Device and method for judging fluid of strike-slip breaking control carbonate fractured-vuggy reservoir
By constructing a cost-sensitive random forest model and combining it with seismic and well logging data, the problem of misjudgment in identifying fluid properties in strike-slip fault-controlled fracture-vuggy carbonate reservoirs was solved, improving the accuracy of identification and economic benefits.
Patent Information
- Application Number
- CN202511388699.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-26
- Publication Date
- 2026-01-09
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing technologies are insufficient to accurately identify the fluid properties in strike-slip fault-controlled fracture-vuggy carbonate reservoirs, leading to misjudgments and economic losses, especially when oil layers are mistakenly identified as water layers, resulting in huge losses.
An apparatus and method are employed to identify fluid properties by combining seismic data and multi-type well logging data through data acquisition, preprocessing, feature reduction, and fluid discrimination modules. This method considers the cost differences in misclassification of different fluid types and improves the discrimination accuracy.
This improves the targeting and accuracy of fluid identification and reduces economic losses caused by misjudgments.
Smart Images
Figure CN121302002A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of fluid discrimination, in particular to a strike-slip fault controlled carbonate fracture-cave type reservoir fluid discrimination device and method. BACKGROUND
[0002] The strike-slip fault controlled carbonate fracture-cave type reservoir is an important target of current deep oil and gas exploration. Its reservoir space is mainly composed of fractures and caves, and has strong heterogeneity, which leads to extremely complex fluid (oil, gas and water) distribution rule and extremely difficult accurate discrimination of fluid properties.
[0003] Traditional reservoir fluid identification methods are mostly based on statistical analysis of logging data or machine learning models such as support vector machine and BF neural network, but these methods all assume that the misclassification cost of each type of sample is equal. However, in the actual exploration and development scene, the economic cost of two types of misclassification is significantly different - misjudging oil layer as water layer may lead to missed production of high-quality reservoir, causing economic loss of tens of millions or even hundreds of millions of yuan.
[0004] Therefore, how to reduce the loss caused by misjudgment and more accurately identify the type of reservoir fluid is a difficult problem to be solved. SUMMARY
[0005] The purpose of the present application is to provide a strike-slip fault controlled carbonate fracture-cave type reservoir fluid discrimination device and method to solve the technical problems mentioned in the background.
[0006] To achieve the above purpose, the present application provides a strike-slip fault controlled carbonate fracture-cave type reservoir fluid discrimination device, comprising a data acquisition module, a data preprocessing module, a strike-slip fault construction module, a feature dimension reduction module and a fluid discrimination module.
[0007] The data acquisition module is used for acquiring seismic data, multi-type logging data and oil testing results of multi-type logging.
[0008] The strike-slip fault construction module is used for constructing a reservoir model based on seismic data and multi-type logging data, and determining a fluid accumulation target area based on the reservoir model.
[0009] The data preprocessing module is used for acquiring multi-type logging data of the fluid accumulation target area, and preprocessing the data acquired by the data acquisition module to optimize the original data.
[0010] The feature dimension reduction module is used for filtering and screening features using a filtering method and a wrapping method based on the optimized multi-type logging data, and performing data dimension reduction.
[0011] The fluid discrimination module is used for constructing a sample set, and discriminating the properties of carbonate fracture-cave type reservoir fluid using a cost-sensitive learning method.
[0012] Preferably, the seismic data acquired by the data acquisition module is three-dimensional seismic data, and the multi-type logging data includes conventional logging data and logging parameter data, wherein the logging parameter data includes seven logging curves of natural gamma, natural potential, compensated neutron, compensated density, acoustic time difference, formation resistivity and flushed zone resistivity, five physical parameter data of permeability, porosity, pressure, shale content and water saturation, and the category of the oil testing result is oil layer, water layer and dry layer.
[0013] Preferably, the strike-slip fault construction module identifies a strike-slip fault zone controlling a carbonate fracture-cave type reservoir based on three-dimensional seismic data and conventional logging data, divides the strike-slip fault zone into different levels of a main fault zone, a secondary fault zone and a derived fracture zone according to the scale, activity period and relationship with oil and gas accumulation of the fault, and establishes a three-dimensional reservoir model to determine the fluid accumulation target area of each level of the reservoir model.
[0014] Preferably, the data preprocessing module preprocesses the logging parameter data, specifically: based on the logging curves and the physical parameters, 12 kinds of parameter data are acquired, and the parameter data are cleaned, missing value processed, and standardized and normalized.
[0015] Preferably, the feature dimension reduction module uses the correlation coefficient method and the recursive feature elimination method to screen features, and uses the principal component analysis method to reduce the dimension of the data.
[0016] Preferably, the fluid discrimination module includes a feature construction and fusion module and a cost-sensitive learning model training module.
[0017] The feature construction module combines or transforms the original features to generate new features, and constructs a comprehensive feature data set of multi-source information based on the new features, the screened features and the oil testing result.
[0018] The cost-sensitive learning model training module constructs and trains a cost-sensitive random forest model, and uses the trained cost-sensitive random forest model to discriminate the fluid properties of the carbonate fracture-cave type reservoir to be tested.
[0019] Preferably, the construction and training of the cost-sensitive random forest model includes:
[0020] The basic parameters of the random forest model are determined through grid search and cross-validation, including the number of decision trees, the maximum tree depth, and the minimum number of samples for node splitting;
[0021] The cost-sensitive random forest model is constructed by introducing the misclassification cost, wherein the cost-sensitive random forest model includes a plurality of cost-sensitive decision trees.
[0022] The comprehensive feature dataset is taken as a training set, and multiple sub-sample sets are generated from the training set by sampling with replacement, each of which is used to train a cost-sensitive decision tree; for a sample in the test set, the sample is input into all decision trees, each of which outputs a classification result, and a final discrimination result is determined by weighted voting.
[0023] Preferably, the introducing the misclassification cost comprises:
[0024] Based on the fluid type, a misclassification cost matrix of the oil layer, the water layer and the dry layer is constructed, specifically:
[0025]
[0026] In the formula, R o represents the loss benefit of misjudging as a dry layer, C T represents the oil testing cost, C D represents the invalid development cost, 0 represents no cost, and i in C(i,j) represents the true class, and j represents the discrimination class.
[0027] Based on the misclassification cost matrix, a node splitting criterion is modified, and the formula is:
[0028]
[0029] In the formula, I represents a node, p i represents the proportion of samples of the true class i in the current node, p j represents the probability that the sample is predicted to be the class j in the current node.
[0030] A loss function L T of a single tree is set, and the formula is:
[0031]
[0032] In the formula, y k represents the true class of the kth sample, represents the predicted class of the kth sample by the decision tree T, and k represents a sample index, from 1 to N.
[0033] An overall loss function L A is L
[0034]
[0035] In the formula, represents the final predicted class of the kth sample by the forest, s k represents the true class of the kth sample, represents the voting proportion of the kth sample being predicted to be the class j, Ym,k represents the predicted class of the mth tree for the kth sample;
[0036] The overall objective function is:
[0037]
[0038] where T1, T2,..., T M are M decision trees, Ω({T m}) is a regular direction, and λ≥0 is a regularization coefficient.
[0039] Preferably, the cost-sensitive random forest model uses an Adam optimizer for parameter optimization.
[0040] The application also provides a strike-slip fault-controlled carbonate fracture-cave type reservoir fluid discrimination method, comprising the following steps:
[0041] S1, obtaining seismic data, multi-type logging data, and oil testing results of multi-type logging of the strike-slip fault-controlled carbonate fracture-cave type reservoir;
[0042] S2, constructing a strike-slip fault-controlled carbonate fracture-cave type reservoir model and determining a reservoir model fluid accumulation target area;
[0043] S3, preprocessing logging parameter data of the fluid accumulation target area, including data cleaning, missing value processing, data standardization, and normalization;
[0044] S4, performing feature screening and dimension reduction on the preprocessed data;
[0045] S5, constructing a carbonate fracture-cave type reservoir data set;
[0046] S6, constructing a cost-sensitive random forest model and training using the data set, and using the trained cost-sensitive random forest model to discriminate the fluid properties of a carbonate fracture-cave type reservoir to be tested.
[0047] Therefore, the application has the following beneficial effects by adopting the above strike-slip fault-controlled carbonate fracture-cave type reservoir fluid discrimination device and method:
[0048] (1) By constructing a three-dimensional reservoir model using seismic data and logging data and determining a fluid accumulation target area, subsequent data preprocessing and fluid discrimination are focused on the most favorable reservoir area in geology, improving the pertinence and overall accuracy of fluid discrimination work;
[0049] (2) By considering the cost difference of misjudgment of different fluid types, the discrimination accuracy is improved.
[0050] The technical solutions of the application will be further described in detail below with reference to the drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS
[0051] Figure 1 A device structure diagram of an embodiment of the present application;
[0052] Figure 2 A method flowchart of an embodiment of the present application. DETAILED DESCRIPTION
[0053] To make the objectives, technical solutions and advantages of embodiments of the present application clearer, the following will be combined with the accompanying drawings of embodiments of the present application to make a clear and complete description of the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. The components of the embodiments of the present application described and shown in the accompanying drawings can be arranged and designed in various different configurations, and therefore should not be understood as a limitation of the present application.
[0054] Embodiment 1
[0055] With reference to Figure 1 , the present application provides a strike-slip fault controlled carbonate fracture-cave type reservoir fluid discrimination device, comprising a data acquisition module, a data preprocessing module, a strike-slip fault construction module, a feature dimension reduction module and a fluid discrimination module. The functions of each module are as follows:
[0056] The data acquisition module is used for acquiring seismic data, multi-type logging data and oil testing results of multi-type logging.
[0057] The strike-slip fault construction module is used for constructing a reservoir model based on the seismic data and the multi-type logging data, and determining a fluid accumulation target area based on the reservoir model.
[0058] The data preprocessing module is used for acquiring multi-type logging data of the fluid accumulation target area, and preprocessing the data acquired by the data acquisition module to optimize the original data.
[0059] The feature dimension reduction module is used for screening features using a filtering method and a wrapping method based on the optimized multi-type logging data, and performing data dimension reduction.
[0060] The fluid discrimination module is used for constructing a sample set, and discriminating the fluid properties of the carbonate fracture-cave type reservoir using a cost-sensitive learning method.
[0061] In this embodiment, the seismic data acquired by the data acquisition module is three-dimensional seismic data, the multi-type logging data includes conventional logging data and logging parameter data, the logging parameter data includes seven logging curves of natural gamma, natural potential, compensated neutron, compensated density, acoustic time difference, formation resistivity and flushed zone resistivity, five physical parameter data of permeability, porosity, pressure, shale content and water saturation, the conventional logging data includes formation data, oil testing and production dynamic data, drilling and logging data and geological data, and the categories of the oil testing results are oil layer, water layer and dry layer.
[0062] In the embodiment, the strike-slip fault modeling module identifies the carbonate fracture-cave type reservoir controlled by the strike-slip fault zone based on three-dimensional seismic data and conventional logging data, divides the strike-slip fault zone into different levels of main fault zone, secondary fault zone and derived fracture zone according to the scale, activity period and relationship with hydrocarbon accumulation of the fault, and establishes a three-dimensional reservoir model to determine the fluid accumulation target area of each level of the reservoir model.
[0063] In the embodiment, the data preprocessing module pre-processes the logging parameter data, specifically: based on the logging curve and physical property parameters, 12 kinds of parameter data are obtained, and the parameter data is cleaned, missing value processed, and standardized and normalized. The cleaning specifically uses the k-neighbor algorithm to detect abnormal points, and uses the mean, median and mode to fill or delete abnormal values. For missing values, appropriate filling strategies need to be selected according to the missing mechanism (such as random missing, completely random missing, etc. Normalization scales the data to the interval [0, 1] or [-1, 1].
[0064] In the embodiment, the feature dimension reduction module uses the correlation coefficient method and the recursive feature elimination method to screen features, and uses the principal component analysis method to reduce the dimension of the data.
[0065] In the embodiment, the fluid discrimination module includes a feature construction and fusion module and a cost-sensitive learning model training module, wherein:
[0066] The feature construction module combines or transforms the original features to generate new features, and constructs a comprehensive feature data set of multi-source information based on the new features, the screened features and the oil test results;
[0067] The cost-sensitive learning model training module constructs and trains a cost-sensitive random forest model, and uses the trained cost-sensitive random forest model to discriminate the fluid properties of the carbonate fracture-cave type reservoir to be tested.
[0068] In the embodiment, constructing and training the cost-sensitive random forest model includes:
[0069] The base parameters of the random forest model are determined through grid search and cross-validation, including the number of decision trees, the maximum tree depth, and the minimum number of samples for node splitting;
[0070] The misclassification cost is introduced to construct the cost-sensitive random forest model, wherein the cost-sensitive random forest model includes a plurality of cost-sensitive decision trees;
[0071] The comprehensive feature dataset is taken as a training set, and multiple sub-sample sets are generated from the training set by sampling with replacement, each of which is used to train a cost-sensitive decision tree; for a sample in the test set, the sample is input into all decision trees, each decision tree outputs a classification result, and the final discrimination result is determined by weighted voting.
[0072] In this embodiment, the misclassification cost is introduced, and the cost-sensitive random forest model is constructed, including:
[0073] Based on the fluid type, the misclassification cost matrix of the oil layer, the water layer and the dry layer is constructed, specifically:
[0074]
[0075] In the formula, R o represents the loss of misjudging as a dry layer, C T represents the oil testing cost, C D represents the invalid development cost, 0 represents no cost, and i in C(i,j) represents the true class, and j represents the discrimination class.
[0076] Based on the misclassification cost matrix, the node splitting criterion is modified, and the formula is:
[0077]
[0078] In the formula, I represents a node, p i represents the proportion of samples of the true class i in the current node, p j represents the probability that the sample is predicted to be class j in the current node.
[0079] The loss function L T of a single tree is set, and the formula is:
[0080]
[0081] In the formula, y k represents the true class of the kth sample, represents the predicted class of the kth sample by the decision tree T, and k represents the sample index, from 1 to N.
[0082] The overall loss function L A is L
[0083]
[0084] In the formula, represents the final predicted class of the kth sample by the forest, s k represents the true class of the kth sample, represents the voting proportion of the kth sample being predicted to be class j, Ym,k represents a predicted class of the mth tree for the kth sample;
[0085] The overall objective function is:
[0086]
[0087] where T1, T2,..., T M are M decision trees, Ω({T m}) is a regular direction, and λ≥0 is a regularization coefficient.
[0088] In this embodiment, the cost-sensitive random forest model uses an Adam optimizer for parameter optimization.
[0089] Embodiment 2
[0090] With reference to Figure 2 The present application provides a strike-slip fault-controlled carbonate fracture-cave type reservoir fluid discrimination method, comprising the following steps:
[0091] S1, obtaining seismic data, multi-type logging data and oil testing results of the multi-type logging of the strike-slip fault-controlled carbonate fracture-cave type reservoir;
[0092] S2, constructing a strike-slip fault-controlled carbonate fracture-cave type reservoir model to determine a reservoir model fluid accumulation target area;
[0093] S3, preprocessing the logging parameter data of the fluid accumulation target area, including data cleaning, missing value processing, data standardization and normalization;
[0094] S4, performing feature screening and dimensionality reduction on the preprocessed data;
[0095] S5, constructing a carbonate fracture-cave type reservoir data set;
[0096] S6, constructing a cost-sensitive random forest model and training using the data set, and using the trained cost-sensitive random forest model to discriminate the fluid properties of the carbonate fracture-cave type reservoir to be tested.
[0097] Therefore, the present application adopts the above-mentioned strike-slip fault-controlled carbonate fracture-cave type reservoir fluid discrimination device and method, which considers the cost difference of misjudgment of different fluid types, thereby improving the discrimination accuracy.
[0098] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application and not to limit them, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that: it can still modify or equivalently replace the technical solutions of the present application, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present application.
Claims
1. A device for distinguishing fluids in strike-slip fault-controlled carbonate fractured-vuggy reservoirs, characterized in that: The method comprises a data acquisition module, a data preprocessing module, a strike-slip fault construction module, a feature dimension reduction module, and a fluid discrimination module. The data acquisition module is used for acquiring seismic data, multi-type logging data, and oil testing results of multi-type logging. The strike-slip fault construction module is used for constructing a reservoir model based on the seismic data and the multi-type logging data, and determining a fluid accumulation target area based on the reservoir model. The data preprocessing module is used for acquiring multi-type logging data of the fluid accumulation target area, and preprocessing the data acquired by the data acquisition module to optimize the original data. The feature dimension reduction module is used for filtering and screening features by using a filtering method and a wrapping method based on the optimized multi-type logging data, and performing data dimension reduction. The fluid discrimination module is used for constructing a sample set, and discriminating fluid properties of a carbonate fracture-cave type reservoir by using a cost-sensitive learning method.
2. The apparatus according to claim 1, wherein the apparatus is characterized in that: The seismic data acquired by the data acquisition module is three-dimensional seismic data, the multi-type logging data comprises conventional logging data and logging parameter data, the logging parameter data comprises seven logging curves of natural gamma, natural potential, compensated neutron, compensated density, acoustic time difference, formation resistivity, and flushed zone resistivity, and five physical parameter data of permeability, porosity, pressure, shale content, and water saturation, and the oil testing result is classified into an oil layer, a water layer, and a dry layer.
3. The apparatus according to claim 2, wherein the apparatus is characterized in that: The strike-slip fault construction module is used for identifying a strike-slip fault zone controlling a carbonate fracture-cave type reservoir based on three-dimensional seismic data and conventional logging data, dividing the strike-slip fault zone into different levels of a main fault zone, a secondary fault zone, and a derived fracture zone according to the size, activity period, and relationship with oil and gas accumulation of the fault, and establishing a three-dimensional reservoir model to determine fluid accumulation target areas of different levels of the reservoir model.
4. The apparatus according to claim 2, wherein the apparatus is characterized by: The data preprocessing module is used for preprocessing the logging parameter data, specifically, acquiring 12 kinds of parameter data based on logging curves and physical parameters, and performing cleaning, missing value processing, data standardization, and normalization on the parameter data.
5. The apparatus according to claim 1, wherein the apparatus is characterized in that: The feature dimension reduction module is used for screening features by using a correlation coefficient method and a recursive feature elimination method, and performing dimension reduction on the data by using a principal component analysis method.
6. The apparatus according to claim 1, wherein the apparatus is characterized by: The fluid discrimination module comprises a feature construction and fusion module and a cost-sensitive learning model training module. The feature construction module is used for combining or transforming original features to generate new features, and constructing a comprehensive feature data set of multi-source information based on the new features, screened features, and oil testing results. The cost-sensitive learning model training module is used for constructing and training a cost-sensitive random forest model, and discriminating fluid properties of a carbonate fracture-cave type reservoir to be tested by using the trained cost-sensitive random forest model.
7. The apparatus according to claim 6, wherein, The construction and training of the cost-sensitive random forest model comprises: determining basic parameters of the random forest model by grid search and cross-validation, including the number of decision trees, the maximum tree depth, and the minimum number of samples for node splitting; introducing a misclassification cost to construct a cost-sensitive random forest model, wherein the cost-sensitive random forest model comprises a plurality of cost-sensitive decision trees. The integrated feature dataset is taken as a training set, and multiple sub-sample sets are generated from the training set through sampling with replacement, each of which is used to train a cost-sensitive decision tree; for a sample in the test set, the sample is input into all decision trees, each of which outputs a classification result, and a final discrimination result is determined through weighted voting.
8. The apparatus according to claim 7, wherein, The cost-sensitive random forest model is constructed by introducing the misclassification cost, and the cost-sensitive random forest model comprises the following steps: Based on the fluid type, a misclassification cost matrix of the oil layer, water layer and dry layer is constructed, and the misclassification cost matrix is specifically constructed as follows: where R o represents the loss of revenue from misclassification as dry, C T represents the cost of testing, C D represents the cost of ineffective development, 0 represents no cost, and i in C(i,j) represents the true class and j represents the classification class; Based on the misclassification cost matrix, the node splitting criterion is modified, and the formula is as follows: In the formula, I represents a node, p i represents the proportion of samples with the true class i in the current node, p j represents the probability that a sample is predicted to be of class j in the current node; Set the loss function L of a single tree T The formula is: where y k represents the true class of the kth sample, represents the predicted class of the kth sample by the decision tree T, k represents the sample index, from 1 to N; The overall loss function L A The formula is: wherein, denotes the final predicted class of the forest for the kth sample, s k denotes the true class of the kth sample, denotes the fraction of votes that the kth sample is predicted to be class j, denotes the predicted class of the mth tree for the kth sample; The overall objective function is as follows: In the formula, T1, T2, …, T M are M decision trees, Ω({T m}) is a regular direction, and λ≥0 is a regularization coefficient.
9. The apparatus according to claim 7, wherein the apparatus is characterized by: The cost-sensitive random forest model uses an Adam optimizer to optimize parameters.
10. A method for determining the fluid in a strike-slip fault-controlled carbonate fractured-vuggy reservoir, characterized by the steps of: using the device for determining the fluid in a strike-slip fault-controlled carbonate fractured-vuggy reservoir according to any one of claims 1-9. The method comprises the following steps: S1, acquiring seismic data, multiple types of logging data and oil testing results of the strike-slip fault-controlled carbonate fracture-vug reservoir; S2, constructing a strike-slip fault-controlled carbonate fracture-vug reservoir model to determine a fluid accumulation target area of the reservoir model; S3, preprocessing logging parameter data of the fluid accumulation target area, including data cleaning, missing value processing, data standardization and normalization; S4, performing feature screening and dimension reduction on the preprocessed data; S5, constructing a carbonate fracture-vug reservoir dataset; S6, constructing a cost-sensitive random forest model and training the model using the dataset, and using the trained cost-sensitive random forest model to discriminate fluid properties of a carbonate fracture-vug reservoir to be tested.