Oil and gas reservoir analogy similarity calculation method based on statistical test

Abstract

The invention relates to the technical field of petroleum data statistical test, and discloses a statistical test-based oil and gas reservoir analogy similarity calculation method, which comprises the following steps of: obtaining a plurality of historical mixed type samples and carrying out data cleaning on the plurality of historical mixed type samples to obtain original mixed type samples, each original mixed type sample comprises original text type geological data and original numerical type geological data; converting the original text type geological data into intermediate numerical type geological data, wherein the intermediate numerical type geological data and the original numerical type geological data jointly form an original full numerical type sample; grouping data fields of the original full numerical type samples to obtain grouped groups and establishing an initial total similarity model; and respectively giving an optimal weight coefficient to the similarity corresponding to each group so as to optimize the initial total similarity model to obtain an optimal total similarity model. According to the method, more accurate reference significance is provided for the recovery ratio and the reserves.

Description

Technical Field

[0001] The invention relates to the technical field of petroleum data statistical inspection, and in particular to a method for calculating oil and gas reservoir analog similarity based on statistical inspection. Background Art

[0002] At present, the industry standard "Methods for Calculating Recoverable Petroleum Reserves" (EN 1014-1-2010 "SY / T 5367-2010") has established corresponding empirical formulas for different types of reservoirs for the evaluation of recoverable reserves in new areas. However, with the continuous deepening of oil field development and the change of development objects, the limitations of the empirical formula method in reserve evaluation have gradually emerged. The empirical formula method establishes a regression formula through multiple samples, but when applied to specific reservoirs, it has the problem of poor adaptability. Secondly, with the advancement of development, some parameter ranges may no longer be applicable. Therefore, in recent years, China has gradually referred to the SEC (Securities and Exchange Commission) reserve assessment rules when calibrating recoverable reserves, and advocated increasing the application of analogy methods in the evaluation of newly-used reserve blocks.

[0003] However, existing analogy methods have difficulties in processing textual data such as reservoir type, drive type, and trap type, while these parameters are of great reference significance to key indicators such as recovery rate and reserves.

[0004] Therefore, how to jointly establish similarity quantification of textual geological data and numerical geological data so as to more accurately evaluate the similarity between oil and gas reservoirs is an urgent problem to be solved. Summary of the invention

[0005] The present invention provides a method for calculating the analog similarity of oil and gas reservoirs based on statistical testing. The method introduces a distribution test model to ensure the scientific nature of the similarity quantification method and has strong versatility and rationality. Through this method, the similarity between oil and gas reservoirs can be more accurately evaluated, and analogous oil and gas reservoirs can be determined based on the similarity to estimate indicators such as recoverable reserves and recovery rate. The present invention helps to improve the accuracy and reliability of the initial evaluation of oil and gas field development, the design and optimization of development plans, and improve the economic benefits and resource utilization of oil field development.

[0006] The present invention is achieved through the following technical solutions:

[0007] A method for calculating oil and gas reservoir analogy similarity based on statistical testing, comprising:

[0008] Acquire a number of historical mixed samples from a general oil and gas reservoir analogy database and perform data cleaning on the historical mixed samples to obtain original mixed samples, each of which includes original text-type geological data and original numerical-type geological data;

[0009] Converting the original text-type geological data into intermediate numerical geological data, wherein the intermediate numerical geological data and the original numerical geological data together constitute an original full numerical sample;

[0010] Grouping the data fields of the original all-numeric samples according to the static parameter categories to obtain grouping groups, and respectively calculating the similarity of each of the original all-numeric samples in each group to establish an initial total similarity model;

[0011] An optimal weight coefficient is respectively given to the similarity corresponding to each group of the groups, so as to optimize the initial total similarity model to obtain an optimal total similarity model.

[0012] As an optimization, the data fields corresponding to each of the original mixed samples include reservoir type, reservoir type, drive type, trap type, original formation pressure, porosity, permeability, gas saturation, gas reservoir medium depth and effective thickness.

[0013] As an optimization, the data fields corresponding to the original text-type geological data include reservoir type, reservoir type, drive type and trap type.

[0014] As an optimization, before the intermediate numerical geological data and the original numerical geological data together form an original full numerical sample, the original numerical geological data is standardized or normalized.

[0015] As an optimization, data cleaning is performed on several of the historical mixed samples to obtain original mixed samples, specifically removing the historical mixed samples containing null values, duplicate values ​​and invalid values, and the remaining historical mixed samples are the original mixed samples.

[0016] As an optimization, the way to convert the original text-type geological data into intermediate numerical geological data is to perform One-Hot Encoding on the original text-type geological data. The specific process is as follows:

[0017] Determine the unique value of the original text-type geological data according to the text name of the original text-type geological data;

[0018] Constructing an index dictionary, wherein the index dictionary stores an integer value corresponding to each unique value of the original text-type geological data;

[0019] Convert the text names corresponding to the original text-type geological data into corresponding integer values ​​through the index dictionary;

[0020] Convert the integer value corresponding to the original text-type geological data into a One-Hot Encoding;

[0021] The original text-type geological data in the original mixed sample is replaced with the corresponding converted One-Hot Encoding, thereby converting the original text-type geological data into intermediate numerical geological data.

[0022] As an optimization, the categories of data fields contained in each of the groups are not less than 2.

[0023] As an optimization, the specific formula of the initial total similarity model is:

[0024]

[0025] Among them, i, j are the numbers of the original all-numeric samples, i, j = 1, 2, ..., N; N is the number of the original all-numeric samples, S i,j is the total similarity between the original all-numeric sample i and the original all-numeric sample j; α k is the weight of the kth group of static parameters; is the similarity between the original all-numeric sample i and the original all-numeric sample j under the kth group of static parameters, and K is the number of groups.

[0026] As an optimization, the specific process of giving each group of the similarities corresponding to the groups an optimal weight coefficient is:

[0027] A1. Using symmetric vector similarity to calculate the similarity of each of the original full-value samples in each group

[0028] A2. Initialize the weight α of each group k Get the initial weight coefficient vector α, α = [α k ], based on the initial weight coefficient vector α, the similarity of each of the original full-value samples in each group is calculated to obtain a similarity matrix S = [S i，j ], i, j = 1, 2, ..., N, a total of N 2 The original fully numerical samples;

[0029] A3, treating the similarity matrix S as an unordered one-dimensional array and performing standardization processing to obtain a standardized similarity sequence S′, Among them, μ S is the mean of the similarity matrix S, σ S is the standard deviation of the similarity matrix S;

[0030] A4. Calculate the empirical distribution function F for the standardized similarity matrix S′ n (s′ m ), the empirical distribution function F n (s′ m ) is used to represent the proportion of data less than or equal to s′ in the original all-numeric sample, s′ m represents the mth variable in the standardized similarity sequence S′;

[0031] A5. Calculate at each standardized sample point s′ m The theoretical standard normal distribution function F(s′ m ), where m is the sample number in S′, m = 1, 2, ..., N 2 ;

[0032] A6. For each of the standardized sample points s′ m , calculate the empirical distribution function F n (s′ m ) and the standard normal distribution function F(s′ m ) to obtain the maximum difference D, D = sup s′ |F n (s′ m )-F(s′ m )|;

[0033] A7, the initial weight coefficient vector α is set as the variable to be solved, and the optimization problem is iteratively optimized using a simulated annealing algorithm, a particle swarm optimization algorithm, or a genetic algorithm to obtain the optimal weight coefficient vector α * , the expression of the optimization problem is:

[0034] α * =armin(D(α)).

[0035] As an optimization, the similarity of each of the original full-value samples in each group The calculation method of is the cosine similarity calculation method or the Pearson coefficient calculation method.

[0036] As an optimization, it also includes converting the newly acquired new mixed sample into a new full numerical sample, and calculating the total similarity between the new full numerical sample and each of the original full numerical samples through the optimal total similarity model, so as to find the original full numerical sample with the highest total similarity with the new full numerical sample from each of the original full numerical samples.

[0037] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0038] 1. The present invention proposes to convert static parameters into numerical types through one-hot encoding, so that the subsequent unified processing of similarity metrics becomes possible, thereby providing a more accurate reference for recovery factor and reserves.

[0039] 2. The present invention groups the given static parameters and assigns different weights to different groups, thereby establishing an interpretable similarity measurement model.

[0040] 3. The present invention is based on the normal distribution assumption, combined with the optimization algorithm, and solves the parameters of the similarity measurement model according to the existing database, and has the characteristics of being data-driven. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0042] Figure 1 Schematic diagram of the single well numerical model;

[0043] Figure 2 Schematic diagram of the permeability of the near-wellbore zone obtained by logging before acidizing;

[0044] Figure 3 Waterfall diagram of DAS response during the whole water injection process (intensity is logarithmic to base 2);

[0045] Figure 4 Waterfall diagram of DAS response at the end of water injection process (intensity logarithm to base 2);

[0046] Figure 5 Optimization of the permeability multiple during the outer layer simulated annealing algorithm (red is the final result);

[0047] Figure 6 The optimized DAS decomposition effect;

[0048] Figure 7 Interpret the water absorption profile obtained for the horizontal section. DETAILED DESCRIPTION

[0049] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0050] The goal of the present invention is to provide a method for establishing a universal oil and gas reservoir analogy database for text and numerical geological data and a similarity quantification method. The text data is encoded into numerical data through One-Hot Encoding to solve the problem of text digitization; the data is reasonably classified into groups such as geological structural background, reservoir physical property parameters, and reservoir reserve parameters, and the similarity is calculated for each group, and different weights are assigned to the similarity of each group. Based on the fact that the weighted similarity should conform to the normal distribution, the solution is obtained by introducing the normal distribution statistical test model and the simulated annealing algorithm. The similarity quantification method established by the present invention is scientific and universal, and can more accurately evaluate the similarity between oil and gas reservoirs, thereby estimating the reserves and recovery rate of unknown oil and gas reservoirs based on analogous oil and gas reservoirs, solving the limitations of existing analogy methods, especially the difficulties in processing text data such as reservoir type, drive type, and closure type, thereby providing a more comprehensive and accurate reserve evaluation.

[0051] The present invention realizes the establishment of an oil and gas reservoir analogy database for text and numerical geological data and a similarity quantification method. The present invention integrates one-hot encoding, statistical test, and simulated annealing algorithm, establishes a universal, quantitative, and interpretable similarity model for oil and gas reservoir analogy, and provides a solution method.

[0052] Next, the process of the present invention is specifically introduced.

[0053] This embodiment 1 provides a method for calculating the oil and gas reservoir analogy similarity based on statistical testing, such as Figure 1 As shown, including:

[0054] S1. Acquire a number of historical mixed samples from a general oil and gas reservoir analogy database and perform data cleaning on the historical mixed samples to obtain original mixed samples, each of which includes original text-type geological data and original numerical-type geological data;

[0055] The (historical mixed type, original mixed type) samples are static data of oil and gas reservoirs. Therefore, this step is the preprocessing of static data of oil and gas reservoirs:

[0056] In some embodiments, the data fields corresponding to each of the original mixed samples include reservoir type, reservoir type, drive type, trap type, original formation pressure, porosity, permeability, gas saturation, gas reservoir depth and effective thickness. Each data field is a static parameter.

[0057] 1. Collect original geological data, which may include but are not limited to fields such as reservoir type, drive type, trap type, original formation pressure, porosity, permeability, etc., forming a table data of size N×F.

[0058] In this embodiment, the data fields corresponding to the original text geological data include reservoir type, reservoir type, drive type and trap type. The original numerical geological data include original formation pressure, porosity, permeability, gas saturation, gas reservoir depth and effective thickness.

[0059] 2. Clean the data, remove null values, duplicate values ​​and invalid data, and ensure the consistency and integrity of the data.

[0060] If it is a text type, you can consider null values ​​as a special text. For table data, if there is a null value in a row, it should be filled or deleted, which is necessary for standardized data cleaning.

[0061] In some embodiments, data cleaning is performed on several of the historical mixed samples to obtain original mixed samples, specifically removing historical mixed samples containing null values, duplicate values, and invalid values, and the remaining historical mixed samples are the original mixed samples.

[0062] S2, converting the original text-type geological data into intermediate numerical geological data, wherein the intermediate numerical geological data and the original numerical geological data together constitute an original full numerical sample;

[0063] In some embodiments, the original text-type geological data is converted into the intermediate numerical geological data by performing One-Hot Encoding on the original text-type geological data (performing One-Hot Encoding for each text-type field), and the specific process is as follows:

[0064] S2.1, determining the unique value of the original text-type geological data according to the text name of the original text-type geological data;

[0065] That is, determine all unique values ​​(categories) in each text field. For example, reservoir types may include sandstone, carbonate rock, etc. "Sandstone" and "Carbonate rock" are the text names of the original text geological data, so "Sandstone" and "Carbonate rock" can also be used as the unique values ​​of the corresponding original text geological data.

[0066] S2.2, constructing an index dictionary, wherein the index dictionary stores an integer value corresponding to each unique value of the original text-type geological data;

[0067] This step is to create a dictionary mapping: assign a unique integer index to each unique category (unique value). Build an index dictionary to map each text category (unique value) to the corresponding integer value.

[0068] S2.3, converting the text names corresponding to the original text-type geological data into corresponding integer values ​​through the index dictionary;

[0069] This step is encoding conversion: use the index dictionary to convert the text data into integer index data, and each text category will be converted into a corresponding integer value.

[0070] S2.4, converting the integer value corresponding to the original text-type geological data into a One-Hot Encoding;

[0071] This step is One-Hot Encoding conversion: convert the integer index data into One-Hot encoding. For each data point, create a vector with a length of all possible categories, set the corresponding category position to 1, and set the remaining positions to 0.

[0072] S2.5. Replace the original text-type geological data in the original mixed sample with the corresponding converted One-Hot Encoding, thereby converting the original text-type geological data into intermediate numerical geological data.

[0073] This step generates a coding matrix: the One-Hot coding vectors of all data points (integer values ​​corresponding to the text names corresponding to the original text-type geological data) are combined into a matrix to form the final coding data set. That is, after converting the text names corresponding to the original text-type geological data into One-Hot codes, the corresponding One-Hot codes are filled into the cells where the text names corresponding to the original text-type geological data are located.

[0074] In some embodiments, before the intermediate numerical geological data and the original numerical geological data are combined to form an original full numerical sample, the original numerical geological data is standardized or normalized, which may be but is not limited to using standard processing procedures such as Min-Max and Standlization.

[0075] S3, grouping the data fields of the original all-numeric samples according to the static parameter categories to obtain grouping groups, and respectively calculating the similarity of each of the original all-numeric samples in each group to establish an initial total similarity model;

[0076] This step is mainly to group static parameters and establish an initial overall similarity model.

[0077] If there is no grouping, only one similarity can be calculated to get the total similarity, and its distribution is fixed, and it is highly likely not a normal distribution. Since multiple similarity components can be calculated after grouping, it is possible to optimize the weight of each group in the total similarity, thereby adjusting the distribution of the total similarity.

[0078] In some embodiments, the categories of data fields included in each of the groups are no less than 2.

[0079] The fully numerical data (original fully numerical samples) processed in step S2 are reasonably classified into K groups, ensuring that there are no less than 2 parameter classes in each group. For reference: for example, ['trap type', 'lithology', 'reservoir space type'] are combined into "geological background parameters", ['porosity', 'permeability', 'original gas saturation'] are combined into "reservoir physical parameters", ['effective thickness', 'original formation pressure'] are combined into "geological quantitative parameters", a total of 3 categories (3 groups, grouped into 3 categories of "geological background parameters", "reservoir physical parameters", and "geological quantitative parameters").

[0080] Construct an initial total similarity model. In some embodiments, the specific formula of the initial total similarity model is:

[0081]

[0082] Among them, i, j are the numbers of the original all-numeric samples, i, j = 1, 2, ..., N; N is the number of the original all-numeric samples, S i,j is the total similarity between the original all-numeric sample i and the original all-numeric sample j; α k is the weight of the kth group of static parameters; is the similarity between the original all-numeric sample i and the original all-numeric sample j under the kth group of static parameters, and K is the number of groups.

[0083] Here, an original fully numerical sample is an oil / gas reservoir data, and an oil / gas reservoir data includes but is not limited to fields such as reservoir type, drive type, trap type, original formation pressure, porosity, and permeability.

[0084] S4. Assign an optimal weight coefficient to each of the similarities corresponding to the groups, thereby optimizing the initial total similarity model to obtain an optimal total similarity model.

[0085] In some embodiments, an optimal weight coefficient α is given to the similarity corresponding to each group, and the Kolmogorov-Smirnov statistic is calculated. The specific process is:

[0086] A1. Using symmetric vector similarity to calculate the similarity of each of the original full-value samples in each group

[0087] Here, it can be cosine similarity, Pearson coefficient, etc. That is, the similarity of each of the original full-value samples in each group The calculation method of is the cosine similarity calculation method or the Pearson coefficient calculation method.

[0088] A2. Initialize the weight α of each group k Get the initial weight coefficient vector α, α = [α k ], based on the initial weight coefficient vector α, the similarity of each of the original full-value samples in each group is calculated to obtain a similarity matrix S = [S i，j ], i, j = 1, 2, ..., N, a total of N 2 The original fully numerical samples;

[0089] S is the NxN similarity matrix obtained by calculating the similarity between each pair of N samples.

[0090] This step calculates the weighted similarity matrix S = [S i，j ], i, j=1, 2,...,N.

[0091] A3, treating the similarity matrix S as an unordered one-dimensional array and performing standardization processing to obtain a standardized similarity sequence S′, Among them, μ S is the mean of the similarity matrix S, σ S is the standard deviation of the similarity matrix S;

[0092] A4. Calculate the empirical distribution function F for the standardized similarity sequence S′ n (s′ m ), the empirical distribution function F n (s′ m ) is used to represent the proportion of data less than or equal to s′ in the original all-numeric sample, s′ m represents the mth variable in the standardized similarity sequence S′;

[0093] A5. Calculate at each standardized sample point s′ m The theoretical standard normal distribution function F(s′ m ), where m is the sample number in S′ (the number of the original fully numerical sample after standardization), m = 1, 2, ..., N 2 ;

[0094] F(s′ m ) for specific formula refer to standard normal distribution function.

[0095] A6. For each of the standardized sample points s′ m , calculate the empirical distribution function F n (s′) and the standard normal distribution function F(s′ m) to obtain the maximum difference D, D = sup s′ |F n (s′ m )-F(s′ m )|;

[0096] The maximum difference D is the Kolmogorov-Smimov statistic.

[0097] A7, the initial weight coefficient vector α is set as the variable to be solved, and the optimization problem is iteratively optimized using a simulated annealing algorithm, a particle swarm optimization algorithm, or a genetic algorithm to obtain the optimal weight coefficient vector α * , the expression of the optimization problem is:

[0098] α * =armin(D(α)).

[0099] That is, α is set as the variable to be solved, and the simulated annealing algorithm (optionally, particle swarm optimization algorithm, genetic algorithm can also be used) is used for the following optimization problem. k Random perturbation is performed, and the search range of each element is set in the interval [10 - 3 1e-, 1], repeat steps A1-A6 to perform iterative optimization solution:

[0100] α * =armin(D(α)) to obtain the optimal weight coefficient vector α for each group * , so that in a given N samples, the calculated N 2 The Kolmogorov-Smimov statistic of the similarity sequence S composed of the total similarities between each pair is the smallest. In this case, the similarity sequence S is closest to the normal distribution.

[0101] The optimal total similarity model is as follows:

[0102]

[0103] In some embodiments, it also includes, S5, converting the newly acquired new mixed sample into a new full numerical sample, and calculating the total similarity between the new full numerical sample and each of the original full numerical samples through the optimal total similarity model, so as to find the original full numerical sample with the highest total similarity with the new full numerical sample from each of the original full numerical samples.

[0104] Based on α * , is the optimal weight coefficient of the kth group. When an unknown sample is given, its N total similarities with the N samples (original full numerical samples) of the known database in step S2 are calculated based on the optimal total similarity model:

[0105]

[0106] X i Represents the total similarity between the unknown sample and the i-th original full-numeric sample.

[0107] X i The greater the total similarity, the more suitable it is as an analog oil and gas reservoir for unknown samples, in order to refer to indicators such as reserves and recoverable rate.

[0108] Next, the present invention is described by taking a gas reservoir database as an example.

[0109] 1. Data preprocessing

[0110] The database (general oil and gas reservoir analogy database) contains 342 samples in total, such as Figure 2 As shown in the figure, it includes reservoir lithology, reservoir type, trap type, drive type, porosity, permeability, gas saturation, medium depth of gas reservoir, effective thickness, and original formation pressure, a total of 10 features. Among them, the first 4 features are text data, and the rest are numerical data.

[0111] Follow step A to obtain the full numeric table data after one-hot encoding.

[0112] 2. Group static parameters and establish a total similarity model:

[0113] Taking into account the attributes of the given data, the full numerical data tables are rationally grouped into the following three groups: [‘trap type’, ‘lithology’, ‘reservoir space type’] are combined as “geological background parameters”, [‘porosity’, ‘permeability’, ‘original gas saturation’] are combined as “reservoir physical parameters”, and [‘effective thickness’, ‘original formation pressure’] are combined as “geological quantitative parameters”.

[0114] Follow step A to obtain the full numeric table data after one-hot encoding.

[0115] The similarity model is constructed as:

[0116]

[0117] Where i, j are the oil / gas reservoir numbers, i, j = 1, 2, ..., N; S i，j is the total similarity between oil / gas reservoir i and oil / gas reservoir j; α k is the weight of the kth group of static parameters; is the similarity between oil / gas reservoir i and oil / gas reservoir j under the kth set of parameters; K=3.

[0118] 3. Solving α based on statistical tests

[0119] a. Use symmetric vector similarity to calculate the similarity between samples in each group using the initial total similarity model As an example, cosine similarity is used here. According to the given α, the weighted similarity sequence S between each sample is calculated = [S i，j ], i, j = 1, 2, ..., N, a total of N 2 samples.

[0120] b. Standardize the similarity sequence S to obtain Among them, μS is the mean of sequence S, σ S is the standard deviation of S.

[0121] c. Calculate the empirical distribution function F for S′ n (s′) is expressed as the proportion of samples that are less than or equal to s′.

[0122] d. Consider the standard normal distribution and calculate s′ at each standardized sample point m The theoretical distribution function value F(s′) m ), where m is the sample number in S′, m = 1, 2, ..., N 2 .

[0123] e. For each standardized sample point s′ m , calculate the empirical distribution function F n (s′) and the standard normal distribution function F(s′ m ) is the difference between .

[0124] f. Get the Kolmogorov-Smimov statistic, that is, the maximum difference:

[0125]

[0126] g. Set α as the variable to be solved, use the simulated annealing algorithm (optionally, the particle swarm optimization algorithm or the genetic algorithm) for the following optimization problem, perform random perturbations on the variable α, set the search range of each element in the interval [1e-3, 1], and use α as the initial solution # =[1.0, 1.0, 1.0], the distribution based on the initial solution is as follows Figure 3 As shown, it is obviously far from the normal distribution. Repeat steps a to f and perform iterative optimization to solve:

[0127] α * =armin(D(α))

[0128] That is, to obtain the optimal weight coefficient vector α * , so that in a given N samples, the calculated N 2 The Kolmogorov-Smirnov statistic of the similarity sequence S composed of the total similarities between each pair is the smallest. In this case, S is closest to the normal distribution.

[0129] Through iterative solution, it finally converges to:

[0130] α * =[0.79222765, 0.70382911, 0.21127358]

[0131] According to the optimal weight coefficient vector α * The calculated similarity distribution is as follows Figure 4 As shown, it can be seen that it is very close to the normal distribution and achieves the expected effect.

[0132] 4. Calculate the similarity and make analogies for unknown gas reservoir samples:

[0133] Given static geological parameters used as samples of unknown gas reservoirs (such as Figure 5 As shown in the figure, the same data preprocessing (including one-hot encoding, standardization or normalization, etc.) is performed as with the training database.

[0134] Based on the optimized vector and the established total similarity measurement model, the similarity between the sample and N = 342 samples in the database is calculated. The similarity distribution is as follows: Figure 6 As shown, it is also closer to the normal distribution and is more ideal.

[0135] Among them, the five database samples with the closest similarity to the unknown sample are as follows: Figure 7 As shown in the figure, there are three gas reservoirs that can be used as analogs, and their sample similarity exceeds 85%. The recovery rates of these three gas reservoirs are between 66% and 70%. Based on this, it can be reasonably inferred that the recovery rate of the unknown gas reservoir should be 66% to 70%. The weighted recovery rate can also be calculated based on the similarity of these three gas reservoirs as weights.

[0136] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for calculating oil and gas reservoir analogy similarity based on statistical testing, characterized in that: include: Acquire a number of historical mixed samples from a general oil and gas reservoir analogy database and perform data cleaning on the historical mixed samples to obtain original mixed samples, each of which includes original text-type geological data and original numerical-type geological data; Converting the original text-type geological data into intermediate numerical geological data, wherein the intermediate numerical geological data and the original numerical geological data together constitute an original full numerical sample; Grouping the data fields of the original all-numeric samples according to the static parameter categories to obtain grouping groups, and respectively calculating the similarity of each of the original all-numeric samples in each group to establish an initial total similarity model; An optimal weight coefficient is respectively given to the similarity corresponding to each group of the groups, so as to optimize the initial total similarity model to obtain an optimal total similarity model.

2. The method for calculating oil and gas reservoir analogy similarity based on statistical testing according to claim 1, characterized in that: The data fields corresponding to each of the original mixed-type samples include reservoir type, reservoir type, drive type, trap type, original formation pressure, porosity, permeability, gas saturation, medium depth of gas reservoir and effective thickness, and the data fields corresponding to the original text-type geological data include reservoir type, reservoir type, drive type and trap type.

3. The method for calculating oil and gas reservoir analogy similarity based on statistical testing according to claim 1, characterized in that: Before the intermediate numerical geological data and the original numerical geological data together form an original full numerical sample, the original numerical geological data is standardized or normalized.

4. The method for calculating oil and gas reservoir analogy similarity based on statistical testing according to claim 1, characterized in that: Data cleaning is performed on several of the historical mixed samples to obtain original mixed samples, specifically removing the historical mixed samples containing null values, duplicate values ​​and invalid values, and the remaining historical mixed samples are the original mixed samples.

5. The method for calculating oil and gas reservoir analogy similarity based on statistical testing according to claim 1, characterized in that: The method of converting the original text-type geological data into intermediate numerical geological data is to perform One-Hot Encoding on the original text-type geological data. The specific process is as follows: Determine the unique value of the original text-type geological data according to the text name of the original text-type geological data; Constructing an index dictionary, wherein the index dictionary stores an integer value corresponding to each unique value of the original text-type geological data; Convert the text names corresponding to the original text-type geological data into corresponding integer values ​​through the index dictionary; Convert the integer value corresponding to the original text-type geological data into a One-Hot Encoding; The original text-type geological data in the original mixed sample is replaced with the corresponding converted One-HotEncoding, thereby converting the original text-type geological data into intermediate numerical geological data.

6. The method for calculating oil and gas reservoir analogy similarity based on statistical testing according to claim 1, characterized in that: Each of the groups contains no less than 2 categories of data fields.

7. The method for calculating oil and gas reservoir analogy similarity based on statistical testing according to claim 1, characterized in that: The specific formula of the initial total similarity model is: Among them, i, j are the numbers of the original all-numeric samples, i, j = 1, 2, ..., N; N is the number of the original all-numeric samples, S i,j is the total similarity between the original all-numeric sample i and the original all-numeric sample j; α k is the weight of the kth group of static parameters; is the similarity between the original all-numeric sample i and the original all-numeric sample j under the kth group of static parameters, and K is the number of groups.

8. The method for calculating oil and gas reservoir analogy similarity based on statistical testing according to claim 1, characterized in that: The specific process of giving each group of the similarities corresponding to the groups an optimal weight coefficient is: A1. Using symmetric vector similarity to calculate the similarity of each of the original full-value samples in each group A2. Initialize the weight α of each group k Get the initial weight coefficient vector α, α = [α k ], based on the initial weight coefficient vector α, the similarity of each of the original full-value samples in each group is calculated to obtain a similarity matrix S = [S i,j ],i,j=1,2,…,N, total N 2 The original fully numerical samples; A3, treating the similarity matrix S as an unordered one-dimensional array and performing standardization processing to obtain a standardized similarity sequence S′, Among them, μ S is the mean of the similarity matrix S, σ S is the standard deviation of the similarity matrix S; A4. Calculate the empirical distribution function F for the standardized similarity sequence S′ n (s′ m ), the empirical distribution function F n (s′ m ) is used to indicate that in the original full-value sample, it is less than or equal to s′ m The proportion of data, s′ m represents the mth variable in the standardized similarity sequence S′; A5. Calculate at each standardized sample point s′ m The theoretical standard normal distribution function F(s′ m ), where m is the sample number in S′, m=1,2,…,N 2 ; A6. For each of the standardized sample points s′ m , calculate the empirical distribution function F n (s′ m ) and the standard normal distribution function F(s′ m ) to obtain the maximum difference D, D = sup s′ |Fn(s′ m )-F(s′ m )|; A7, the initial weight coefficient vector α is set as the variable to be solved, and the optimization problem is iteratively optimized using a simulated annealing algorithm, a particle swarm optimization algorithm, or a genetic algorithm to obtain the optimal weight coefficient vector α * , the expression of the optimization problem is: a * = armin(D(α)).

9. The method for calculating oil and gas reservoir analog similarity based on statistical testing according to claim 7 or 8, characterized in that: The similarity of each of the original full-value samples in each group The calculation method of is the cosine similarity calculation method or the Pearson coefficient calculation method.

10. The method for calculating oil and gas reservoir analogy similarity based on statistical testing according to claim 8, characterized in that: It also includes converting the newly acquired new mixed sample into a new full numerical sample, and calculating the total similarity between the new full numerical sample and each of the original full numerical samples through the optimal total similarity model, so as to find the original full numerical sample with the highest total similarity to the new full numerical sample from each of the original full numerical samples.

Images