discloses a method for classifying and identifying flow unit information. This method includes the following steps: determining the sample set and preprocessing the data, and using the mapminmax function in MATLAB to normalize the sample data; using C-SVM classification technology to build the model and optimize the parameters; and using test samples to predict and verify the established prediction model. This method employs a flow unit information classification method based on the support vector machine algorithm, which has strong capabilities in handling nonlinear problems. It can, to some extent, solve the nonlinear problems between flow unit types and multiple factors, and is also a type of machine learning method. However, this method differs from the classification method in this application.

[0011] In summary, this type of method is an improvement upon the second type of method mentioned above. Its advantages include using more feature parameters and, to some extent, addressing the nonlinearity issues encountered by the second type of method when the thresholds between classification samples are relatively clear, thus improving the accuracy of flow unit segmentation. The disadvantage is that existing classification methods are all single-model or weak classifiers. When the thresholds for classification samples are unclear and nonlinearity between categories is prominent, these single-classification models cannot achieve satisfactory segmentation results, exhibiting poor model generalization ability. Currently, fluid flow unit segmentation methods are developing towards a shift from qualitative to quantitative, from linear to nonlinear, and from low-dimensional to high-dimensional spaces. Therefore, providing a fine-grained segmentation method for flow units in strongly heterogeneous clastic rock formations based on a machine learning ensemble model, along with computer equipment and a readable storage medium, is of great significance for improving the segmentation accuracy of flow units in strongly heterogeneous clastic rock formations. Summary of the Invention

[0012] In view of the shortcomings of the prior art, the purpose of this invention is to solve one or more problems existing in the prior art. For example, one objective of this invention is to provide a method for finely dividing the flow units of formations with strong heterogeneity and many flow unit types based on a machine learning ensemble model, as well as a computer device and a readable storage medium.

[0013] To achieve the above objectives, the present invention provides a method for finely delineating flow units in highly heterogeneous clastic rock formations, the method comprising the following steps:

[0014] Based on the experimental data of the core physical properties of the target area, the flow zone index of the core sample was calculated;

[0015] Based on the mercury intrusion porosimetry data of the core samples in the target area, the pore throat radius parameter corresponding to 35% saturation of the non-wetting phase on the mercury intrusion porosimetry curve of the core samples was obtained.

[0016] Based on the cross-plot analysis results of the flow zone index and pore throat radius parameters, the flow unit types are classified, and the core data samples are labeled with flow unit classification tags as output data for the fine division model of flow units.

[0017] Core repositioning refers to repositioning the depth of core sample data to the logging depth.

[0018] High-dimensional characteristic parameters are constructed based on conventional well logging curves;

[0019] Principal component analysis was used to reduce the dimensionality of high-dimensional feature parameters and extract the feature parameters that characterize the flow unit type as input data for the flow unit fine partitioning model.

[0020] The feature parameters characterizing the flow unit type are combined with the flow unit type label data to form a machine learning sample dataset, which includes a training dataset and a validation dataset.

[0021] For the training dataset, a stochastic gradient boosting decision tree algorithm is used to establish a fine-grained partitioning model for flow units;

[0022] The established flow unit fine division model was tested using a validation dataset. After the calculated accuracy and recall met the accuracy requirements, the model was extended to all single wells in the target area to obtain the fine discrimination results of flow units for each well.

[0023] According to an exemplary embodiment of one aspect of the present invention, the formula for calculating the core flow zone index is as follows:

[0024]

[0025]

[0026]

[0027] Where, Φ z Φ is the ratio of pore volume to particle volume, dimensionless; e 1 represents core porosity, a decimal; K represents core permeability, in mD; RQI represents reservoir quality index, dimensionless; FZI represents core flow zone index, dimensionless.

[0028] According to an exemplary embodiment of one aspect of the present invention, the calculation formula for dimensionality reduction of high-dimensional feature parameters by the principal component analysis method can be as follows:

[0029]

[0030] Where var(X) is the covariance matrix, dimensionless; X is the eigenvalue, dimensionless; m is the number of samples; a i μ is the i-th sample value when X is a feature, and it is dimensionless. x This is the average value, dimensionless.

[0031] According to an exemplary embodiment of one aspect of the present invention, the calculation formula of the stochastic gradient boosting decision tree algorithm can be as follows:

[0032] Y m+1 (X)=fY m (X)+ρ m h(x), 1≤m≤M (5)

[0033] Among them, Y m+1 (X) represents the new learning model; f is a random factor, dimensionless; Y m (X) represents the current learning model; ρ m is the learning rate, which is dimensionless; h(x) is the base learning model fitted along the negative gradient direction of the current loss function; M is the maximum number of iterations; m is the number of samples.

[0034] According to an exemplary embodiment of one aspect of the present invention, 70-80% of the data in the machine learning sample dataset can be used as a training dataset, and the remaining data can be used as a validation dataset.

[0035] According to an exemplary embodiment of one aspect of the present invention, the formulas for calculating the precision and recall are as follows:

[0036]

[0037]

[0038] Where Accuracy is the model's prediction of a positive sample; Recall is the model's prediction of a positive sample; FP is the model's prediction of a positive sample; FN is the model's prediction of a negative positive sample; and TN is the model's prediction of a negative negative sample.

[0039] According to an exemplary embodiment of one aspect of the present invention, the high-dimensional characteristic parameters may include natural gamma, natural potential, compensated neutron, compensated acoustic wave, density curve, deep lateral resistivity, shallow lateral resistivity, mass photoelectric absorption cross section index, deep-shallow resistivity ratio, and volume photoelectric absorption cross section index; the characteristic parameters characterizing the flow unit type may include the first principal component to the i-th principal component.

[0040] According to an exemplary embodiment of one aspect of the present invention, the identification accuracy of the fine division method of the flow unit can be improved by more than 10% compared with the support vector machine classification model.

[0041] Another aspect of the present invention provides a computer device, the computer device comprising:

[0042] At least one processor;

[0043] A memory storing program instructions, wherein the program instructions are configured to be executed by the at least one processor, and the program instructions may include instructions for performing any of the methods described above.

[0044] Another aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions which, when executed by a processor, can implement the method described in any of the above-described embodiments.

[0045] Compared with the prior art, the beneficial effects of the present invention include at least one of the following:

[0046] (1) The method for fine division of flow units in strongly heterogeneous clastic rock formations proposed in this invention can combine conventional logging methods with artificial intelligence algorithms, establish a more efficient integrated classification machine learning model for dividing flow units, and improve the accuracy of flow unit division in strongly heterogeneous clastic rock formations.

[0047] (2) The method for fine division of flow units in strongly heterogeneous clastic rock formations proposed in this invention overcomes the limitation of conventional methods that require complete reliance on core data compared to traditional flow unit division methods.

[0048] (3) The method for fine division of flow units in strongly heterogeneous clastic rock formations proposed in this invention can accurately divide the multiple flow unit types in strongly heterogeneous clastic rock formations. The division results are consistent with regional geological patterns and production realities, and have broad application and promotion prospects. Attached Figure Description

[0049] The above and other objects and features of the present invention will become clearer from the following description taken in conjunction with the accompanying drawings, in which:

[0050] Figure 1 A flowchart of the method for finely dividing flow units in strongly heterogeneous clastic rock formations according to the present invention is shown;

[0051] Figure 2 A flow cell type distribution diagram of an exemplary embodiment of the present invention is shown;

[0052] Figure 3 A schematic diagram of an exemplary embodiment of the present invention, illustrating a flow unit evaluation method based on PCA-SGBDT artificial intelligence, is shown.

[0053] Figure 4 A comprehensive diagram showing the flow unit division of the PRHC-22 well, an exemplary embodiment of the present invention, is illustrated.

[0054] Figure 5 A comparative diagram showing the effect of flow unit division in the PRHC-22 well, an exemplary embodiment of the present invention, is presented. Detailed Implementation

[0055] In the following, a method for finely dividing formation flow units, a computer device, and a readable storage medium according to the present invention will be described in detail with reference to the accompanying drawings and exemplary embodiments.

[0056] It should be noted that "first," "second," "third," etc., are merely for the convenience of description and distinction, and should not be interpreted as indicating or implying relative importance.

[0057] Figure 1 A flowchart of the method for finely dividing flow units in strongly heterogeneous clastic rock formations according to the present invention is shown; Figure 2 A flow cell type distribution diagram of an exemplary embodiment of the present invention is shown;

[0058] Figure 3 A schematic diagram of an exemplary embodiment of the present invention, illustrating a flow unit evaluation method based on PCA-SGBDT artificial intelligence, is shown. Figure 4 A comprehensive diagram showing the flow unit division of the PRHC-22 well, an exemplary embodiment of the present invention, is illustrated. Figure 5 A comparative diagram showing the effect of flow unit division in the PRHC-22 well, an exemplary embodiment of the present invention, is presented.

[0059] In a first exemplary embodiment of the present invention, the method for finely classifying flow units in strongly heterogeneous clastic rock formations first requires combining core physical property experimental data and mercury intrusion porosimetry data to classify the flow unit types in the study area, using these as label data. This label data serves as the output data of the fine-grained flow unit classification model (machine learning model). Based on core relocation and considering the main controlling factors of the reservoir, principal component analysis (PCA) is used to extract characteristic parameters representing the flow units, which are then used as input data for the fine-grained flow unit classification model. These characteristic parameters, along with the flow unit type label data, constitute a machine learning sample dataset. 80% of the samples are used as the training dataset for model learning, and the remaining 20% ​​are used as the validation dataset for model testing. Finally, a stochastic gradient boosting decision tree (SGBDT) algorithm is used to establish a flow unit classification model for the training dataset, and the model is then tested using the validation dataset. Once the accuracy requirements are met, the model is then widely applied. This method can accurately classify multiple flow unit types in strongly heterogeneous clastic rock formations, and the classification results conform to regional geological patterns and actual production conditions, showing broad application and promotion prospects. Figure 1 As shown, the method for finely dividing flow units in strongly heterogeneous clastic rock formations mainly includes the following steps:

[0060] Based on the core physical property experimental data of the target area, the porosity (Φ) eThe flow zone index (FZI) was calculated from the core permeability (K). Mercury intrusion porosimetry (MIP) data from the target area was processed, and the pore throat radius (R35) corresponding to 35% saturation of the unwetting phase was obtained using MIP curves. The core data used in the above steps were repositioned to the logging depth to reduce the depth error between the core analysis data and the logging curves. Cross-plot analysis was performed using the obtained flow zone index and pore throat radius data to classify flow unit types. The identified flow unit types were then used to label the core data of the target area. Based on the logging response of the reservoir's main controlling factors, high-dimensional feature parameters were extracted and constructed from conventional logging curves. Principal component analysis (PCA) was used to reduce the dimensionality of the high-dimensional feature parameters, obtaining feature parameters for flow unit modeling (i.e., feature parameters characterizing flow units). The labeled core data and the feature parameters used for flow unit modeling were combined to form a machine learning sample dataset, which includes a training dataset and a validation dataset. For the training dataset, the Stochastic Gradient Boosting Decision Tree (SGBDT) algorithm, which has excellent performance in classifying imbalanced data, was selected. This method was used to train the model on the training dataset to establish a fine-grained flow unit partitioning model. SGBDT is an artificial intelligence algorithm with high prediction accuracy, capable of handling various data types, including continuous and discrete values, and able to more accurately partition flow unit types. The fine-grained flow unit partitioning model was tested using a validation dataset to verify its suitability. If successful, the model can be applied to all single wells in the target area to identify flow units in uncorked wells, obtaining fine-grained flow unit identification results for each well. This provides technical support for subsequent precise well placement, oil well productivity evaluation, prediction of remaining oil favorable zones, and evaluation of water injection development potential. If unsuccessful, the model parameters were fine-tuned until successful.

[0061] In this exemplary embodiment, based on the experimental data of the core physical properties of the target area, the formula for calculating the core flow zone index is as follows:

[0062]

[0063]

[0064]

[0065] Where, Φ z Φ is the ratio of pore volume to particle volume, dimensionless; e 1 represents core porosity, a decimal; K represents core permeability, in mD; RQI represents reservoir quality index, dimensionless; FZI represents core flow zone index, dimensionless.

[0066] In this exemplary embodiment, the constructed high-dimensional feature parameters are processed using principal component analysis (PCA), and the calculation formula is as follows:

[0067]

[0068] Where var(X) is the covariance matrix, dimensionless; X is the eigenvalue, dimensionless; m is the number of samples; a i μ is the i-th sample value when X is a feature, and it is dimensionless. x This is the average value, dimensionless.

[0069] In this exemplary embodiment, the parameters from the training dataset are input into the SGBDT decision tree ensemble, and the calculation formula for the stochastic gradient boosting decision tree algorithm can be as follows:

[0070] Y m+1 (X)=fY m (X)+ρ m h(x), 1≤m≤M (5)

[0071] Among them, Y m+1 (X) represents the new learning model; f is a random factor, dimensionless; Y m (X) represents the current learning model; ρ m is the learning rate, which is dimensionless; h(x) is the base learning model fitted along the negative gradient direction of the current loss function; M is the maximum number of iterations; m is the number of samples.

[0072] In this exemplary embodiment, the training dataset may include 70-80% of the data from the machine learning sample dataset; for example, 70%, 75%, or 80% of the data may be selected as the training dataset. The validation dataset may include 20-30% of the data from the machine learning sample dataset; for example, 20%, 25%, or 30% of the data may be selected as the validation dataset. Both the training and validation datasets may include labeled core data and well logging characteristic parameters used for flow element modeling.

[0073] In this exemplary embodiment, the fine-grained partitioning method may further include testing the flow unit fine-grained partitioning model using a validation dataset, and determining whether it meets the accuracy requirements by calculating accuracy and recall. The formulas for calculating accuracy and recall are as follows:

[0074]

[0075]

[0076] Where Accuracy is the model's prediction of a positive sample; Recall is the model's prediction of a positive sample; FP is the model's prediction of a positive sample; FN is the model's prediction of a negative positive sample; and TN is the model's prediction of a negative negative sample.

[0077] In this exemplary embodiment, the high-dimensional characteristic parameters characterizing the flow unit may include natural gamma (GR), natural potential (SP), compensated neutron (CNL), compensated acoustic wave (AC), density curve (DEN), deep lateral resistivity (Rt), shallow lateral resistivity (Rxo), mass photoelectric absorption cross-section index (Pe), deep / shallow resistivity ratio (Rt / Rxo), and volume photoelectric absorption cross-section index (U), etc. After dimensionality reduction processing of the high-dimensional characteristic parameters by principal component analysis, the characteristic parameter (Xi) characterizing the flow unit type may include the first principal component (X1), the second principal component (X2), the third principal component (X3), the fourth principal component (X4), ... the i-th principal component (Xi).

[0078] In this exemplary embodiment, the identification accuracy of the fine division method of flow units in strongly heterogeneous clastic rock formations can be improved by more than 10% compared with the identification accuracy of the support vector machine classification model.

[0079] To better understand the exemplary embodiments of the present invention described above, further explanation is provided below with reference to specific examples.

[0080] Example 1

[0081] Taking the highly heterogeneous Cretaceous Napo Formation (T sandstone layer) in the Orient Basin of South America as an example, the strata of this target area are finely divided.

[0082] (1) Based on the experimental data of core physical properties, the core flow zone index is calculated by formula (1), formula (2) and formula (3).

[0083] (2) The mercury intrusion porosimetry experimental data were organized, and the pore throat radius data corresponding to 35% saturation of the unwetting phase on the mercury intrusion porosimetry curve were read. Representative data are shown in Table 1:

[0084] Table 1. Representative sample data of core physical properties and mercury porosimetry parameters.

[0085]

[0086] (3) Return the core data used in steps (1) and (2) to the depth, that is, return it to the logging depth.

[0087] (4) Use the flow zone index and pore throat radius data obtained in steps (1) and (2) to perform cross-analysis, classify the flow unit types and label the core data. Figure 2The diagram shows the distribution of flow unit types.

[0088] (5) High-dimensional characteristic parameters characterizing the flow unit may include natural gamma (GR), natural potential (SP), compensated neutron (CNL), compensated acoustic wave (AC), density curve (DEN), deep lateral resistivity (Rt) and shallow lateral resistivity (Rxo), mass photoelectric absorption cross-section index (Pe), deep / shallow resistivity ratio (Rt / Rxo), and volume photoelectric absorption cross-section index (U). Principal component analysis is performed on the high-dimensional characteristic parameters characterizing the flow unit to obtain characteristic parameters used for flow unit modeling, including the first principal component (X1), the second principal component (X2), the third principal component (X3), and the fourth principal component (X4).

[0089] (6) Combine the labeled core data obtained in step (4) and the well logging feature parameters obtained in step (5) for flow unit modeling into a machine learning sample dataset. Use 80% of the data as the training dataset and the remaining 20% ​​as the validation dataset.

[0090] (7) Input the parameters in the training dataset into the SGBDT decision tree integrator and use the stochastic gradient boosting decision tree algorithm to establish a fine partitioning model of the flow unit using equations (4) and (5). Figure 3 This paper presents an AI-based method for evaluating flow cells using PCA-SGBDT, such as... Figure 3 As shown, the input data first calculates a data subset D1 using the algorithm's default weights. A considerable proportion of the subset data is then extracted from a randomly set range of factors (f ranges from 0 to 1, and in this case, the range is 0.5 to 0.8) to obtain a base learning model G1 for model learning. Then, the residual e is used to determine whether to enter the next cycle, and so on, until the maximum number of cycles M is reached to obtain a strong ensemble learning model G.

[0091] (8) Using the validation dataset, the accuracy and recall are calculated using equations (6) and (7) to verify the fine-grained flow unit partitioning model and test whether it meets the accuracy requirements. If it is qualified, the model is then applied to all single wells in the target area to obtain the fine-grained flow unit discrimination results for each well. The fine-grained flow unit partitioning model established in this example is applied to single wells in the target area to obtain the fine-grained flow unit discrimination results. Figure 4 The results of the flow unit division of well PRHC-22 in this target area are shown. The flow units predicted by the PCA-SGBDT algorithm are highly correlated with the flow units in the core. Figure 5The comparison of flow unit segmentation in well PRHC-22 within the target area (comparison of actual and predicted values) is shown, with the correlation of each flow unit type exceeding 88%. In this example, using the confusion matrix (based on precision and recall) to analyze the correlation between the core flow units and the algorithm-predicted flow units, the overall correlation reaches 93.72%, and the accuracy of flow unit identification is more than 10% higher than that of the support vector machine classification model.

[0092] A second exemplary embodiment of the present invention provides a computer device, which mainly includes at least one processor and a memory storing program instructions, wherein the program instructions are configured to be executed by at least one processor, and the program instructions may include instructions for performing the method described in the first exemplary embodiment above.

[0093] A third exemplary embodiment of the present invention provides a computer-readable storage medium that primarily stores computer program instructions, which, when executed by a processor, can implement the method described in the first exemplary embodiment.

[0094] In summary, the advantages proposed by this invention include at least one of the following:

[0095] (1) The method for fine division of flow units in strongly heterogeneous clastic rock formations proposed in this invention has high accuracy in dividing formations with strong heterogeneity and many types of flow units.

[0096] (2) The novel artificial intelligence algorithm used in the fine division method of flow units in strongly heterogeneous clastic rock formations proposed in this invention can better solve the problem of accuracy in fine division of reservoir flow units caused by insufficient core analysis data in complex lithological environments.

[0097] (3) The stochastic gradient boosting decision tree algorithm used in the fine division method of flow units in strongly heterogeneous clastic rock formations proposed in this invention is a state-of-the-art artificial intelligence algorithm with the advantages of high prediction accuracy, ability to process nonlinear data, and flexible processing of various types of data including continuous and discrete values, and can more accurately divide the flow unit types.

[0098] (4) The method for fine division of flow units in strongly heterogeneous clastic rock formations proposed in this invention improves the accuracy of fluid identification by more than 10%.

[0099] (5) The computer equipment and computer-readable storage medium proposed in this invention can help solve problems such as unclear understanding of reservoir fluid interface division and large contradiction between single-well division and overall reservoir law.

[0100] Although a method for finely dividing formation flow units, a computer device, and a readable storage medium of the present invention have been described above in conjunction with exemplary embodiments, those skilled in the art will understand that various modifications and changes can be made to the exemplary embodiments of the present invention without departing from the spirit and scope defined by the claims.

Claims

1. A method for finely dividing flow units in strongly heterogeneous clastic rock formations, characterized in that, The refined partitioning method includes the following steps: Based on the experimental data of the core physical properties of the target area, the flow zone index of the core sample was calculated; Based on the mercury intrusion porosimetry data of the core samples in the target area, the pore throat radius parameter corresponding to 35% saturation of the non-wetting phase on the mercury intrusion porosimetry curve of the core samples was obtained. Based on the cross-plot analysis results of the flow zone index and pore throat radius parameters, the flow unit types are classified, and the core data samples are labeled with flow unit classification tags as output data for the fine division model of flow units. Core repositioning refers to repositioning the depth of core sample data to the logging depth. High-dimensional characteristic parameters are constructed based on conventional well logging curves; Principal component analysis was used to reduce the dimensionality of high-dimensional feature parameters and extract the feature parameters that characterize the flow unit type as input data for the flow unit fine partitioning model. The feature parameters characterizing the flow unit type are combined with the flow unit type label data to form a machine learning sample dataset, which includes a training dataset and a validation dataset. For the training dataset, a stochastic gradient boosting decision tree algorithm is used to establish a fine-grained partitioning model for flow units; The established flow unit fine division model was tested using a validation dataset. After the calculated accuracy and recall met the accuracy requirements, the model was extended to all single wells in the target area to obtain the fine discrimination results of flow units for each well. The high-dimensional characteristic parameters include natural gamma, natural potential, compensated neutron, compensated acoustic wave, density curve, deep lateral resistivity, shallow lateral resistivity, mass photoelectric absorption cross section index, deep-shallow resistivity ratio, and volume photoelectric absorption cross section index; the characteristic parameters characterizing the flow unit type include the first principal component to the i-th principal component.

2. The method for finely dividing flow units in strongly heterogeneous clastic rock formations according to claim 1, characterized in that, The formula for calculating the core flow zone index is as follows: （1） （2） （3） in, Φ z The ratio of pore volume to particle volume is dimensionless. Φ e Core porosity, decimal; K Core permeability, in mD; RQI This is a reservoir quality indicator, dimensionless. FZI The core flow zone index is dimensionless.

3. The method for finely dividing flow units in strongly heterogeneous clastic rock formations according to claim 1, characterized in that, The calculation formula for dimensionality reduction of high-dimensional feature parameters using the principal component analysis method is as follows: （4） in, The covariance matrix is ​​dimensionless. X These are eigenvalues, dimensionless; m The number of samples; Let be the i-th sample value when is feature X, which is dimensionless; This is the average value, dimensionless.

4. The method for finely dividing flow units in strongly heterogeneous clastic rock formations according to claim 1, characterized in that, The calculation formula for the stochastic gradient boosting decision tree algorithm is as follows: ,1≤ m ≤ M （5） in, Y m+1 (X) For the new learning model; f The random factor is dimensionless. This is the current learning model; The learning rate is dimensionless. This is the base learning model fitted along the negative gradient direction of the current loss function; M This represents the maximum number of loops. m This represents the number of samples.

5. The method for finely dividing flow units in strongly heterogeneous clastic rock formations according to claim 1, characterized in that, 70-80% of the data in the machine learning sample dataset is used as the training dataset, and the remaining data is used as the validation dataset.

6. The method for finely dividing flow units in strongly heterogeneous clastic rock formations according to claim 5, characterized in that, The formulas for calculating precision and recall are as follows: （6） （7） in, Accuracy For accuracy; Recall Recall rate; TP For positive samples that the model predicts to be positive; FP For negative samples that the model predicts to be positive; FN For positive samples that the model predicts to be negative; TN For negative samples that the model predicts to be negative.

7. A computer device, characterized in that, include: At least one processor; A memory storing program instructions, wherein the program instructions are configured to be executed by the at least one processor, the program instructions including instructions for performing the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images