A reservoir fluid property identification method based on KNN

Abstract

The application discloses a reservoir fluid property identification method based on KNN, and steps include: S1, obtaining the logging data of all tested oil wells in the same block range as the unknown reservoir as interpretation attributes; S2, normalizing the interpretation attributes to generate a multi-dimensional cubic attribute space; S3, defining a plurality of fluid labels for representing the reservoir fluid property; S4, each sample data under each attribute corresponds to a fluid label; S5, calculating the attribute sample mean of each attribute sample data under different labels; S6, determining the sensitive attribute for identifying the fluid property; S7, constructing a fluid property identification model; S8, obtaining the prediction result of the unknown reservoir fluid property; the reservoir fluid property identification method based on KNN can realize self-adaptive determination of the number K of the near neighbor sample points of the KNN method and the discrimination criterion with less manual intervention, and improve the identification precision of the complex reservoir fluid property.

Description

Technical Field

[0001] This invention relates to the field of conventional and complex reservoir fluid property identification, as well as sweet spot evaluation technology for unconventional reservoirs such as shale oil, and particularly to a KNN-based reservoir fluid property identification method. Background Technology

[0002] In the field of oil and gas exploration, identifying reservoir fluid properties using well logging data is a crucial aspect of reservoir evaluation. Fluid property identification often involves constructing fluid identification charts by selecting sensitive parameters. Traditional methods rely heavily on experienced experts for manual extraction, which demands extensive experience in regional well logging interpretation, leading to uneven interpretation and evaluation techniques across oilfields and low efficiency. To improve the accuracy of reservoir hydrocarbon identification, enhance interpretation and development efficiency, and reduce labor costs, an intelligent reservoir fluid property identification method is needed. This method aims to improve the accuracy of reservoir hydrocarbon identification, achieve rapid reservoir evaluation, and guide subsequent development and construction plans. Summary of the Invention

[0003] The purpose of this invention is to provide a KNN-based reservoir fluid property identification method that can accurately identify the fluid properties of untested reservoirs.

[0004] Therefore, the technical solution of the present invention is as follows:

[0005] A reservoir fluid property identification method based on KNN, comprising the following steps:

[0006] S1. Obtain logging data from all tested wells located in the same block as the unknown reservoir to obtain N types of logging data for predicting the reservoir properties of the unknown wells. Each type of logging data represents an interpretation attribute.

[0007] S2. Using the max-min standardization method, all sample data within each explanatory attribute are projected into the interval [-10, 10] to generate a multidimensional cube attribute space for searching neighboring samples;

[0008] S3. Based on the reservoir fluid properties of all tested wells, define multiple fluid labels to characterize the reservoir fluid properties;

[0009] S4. Substitute the fluid labels defined in step S3 into the multidimensional cube attribute space obtained in step S2, so that each sample data under each attribute corresponds to a fluid label.

[0010] S5. Calculate the mean of the sample data for each attribute under different labels;

[0011] S6. Sort the mean values ​​of attribute samples for each attribute under different labels in ascending order, and use the t-test to check whether the mean values ​​of attribute samples for the two types of labels in the same sequence before and after are equal: if they are not equal, then the attribute is a sensitive attribute for identifying fluid properties; if they are equal, then the attribute is not a sensitive attribute.

[0012] S7. Construct a fluid property identification model, including:

[0013] 1) Use the sample data corresponding to the sensitive attributes as the total sample data for modeling, and randomly divide them into two groups: the explanatory sample group and the explained sample group;

[0014] 2) Set the maximum number of nearest neighbor samples input by the user to K. max ;

[0015] 3) The judgment criteria for the model include simple voting, weighted voting, and attribute preference weighting.

[0016] 4) For any explained sample, search for K in the explained sample group. max Using the nearest neighbor samples, the label attribution of the explained sample is determined iteratively, and the error vector is calculated:

[0017] ,

[0018] In the formula, Record the number of errors when using the j-th discrimination criterion to select the k-th nearest neighbor sample point; k=1, 2,3, …, K max j=1, 2, 3;

[0019] Based on this, the error vector calculation result is obtained by iterating through all the sample data in the sample group to be explained.

[0020] 5) Using the method described in step 4) above, obtain the calculation results of the error vector under different combinations of K and j values. Take the K value and j value corresponding to the minimum value of the error vector calculation result, which are the best nearest neighbor sample point K and the optimal discrimination criterion j in the fluid property identification model, respectively.

[0021] S8. Take the total sample data in step S7 as the interpretation sample group, and standardize the interpretation attributes of the reservoir to be evaluated according to the method in step S2 as the interpreted sample group. In the interpretation sample group, search for the number of k nearest neighbor sample points in the attribute space, and obtain the prediction results of reservoir fluid properties based on the optimal discrimination criterion determined in step S7.

[0022] Furthermore, in step S1, the logging data is selected from natural gamma, spontaneous potential, deep resistivity, photoelectric absorption cross section index, compensated acoustic wave, compensated density, compensated neutron and gas-based total hydrocarbons.

[0023] Furthermore, in step S2, the maximum-min normalization method calculation formula for each type of sensitive logging data is as follows:

[0024] ,

[0025] In the formula, Let be the maximum value of the i-th attribute. Let x be the minimum value of the i-th attribute. i Let i be the measured value of the i-th attribute. This is the standardized result for the i-th attribute.

[0026] Furthermore, in step S5, the formula for calculating the mean of the attribute samples is:

[0027] ,

[0028] In the formula, i is the attribute index and k is the tag index. Let be the sample mean of the k-th label and the i-th attribute. denoted as the number of sample data for the k-th label.

[0029] Furthermore, in step S7, the maximum number of nearest neighbor sample points K max The range of values ​​for is: 1 ≤ K max <30.

[0030] Furthermore, in step S7, the judgment criterion for the simple voting method is: the label of an explained sample is the same as the label of the most nearby sample points. The judgment criterion for the weighted voting method is: assuming that each nearby point contributes differently to the determination of the label of the explained sample point, and that its contribution decreases as the distance in the attribute space increases, that is, the weight of the nearby sample is a function of distance, the cumulative weight of each label sample in the nearby samples is calculated, and the label of the explained sample point is the same as the label with the largest cumulative weight. The judgment criterion for the attribute preference weighted method is: when calculating the weight, the contribution rate of a certain attribute can be increased based on expert experience, and its judgment method is consistent with the weighted voting method.

[0031] Furthermore, in step S7, the number of sample data in the explanatory sample group is 80% of the total sample data, and the number of sample data in the explained sample group is 20% of the total sample data.

[0032] Compared with existing technologies, this KNN-based reservoir fluid property identification method can adaptively determine the number of nearest neighbor sample points K and the discrimination criterion of the KNN method with less manual intervention, thereby improving the identification accuracy of complex reservoir fluid properties. Attached Figure Description

[0033] Figure 1This is a flowchart of the reservoir fluid property identification method based on KNN of the present invention;

[0034] Figure 2 This is a comparison image of the neighboring sample point picking areas using conventional methods and the method of this application;

[0035] Figure 3 This is a comparison diagram of the unknown reservoir fluid properties determined using conventional methods and the methods described in this application. Detailed Implementation

[0036] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the following embodiments are by no means intended to limit the present invention.

[0037] See Figure 1 The specific implementation steps for identifying the reservoir properties of a shale oil sweet spot layer in the QB101 well area of ​​the Shahekou Depression Qibei Slope Shahekou section using the reservoir fluid property identification method based on KNN proposed in this application are as follows:

[0038] S1. Obtain logging data from all tested wells located in the same block as the unknown reservoir to obtain N types of logging data for predicting the reservoir properties of the unknown wells. Each type of logging data represents an interpretation attribute. The logging data is selected from natural gamma, spontaneous potential, deep resistivity, photoelectric absorption cross section index, compensated acoustic wave, compensated density, compensated neutron, and gas-based total hydrocarbons.

[0039] S2. Using the max-min standardization method, all sample data within each explanatory attribute are projected into the interval [-10, 10] to generate a multidimensional cube attribute space for searching neighboring samples;

[0040] Specifically, the maximum-min normalization formula for each type of sensitive logging data is as follows:

[0041] ,

[0042] In the formula, Let be the maximum value of the i-th attribute. Let x be the minimum value of the i-th attribute. i Let i be the measured value of the i-th attribute. This represents the standardized result for the i-th attribute;

[0043] S3. Based on the reservoir fluid properties of all tested wells, define multiple fluid labels to characterize the reservoir fluid properties. Among them, the fluid property types in oil fields usually include: oil layer, poor oil layer, oil-water co-containment layer, oil-water layer, water layer, dry layer, etc. In this step, data processing personnel can add fluid property types according to the degree of research refinement, such as subdividing oil layers into Class I oil layers, Class II oil layers, Class III oil layers, etc.

[0044] In this embodiment, the QB101 well area has a well-developed shale oil reservoir in the Shaxia section. Statistical analysis of the oil well data in this block determined that the fluid labels were set into six categories: oil layer (k=1), poor oil layer (k=2), oil-water co-containment layer (k=3), oil-water-bearing layer (k=4), water layer (k=5), and dry layer (k=6). The classification criteria for each label are different based on the range of daily oil production.

[0045] S4. Substitute the fluid labels defined in step S3 into the multidimensional cube attribute space obtained in step S2, so that each sample data under each attribute corresponds to a fluid label.

[0046] S5. Calculate the mean of the sample data for each attribute under different labels;

[0047] Specifically, the formula for calculating the mean of attribute samples is as follows:

[0048] ,

[0049] In the formula, i is the attribute number, k is the tag number, and k = 1, 2, …, 6; Let n be the sample mean of the k-th label and the i-th attribute. k The number of sample data for the k-th class label;

[0050] S6. Sort the mean values ​​of attribute samples for each attribute under different labels in ascending order, and use the t-test to check whether the mean values ​​of attribute samples for the two types of labels in the same sequence before and after are equal: if they are not equal, the attribute is a sensitive attribute for identifying fluid properties; if they are equal, the attribute is not a sensitive attribute; in this embodiment, the data determined to be sensitive attributes after step S6 are natural gamma and resistivity.

[0051] S7. Construct a fluid property identification model, including:

[0052] 1) The sample data corresponding to the sensitive attributes are used as the total sample data for modeling, and are randomly divided into two groups: the explanatory sample group and the explained sample group; the sample data of the explanatory sample group is 80% of the total sample data, and the sample data of the explained sample group is 20% of the total sample data.

[0053] 2) Set the maximum number of nearest neighbor samples input by the user to K. max =30;

[0054] 3) The judgment criteria for the model include simple voting, weighted voting, and attribute preference weighting. The simple voting criterion is that the label of an explained sample is the same as the label of the nearest neighbor sample with the most samples. The weighted voting criterion is that each nearest neighbor contributes differently to determining the label of the explained sample, and its contribution decreases as the distance in the attribute space increases; that is, the weight of the nearest neighbor sample is a function of distance. The cumulative weight of each label sample in the nearest neighbor sample is calculated, and the label of the explained sample is the same as the label with the largest cumulative weight. The attribute preference weighting criterion is that the contribution rate of a certain attribute can be increased based on expert experience when calculating the weights; its judgment method is consistent with the weighted voting method.

[0055] 4) For any explained sample, search for K in the explained sample group. max Using the nearest neighbor samples, the label attribution of the explained sample is determined iteratively, and the error vector is calculated:

[0056] ,

[0057] In the formula, Record the number of errors when using the j-th discrimination criterion to select the k-th nearest neighbor sample point; K=1, 2,3, …, K max j=1, 2, 3;

[0058] Based on this, the error vector calculation result is obtained by iterating through all the sample data in the sample group to be explained.

[0059] 5) Using the method described in step 4) above, obtain the calculation results of the error vector under different combinations of K and j values. Take the K value and j value corresponding to the minimum value of the error vector calculation result, which are the best nearest neighbor sample point K and the optimal discrimination criterion j in the fluid property identification model, respectively. In this embodiment, K=20, j=2, that is, the best nearest neighbor sample point is 20, and the optimal discrimination criterion is the weighted voting method.

[0060] S8. Take the total sample data in step S7 as the interpretation sample group, and standardize the interpretation attributes of the reservoir to be evaluated according to the method in step S2 as the interpretation sample group. In the interpretation sample group, search for the number of 20 nearest neighbor sample points in the attribute space, and obtain the prediction results of reservoir fluid properties based on the weighted voting method determined in step S7 as the discrimination criterion.

[0061] like Figure 2 The image shows a comparison of the nearest neighbor sample point picking regions using conventional methods and the method described in this application. Figure 2The fifth line in the figure is the layer boundary manually determined by technicians using half-amplitude points of logging curves such as natural gamma and resistivity; the sixth line is the layer boundary determined using the method of this application; it can be seen from the layer boundaries in the figure that the two lines are in good agreement.

[0062] like Figure 3 The figure shown is a comparison of the unknown reservoir fluid properties determined using conventional methods and the method described in this application. Among them, Figure 3 The fifth line represents the interpretation conclusion determined by conventional methods, obtained by interpreters over several days by combining regional oil testing data to establish corresponding well logging interpretation charts. The sixth line represents the conclusion determined by this application, and the interpretation conclusions determined by the two methods are consistent. Simultaneously, oil testing was conducted on layers 7-12 of this well, followed by fracturing and blowout, using a 6mm nozzle, resulting in a daily oil production of 8.0 tons. This further demonstrates that the prediction results of the method in this application have a high degree of consistency with the oil testing results. Compared with conventional interpretation methods, the method in this application has the characteristics of high evaluation accuracy, convenience, speed, and strong practicality.

Claims

1. A method for identifying reservoir fluid properties based on KNN, characterized in that, The steps are as follows: S1. Obtain logging data from all tested wells located in the same block as the unknown reservoir to obtain N types of logging data for predicting the reservoir properties of the unknown wells. Each type of logging data represents an interpretation attribute. S2. Using the max-min standardization method, all sample data within each explanatory attribute are projected into the interval [-10, 10] to generate a multi-dimensional cubic attribute space for searching neighboring samples; S3. Based on the reservoir fluid properties of all tested wells, define multiple fluid labels to characterize the reservoir fluid properties; S4. Substitute the fluid labels defined in step S3 into the multidimensional cube attribute space obtained in step S2, so that each sample data under each attribute corresponds to a fluid label. S5. Calculate the mean value of the sample data for each attribute under different labels; S6. Sort the mean values ​​of attribute samples for each attribute under different labels in ascending order, and use the t-test to check whether the mean values ​​of attribute samples for the two types of labels in the same sequence before and after are equal: if they are not equal, then the attribute is a sensitive attribute for identifying fluid properties. If they are equal, then the attribute is not a sensitive attribute; S7. Construct a fluid property identification model, including: 1) The sample data corresponding to the sensitive attributes are used as the total sample data for modeling, and are randomly divided into two groups: the explanatory sample group and the explained sample group; wherein, the sample data of the explanatory sample group accounts for 80% of the total sample data, and the sample data of the explained sample group accounts for 20% of the total sample data; 2) Set the maximum number of nearest neighbor samples input by the user to K. max ; 3) The judgment criteria for the model include simple voting, weighted voting, and attribute preference weighting. 4) For any explained sample, search for K in the explained sample group. max Using the nearest neighbor samples, the label attribution of the explained sample is determined iteratively, and the error vector is calculated: ， In the formula, Record the number of errors when using the j-th discrimination criterion to select the k-th nearest neighbor sample point; k=1, 2, 3,…, K max j=1, 2, 3; Based on this, the error vector calculation result is obtained by iterating through all the sample data in the sample group to be explained. 5) Using the method described in step 4) above, obtain the calculation results of the error vector under different combinations of K and j values. Take the K value and j value corresponding to the minimum value of the error vector calculation result, which are the best nearest neighbor sample point K and the optimal discrimination criterion j in the fluid property identification model, respectively. S8. Take the total sample data in step S7 as the interpretation sample group, and standardize the interpretation attributes of the reservoir to be evaluated according to the method in step S2 as the interpreted sample group. In the interpretation sample group, search for the number of k nearest neighbor sample points in the attribute space, and obtain the prediction results of reservoir fluid properties based on the optimal discrimination criterion determined in step S7.

2. The reservoir fluid property identification method based on KNN according to claim 1, characterized in that, In step S1, the logging data is selected from natural gamma, spontaneous potential, deep resistivity, photoelectric absorption cross section index, compensated acoustic wave, compensated density, compensated neutron and gas-based total hydrocarbons.

3. The reservoir fluid property identification method based on KNN according to claim 1, characterized in that, In step S2, the maximum-min normalization method calculation formula for each type of sensitive logging data is as follows: ， In the formula, Let be the maximum value of the i-th attribute. Let x be the minimum value of the i-th attribute. i Let i be the measured value of the i-th attribute. This is the standardized result for the i-th attribute.

4. The reservoir fluid property identification method based on KNN according to claim 1, characterized in that, In step S5, the formula for calculating the mean of the attribute samples is: ， In the formula, i is the attribute index and k is the tag index. Let be the sample mean of the k-th label and the i-th attribute. denoted as the number of sample data for the k-th label.

5. The reservoir fluid property identification method based on KNN according to claim 1, characterized in that, In step S7, the maximum number of nearest neighbor sample points K max The range of values ​​for is: 1 ≤ K max <30.

6. The reservoir fluid property identification method based on KNN according to claim 1, characterized in that, In step S7, the simple voting method's judgment criterion is: the label of an explained sample is the same as the label of the most nearby sample points. The weighted voting method's judgment criterion is: assuming that each nearby point contributes differently to determining the label of the explained sample point, and that its contribution decreases as the distance in the attribute space increases, i.e., the weight of the nearby sample is a function of distance, the cumulative weight of each label sample in the nearby samples is calculated, and the label of the explained sample point is the same as the label with the largest cumulative weight. The attribute preference weighted method's judgment criterion is: when calculating the weight, the contribution rate of a certain attribute is increased based on expert experience, and its judgment method is consistent with the weighted voting method.

7. The reservoir fluid property identification method based on KNN according to claim 1, characterized in that, In step S7, the number of sample data in the explanatory sample group is 80% of the total sample data, and the number of sample data in the explained sample group is 20% of the total sample data.

Images