A new drilling potential evaluation method based on hybrid feature selection

Abstract

The application discloses a new drilling potential evaluation method based on a mixed feature selection, and is characterized in that the number of newly produced wells in a historical interval is calculated, and reservoir development history data is marked; the correlation influence of each feature in the reservoir development history data is analyzed by using a Pearson product-moment correlation coefficient, individual fitness values are calculated, an elite selection mechanism is adopted to add individuals with larger fitness values and individuals generated after random selection and crossover to a sub-population, the population number is kept stable, and iteration is performed until an optimal feature combination of the oilfield history data is obtained; a support vector regression machine is used to mine deep information of the oilfield history data, and new drilling potential evaluation is realized. The application calculates the correlation influence of features by using the Pearson product-moment correlation coefficient, guides the selection of the optimal feature combination of the oilfield history data, speeds up the execution speed of the feature selection, adds the elite selection mechanism, increases the search precision of the feature combination of the oilfield history data, and improves the accuracy of the new drilling potential evaluation.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of geophysical exploration and artificial intelligence, and particularly relates to a new drilling potential evaluation method based on mixed feature selection. BACKGROUND

[0002] Analyzing the data information of the developed oilfield history and evaluating the new drilling potential of the oilfield can guide the new well production status in the next few years, and significantly improve the recovery rate of the old oilfield and expand the utilization rate of the proven resources. The historical data related to the new drilling potential evaluation of the oilfield are numerous and complex in relation, and a single attribute cannot reflect the change law of the new drilling potential, and the redundancy between multiple attributes cannot be directly used. Using attribute analysis technology can effectively analyze the mutual relationship between multiple attributes, extract a feature combination with high correlation degree to the new drilling potential evaluation, and then use multiple attributes to reflect the change law of the new drilling potential.

[0003] The traditional new drilling potential evaluation generally only considers several related factors such as cumulative oil production for estimation, and it is difficult to fully consider many factors affecting the new drilling potential evaluation. In the complex geological conditions, the evaluation result has deviation. With the rapid development of computer technology, it has become more and more common to introduce the related knowledge of feature selection to analyze the correlation between attributes. Through feature selection of the oilfield development history data, the oilfield development history data features irrelevant to the new drilling potential evaluation and redundant are removed, and the data features with high correlation degree are retained. In full consideration of many factors affecting the new drilling potential evaluation, the support vector regression machine method is used to mine the deep information of the oilfield development history data, and the new drilling potential evaluation is realized. SUMMARY

[0004] In order to overcome the problem that the traditional new drilling potential evaluation method cannot fully consider the influencing factors, resulting in deviation of the evaluation result, the present application proposes a new drilling potential evaluation method based on mixed feature selection. The filter selection operator is used to analyze the influence of each dimension feature of the oilfield development history data on the correlation of the new drilling potential evaluation, to guide the selection of the optimal feature combination of the oilfield history data by the genetic algorithm, and to add an elite selection mechanism to increase the search precision of the feature combination of the oilfield history data with high correlation degree. Combined with the support vector regression machine for deep information mining, the new drilling potential evaluation is realized.

[0005] To achieve the above purpose, the technical scheme of the present application mainly includes the following steps:

[0006] The reservoir development history data and the block new production well data are extracted from the seismic database, reservoir history database and other data sources. According to the number of new production wells in the historical interval in the block new production well data, the block new drilling potential evaluation value is calculated. The reservoir history data and the block new drilling potential evaluation value are combined, and after preprocessing, the initial data set of the new drilling potential with labels is obtained.

[0007] A. The optimal feature combination is extracted by using a hybrid feature selection method:

[0008] (1) The correlation degree of each dimension feature in the initial data set of the new drilling potential to the influence of the new drilling potential evaluation is calculated by using the Pearson product-moment correlation coefficient, which is denoted as r value. The r value is between-1 and 1, and the greater the absolute value, the stronger the correlation. According to the Pearson product-moment correlation coefficient, a weight is assigned to each dimension attribute, and the weight is proportional to the correlation coefficient;

[0009] (2) The feature dimension in the initial data set of the new drilling potential is N, the population size in the genetic algorithm is m, the mutation probability is P, the maximum iteration number is t, and the chromosome X with a length of D is used to encode the feature combination, wherein the encoding 0 indicates that the feature at the corresponding position in the reservoir history data is selected, and the encoding 1 indicates that the feature at the corresponding position in the reservoir history data is not selected. The weighted attributes are extracted by using the roulette wheel method to generate the initial population S. The feature combination represented by the individual X is taken as the input, and the modified value MAE(X) of the average absolute error of the evaluation result of the deep belief network is taken as the fitness function F;

[0010] (3) The fitness values F of all individuals in the population are calculated i The individuals in the population are sorted in descending order according to their fitness values. The individuals with larger fitness values in the population are selected to form a sub-population A, and the individuals with smaller fitness values in the population are selected to form a sub-population B;

[0011] (4) The selected population A and population B are crossed to form a population C by using single-point crossover, and the individuals in A and C are mixed to generate the offspring population S';

[0012] (5) The individuals in the population are randomly selected according to the mutation probability P for mutation operation;

[0013] (6) Steps (3)-(5) are repeated until the maximum iteration number is reached or the threshold is reached. The individual X with the largest fitness value is selected max , and the chromosome encoding value corresponding to X max is the optimal feature combination of the oilfield history data. The corresponding features are extracted according to the optimal feature combination of the oilfield history data to obtain the optimal data set of the new drilling potential.

[0014] B. The support vector regression machine is used to evaluate the new drilling potential

[0015] (1) The optimal dataset of new drilling potential is randomly divided into a training sample set, a test sample set and a validation sample set, with a ratio of 7:2:1.

[0016] (2) Use the support vector regression machine method to mine the deep information of the training sample set data until the effect is stable and converged. Use the mean absolute error as the evaluation standard to test the evaluation effect of the support vector regression machine on the test sample set.

[0017] (3) Retain the validation sample set, re-divide the training sample set and the test sample set, and repeat the process in (2). Compare the evaluation results of multiple support vector regression machines, retain the support vector regression machine with the smallest mean absolute error, process the data of the block to be evaluated using the support vector regression machine, and realize the evaluation of new drilling potential.

[0018] The beneficial effects of this invention are: the filtering selection operator calculates the correlation influence of each feature on the evaluation of new drilling potential, accelerates the execution speed of feature selection, and guides the selection of the optimal feature combination of oilfield historical data; the addition of an elite selection mechanism to the genetic algorithm retains a portion of individuals with higher fitness values ​​in each iteration, increases the search accuracy for feature combinations of oilfield historical data with high correlation, and increases the accuracy of new drilling potential evaluation. Attached Figure Description

[0019] Figure 1 This is a model structure diagram of the present invention. Detailed Implementation

[0020] The following is combined Figure 1 The present invention will be described in further detail as follows:

[0021] Historical reservoir development data and newly commissioned well data for the block were extracted from data sources such as seismic databases and reservoir history databases. Historical reservoir development data includes reservoir geological parameters and development status data. Based on the number of newly commissioned wells within a historical period in the newly commissioned well data for the block, the new drilling potential evaluation value for the block was labeled. The historical reservoir data and the new drilling potential evaluation values ​​for the block were merged, and after preprocessing, an initial dataset of labeled new drilling potential was obtained.

[0022] A. Extracting the optimal feature combination using a hybrid feature selection method:

[0023] (1) Feature correlation pre-analysis

[0024] The correlation between the central features of the initial dataset of new drilling potential and the evaluation of new drilling potential is calculated using the Pearson product-moment correlation coefficient, denoted as the r-value. The r-value ranges from -1 to 1, with a larger absolute value indicating a stronger correlation. Each attribute is assigned a weight based on the absolute value of the Pearson product-moment correlation coefficient, with the weight proportional to the correlation coefficient. The formula for calculating the r-value is:

[0025]

[0026] Where Mi is the eigenvalue. Let Yi be the characteristic mean, and Y be the corresponding new drilling potential evaluation value.

[0027] (2) Population initialization

[0028] The reservoir history data has a feature dimension of N. The genetic algorithm uses a population size of m, a mutation probability of P, and a maximum number of iterations of t. A chromosome X of length D is used to encode the feature combinations, where a code of 0 indicates that the feature at the corresponding position in the reservoir history data is selected, and a code of 1 indicates that the feature at the corresponding position in the reservoir history data is not selected. A roulette wheel approach is used to extract weighted attributes. The initial population is S = {X1, X2, ..., X...}. m The feature combination represented by individual X is used as input, and the corrected value of the mean absolute error (MAE(X)) of the evaluation results of the deep belief network is used as the fitness function F;

[0029] (3) Select operation

[0030] Calculate the fitness value F of all individuals in the population. i F i =MAE(X i Given f = 1, 2, ..., m, sort the individuals in the population in descending order of their fitness values ​​{X′1, X′2, ..., X′}. m}, X′ i Let represent the individual with rank i. Select half of the individuals with the higher fitness values ​​from the population to form a subpopulation A = {X′1, X′2, ..., X′}. m / 2}, the half of the individuals with the smaller fitness values ​​constitute the subpopulation B={Y′1,Y′2,…,Y′ m / 2};

[0031] (4) Cross operation

[0032] The crossover operation, based on a set crossover probability, exchanges partial genes between any two chromosomes. This invention retains the better individuals from each iteration as the next generation population during the iteration process, and selects the subpopulations of the worse individuals and the subpopulations of the best individuals for crossover operations. The newly generated individuals are then combined with the subpopulations of the better individuals to form a new population, thereby improving the algorithm's global search capability while retaining the better individuals.

[0033] The specific steps are as follows:

[0034] Using a single-point crossover method, crossover operations are performed on selected populations A and B to form population C = {Z′1, Z′2, ..., Z′}. m / 2}, mix the individuals from A and C together to generate a progeny population S' = {X′1, X′2, ..., X′}. m / 2 Z′1, Z′2, ... Z′ m / 2};

[0035] (5) Mutation operation

[0036] The mutation operation is based on a set mutation probability to mutate a chromosome or a segment of a gene on a chromosome. This invention causes an individual to mutate according to the mutation probability P, forming a new individual.

[0037] (6) Repeat steps (3)-(5) until the maximum number of iterations is met or the set threshold is reached, indicating that the population has stabilized and is no longer evolving. Stop the operation. Select the individual X with the highest fitness value. max X max The corresponding chromosome coding value is the optimal feature combination of the oilfield's historical data. The corresponding features are extracted according to the optimal feature combination of the oilfield's historical data to obtain the optimal dataset for new drilling potential.

[0038] B. Using support vector regression to evaluate the potential of new drilling wells.

[0039] (1) The optimal dataset of new drilling potential is randomly divided into training sample set, test sample set and validation sample set in blocks, with a ratio of 7:2:1.

[0040] (2) Use support vector regression machine to mine the deep information of the training sample set data until the effect is stable and converged. Use the mean absolute error as the evaluation standard to test the performance of support vector regression machine on the test sample set.

[0041] (3) Retain the validation sample set, re-divide the training sample set and the test sample set, and repeat the process in (2). Compare the evaluation results of multiple support vector regression machines, retain the support vector regression machine with the smallest mean absolute error, process the validation sample set with the support vector regression machine, and realize the evaluation of the new drilling potential of the block.

[0042] The above description is merely a preferred embodiment of the present invention. Any person skilled in the art may modify or alter the above-described technical solutions to create equivalent examples. Any simple modifications, alterations, or modifications made to the above embodiments based on the technical solutions of the invention without departing from the scope of the invention's technical solutions shall fall within the protection scope of the invention's technical solutions.

Claims

1. A novel drilling potential evaluation method based on hybrid feature selection, characterized in that, Includes the following steps: Historical reservoir development data and new production well data for blocks are extracted from data sources such as seismic databases and reservoir history databases. Historical reservoir development data includes reservoir geological parameters and development status data. Based on the number of new production wells in the historical interval in the new production well data of blocks, the new drilling potential evaluation value of blocks is marked. The historical reservoir data and the new drilling potential evaluation value of blocks are merged and preprocessed to obtain the initial dataset of new drilling potential with labels. A. Use a hybrid feature selection method to extract the optimal feature combination; (1) Feature correlation pre-analysis The correlation between the central features of the initial dataset of new drilling potential and the evaluation of new drilling potential is calculated using the Pearson product-moment correlation coefficient, denoted as the r-value. The r-value ranges from -1 to 1, with a larger absolute value indicating a stronger correlation. Each attribute is assigned a weight based on the absolute value of the Pearson product-moment correlation coefficient, with the weight proportional to the correlation coefficient. The formula for calculating the r-value is as follows: Where Mi is the eigenvalue. For special The mean value, Yi is the corresponding new drilling potential evaluation value, and Y is the mean value of the new drilling potential evaluation; (2) Population initialization The feature dimension of the reservoir history data is N. The number of populations in the genetic algorithm is set to m, the mutation probability is P, and the maximum number of iterations is t. The feature combinations are encoded using chromosomes X of length D, where 0 indicates that the feature at the corresponding position in the reservoir history data is selected, and 1 indicates that the feature at the corresponding position in the reservoir history data is not selected. The weighted attributes are extracted using a roulette wheel method. The population S = {X1, X2, ..., Xm} is initialized. The feature combination represented by individual X is used as input, and the corrected value of the mean absolute error of the evaluation results of the deep belief network, MAE(X), is used as the fitness function F. (3) Select operation Calculate the fitness value F of all individuals in the population. i F i =MAE(X i Given f = 1, 2, ..., m, sort the individuals in the population in descending order of their fitness values ​​{X′1, X′2, ..., X′}. m }, X′ i This represents the individual with rank number i. The subpopulation A = {X′1, X′2, ..., X′} is formed by selecting half of the individuals with the higher fitness values ​​from the population. m / 2}, the half of the individuals with the smaller fitness values ​​constitute the subpopulation B={Y′1,Y′2,…,Y′ m / 2}; (4) Cross operation The crossover operation uses the set crossover probability as a basis to exchange some genes of any two chromosomes. In the iteration process, this invention retains the better individuals in the current iteration as the next generation population, and selects the subpopulation of the worse individuals and the subpopulation of the best individuals to perform crossover operation. The new individuals generated by the crossover and the subpopulation of the better individuals are then combined to form a new population, thereby improving the global search capability of the algorithm while retaining the better individuals. Specific steps: Using a single-point crossover method, crossover operations are performed on the selected population A and population B to form population C = {Z′1, Z′2, ..., Z′}. m / 2}, mix the individuals from A and C together to generate a offspring population S'={X′1,X′2,…X′ m / 2, Z′1, Z′2, ... Z′ m / 2}; (5) Mutation operation The mutation operation is based on a set mutation probability to mutate a chromosome or a segment of a gene on a chromosome. This invention causes an individual to mutate according to the mutation probability P, forming a new individual. (6) Repeat steps (3)-(5) until the maximum number of iterations is met or the set threshold is reached, indicating that the population has stabilized and is no longer evolving. Stop the operation and select the individual X with the highest fitness value. max X max The corresponding chromosome coding value is the optimal feature combination of historical oilfield data. The corresponding features are extracted according to the optimal feature combination of historical oilfield data to obtain the optimal dataset for new drilling potential. B. Using support vector regression to evaluate the potential of new drilling wells. (1) The optimal dataset of new drilling potential is randomly divided into training sample set, test sample set and validation sample set in blocks, with a ratio of 7:2:

1. (2) Use support vector regression machine to mine the deep information of the training sample set data until the effect is stable and converged. Use the mean absolute error as the evaluation standard to test the performance of support vector regression machine on the test sample set. (3) Retain the validation sample set, redivide the training sample set and the test sample set, repeat the process (2), compare the evaluation effects of multiple support vector regression machines, retain the support vector regression machine with the smallest mean absolute error, process the support vector regression machine to evaluate the block data to be evaluated, and realize the evaluation of new drilling potential.

Images