A method for identifying low-resistivity oil reservoirs during drilling

Abstract

This invention discloses a method for identifying low-resistivity oil layers based on drilling, comprising the following steps: Step 1: Acquire historical gas logging data and preprocess the data samples; Step 2: Obtain drilling parameter tables and perform data correction on gas logging parameters such as total gas content (TG) and methane C1 based on the drilling parameters; Step 3: To highlight the differences between oil layers, low-resistivity oil layers, and water layers, feature extraction is performed on the data from Step 1 and Step 2, and stepwise multivariate regression analysis is used to optimize the parameters of the extracted features; Step 4: The parameters from Step 3 are combined with the drilling parameter gamma ray (GR) as input to the model to establish a GRNN neural network classification model based on the goose flocking optimization algorithm to classify the oil layers. This invention solves the inaccuracy of manual judgment of fluid properties and eliminates the need to identify fluid properties by measuring resistivity, thereby greatly reducing the cost and complexity of fluid identification.

Description

Technical Field

[0001] This invention belongs to the field of petroleum exploration technology, specifically relating to a method for identifying low-resistivity oil reservoirs during drilling. Background Technology

[0002] Low-resistivity oil reservoirs are often defined as reservoirs with resistivity lower than normal oil reservoirs, similar to or even lower than water reservoirs. Their formation is also diverse, and their identification accuracy is usually low. The interpretation of reservoir fluids in the Bohai Sea region is becoming increasingly difficult. Moreover, most well logging evaluations of reservoir oil and gas content are still in the qualitative stage. Therefore, it is urgent to establish a quantitative interpretation model for well logging.

[0003] Currently, there are many methods for identifying low-resistivity oil layers, and the methods used vary depending on the available data. Research shows that, for example, Hu Yun et al., in their article "Deconstruction Method of Gas-Phase Hydrocarbon Derivative Parameters and Its Application in Bohai Oilfield," established the intrinsic relationship between gas-phase hydrocarbon derived parameters and reservoir characteristics, using a plotting method to identify fluid properties. Zhang Jianbin et al., in their article "A New Method for Quantitative Evaluation of Hydrocarbon-Bearing Properties of Bohai Oilfield Reservoirs Based on Gas-Phase Hydrocarbons and Its Application Effect," used gas logging data to construct hydrocarbon indices and abundance, and identified low-resistivity oil layers by drawing cross-plots. Among the published invention patents, CN113969781A proposes a method for identifying low-resistivity oil layers based on the microporous pore structure of marine carbonate rocks, establishing first and second identification models by determining their origin and acquiring data. CN106285661A discloses a method and device for identifying low-resistivity oil layers based on a judgment index, using data to derive an empirical judgment index to locate suspicious low-resistivity oil layers.

[0004] While numerous methods, such as resistivity methods, spontaneous potential methods, special imaging logging methods, chart methods, and curve overlay methods, have been applied to the identification of low-resistivity oil reservoirs, many require significant human and financial resources. There is still a lack of technologies that utilize gas logging data combined with drilling parameters and machine learning to quantitatively identify low-resistivity oil reservoirs, especially in the Bohai Oilfield. Therefore, providing an artificial intelligence-based identification method has become a pressing technical problem to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide a method for identifying low-resistivity oil layers based on drilling. This method combines gas logging data and drilling parameters with machine learning to identify low-resistivity oil layers, which greatly improves the accuracy.

[0006] The present invention adopts the following technical solution:

[0007] A method for identifying low-resistivity oil reservoirs during drilling includes the following steps:

[0008] Step 1: Obtain the basic gas measurement parameters: total gas volume, methane, ethane, propane, n-butane, isobutane, n-pentane, and isopentane. Preprocess the data samples: use the quartile method to remove outliers for each basic parameter, and fill out the outliers with nearby interpolation.

[0009] Step 2: Obtain the drilling parameter table and perform data correction on the gas logging parameters based on the drilling parameters; the gas logging parameters include: total gas volume. Alkane volume fraction;

[0010] Step 3: To highlight the differences between oil layers, low-resistivity oil layers, and water layers, feature extraction is performed on the preprocessed data from Step 1 and the corrected data from Step 2, and stepwise multivariate regression analysis is used to optimize the parameters of the extracted features.

[0011] Step 4: Combine the optimized parameters obtained in Step 3 with the drilling parameter gamma. As input to the model, a GRNN neural network classification model based on the goose flocking optimization algorithm is established to classify oil layers;

[0012] The steps for establishing a GRNN neural network model based on the goose flocking optimization algorithm are as follows:

[0013] S1: Population parameter GSO initialization, randomly initialize a goose flock M, and set the initial position of the i-th goose in N-dimensional space as follows. The speed is And the individual extreme value and global extreme value of the i geese are respectively and ;

[0014] The parameter GSO refers to the size of the population, as well as the initial position, velocity, individual extreme value, and global extreme value of a goose in the population.

[0015] This represents a set of the initial positions of the i-th goose in each dimension;

[0016] This represents the position of the i-th goose in each dimension. This represents a set of individual extreme values ​​for the i-th goose in each dimension; It represents a set of global extrema for the i-th goose in each dimension;

[0017] S2: Map the position of each goose to the GRNN neural network, construct the structure of the GRNN neural network, and initialize the network smoothing factor;

[0018] S3: Train the GRNN neural network, uniformly arranging the geese in the flock in each iteration to calculate the flock fitness value. When the GSO fitness value... E When the set number of iterations is reached, the minimum fitness value is compared with the minimum fitness value in each iteration, the algorithm ends, and the fitness value is saved. E The smoothing factor at the minimum value is finally saved into the training model of the GRNN neural network; when the GSO fitness value... E If the set number of iterations is reached, but the iteration fitness value is not the minimum, then the value is continuously adjusted and optimized through steps S4-S5.

[0019] The function value for calculating fitness is: ;

[0020] In the formula, The number of samples is represented by m, and the number of categories is represented by m. Indicates the first The label value corresponding to the category of each sample. Indicates the first The probability values ​​obtained by applying the Softmax activation function to each sample;

[0021] S4: Calculate the first [number] within the field of view using a simple averaging method. i The individual extreme value of a single goose is taken as the global extreme value perceived by the goose at that moment, that is: ;

[0022] Where k represents the number of iterations. Let represent the individual extreme value of the s-th goose in the k-th iteration, where the positive integer s takes values ​​in the range [1, ..., ...]. i ];

[0023] Then, the individual extreme value of the previous goose is directly used as the individual extreme value of the current goose, that is: ;

[0024] This represents the individual extreme value of the goose in the k-th iteration. This represents the individual extreme value of the previous goose in the k-th iteration;

[0025] S5: Based on the improvement of S4, the first... The speed of Zhiyan updates is as follows: ;

[0026] No. The formula for updating the position of a single goose is: ;

[0027] in, This is the inertia weight at the k-th iteration, used to balance the global and local convergence capabilities of the algorithm. , They are the velocity and position of the i-th wild goose at the k-th iteration respectively;

[0028] The dimension d of the wild geese is 1, 2, …, N; and are random numbers obeying the uniform distribution on [0, 1]; and are non-negative acceleration learning factors, which are used to adjust the maximum moving step lengths in the directions of the individual extreme value and the global extreme value of the current wild goose flock respectively; The weight coefficient is updated according to the formula: ;

[0029] where and are respectively the maximum value and the minimum value of the weight coefficient, is the current iteration number, is the maximum iteration number;

[0030] S6: Determine whether the wild goose flock meets the end condition. If it does, enter S7; otherwise, return to S3 and continue to execute;

[0031] S7: Map each dimension of the global extreme value to the smoothing factor in the GRNN, train the network, and output the classification result.

[0032] Preferably, the preprocessing of the basic gas logging historical data in step 1 includes the following steps:

[0033] Use the quartile method to剔除异常值 (remove outliers) for each basic parameter, and fill the outliers with nearby interpolation. The nearby interpolation is specifically divided into the following three cases:

[0034] (0) When < 3 and the th data is an outlier, at this time the th parameter's th data is ; where represents the th parameter's th data, represents the th parameter's th data; represents the serial number of a certain sample, and its value range is [1, n], where n represents the number of samples;

[0035] (1) When 3 < < n - 3 and the th data is an outlier, at this time the th parameter's th data is ;

[0036] where n represents the number of samples, the th data of the th parameter, the th data of the th parameter, the th data of the th parameter, the th data of the th parameter;

[0037] (3) When n - 3 < < n and the th data is an outlier, then the th data of the th parameter is .

[0038] Preferably, in step 2, the gas total volume and methane volume fraction in the gas logging data are corrected in combination with the drilling parameters. The correction formula for the methane volume fraction is as follows: ;

[0039] where, represents the corrected methane value, is an empirical constant,​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​Indicates the sample gas flow rate;

[0043] The volume fractions of ethane, propane, n-butane, isobutane, n-pentane, and isopentane were also corrected according to the volume fraction correction formula for methane, resulting in the corrected volume fractions of ethane, propane, n-butane, isobutane, n-pentane, and isopentane as follows: , , , , , ; value, Represents the corrected propane number, Represents the corrected butane number, value, This represents the corrected n-pentane number. Isopentane number.

[0044] Preferably, in step 3, feature extraction is performed on the preprocessed and corrected data. Based on the different trends of gas logging parameters in different oil layers, the methane volume fraction of the gas logging parameters is corrected. Corrected ethane The corrected propane volume fraction is Corrected n-butane volume fraction Corrected volume fraction of isobutane Corrected n-pentane volume fraction Corrected volume fraction of isopentane Total amount of gas The corrected values ​​are combined with different features to highlight the differences between oil layers, low-resistivity oil layers, and water layers. These feature parameters include... , , , , , , ;in , The components are light components. The above components are recombinant components. .

[0045] To reduce the redundancy of the extracted feature parameters and the complexity of the model input, stepwise regression analysis was used to optimize the parameters. The parameters ultimately optimized using this method include... , , .

[0046] Preferably, step 4 includes:

[0047] The parameters selected in step 3 are organized and combined with the drilling parameter gamma. Establish data samples to ensure that the proportion of each sample is uniform, quantize the labels using binary encoding, and normalize the data.

[0048] A GRNN neural network classification model based on the goose flocking optimization algorithm was established. The normalized feature parameter values ​​were then fed into the model for training, thereby obtaining a low-resistivity oil reservoir classification and prediction model.

[0049] The beneficial effects of this invention are:

[0050] For the Bohai Oilfield, low-resistivity oil layers can be quickly identified by combining gas logging data with drilling parameters, replacing the inaccuracies of traditional identification methods. At the same time, low-resistivity oil layers are no longer identified by resistivity, which greatly improves economic efficiency.

[0051] By fully exploring the inherent characteristics in gas logging data and combining it with gamma-ray logging to establish a goose flock optimized GRNN classification model, the accuracy of low-resistivity oil layers can be quantitatively identified using artificial intelligence. This invention can provide a scientific basis for the selection of subsequent logging methods and testing schemes. Attached Figure Description

[0052] Figure 1 A flowchart of an embodiment provided for an embodiment of the present invention.

[0053] Figure 2 A flowchart is established for the model of this invention.

[0054] Figure 3 The structural topology diagram for classifying GRNN neural networks.

[0055] Figure 4 This is a diagram showing the comparison between the classification results obtained from the test set using the established model and the actual results. Detailed Implementation

[0056] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] This invention is mainly based on gas logging data and drilling parameters, combined with gamma-ray while drilling to identify low-resistivity oil layers. The application principle of this invention will be described in detail below with reference to the accompanying drawings.

[0058] like Figure 1 As shown, the present invention provides a method for identifying low-resistivity oil layers based on drilling, and the specific steps are as follows:

[0059] S1: Preprocess the data samples according to the basic gas logging parameters, namely total gas volume, methane, ethane, propane, n-butane, isobutane, n-pentane and isopentane. The specific steps include: eliminating outliers for each basic parameter using the quartile method and filling the outliers with nearby interpolation. The nearby interpolation specifically includes the following three cases:

[0060] (1) When < 3 and the -th data is an outlier, at this time the -th data of the -th parameter is ; where represents the -th data of the -th parameter, represents the -th data of the -th parameter; represents the serial number of a certain sample, and the value range is [1, n], where n represents the number of samples;

[0061] (2) When 3 < < n - 3 and the -th data is an outlier, at this time the -th data of the -th parameter is ;

[0062] where n represents the number of samples, the -th data of the -th parameter, the -th data of the -th parameter, the -th data of the -th parameter, the -th data of the -th parameter;

[0063] (3) When n - 3 < < n and the -th data is an outlier, at this time the -th data of the -th parameter is .

[0064] S2: Obtain the drilling parameter table, and combine the drilling parameters to correct the gas logging parameters such as total gas volume , methane , etc. The correction formula for methane is as follows: ;

[0065] In the formula This represents the corrected methane value. It is an empirical constant. For drilling time, This refers to the drilling fluid discharge rate. The diameter of the drill bit. Equivalent density, Density of drilling fluid; Total gas volume, the correction formula for total gas volume is: ;

[0066] In the formula This represents the corrected total gas volume. It is an empirical constant. , This refers to the drilling fluid discharge rate. The diameter of the drill bit. For formation pressure gradient, This represents the rupture pressure gradient value.

[0067] Taking into account the influence of factors such as the degasser itself and the drilling process, the empirical constants in the formula are adjusted. The quantification is performed, and the specific quantification formula is as follows: ;

[0068] in Indicates the drilling fluid flow rate of the degasser. Indicates the sample gas flow rate;

[0069] The volume fractions of ethane, propane, n-butane, isobutane, n-pentane, and isopentane were also corrected according to the volume fraction correction formula for methane, resulting in the corrected volume fractions of ethane, propane, n-butane, isobutane, n-pentane, and isopentane as follows: , , , , , ; value, Represents the corrected propane number, Represents the corrected butane number, value, This represents the corrected n-pentane number. Isopentane number; other hydrocarbons can also be corrected according to the correction formula for methane.

[0070] S3: Feature extraction is performed on the preprocessed data. Based on the different trends of gas logging parameters in different oil layers, the methane volume fraction after correction of the gas logging parameters is adjusted. Corrected ethane The corrected propane volume fraction is Corrected n-butane volume fraction Corrected volume fraction of isobutane Corrected n-pentane volume fraction Corrected volume fraction of isopentane Total amount of gas The corrected values ​​are combined with different features to highlight the differences between oil layers, low-resistivity oil layers, and water layers. These feature parameters include... , , , , , , ;in , The components are light components. The above components are recombinant components. ;

[0071] To reduce the redundancy of extracted feature parameters and the complexity of model input, stepwise regression analysis was used to optimize the parameters. The parameters ultimately optimized using this method include... , , .

[0072] S4: Organize the selected parameters and combine them with the drilling parameters gamma. A total of 600 data samples were established, ensuring a uniform proportion among the samples. Labels were quantified, and the data was normalized using the following formula: .

[0073] in After normalization The first in One value, , , These are the characteristics of the parameter. The first in A set of values, including the maximum and minimum values.

[0074] The parameters selected in step 3 are organized and combined with the drilling parameter gamma. A total of 600 data samples were established, ensuring that the proportion of each sample was uniform. The labels were quantized using binary encoding, and the quantization results are shown in Table 1 below.

[0075] Table 1 Label Quantification Table

[0076]

[0077] The sample set is divided into a training set and a test set, with 480 samples in the training set and 120 samples in the test set. These samples are then used in the GRNN neural network classification model based on the goose flocking optimization algorithm.

[0078] like Figure 2 As shown, the specific steps for establishing a GRNN neural network model based on the goose flocking optimization algorithm are as follows:

[0079] S1: Initialize GSO (Goose Optimization Algorithm) parameters, randomly initialize a goose flock M, and set the parameters of the first... i The initial position of the goose in N-dimensional space is The speed is And the individual extreme value and global extreme value of the i geese are respectively and .

[0080] S2: Map the position of each goose to the GRNN neural network, construct the structure of the GRNN neural network, and initialize the network smoothing factor.

[0081] like Figure 3 As shown, after initializing the population, the smoothing factor of the GRNN model is used as the position of each goose. The data is then imported into the GRNN model, and the fitness value is calculated based on the initial position of each goose.

[0082] Since the initial position of each goose is not necessarily optimal and it needs to move according to certain rules, it is necessary to update its position and velocity so that it gradually moves towards the global optimum by continuously comparing its position size, thereby obtaining the desired optimal smoothness factor.

[0083] S3: Train the GRNN neural network, uniformly arranging the geese in the flock in each iteration to calculate the flock fitness value. When the GSO fitness value... E When the set number of iterations is reached, the minimum fitness value is compared with the minimum fitness value in each iteration, the algorithm ends, and the fitness value is saved. E The smoothing factor at the minimum value is finally saved into the training model of the GRNN neural network; when the GSO fitness value... E If the set number of iterations is reached, but the iteration fitness value is not the minimum, then the value is continuously adjusted and optimized through steps S4-S5.

[0084] The function value for calculating fitness is: ;

[0085] In the formula, The number of samples is represented by m, and the number of categories is represented by m. Indicates the first The label value corresponding to the category of each sample. Indicates the first The probability values ​​obtained by applying the Softmax activation function to each sample;

[0086] S4: Calculate the first [number] within the field of view using a simple averaging method. i The individual extreme value of a single goose is taken as the global extreme value perceived by the goose at that moment, that is: ;

[0087] Where k represents the number of iterations. Let represent the individual extreme value of the s-th goose in the k-th iteration, where the positive integer s takes values ​​in the range [1, ..., ...]. i ];

[0088] Then, the individual extreme value of the previous goose is directly used as the individual extreme value of the current goose, that is: ;

[0089] This represents the individual extreme value of the goose in the k-th iteration. This represents the individual extreme value of the previous goose in the k-th iteration.

[0090] S5: Based on the improvements made to S4, the... i The speed of Zhiyan updates is as follows: ;

[0091] No. The formula for updating the position of a single goose is: ;

[0092] in, This is the inertia weight at the k-th iteration, used to balance the global and local convergence capabilities of the algorithm. , Let $i$ be the velocity and position of the i-th goose at the k-th iteration.

[0093] The dimensions of the gossamer are d = 1, 2, ..., N; and These are random numbers that follow a uniform distribution in the range [0, 1]. and The non-negative acceleration learning factor is used to adjust the maximum step size in the directions of the individual extremum and the global extremum of the current flock, respectively; the weighting coefficients... The update formula is: ;

[0094] in and They are respectively The maximum and minimum values ​​of the weighting coefficients, This represents the current iteration number. This represents the maximum number of iterations.

[0095] S6: Determine if the flock of geese meets the termination condition. If it does, proceed to S7; otherwise, return to continue executing S3.

[0096] S7: Set global extremum Each dimension of the algorithm is mapped to a smoothing factor in the GRNN, the network is trained, and the classification result is output.

[0097] The final result obtained through code using MATLAB software is as follows: Figure 4 The image shows the comparison between the classification results obtained from the model built on the test set and the actual results.

[0098] Although the present invention has been described in detail through the preferred embodiments above, it should be understood that the above description should not be considered as a limitation of the present invention. Various modifications and substitutions to the present invention will be apparent to those skilled in the art after reading the above description. Therefore, the scope of protection of the present invention should be defined by the appended claims.

Claims

1. A method for identifying low-resistivity oil reservoirs during drilling, characterized in that, Comprising the following steps: Step 1: Obtain gas logging basic parameters of gas total amount, methane, ethane, propane, n-butane, isobutane, n-pentane and isopentane, and pretreat the data sample: adopt quartile method to eliminate abnormal values for each basic parameter, and fill the abnormal values by nearby interpolation; Step 2: Obtain a drilling parameter table, and correct the gas logging parameters in combination with the drilling parameters; wherein the gas logging parameters include: gas total amount , and alkane volume fraction; Step 3: In order to highlight the differences between oil layers, low-resistance oil layers and water layers, feature extraction is performed on the data pretreated in the step 1 and the corrected data in the step 2, and parameter optimization is performed on the extracted features by using stepwise multiple regression analysis; Step 4: combining the parameters preferably obtained in step 3 with the while-drilling parameter gamma As the input of the model, the GRNN neural network classification model based on the flock optimization algorithm is established to classify the oil layers. Wherein, the steps of establishing the GRNN neural network model based on the flock optimization algorithm are as follows: S1: population parameter GSO initialization, randomly initialize a flock M, set the initial position of the i-th goose in the N-dimensional space as , the speed as , and the individual extreme value and global extreme value of the i-th goose as and ; The parameter GSO refers to the size of the population, and the initial position, speed, individual extreme value and global extreme value of a goose in the population; represents a set of initial positions of the i-th goose in each dimension; denotes a set of individual extrema of the i-th goose in each dimension denotes a set of individual extrema of the i-th goose in each dimension denotes a set of global extrema of the i-th goose in each dimension S2: Map the position of each goose to the GRNN neural network, construct the structure of the GRNN neural network, and initialize the network smoothing factor; S3: training GRNN neural network, the geese in the flock are arranged in each iteration to calculate the fitness value of the flock, and when the GSO fitness value E reaches the set number of iterations, the minimum value of the fitness value of each iteration is compared, the algorithm ends, and the smooth factor when the minimum value of the fitness value is saved E is saved to the training model of the GRNN neural network; when the GSO fitness value E reaches the set number of iterations, but the iteration fitness value is not the minimum value, the optimization is continuously adjusted through steps S4-S5; where the fitness is calculated as the value of the function: ; In the formula, The number of samples is represented by m, and the number of categories is represented by m. Indicates the first The label value corresponding to the category of each sample. Indicates the first The probability values ​​obtained by applying the Softmax activation function to each sample; S4: Calculate the individual extremum of the current goose as the global extremum perceived by the current goose using the simple average method. i Only the individual extremum of the goose is taken as the global extremum perceived by the current goose, i.e. ; where k represents the iteration number, denotes the individual extremum of the s-th goose at the k-th iteration, where the positive integer s has a value range of [1, i ]; Then directly using the previous individual extreme value of the geese as the individual extreme value of the current geese, namely: ; denotes the individual extremum of the current goose at the kth iteration, denotes the individual extremum of the previous goose at the kth iteration; S5: After the improvement according to the step S4, the first The velocity of the only goose is updated as: ; The position update formula for the only goose is: ;​ wherein, is the inertia weight at the kth iteration, used to balance the global and local search ability of the algorithm, , is the velocity and position of the ith goose at the kth iteration, respectively; N is the number of geese; and is a random number obeying uniform distribution in [0, 1]; and is a non-negative acceleration learning factor, respectively used to adjust the maximum moving step length in the direction of the individual extremum and the global extremum of the current flock; the weight coefficient The update formula of is: ; wherein and are respectively the maximum and minimum values of the weight coefficient, is the current iteration number, is the maximum iteration number; S6: Determine whether the flock meets the end condition, if yes, go to S7, otherwise, return to continue executing S3; S7: Set global extremum Each dimension of the algorithm is mapped to a smoothing factor in the GRNN, the network is trained, and the classification result is output. The step 2 combines the drilling parameters to correct the total gas and methane volume fraction in the gas logging data, wherein the correction formula of the methane volume fraction is as follows: ; wherein, represents the corrected methane number, is an empirical constant, is the time while drilling, is the drilling fluid discharge, is the bit diameter, is the equivalent density, is the drilling fluid density; the total gas amount, and the correction formula of the total gas amount is: ; wherein, represents the corrected gas volume, is an empirical constant, , is the drilling fluid flow rate, is the drill bit diameter, is the formation pressure gradient, is the fracture pressure gradient value; In the formula, the empirical constant is quantified considering the influence of the degasser itself and drilling engineering factors, and the quantified formula is as follows: ;​ wherein Qd = represents the degasser drilling fluid flow rate, Qs = represents the sample gas flow rate; The volume fractions of ethane, propane, n-butane, isobutane, n-pentane, isopentane are also corrected according to the methane volume fraction correction formula, and the corrected volume fractions of ethane, propane, n-butane, isobutane, n-pentane, isopentane are respectively , , , , , ; value, representing the corrected propane value, representing the corrected butane value, value, representing the corrected n-pentane value, isopentane value.

2. The method according to claim 1, wherein, The pretreatment of the gas logging historical basic data in the step 1 comprises the following steps: Abnormal values are eliminated by quartile method for each basic parameter, and the abnormal values are filled by nearby interpolation, wherein the nearby interpolation is specifically the following three cases: (1) When <3 and the first data is an outlier, the first data of the first parameter is ; wherein represents the first data of the first parameter, represents the first data of the first parameter; represents the serial number of a certain sample, and the value range is [1, n], wherein n represents the number of samples. (2) when 3 < n-3 and the (n-3)th data is an outlier, the (n-3)th data of the (n-3)th parameter is ;​​​ where n represents the number of samples, the first parameter of the first data, the first parameter of the first data, the first parameter of the first data, the first parameter of the first data; (3) when n-3 < n and the first data is an outlier, then the first data of the first parameter is .

3. The method according to claim 1, wherein, The step 3 carries out feature extraction on the pretreated data and the corrected data, and different feature combinations are carried out on the corrected methane volume fraction , the corrected ethane volume fraction , the corrected propane volume fraction , the corrected n-butane volume fraction , the corrected isobutane volume fraction , the corrected n-pentane volume fraction , the corrected isopentane volume fraction , and the corrected gas total value to highlight the differences among the oil layers, low-resistance oil layers and water layers, wherein the feature parameters include , , , , , , ; wherein , the light component is , and the above components are the heavy component ; In order to reduce the redundancy of the extracted feature parameters and the complexity of the model input, stepwise regression analysis is used to optimize the parameters, and the parameters finally optimized by the method include , , .

4. The method according to claim 1, wherein, The step 4 comprises: The preferred parameters in step 3 are arranged, and a while-drilling parameter gamma is combined The data samples are established, the proportions of each sample are ensured to be uniform, the labels are quantified by using binary coding, and the data is normalized. A GRNN neural network classification model based on a flock optimization algorithm is established, the normalized characteristic parameter values are brought into the model for training, and thus a low-resistivity oil layer classification and prediction model is obtained.

Images