Lithology identification method and device for unbalanced logging data

Abstract

The invention discloses a lithology identification method and device for unbalanced logging data. The method comprises the following steps: acquiring logging data of an oil well and a lithology category to be identified; inputting to-be-identified logging data into the trained lithology identification model to obtain a lithology identification result; wherein the trained lithology identification model is obtained by preprocessing training set data through an ADASYN algorithm and training based on a cascade ELM model of a multi-granularity scanning module.

Description

Technical Field

[0001] The present invention relates to the field of petroleum engineering technology, and in particular to a lithology identification method and device for unbalanced logging data. Background Art

[0002] As the development of industry leads to global energy shortage, oil exploration and development is essential to meet the demand. Lithology reflects the energy storage status of adjacent reservoirs, so lithology identification is the key to reservoir quality evaluation in oilfield exploration.

[0003] Traditional lithology identification methods include rock logging and drilling coring, but due to external interference and heavy workload, the measurement depth error is large and the cost is high. Since logging can objectively, continuously and indirectly describe the properties of rocks and reflect the lithology information at different depths, logging interpretation has been widely used in lithology identification as an alternative to traditional methods. Early logging interpretation research mainly focused on establishing theoretical models such as cross plots and statistical analysis. However, due to the large differences in geological characteristics in different regions, the theoretical model relies on the limitations of subjective experience, resulting in poor generalization effect. Machine learning technology can automatically train models based on given inputs and achieve good prediction results for different types of logging data. Summary of the invention

[0004] The inventors of this application have found that the existing lithology identification algorithms based on machine learning have the following deficiencies:

[0005] (1) Current machine learning methods still have some problems related to well logging data. For example, due to rock heterogeneity, the number of samples of different lithology types varies greatly, which causes the model to tend to identify majority class samples;

[0006] (2) There is complementary logging information between the logging curves collected by different sensors, but the lithology identification algorithm based on machine learning ignores the interaction of logging curves;

[0007] In view of the problems existing in machine learning methods, it is urgent to propose a lithology identification method for unbalanced logging data to solve the above problems.

[0008] The present application provides a lithology identification method and device for unbalanced logging data. The method preprocesses the training set data through the ADASYN algorithm, and obtains a trained lithology identification model based on the cascade ELM model of the multi-granularity scanning module. The trained lithology identification model is used for lithology identification, which effectively improves the accuracy of lithology identification.

[0009] In a first aspect, the present application provides a lithology identification method for unbalanced logging data, the method comprising:

[0010] Obtain well logging data of oil wells and the lithology categories to be identified;

[0011] Input the well logging data to be identified into the trained lithology identification model to obtain the lithology identification result;

[0012] The trained lithology identification model is obtained by preprocessing the training set data through the ADASYN algorithm and training based on the cascade ELM model of the multi-granularity scanning module.

[0013] In an exemplary embodiment, the training process of the lithology identification model includes:

[0014] Divide the logging sample data and their corresponding lithology categories into a training set and a test set according to a predetermined ratio;

[0015] Preprocessing the well logging sample data in the training set using the ADASYN algorithm;

[0016] The pre-processed logging sample data is input into the cascade ELM model based on the multi-granularity scanning module to obtain the characterization vector of lithology;

[0017] The characterization vector of the lithology is input into the parameter optimization module, and the PSO algorithm is used to determine the optimal parameters of the lithology identification model;

[0018] A trained lithology identification model is obtained according to the determined optimal parameters.

[0019] In an exemplary embodiment, the preprocessing of the well logging sample data in the training set by using the ADASYN algorithm includes:

[0020] Classifying the well logging sample data in the training set according to lithology categories, and determining a minority class and a majority class;

[0021] Calculate each minority class separately to determine the number of samples to be synthesized for that minority class;

[0022] According to the sample synthesis method of the minority class and the number of samples to be synthesized, new sample data corresponding to each minority class is obtained.

[0023] In an exemplary embodiment, obtaining new sample data corresponding to each minority class according to the sample synthesis method of the minority class and the number of samples to be synthesized includes:

[0024] Extract K nearest neighbors of each minority class sample according to the Euclidean distance;

[0025] Determine the proportion of each majority class in the K nearest neighbors of each minority class sample;

[0026] Normalize the proportion of all majority classes to obtain the sample weight value;

[0027] Calculate the number of new samples to be synthesized for each minority class sample according to the sample weight value;

[0028] From the K nearest neighbors of the minority class sample, a minority class sample is randomly selected to synthesize new sample data of the minority class sample.

[0029] In an exemplary embodiment, the pre-processed logging sample data is input into a cascade ELM model based on a multi-granularity scanning module to obtain a lithology characterization vector, including:

[0030] Inputting the preprocessed logging sample data into the multi-granularity scanning module to generate a conversion feature vector;

[0031] The converted feature vector is input into the cascade ELM module to obtain a characterization vector of lithology.

[0032] In an exemplary embodiment, the step of inputting the preprocessed logging sample data into the multi-granularity scanning module to generate a conversion feature vector includes:

[0033] Using a sliding window for the preprocessed logging sample data, a plurality of feature vectors of different sizes are generated;

[0034] Input each feature vector into the corresponding first learner and second learner respectively to generate a first type probability distribution vector and a second type probability distribution vector;

[0035] The first probability distribution vector and the second probability distribution vector are concatenated into a transformed feature vector tf i .

[0036] In an exemplary embodiment, the first learner and the second learner are ELM models;

[0037] Wherein, the ELM model is:

[0038] Hβ=T;

[0039]

[0040] β=[β1,β2,…,β L ] T ;

[0041] T=[t1,t2,…,t n ] T ;

[0042] In the above formula, H is the hidden layer output matrix, β is the weight matrix, T is the target matrix, and x nis the nth logging data, L is the number of hidden layer nodes, h(x n ) represents the hidden layer pair x n The output matrix, h L (x n ) represents the hidden layer node L for x n The output value of g(·) is the hidden layer activation function, w L is the weight of the hidden layer node L, b L is the bias of the hidden layer node L, β L represents the weight between the hidden layer node L and the output layer, t n Represents the ELM model for x n Output.

[0043] In an exemplary embodiment, the step of inputting the converted feature vector into a cascade ELM module to obtain a lithology characterization vector includes:

[0044] The converted feature vector is input into the cascade ELM module to obtain the lithology characterization vector, including:

[0045] The cascade ELM module is composed of N layers of cascade ELM models, each layer consisting of multiple parallel ELM models;

[0046] Inputting the converted feature vector into the first layer of the cascaded ELM model to obtain a three-dimensional vector corresponding to the first layer;

[0047] splicing the converted feature vector and the three-dimensional vector corresponding to the first layer and inputting the concatenated vector into the second-layer cascaded ELM model;

[0048] Until the conversion feature vector and the three-dimensional vector corresponding to the N-1th layer are concatenated and input into the ELM model of the Nth layer cascade, the three-dimensional vector corresponding to the Nth layer is obtained;

[0049] The three-dimensional vectors corresponding to the Nth layer are averaged to obtain a characterization vector of the lithology.

[0050] In an exemplary embodiment, the optimal parameters of the lithology identification model include: the number of base models, the number of hidden layer neurons and the maximum number of cascade layers.

[0051] In a second aspect, an embodiment of the present invention provides a lithology identification device for unbalanced well logging data, characterized in that the device comprises: a memory and a processor; the memory is used to store a program for a lithology identification method for unbalanced well logging data, and the processor is used to read and execute the program for the lithology identification method for unbalanced well logging data, and execute the lithology identification method for unbalanced well logging data described in any one of the above embodiments.

[0052] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a data processing program stored thereon, and the data processing program is executed by a processor according to any one of the lithology identification methods for unbalanced logging data described in the above embodiments.

[0053] Compared with the related art, the present application provides a lithology identification method and device for unbalanced well logging data, the method comprising: obtaining well logging data of an oil well and the lithology category to be identified; inputting the well logging data to be identified into a trained lithology identification model to obtain a lithology identification result; wherein the trained lithology identification model is obtained by preprocessing the training set data through the ADASYN algorithm and training based on the cascade ELM model of a multi-granularity scanning module. The present application preprocesses the training set data through the ADASYN algorithm and trains based on the cascade ELM model of a multi-granularity scanning module to obtain a trained lithology identification model; the trained lithology identification model is used for lithology identification, which can effectively improve the accuracy of lithology identification for low-quality unbalanced well logging data.

[0054] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or be understood by implementing the present application. Other advantages of the present application can be realized and obtained by the schemes described in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The accompanying drawings are used to provide an understanding of the technical solution of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the technical solution of the present application and do not constitute a limitation on the technical solution of the present application.

[0056] Figure 1 This is a flow chart of a lithology identification method for unbalanced logging data in an embodiment of the present application;

[0057] Figure 2 This is a schematic diagram of a lithology identification device for unbalanced logging data in an embodiment of the present application;

[0058] Figure 3 A flow chart of a lithology identification method for unbalanced logging data in some exemplary embodiments;

[0059] Figure 4 is a flowchart of an ADASYN algorithm in some exemplary embodiments;

[0060] Figure 5 is a schematic diagram of an ELM model in some exemplary embodiments;

[0061] Figure 6 A schematic diagram of constructing a multi-granularity scanning model in some exemplary embodiments;

[0062] Figure 7 is a schematic diagram of constructing a cascade ELM model in some exemplary embodiments;

[0063] Figure 8 is a schematic diagram of five-fold cross validation in some exemplary embodiments;

[0064] Fig. 9 Graphs showing lithology identification results of model ablation experiments in some exemplary embodiments. DETAILED DESCRIPTION

[0065] The present application describes multiple embodiments, but the description is exemplary rather than restrictive, and it is obvious to those skilled in the art that there may be more embodiments and implementations within the scope of the embodiments described in the present application. Although many possible feature combinations are shown in the drawings and discussed in the specific embodiments, many other combinations of the disclosed features are also possible. Unless specifically limited, any feature or element of any embodiment may be used in combination with any other feature or element in any other embodiment, or may replace any other feature or element in any other embodiment.

[0066] The present application includes and contemplates combinations of features and elements known to those of ordinary skill in the art. The embodiments, features and elements disclosed in the present application may also be combined with any conventional features or elements to form a unique invention scheme defined by the claims. Any features or elements of any embodiment may also be combined with features or elements from other invention schemes to form another unique invention scheme defined by the claims. Therefore, it should be understood that any feature shown and / or discussed in the present application may be implemented individually or in any appropriate combination. Therefore, except for the limitations made according to the attached claims and their equivalents, the embodiments are not subject to other restrictions. In addition, various modifications and changes may be made within the scope of protection of the attached claims.

[0067] In addition, when describing representative embodiments, the specification may have presented the method and / or process as a specific sequence of steps. However, to the extent that the method or process does not rely on the specific order of the steps described herein, the method or process should not be limited to the steps of the specific order described. As will be understood by those of ordinary skill in the art, other sequences of steps are also possible. Therefore, the specific sequence of the steps set forth in the specification should not be interpreted as a limitation to the claims. In addition, the claims for the method and / or process should not be limited to the steps of performing them in the order written, and those skilled in the art can easily understand that these sequences can be changed and still remain within the spirit and scope of the embodiments of the present application.

[0068] The embodiment of the present invention provides a lithology identification method for unbalanced logging data, such as Figure 1 As shown, the method includes steps S100-S110, which are specifically as follows:

[0069] S100: Obtaining well logging data of the oil well and the lithology category to be identified;

[0070] S110 inputs the lithology category to be identified and the logging data into the trained lithology identification model to obtain a lithology identification result.

[0071] In this embodiment, the trained lithology recognition model is obtained by preprocessing the logging data samples using the ADASYN algorithm and training based on the cascade ELM model of the multi-granularity scanning module.

[0072] In an exemplary embodiment, the well logging data is a characteristic curve reflecting lithology;

[0073] The logging data includes one or more of the following: natural gamma logging curve, sonic time difference logging curve, compensated neutron logging curve, density logging curve, deep lateral resistivity logging curve and shallow lateral resistivity logging curve.

[0074] In an exemplary embodiment, the training process of the lithology identification model includes:

[0075] Step 1. Divide the logging data and its corresponding lithology categories into a training set and a test set according to a predetermined ratio;

[0076] Step 2. Preprocess the well logging sample data in the training set using the ADASYN algorithm;

[0077] Step 3. Input the preprocessed logging sample data into the cascade ELM model based on the multi-granularity scanning module to obtain the characterization vector of the lithology;

[0078] Step 4. Input the lithology characterization vector into the parameter optimization module, and use the PSO algorithm to determine the optimal parameters of the lithology identification model;

[0079] Step 5. Obtain a trained lithology identification model based on the determined optimal parameters.

[0080] In an exemplary embodiment, the preprocessing of the well logging sample data in the training set by using the ADASYN algorithm includes:

[0081] Classifying the well logging sample data in the training set data according to lithology categories, and determining a minority class and a majority class;

[0082] Calculate each minority class separately to determine the number of samples to be synthesized for that minority class;

[0083] According to the sample synthesis method of the minority class and the number of samples to be synthesized, new sample data corresponding to each minority class is obtained.

[0084] In an exemplary embodiment, obtaining new sample data corresponding to each minority class according to the sample synthesis method of the minority class and the number of samples to be synthesized includes:

[0085] Extract K nearest neighbors of each minority class sample according to the Euclidean distance;

[0086] Determine the proportion of the majority class in the K nearest neighbors of each minority class sample;

[0087] Normalize the proportion of all majority classes to obtain the sample weight value;

[0088] Calculate the number of new samples to be synthesized for each minority class sample according to the sample weight value;

[0089] From the K nearest neighbors of the minority class sample, a minority class sample is randomly selected to synthesize a new sample of the minority class sample.

[0090] In an exemplary embodiment, the step of inputting the pre-processed logging sample data into a cascade ELM model based on a multi-granularity scanning module to obtain a lithology characterization vector includes:

[0091] The pre-processed logging data is input into the multi-granularity scanning module to generate a transformation feature vector tf i ;

[0092] The transformed feature vector tf i Input the cascade ELM module to obtain the lithology characterization vector.

[0093] In an exemplary embodiment, the pre-processed logging data is input into the multi-granularity scanning module to generate a conversion feature vector tf i ,include:

[0094] A sliding window is used to generate multiple feature vectors fv of different sizes for the preprocessed training data. i ;

[0095] Each feature vector fv i Input into the corresponding first learner and second learner respectively to generate the first type probability distribution vector lf1 i and the second probability distribution vector lf2 i ;

[0096] The first probability distribution vector and the second probability distribution vector are concatenated into a transformed feature vector tf i .

[0097] In an exemplary embodiment, the first learner and the second learner are ELM models;

[0098] Wherein, the ELM model is:

[0099] Hβ=T;

[0100]

[0101] β=[β1,β2,…,β L ] T ;

[0102] T=[t1,t2,…,t n ] T ;

[0103] In the above formula, H is the hidden layer output matrix, β is the weight matrix, T is the target matrix, and x n is the nth logging data, L is the number of hidden layer nodes, h(x n ) represents the hidden layer pair x n The output matrix, h L (x n ) represents the hidden layer node L for x n The output value of g(·) is the hidden layer activation function, w L is the weight of the hidden layer node L, b L is the bias of the hidden layer node L, β L represents the weight between the hidden layer node L and the output layer, t n Represents the ELM model for x n Output.

[0104] In an exemplary embodiment, the step of inputting the converted feature vector into a cascade ELM module to obtain a lithology characterization vector includes:

[0105] The cascade ELM module is composed of N layers of cascade ELM models, each layer consisting of multiple parallel ELM models;

[0106] The transformed feature vector tf i Input into the first layer of cascade ELM model to obtain the three-dimensional vector corresponding to the first layer;

[0107] The transformed feature vector tf i The three-dimensional vectors corresponding to the first layer are concatenated and input into the second-layer cascaded ELM model;

[0108] Until the transformed feature vector tf i After being concatenated with the three-dimensional vector corresponding to the N-1th layer and input into the ELM model of the Nth layer cascade, the three-dimensional vector corresponding to the Nth layer is obtained;

[0109] The three-dimensional vectors corresponding to the Nth layer are averaged to obtain a characterization vector of the lithology.

[0110] In an exemplary embodiment, the optimal parameters of the lithology identification model include: the number of base models, the number of hidden layer neurons and the maximum number of cascade layers.

[0111] The embodiment of the present disclosure also provides a lithology identification device for unbalanced logging data, such as Figure 2 As shown, the device includes: a memory 200 and a processor 210; the memory is used to store a program for lithology identification, and the processor is used to read and execute the program for lithology identification oriented to unbalanced logging data, and execute any one of the methods in the above embodiments.

[0112] The embodiment of the present disclosure further provides a computer-readable storage medium, on which a data processing program is stored. The data processing program is executed by a processor to implement the lithology identification method for unbalanced logging data described in any one of the above embodiments.

[0113] Example 1

[0114] The present invention provides a lithology identification method for unbalanced logging data, such as Figure 3 As shown, the following steps are included:

[0115] S1: Obtain logging data of oil wells;

[0116] The logging data is a characteristic curve reflecting lithology;

[0117] The logging data includes one or more of the following: natural gamma logging curve, sonic time difference logging curve, compensated neutron logging curve, density logging curve, deep lateral resistivity logging curve and shallow lateral resistivity logging curve.

[0118] S2: Divide the logging data and their corresponding lithology categories into a training set and a test set according to a predetermined ratio;

[0119] S3: preprocessing the sample data in the training set using the ADASYN algorithm to generate new samples and add them to at least several categories of samples;

[0120] In this step, the ADASYN algorithm is used to preprocess the sample data in the training set, such as Figure 4 As shown, including:

[0121] S31: Input imbalanced dataset;

[0122] Input an unbalanced data set, and classify the sample data in the training set data according to the lithology category to obtain the lithology category of the minority class and the lithology category of the majority class;

[0123] S32: Calculate the number of samples required for sampling;

[0124] In this step, the sample data of each minority class is calculated separately to determine the number of samples G to be synthesized;

[0125] S33: Calculate the proportion of the majority lithology;

[0126] In this step, the proportion of most types of lithology is calculated, including:

[0127] S331: For each minority class sample data, extract K nearest neighbors of each minority class sample according to the Euclidean distance;

[0128] S332: Determine the proportion of the majority lithology category among the K nearest neighbors;

[0129] S34: normalized majority lithology proportion;

[0130] Normalizing the proportions of the majority of lithology categories determined in step S332 to obtain sample weight values;

[0131] S35: Calculate the number of new samples required to generate for each minority class sample;

[0132] In this step, the number of new samples to be synthesized for each minority class sample is calculated according to the sample weight value;

[0133] S36: Synthesize minority samples based on K nearest neighbors;

[0134] Randomly select a minority class sample xz from the K nearest neighbors of the minority class sample i , synthesize new samples of the minority class samples;

[0135] S37: Output the balanced data set.

[0136] S4: input the preprocessed sample data into the multi-granularity scanning module to generate a transformation feature vector tfi;

[0137] S5: inputting the converted feature vector into the cascade ELM module to obtain a lithology characterization vector;

[0138] S6: inputting the representation vector output by the cascade ELM module into the parameter optimization module, and using the PSO algorithm to determine the optimal parameters of the lithology identification model; obtaining the trained lithology identification model according to the determined optimal parameters;

[0139] S7: input the test set data into the trained lithology recognition model to obtain the lithology recognition result;

[0140] S8: Input the lithology identification results into the index evaluation module to obtain the evaluation performance of the model.

[0141] Example 2

[0142] A lithology identification method for unbalanced logging data, the method comprising the following steps:

[0143] S1: Obtaining original well logging data and lithology category data of the oil well to be lithology classified;

[0144] For example, six logging curves of an oil well are obtained, namely, Gamma Ray (GR), Acoustic (AC), Compensated Neutron Log (CNL), Density (DEN), Deep Lateral Resistivity Log (LLD) and Shallow Lateral Resistivity Log (LLS);

[0145] The lithology categories include four different lithologies: Salt Rock (SR), Sandy Mudstone (SM), Calcareous Dolomitic Mudstone (CDM) and Clay Mudstone (CM).

[0146] S2: Divide the logging data and their corresponding lithology categories into a training set and a test set according to a predetermined ratio;

[0147] As shown in Table 1, the minority class and the majority class are divided according to the proportion. Here, the largest proportion is determined as the majority class, and the rest are the minority class. Taking well A as an example, SR belongs to the majority class, and CM, SM, and CDM all belong to the minority class. Taking well B as an example, SR belongs to the majority class, and CM, SM, and CDM all belong to the minority class.

[0148] Table 1

[0149]

[0150] S3: Use the ADASYN algorithm to preprocess the training set data and generate new samples to add to the samples of at least several categories;

[0151] Specifically, new samples are generated by the ADASYN algorithm:

[0152] a. Calculate the number of samples G required to synthesize the minority class;

[0153] G=(m l -m s )×β

[0154] Among them, β∈[0,1] is a parameter reflecting the required balance level. l With m s are the number of samples of majority class and minority class respectively.

[0155] b. For the minority class sample x, extract its K nearest neighbors based on the Euclidean distance and calculate the majority class proportion r among the K nearest neighbors i .

[0156] r i =Δ i / K, i = 1, ..., m s

[0157] Among them, Δ i is the number of majority class samples among the K nearest neighbors.

[0158] c. Normalize the majority class proportion and use the normalized value As the sample weight value;

[0159] Among them, the normalized calculation formula is:

[0160]

[0161] d. Calculate each minority class sample x according to the sample weight value i The number of new samples to generate.

[0162]

[0163] e. From the minority class sample x i Among the K nearest neighbors, randomly select a minority class sample x zi , synthesize new samples x of the minority class samples new :

[0164] x new =x i +(x zi -x i )×λ

[0165] Among them, λ∈[0,1] is a random number.

[0166] f. Repeat steps a to e for each minority class.

[0167] S4: Input the preprocessed sample data into the multi-granularity scanning module to generate a transformation feature vector tfi; the learner in the multi-granularity scanning module is an ELM model; wherein the ELM network structure is as follows Figure 5 As shown:

[0168] The ELM model is a single hidden layer feedforward neural network, which consists of L hidden layer neurons and N output layer neurons. The ELM model can be defined as

[0169] Hβ=T;

[0170]

[0171] β=[β1,β2,…,β L ] T ;

[0172] T=[t1,t2,…,t n ] T ;

[0173] Among them, H is the hidden layer output matrix, β is the weight matrix, T is the target matrix, and x n is the nth logging data, L is the number of hidden layer nodes, h(x n ) represents the hidden layer pair x n The output matrix, h L (x n ) represents the hidden layer node L for x n The output value of g(·) is the hidden layer activation function, w L is the weight of the hidden layer node L, b L is the bias of the hidden layer node L, β L represents the weight between the hidden layer node L and the output layer, t n Indicates that ELM is n Output.

[0174] In this step, the multi-granularity scanning module implements the process as follows: Figure 6 As shown:

[0175] 1. In the multi-granularity scanning module, the sliding window is used to scan the sample data after the original input preprocessing; for a given window size d i , generate feature vectors fv of different sizes i ;

[0176] 2. Use lithology category to mark the feature vector fv i, and input them into the corresponding first learner and second learner respectively to generate the first type probability distribution vector lf1 i and the second probability distribution vector lf2 i ;

[0177] For example: fv1 is input into the corresponding first learner A1 and second learner A2; fv i Input into the corresponding first learner B1 and second learner B2;

[0178] 3. Concatenate the first type of probability distribution vector and the second type of probability distribution vector into a conversion feature vector tf i .

[0179] S5: transform the feature vector tf i Input the cascade ELM module to obtain the characterization vector of lithology;

[0180] In this step, the cascade ELM module implementation process is as follows Figure 7 As shown:

[0181] Each cascade layer is composed of different ELM models. Taking three lithologies as an example, the output of each ELM model is a three-dimensional vector. As the input of the next layer, the output of each layer of ELM model is connected with the feature vector from the multi-granularity scanning module to enhance the representation. After each cascade expansion, the entire cascade structure is self-evaluated. If the performance is not significantly improved, the training process is terminated and the average result of the last layer is output. Self-evaluation can be judged by training error, which can be set in advance. For example, the training error can be set to the accuracy of the model on the validation data set. The entire training process is as follows: After the logging data is input, the model automatically selects part of the data as the validation data set. When the model is trained, its accuracy on the validation data set is calculated. The above is a complete model training. When the accuracy no longer improves after 3 repetitions, the iteration is stopped.

[0182] S6: input the representation vector output by the cascade ELM module into the parameter optimization module, and use the PSO algorithm to optimize the model;

[0183] Specifically, the PSO algorithm optimizes the deep ELM model implementation process as follows: Figure 7 As shown, the steps are as follows:

[0184] a. Determine and initialize the particle x and iteration step size v in the PSO algorithm. Particle x represents the hyperparameters of the lithology recognition model, which includes three parameters: the number of base models, the number of neurons in the hidden layer, and the maximum number of cascade layers.

[0185] The number of base models indicates the number of ELM models in each cascade layer; the number of hidden layer neurons indicates the number of neurons in each hidden layer of the ELM model; the maximum number of cascade layers indicates the maximum number of cascade layers of the lithology recognition model. The iteration step v is the amplitude of the increase or decrease of particle x. Particle x is represented by the following formula:

[0186]

[0187] b. Determine the maximum number of iterations iter in the PSO algorithm max , particle swarm size pop and particle parameter upper limit up and lower limit down. At the same time, the evaluation function is set to calculate the local optimal local i , and then select the global optimal global. The evaluation function is used to evaluate the recognition accuracy of the lithology recognition model, and a conventional evaluation function can be used.

[0188] c. Update particle x and iteration step v:

[0189] v i =v i +r1*c1*(local i -x i )+r2*c2*(golbal-x i )

[0190] x i =x i +v i

[0191] x i =max(x i , x down )

[0192] x i =min(x i , x up )

[0193] Among them, the random numbers r1, r2∈[0, 1], c1, c2 represent the learning rate, x i and v i Represents the particle and its step size at time i.

[0194] d. Repeat steps b and c to update the local optimal value of particle x. i and the global optimum global until the number of iterations reaches the maximum iter max At this time, the global optimal corresponding particle value is the optimal hyperparameter of the lithology identification model.

[0195] S7: input the test set data into the trained lithology recognition model to obtain the lithology recognition result;

[0196] Specifically, the logging feature vector of the rock sample to be tested is input into the trained lithology identification model (multi-granularity scanning module + cascade ELM module) to obtain the lithology identification result of the lithology sample to be tested.

[0197] S8: Input the lithology identification results into the index evaluation module to obtain the evaluation performance of the model.

[0198] Specifically, the indicator evaluation module uses two indicators (accuracy and F1 score) to evaluate the performance of the lithology recognition model. Accuracy reflects the proportion of correctly predicted samples, while the F1 score is the harmonic mean of accuracy and recall, which represents the classification performance on an unbalanced data set. The lithology output results are compared with the true lithology labels, and the performance indicators of the model are calculated as follows:

[0199]

[0200]

[0201] Among them, tp represents true positive, fn represents false negative, fp represents false positive, and tn represents true negative.

[0202] Example 3

[0203] To verify the effectiveness of the above lithology identification method, the data used are two shale oil wells (well A and well B) in an oil field in Jianghan area of ​​China, including 2664 and 2745 sets of logging data respectively, with sampling interval of 0.125m and well depths of 1206.125-1539m and 1341-1684m respectively. The data set contains a total of 6 types of logging curves: natural gamma ray (GR), acoustic time difference (AC), compensated neutron log (CNL), density (DEN), deep lateral resistivity log (LLD) and shallow lateral resistivity log (LLS). There are four main lithology types, including Salt Rock (SR), Sandy Mudstone (SM), Calcareous Dolomitic Mudstone (CDM) and Clay Mudstone (CM). The lithology distribution is shown in Table 1.

[0204] Table 1

[0205]

[0206]

[0207] (1) Different data balancing methods specifically verify the impact of different data balancing methods and different classification models on the classification results, and conduct ablation experiments on the models, as shown below:

[0208] In the proposed lithology identification model, the ADASYN algorithm is used to oversample the original logging data to deal with the data imbalance problem. To verify the superiority of the ADASYN algorithm over other data balancing algorithms, comparative experiments were conducted with other data balancing algorithms, including ROS, SMOTE, and MAHAKIL. To avoid the influence of chance, a 5-fold cross validation was used to obtain the evaluation index. The cross validation diagram is shown in the figure. Figure 8 As shown in the figure. First, the well logging data set is divided into 5 folds, each with 20% lithology samples. Each fold is used as a test set and the rest as a training set, and five experimental results are obtained. In each fold, the training and test data are normalized to speed up the training. After oversampling the training data, the deep ELM model is used to identify the lithology. Finally, the mean and standard deviation are calculated based on all experimental results to evaluate the average performance and stability of the method.

[0209] Table 2 shows the model recognition results of different data balancing algorithms. For well A, all the comparison methods improved the performance of lithology recognition. Among them, the ADASYN algorithm improved the accuracy from 94.447% to 95.521%. At the same time, the F1 score also increased from 94.420% to 95.522%. The results show that oversampling can improve the recognition effect of unbalanced lithology data, and ADASYN has the largest improvement effect (more than 1%) among the above comparison methods.

[0210] Well B shows different results from Well A, as most oversampling methods lead to poor lithology identification. This may be due to oversampling leading to model overfitting and blurred class boundaries, while data imbalance has little effect on model classification. Nevertheless, the ADASYN algorithm still improves the accuracy of lithology identification (from 98.543% to 98.621%), indicating that the model trained with samples synthesized by ADASYN has the best generalization.

[0211] Table 2

[0212]

[0213] (2) Different classification models

[0214] In order to verify the lithology recognition effect of the deep ELM model, this embodiment is experimentally compared with the RF model, XGBoost model, LightGBM model, ELM model, and deep RF model, deep XGBoost model, and deep LightGBM model. Table 3 shows the accuracy and F1 scores of different classification models.

[0215] Table 3

[0216]

[0217]

[0218] Overall, the deep ELM model achieved the best results on both datasets, with classification accuracy of 95.521% and 98.621%, and F1 scores of 95.522% and 98.622%, respectively. Compared with the deep RF model, the deep ELM model improved the accuracy of well A and well B by 1% and 0.8%, respectively. As alternative base learners to RF, deep XGBoost and deep LightGBM are not as effective as the original deep RF. In addition, among the other machine learning models, LightGBM performed the best, but despite this, the deep ELM model still had the highest classification accuracy and F1 score compared with other machine learning models without multi-granularity scanning.

[0219] (3) Model ablation experiment

[0220] In order to verify the improvement effect of the proposed model, ablation experiments were carried out on various parts of the model. The deep RF model without ADASYN sampling was regarded as the baseline, and the deep RF model with ADASYN sampling, the deep ELM model without ADASYN sampling and the proposed deep ELM model with ADASYN sampling were compared. In order to show the improvement effect on each lithology category, the comparison results are displayed in the form of a confusion matrix, as shown in Fig. 9 shown.

[0221] contrast Fig. 9 (a) and Fig. 9 (b) The ADASYN algorithm improves the recognition accuracy of minority lithologies. The classification accuracy of SM in well A and well B increased from 85.71% and 95.45% to 90.48% and 98.46%, respectively. The results show that the application of ADASYN oversampling can improve the accuracy of minority lithology recognition. Fig. 9 (a) and Fig. 9 (c) It can be seen that among the eight lithology categories of the two data sets, the deep ELM model has higher recognition accuracy for six of them than the deep RF model. The results show that the deep ELM model can help identify most lithologies. Fig. 9Compared with (d), the deep ELM model combined with the ADASYN algorithm has the highest overall lithology recognition accuracy, and the classification effects of the majority and minority lithologies are improved. Therefore, the model can balance the minority samples and effectively improve the lithology recognition performance.

[0222] The technical solution implemented by the present invention has the following beneficial effects:

[0223] (1) In the embodiments of the present application, the ADASYN algorithm is used to mine the intrinsic connection between minority class samples and majority class samples, and new artificial samples are synthesized near the boundary between the two classes to balance the minority class samples. This overcomes the problem of traditional data balancing algorithms that only analyze the internal structure of minority class samples, and effectively improves the classification model's recognition effect on minority lithology.

[0224] (2) The present invention proposes a lithology identification method and system for low-quality unbalanced logging data, and designs a lithology identification model constructed by a multi-granularity scanning module and a cascade ELM module, which effectively solves the problem that traditional machine learning models ignore the relationship between logging curves, mines the complementary information between logging curves to enhance logging representation, and effectively improves the recognition accuracy of the lithology identification model based on machine learning.

[0225] (3) The present invention proposes a lithology identification method and system for low-quality unbalanced logging data, and designs a lithology identification algorithm based on the PSO algorithm + deep ELM model, which effectively solves the problem of time-consuming model construction and low identification accuracy of traditional lithology identification methods, and effectively overcomes the problem of low efficiency of hyperparameter optimization in conventional machine learning lithology identification models.

[0226] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all components may be implemented as software executed by a processor, such as a digital signal processor or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those of ordinary skill in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

Claims

1. A lithology identification method for unbalanced logging data, characterized in that: The method comprises: Obtain well logging data of oil wells and the lithology categories to be identified; Inputting the lithology category to be identified and the well logging data into the trained lithology identification model to obtain a lithology identification result; The trained lithology identification model is obtained by preprocessing the logging sample data through the ADASYN algorithm and training based on the cascade ELM model of the multi-granularity scanning module.

2. The lithology identification method for unbalanced logging data according to claim 1, characterized in that: The well logging data is a characteristic curve reflecting lithology; The logging data includes one or more of the following: natural gamma logging curve, sonic time difference logging curve, compensated neutron logging curve, density logging curve, deep lateral resistivity logging curve and shallow lateral resistivity logging curve.

3. The lithology identification method for unbalanced logging data according to claim 1, characterized in that: The training process of the lithology identification model includes: Divide the logging sample data and their corresponding lithology categories into a training set and a test set according to a predetermined ratio; Preprocessing the well logging sample data in the training set using the ADASYN algorithm; The pre-processed logging sample data is input into the cascade ELM model based on the multi-granularity scanning module to obtain the characterization vector of lithology; The characterization vector of the lithology is input into the parameter optimization module, and the PSO algorithm is used to determine the optimal parameters of the lithology identification model; A trained lithology identification model is obtained according to the determined optimal parameters.

4. The lithology identification method for unbalanced logging data according to claim 3 is characterized in that: The preprocessing of the well logging sample data in the training set by using the ADASYN algorithm includes: Classifying the well logging sample data in the training set according to lithology categories, and determining a minority class and a majority class; Calculate for each minority class separately to determine the number of samples v to be synthesized for the minority class According to the sample synthesis method of the minority class and the number of samples to be synthesized, new sample data corresponding to each minority class is obtained.

5. The lithology identification method for unbalanced logging data according to claim 4, characterized in that: The method of synthesizing samples of the minority class and the number of samples to be synthesized are used to obtain new sample data corresponding to each minority class, including: Extract K nearest neighbors of each minority class sample according to the Euclidean distance; Determine the proportion of each majority class in the K nearest neighbors of each minority class sample; Normalize the proportion of all majority classes to obtain the sample weight value; Calculate the number of new samples to be synthesized for each minority class sample according to the sample weight value; From the K nearest neighbors of the minority class sample, a minority class sample is randomly selected to synthesize new sample data for the minority class sample.

6. The lithology identification method for unbalanced logging data according to claim 3, characterized in that: The pre-processed logging sample data is input into the cascade ELM model based on the multi-granularity scanning module to obtain the lithology characterization vector, including: Inputting the preprocessed logging sample data into the multi-granularity scanning module to generate a conversion feature vector; The converted feature vector is input into the cascade ELM module to obtain a characterization vector of lithology.

7. The lithology identification method for unbalanced logging data according to claim 6, characterized in that: The step of inputting the pre-processed logging sample data into the multi-granularity scanning module to generate a conversion feature vector comprises: Using a sliding window for the preprocessed logging sample data, a plurality of feature vectors of different sizes are generated; Input each feature vector into the corresponding first learner and second learner respectively to generate a first type probability distribution vector and a second type probability distribution vector; The first type of probability distribution vector and the second type of probability distribution vector are concatenated into a conversion feature vector.

8. The lithology identification method for unbalanced logging data according to claim 7, characterized in that: The first learner and the second learner are ELM models; Wherein, the ELM model is: Hβ=T; β＝[β1，β2，…，β L ] T ； T=[t1,t2,…,t n ] T ; In the above formula, H is the hidden layer output matrix, β is the weight matrix, T is the target matrix, and x n is the nth logging data, L is the number of hidden layer nodes, h(x n ) represents the hidden layer pair x n The output matrix, h L (x n ) represents the hidden layer node L for x n The output value of g(·) is the hidden layer activation function, w L is the weight of the hidden layer node L, b L is the bias of the hidden layer node L, β L represents the weight between the hidden layer node L and the output layer, t n Represents the ELM model for x n Output.

9. The lithology identification method for unbalanced logging data according to claim 7, characterized in that: The step of inputting the converted feature vector into the cascade ELM module to obtain a lithology characterization vector includes: The cascade ELM module is composed of N layers of cascade ELM models, each layer consisting of multiple parallel ELM models; Inputting the converted feature vector into the first layer of the cascaded ELM model to obtain a three-dimensional vector corresponding to the first layer; splicing the converted feature vector and the three-dimensional vector corresponding to the first layer and inputting the concatenated vector into the second-layer cascaded ELM model; Until the conversion feature vector and the three-dimensional vector corresponding to the N-1th layer are concatenated and input into the ELM model of the Nth layer cascade, the three-dimensional vector corresponding to the Nth layer is obtained; The three-dimensional vectors corresponding to the Nth layer are averaged to obtain a characterization vector of the lithology.

10. The lithology identification method for unbalanced logging data according to claim 3, characterized in that: The optimal parameters of the lithology identification model include: the number of base models, the number of hidden layer neurons and the maximum number of cascade layers.

11. A lithology identification device for unbalanced logging data, characterized in that: The device comprises: a memory and a processor; the memory is used to store a program for a lithology identification method for unbalanced logging data, and the processor is used to read and execute the program for a lithology identification method for unbalanced logging data, and execute the method described in any one of claims 1-10.

12. A computer-readable storage medium having a data processing program stored thereon, wherein the data processing program is executed by a processor to implement the lithology identification method for unbalanced logging data according to any one of claims 1 to 10.

Images