Navigation system for oil identification of sand-shale reservoir based on oversampling and ensemble learning

Abstract

The present application belongs to the field of oil and gas reservoir identification, and particularly relates to a sandstone and conglomerate oil and gas reservoir intelligent drawing and oil identification navigation system based on oversampling and ensemble learning, aiming to solve the problem that in the prior art, due to the uneven distribution of various small categories of sandstone and conglomerate lithology samples, direct application of a machine learning model may lead to model overfitting or poor performance. The present application comprises: taking the maximum feature-class information and feature-feature information as sensitive well data according to the round; obtaining a set number Q of random samples with replacement for each heterogeneous ensemble learning model; oversampling according to the cosine similarity to obtain P oversampled data subsets; and obtaining the lithology identification result through the trained heterogeneous ensemble learning model by weighted voting. The present application proposes a method of logging data multi-subset oversampling and multi-model heterogeneous ensemble classification, increases the number and diversity of sandstone and conglomerate logging samples, and greatly improves the accuracy and running efficiency of sandstone and conglomerate reservoir identification.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of oil and gas reservoir identification, and particularly relates to a sandstone and conglomerate oil and gas reservoir intelligent drawing and identification oil navigation system based on oversampling and ensemble learning. BACKGROUND

[0002] Sandstone and conglomerate reservoirs are widely distributed in Mesozoic and Cenozoic continental oil and gas basins in the east and west of China. The characteristics of multiple sources, grain size differentiation and uneven distribution of various small classes of lithology of sandstone and conglomerate mean that the logging data samples are relatively mixed and unevenly distributed, and traditional manual evaluation and interpretation work will face great challenges. Accurate identification of sandstone and conglomerate reservoirs is an important technical requirement for high-precision detection and real-time geosteering of intelligent drawing and identification oil navigation.

[0003] Intelligent lithology identification models based on machine learning have the advantages of high automation, low running cost and strong universality, and usually require high-quality data as support. Due to the uneven distribution of various small classes of lithology of sandstone and conglomerate, direct application of machine learning models may lead to overfitting of the model or poor performance.

[0004] The application proposes using oversampling and sample screening methods to increase the number and diversity of reservoir samples to overcome the problems caused by sample imbalance and scarcity. The oversampling technology can generate simulated data to increase the number of reservoir samples, so that the machine learning model can better learn the characteristics of the reservoir. At the same time, the ensemble learning method can integrate the analysis results of different models on the data from multiple perspectives to improve the overall accuracy and achieve accurate identification of sandstone and conglomerate reservoirs. SUMMARY

[0005] In order to solve the above problems in the prior art, i.e. due to the uneven distribution of various small classes of lithology of sandstone and conglomerate, direct application of machine learning models may lead to overfitting of the model or poor performance, the application provides a sandstone and conglomerate oil and gas reservoir intelligent drawing and identification oil navigation system based on oversampling and ensemble learning, which comprises:

[0006] A logging feature selection module configured to select, from the standardized logging data, the feature with the largest feature-class information amount and feature-feature information amount as sensitive logging data in each round;

[0007] A random sampling module configured to determine the number P of heterogeneous ensemble learning models according to the component to be identified based on the sensitive logging data, and to obtain a set of random sampling data with a set number Q for each heterogeneous ensemble learning model as a sampling data subset;

[0008] An oversampling module configured to perform oversampling according to the cosine similarity based on the sampling data subset, and to obtain P sets of oversampled data subsets;

[0009] The lithology identification module obtains a lithology identification result by the trained heterogeneous ensemble learning model through weighted voting based on the P oversampled data subsets.

[0010] Further, before standardization, a step of removing outliers in the original logging data by a Z-score method is further included.

[0011] Further, the method for obtaining the logging data to be standardized includes:

[0012] selecting logging data of a standard well calculating a mean value estimate of the standard well and a standard deviation estimate of the standard well ;

[0013] logging data of a non-standard well ; calculating a mean value estimate of the non-standard well and a standard deviation estimate of the non-standard well ;

[0014] The standardized logging data is obtained by a target standardization transformation method for logarithmic data. The target standardization transformation method for logarithmic data ensures the consistency of the data dimension, value range and probability distribution between wells by the standard well, retains the readability of the data and the geological significance behind the numerical value, avoids the error of reservoir identification result caused by accidental error, and avoids the data distortion and precision error problem caused by the conventional Z-Score standardization which puts all data into the [0, 1] interval.

[0015] Further, the feature-class information amount and the feature-feature information amount are measured by normalized mutual information.

[0016] Mutual information evaluates the dependency relationship between variables, which reflects the dynamic influence of one variable on another variable; similarity often measures the correlation relationship between variables, which is static and different in meaning. Dependency relationship is the main basis for feature selection or rejection, such as when the dependency relationship between a certain feature and the label is weak, it needs to be rejected, while the correlation relationship is not suitable here.

[0017] Further, the specific method for selecting sensitive logging data is:

[0018] selecting a logging feature with the maximum normalized mutual information with the lithology category as the selected feature set S;

[0019] calculating the mutual information between the unselected feature set and the initial selected feature set S and the mutual information between the unselected feature set and the lithology category, which are collectively used as comprehensive mutual information;

[0020] Select the unselected feature with the maximum integrated mutual information from the unselected feature set and add it to the selected feature set S;

[0021] Iteratively calculate the integrated mutual information and add the unselected feature with the maximum integrated mutual information to the selected feature set S until the difference between the unselected feature with the maximum integrated mutual information in the current iteration round and the unselected feature with the maximum integrated mutual information in the last iteration round is less than a set threshold, and obtain the sensitive well logging data. Selecting the well logging curve sensitive to the lithology information can reduce the number of features, reduce the learning difficulty and calculation cost of the model. That is, which features are more dependent on the lithology category information, which is the embodiment of the feature dependency of mutual information calculation. Compared with the features obtained by dimensionality reduction, the dimensionality reduction mostly uses fusion operation and other methods, and the whole dimensionality reduction process is a black box model, and the interpretability of the dimensionality reduction result is poor, and it cannot be determined what the relationship between the original features and the features after dimensionality reduction is. Selecting the sensitive well logging data is feature selection, which maximizes the preservation of original feature information, and the selected result has strong interpretability and obvious geological significance.

[0022] Further, the number of samples to be collected is: the total number of samples in the current sampling data subset that are not of the target rock class minus the total number of samples in the current data subset that are of the target rock class.

[0023] Further, the oversampling according to the cosine similarity is specifically:

[0024] Randomly select one current sample from the sand and gravel rock samples, and calculate the cosine similarity between the current sample and all samples of the same category;

[0025] Select the K samples of the same category with the highest cosine similarity as the candidate sample set;

[0026] Randomly select a to-be-synthesized sample from the candidate sample set, and synthesize a simulation sample by quadratic spline interpolation of the to-be-synthesized sample and the current sample;

[0027] Randomly select another current sample from the sand and gravel rock samples to obtain a synthesized simulation sample, and repeat the iteration until the number of samples to be collected is reached, to obtain an oversampled data subset.

[0028] Further, the trained heterogeneous ensemble learning model for weighted voting includes a plurality of heterogeneous ensemble learning models, and the heterogeneous ensemble learning models are different types of classifiers.

[0029] Further, the different types of classifiers include at least two of logistic regression, SVM, C4.5, CART and GBDT.

[0030] Further, the training method of the trained heterogeneous ensemble learning model for weighted voting is:

[0031] normalizing the recall rate of each heterogeneous ensemble learning model as a classification weight;

[0032] obtaining a final classification result according to the classification result and the classification weight of each heterogeneous ensemble learning model;

[0033] calculating a loss function according to the final classification result, repeating iteration until the loss function is lower than a set threshold, and obtaining a trained heterogeneous ensemble learning model of weighted voting.

[0034] Advantages of the present application:

[0035] The present application proposes a method of well logging data multi-subset oversampling and multi-model heterogeneous ensemble classification aiming at the problems of multiple sources, grain size differentiation and uneven distribution of small classes of each lithology of sandy conglomerate, increases the number and diversity of sandy conglomerate logging samples, and greatly improves the accuracy and operation efficiency of sandy conglomerate reservoir identification. BRIEF DESCRIPTION OF DRAWINGS

[0036] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the accompanying drawings:

[0037] Figure 1 is a structural block diagram of a sandy conglomerate oil and gas reservoir intelligent drawing and oil identification navigation system based on oversampling and ensemble learning in an embodiment of the present application;

[0038] Figure 2 is a schematic diagram of mutual information calculation in an embodiment of the present application;

[0039] Figure 3 is a schematic diagram of well logging data subset oversampling in an embodiment of the present application;

[0040] Figure 4 is a schematic diagram of a heterogeneous ensemble learning process taking heterogeneous ensemble learning of three classifiers of C4.5, SVM and CART as an example in an embodiment of the present application;

[0041] Figure 5 is a sandy conglomerate reservoir identification result effect diagram of a sandy conglomerate oil and gas reservoir intelligent drawing and oil identification navigation system based on oversampling and ensemble learning in an embodiment of the present application. DETAILED DESCRIPTION

[0042] The present application will be further described in detail below with reference to the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related application, and not to limit the application. In addition, it should be noted that, in order to facilitate description, only the parts related to the application are shown in the drawings.

[0043] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0044] In order to more clearly illustrate the sand and gravel oil and gas reservoir intelligent drawing and oil identification navigation system based on oversampling and ensemble learning of the present application, the following will be described in combination with Figure 1 The functions of each module in the embodiments of the present application will be described in detail.

[0045] The sand and gravel oil and gas reservoir intelligent drawing and oil identification navigation system based on oversampling and ensemble learning of the first embodiment of the present application includes a well logging feature selection module, a random sampling module, an oversampling module and a lithology identification module, which are described in detail as follows:

[0046] In this embodiment, before standardization, it also includes removing outliers in the original well logging data by Z-score method;

[0047] All well logging samples are taken as original well logging data , the sample mean and the sample standard deviation are calculated:

[0048] ;

[0049] ;

[0050] , represents the sample capacity of the original data set, represents the i-th sample in the original well logging data;

[0051] Based on the sample mean and the sample standard deviation , the Z-score score of the sample is calculated:

[0052] ;

[0053] x represents a sample point in the original well logging data;

[0054] The set of samples after setting the Z-score score in the interval is taken as the well logging data set after removing outliers and as the well logging data to be standardized ; wherein .

[0055] In this embodiment, the method for obtaining the well logging data to be standardized includes:

[0056] Selecting standard well logging data Calculate the standard well mean estimate value And the standard well standard deviation estimate value ;

[0057] ;

[0058] ;

[0059] Indicates the sample capacity of the selected standard well logging data, Indicates the i-th sample in the standard well logging data.

[0060] Non-standard well logging data ; Calculate the non-standard well mean estimate value And the non-standard well standard deviation estimate value ;

[0061] Divided standard well logging data And non-standard well logging data Satisfy ;

[0062] Obtain the normalized logging data by the target standardization transformation method of the logarithmic data;

[0063] The target standardization transformation method of the logarithmic data is specifically:

[0064] ;

[0065] Indicates the normalized logging curve, The logging curve value before normalization;

[0066] Because the data between different wells may not match the dimension and data range due to measurement equipment or human factors, in order to avoid such factors causing too large differences in the geological laws reflected by each well, normalization is needed.

[0067] The purpose of using the target standardization transformation method of the logarithmic data is to ensure the consistency of the data dimension, value range and probability distribution between wells by the standard well, avoid accidental errors leading to incorrect reservoir identification results, and the target standardization transformation method of the logarithmic data will not appear the data distortion and precision error problem caused by the commonly used Z-score standardization method which scales all data to the [0, 1] interval.

[0068] The logging feature selection module is configured to select the feature with the largest feature-class information amount and feature-feature information amount as the sensitive logging data in each round of the normalized logging data;

[0069] Before performing reservoir lithology identification tasks, feature selection techniques can be used to select logging curves that are sensitive to lithology information, based on the principle of maximizing the information between features and categories and minimizing the redundancy between features. This can reduce the number of features, lower the learning difficulty of the model, and reduce computational costs.

[0070] In this embodiment, the feature-category information and feature-feature information are measured by normalized mutual information.

[0071] like Figure 2 As shown, the normalized mutual information calculation method is as follows:

[0072] Total number of statistical logging feature dimensions and the set C of all lithological categories;

[0073] For each logging curve and each lithology category Calculate the joint probability distribution ;

[0074] According to the joint probability distribution Calculate well logging curves Marginal probability distribution and lithological categories Marginal probability distribution ;

[0075] Calculate all logging curves With lithological category Mutual information between :

[0076] ;

[0077] Mutual information Perform normalization to obtain normalized mutual information. :

[0078] ;

[0079] In this embodiment, the specific method for selecting sensitive logging data is as follows:

[0080] Select the logging feature with the highest mutual information normalized to the lithology category. As the selected feature set S;

[0081] Unselected feature set for:

[0082] ;

[0083] Calculate the set of unselected features The mutual information of the initial selected feature set S and the mutual information of the unselected feature set and the lithology category are collectively used as comprehensive mutual information.

[0084] Comprehensive mutual information is:

[0085] ;

[0086] represents an alternative feature, , represents an alternative feature and the mutual information of the lithology category, represents an alternative feature and a feature in the selected feature set S; ;

[0087] The comprehensive mutual information represents the mutual information of the well logging curve and the label , minus the mutual information between the well logging curve and other well logging curves. It can be seen that when the mutual information between the well logging curve and the label is larger and the mutual information between the well logging curve and other well logging curves is smaller, the value of the comprehensive mutual information is larger. That is, The larger the mutual information between the well logging curve and the label indicates that the feature has a higher dependence on the label and is more suitable for predicting the label; on the other hand, The smaller the mutual information between the well logging curve and each feature in the selected feature set S indicates that the feature has a low dependence on other selected features, and there is little multiple collinearity or information redundancy between them, which is less likely to have a negative impact on the prediction result of the label. In summary, the larger the value of the comprehensive mutual information indicates that the alternative feature is more suitable to be added to the selected feature set.

[0088] The unselected feature with the largest comprehensive mutual information is selected from the unselected feature set and added to the selected feature set S;

[0089] The comprehensive mutual information is iteratively calculated, and the unselected feature with the largest comprehensive mutual information is added to the selected feature set S until the difference between the unselected feature with the largest comprehensive mutual information in the current iteration round and the unselected feature with the largest comprehensive mutual information in the last iteration round is less than a set threshold value, and the sensitive well logging data is obtained.

[0090] The difference between the unselected feature with the largest comprehensive mutual information in the current iteration round and the unselected feature with the largest comprehensive mutual information in the last iteration round is:

[0091] ;

[0092] wherein, denotes the iteration round, denotes the unselected feature with the maximum integrated mutual information mRmR in the th iteration, denotes the selected feature set in the th iteration; fmax,cnt-1 denotes the unselected feature with the maximum integrated mutual information mRmR in the previous iteration, denotes the selected feature set in the previous iteration;

[0093] When Δ is less than 0.01, it indicates that the correlation between the newly added feature and the lithology category is weak, the iteration is terminated, and the current selected feature set is taken as the feature selection result, denoted as sensitive well logging data. .

[0094] A random sampling module determines the number P of heterogeneous ensemble learning models according to the to-be-identified component based on the sensitive well logging data, and obtains a set of random sampling data with replacement of a set number Q of random sampling data for each heterogeneous ensemble learning model.

[0095] By extracting samples from the original data set with replacement, the diversity of the training set can be increased, thereby helping the model to learn different information in the data, improving the generalization ability of the model, and reducing the risk of overfitting.

[0096] In this embodiment, the set number Q is 80% of the total data set capacity.

[0097] An oversampling module performs oversampling according to the cosine similarity based on the set of sampling data, and obtains P sets of oversampled data. After sampling and oversampling the sensitive well logging data, the data distribution is as shown in Figure 3 .

[0098] In this embodiment, the oversampling requires a number of samples to be collected, which is the total number of samples that are not of the target rock class in the current set of sampling data minus the total number of samples of the target rock class in the current set of sampling data.

[0099] Taking the sandy conglomerate as the target rock class as an example, the total number of samples of the sandy conglomerate in the current set of sampling data is denoted as G, and the total number of samples that are not of the sandy conglomerate in the current set of sampling data is denoted as Z. Then the number of samples to be collected is N=Z-G.

[0100] In this embodiment, the oversampling according to the cosine similarity is specifically:

[0101] A current sample is randomly selected from the sandstone sample, and a cosine similarity between the current sample and all samples in the same category is calculated;

[0102] For the current sample h, the cosine similarity with all samples in the same category is:

[0103] ;

[0104] The cosine similarity between the sample h of the current category and the sample p from the same category is represented, and vech and vecp are the vectorized representations of h and p, respectively;

[0105] The inner product of the vectors vech and vecp is represented:

[0106] ;

[0107] n represents the total number of feature dimensions of each logging sample vector, and k represents the serial number of the current feature dimension, k ranges from 1 to an integer in n;

[0108] The K samples in the same category with the highest cosine similarity are selected as the candidate sample set;

[0109] A to-be-synthesized sample is randomly selected from the candidate sample set, and the to-be-synthesized sample and the current sample are synthesized into a simulation sample by a quadratic spline interpolation method;

[0110] The selected to-be-synthesized sample is q, and the synthesized simulation sample is:

[0111] ;

[0112] The synthesized simulation sample is represented as is a random number, and satisfies 0 <1;

[0113] Another current sample is randomly selected from the sandstone sample to obtain a synthesized simulation sample, and the iteration is repeated until the number of samples N to be collected is reached, thereby obtaining a subset of oversampled data.

[0114] The lithology identification module obtains a lithology identification result by using a trained heterogeneous ensemble learning model based on the P subsets of oversampled data.

[0115] For each subset of data obtained after sampling, the sandstone samples in the subset are selected and oversampled, so as to increase the number of sandstone samples and enable the model to learn data information that can improve the accuracy of sandstone identification. In the oversampling process, a nonlinear interpolation method is used to avoid over-smoothing problems and to conform to the nonlinear relationship between logging features and lithology.

[0116] In the embodiment, the trained weighted voting heterogeneous ensemble learning model includes a plurality of heterogeneous ensemble learning models, and the heterogeneous ensemble learning models are different types of classifiers.

[0117] In the embodiment, the different types of classifiers include at least two of logistic regression, SVM, C4.5, CART and GBDT.

[0118] Heterogeneous ensemble learning refers to combining different types of learning algorithms to expect that the performance of the integrated system is better than the average performance of a single classifier. Since each data set is processed by random sampling and oversampling, the data distribution may be different, and considering the strong concealment of conglomerate distribution, the mixed logging data and the differentiation of each lithology subclass, the advantages of different architectures of classifiers need to be considered to accurately identify Li Yan.

[0119] The heterogeneous ensemble learning for sandstone identification ensures that each model and each data subset is used once in training, each single model is sorted according to the training performance score after training is completed, and the higher ranking has a higher weight in the multi-model voting, which has a greater impact on the final result of the integrated learning model.

[0120] In the embodiment, the training method of the trained weighted voting heterogeneous ensemble learning model is as follows:

[0121] Select one data subset that is not selected to participate in training as the current training data subset, and train the heterogeneous ensemble learning model one by one;

[0122] After each heterogeneous ensemble learning model is trained separately, the sandstone training recall rate is calculated :

[0123] ;

[0124] Wherein represents the total number of sandstone samples in the current training data subset, represents the number of sandstone samples correctly identified by the current classifier;

[0125] In the embodiment, the highest score is selected from the scores of the five classifiers, and the highest score is recorded as , and the lowest score is selected, and the lowest score is recorded as ;

[0126] The normalized recall rate of each heterogeneous ensemble learning model is calculated as the classification weight ; ​

[0127] Calculate the maximum-minimum value normalization score for all classifiers

[0128]

[0129] According to the classification result and the classification weight of each heterogeneous ensemble learning model, a final classification result is obtained.

[0130]

[0131] is a voting score of the ensemble learning model identifying the input data as a class label, is a voting weight of the n-th classifier, is a class predicted by the n-th classifier, denotes the total number of classifiers, is an indicator function, if the indicator function value is 1, otherwise 0;

[0132] The class with the highest voting score is selected as the final classification result of the heterogeneous ensemble learning model:

[0133]

[0134] According to the final classification result, a loss function is calculated, and the iteration is repeated until the loss function is lower than a set threshold, to obtain a trained heterogeneous ensemble learning model with weighted voting.

[0135] Taking the heterogeneous ensemble learning of three classifiers C4.5, SVM and CART as an example, the flow of ensemble learning is shown in Figure 4 The final result of the sand and gravel reservoir identification result of the sand and gravel reservoir intelligent drawing oil identification navigation system based on oversampling and ensemble learning is shown in Figure 5

[0136] It should be noted that the sand and gravel reservoir intelligent drawing oil identification navigation system based on oversampling and ensemble learning provided in the above embodiments is only exemplified by the division of the above functional modules, and in actual application, the above functions can be completed by different functional modules according to needs, i.e. the modules or steps in the embodiments of the present application are further divided or combined, for example, the modules of the above embodiments can be combined into one module, or can be further split into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present application are only for distinguishing the modules and steps, and should not be considered as an improper limitation of the present application.

[0137] ​​​​​Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the storage device and the processing device described above and the related descriptions can refer to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0138] Those skilled in the art will realize that the modules, method steps of the examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or combinations of both. The software modules, method steps, corresponding to the various blocks of the examples can be stored in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. In order to clearly illustrate the interchangeability of electronic hardware and software, the components and steps of the examples have been generally described in terms of their functionality in the above description. Whether such functionality is implemented in electronic hardware or software depends on the particular application and design constraints imposed on the overall architecture. Those skilled in the art can implement the described functionality in varying ways for each particular application, but such implementation should not be considered to depart from the scope of the present application.

[0139] The terms "first", "second", and the like, are used to distinguish similar objects, not to describe or indicate a particular order or sequence.

[0140] The term "comprising" or any other similar term is intended to encompass the inclusion of non-exclusive elements, so that a process, method, article, or apparatus / device including a series of elements includes not only those elements, but also other elements not explicitly listed, or inherent to such a process, method, article, or apparatus / device.

[0141] So far, the technical solutions of the present application have been described in combination with the preferred embodiments shown in the drawings, but those skilled in the art can easily understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical solutions after such changes or replacements will all fall within the protection scope of the present application.

Claims

1. A sand-shale reservoir intelligent drawing and oil identification navigation system based on oversampling and ensemble learning, characterized in that, The system comprises: a well logging feature selection module configured to select, round by round, a feature with the largest feature-class information amount and feature-feature information amount from the standardized well logging data as sensitive well logging data; a random sampling module configured to determine, based on the sensitive well logging data, a number P of heterogeneous ensemble learning models according to a component to be identified, and to obtain, for each heterogeneous ensemble learning model, a set of Q randomly sampled data as a sampled data subset; an oversampling module configured to perform oversampling on the sampled data subset according to a cosine similarity, and to obtain P oversampled data subsets; a lithology identification module configured to obtain a lithology identification result by using a trained heterogeneous ensemble learning model with weighted voting based on the P oversampled data subsets.

2. The sand-shale reservoir intelligent oil identification and navigation system based on oversampling and ensemble learning of claim 1, wherein, Before the standardization, a step of removing outliers in the original well logging data by a Z-score method is further included.

3. The sand-shale reservoir intelligent oil identification and navigation system based on oversampling and ensemble learning of claim 1, wherein, The method for obtaining the well logging data to be standardized comprises: selecting well logging data of a standard well to calculate a standard well mean value estimate and a standard well standard deviation estimate; selecting well logging data of a non-standard well to calculate a non-standard well mean value estimate and a non-standard well standard deviation estimate; and obtaining the standardized well logging data by a target standardization transformation method for the logarithmic data.

4. The sand-shale reservoir intelligent mapping and oil identification system based on oversampling and ensemble learning of claim 1, wherein, The feature-class information amount and the feature-feature information amount are measured by normalized mutual information.

5. The sand-shale reservoir intelligent mapping and oil identification system based on oversampling and ensemble learning of claim 1, wherein, The method for selecting the sensitive well logging data comprises: selecting a well logging feature with the largest normalized mutual information with the lithology category as a selected feature set S; calculating mutual information of a non-selected feature set with the initial selected feature set S and mutual information of the non-selected feature set with the lithology category, and taking the mutual information as comprehensive mutual information; selecting a non-selected feature with the largest comprehensive mutual information from the non-selected feature set and adding the non-selected feature to the selected feature set S; iteratively calculating the comprehensive mutual information and adding the non-selected feature with the largest comprehensive mutual information to the selected feature set S until a difference between the non-selected feature with the largest comprehensive mutual information in a current iteration round and the non-selected feature with the largest comprehensive mutual information in a previous iteration round is less than a set threshold, and obtaining the sensitive well logging data.

6. The oversampling and ensemble learning based sand-shale reservoir intelligent mapping and oil identification navigation system of claim 1, wherein, The number of samples to be collected for the oversampling is a total number of samples that are not of a target rock class in a current sampled data subset minus a total number of samples of the target rock class in the current data subset.

7. The sand-shale reservoir intelligent mapping and oil identification system based on oversampling and ensemble learning of claim 6, wherein, The oversampling according to the cosine similarity specifically comprises: randomly selecting one current sample from the target rock class samples, calculating cosine similarities between the current sample and all samples in the target rock class samples; taking K samples with the highest cosine similarities as a candidate sample set; randomly selecting a to-be-combined sample from the candidate sample set, combining the to-be-combined sample and the current sample into a simulation sample by a quadratic spline interpolation method; randomly selecting another current sample from the sandstone sample to obtain a combined simulation sample, and repeating the iteration until the number of samples to be collected is reached, and obtaining one oversampled data subset.

8. The sand-shale reservoir intelligent mapping and oil identification system based on oversampling and ensemble learning of claim 1, wherein, The trained heterogeneous ensemble learning model with weighted voting comprises a plurality of heterogeneous ensemble learning models, and the heterogeneous ensemble learning models are different types of classifiers.

9. The sand-shale reservoir intelligent oil identification and navigation system based on oversampling and ensemble learning of claim 8, wherein, The different types of classifiers comprise at least two of a logistic regression, a SVM, a C4.5, a CART and a GBDT.

10. The sand-shale reservoir intelligent oil identification and navigation system based on oversampling and ensemble learning of claim 1, wherein, The training method of the trained weighted voting heterogeneous ensemble learning model is: Calculate the normalized recall rate of each heterogeneous ensemble learning model as a classification weight; According to the classification result and the classification weight of each heterogeneous ensemble learning model, the final classification result is obtained; According to the final classification result, the loss function is calculated, and the iteration is repeated until the loss function is lower than the set threshold, and the trained weighted voting heterogeneous ensemble learning model is obtained.

Images