Random forest-based effective crack identification method, apparatus and device, and medium

Abstract

The invention discloses an effective crack identification method and device based on a random forest, equipment and a medium. The method comprises the following steps: establishing a crack effectiveness evaluation standard based on logging information, and determining a crack effectiveness index; establishing a fracture prediction data set according to the seismic attributes representing the fractures of different scales and the inversion results, and labeling by using the effective indexes of the fractures; training a random forest effective fracture prediction model, and predicting an unknown fracture effective index; and based on the trained random forest effective fracture prediction model, predicting three-dimensional whole-area effective fractures. According to the method, the improved random forest model is built on the basis of logging fracture effectiveness evaluation and supported by earthquake multi-attribute analysis, effective fracture automatic identification is achieved, and reference is provided for reservoir effective fracture evaluation.

Description

Technical Field

[0001] The present invention relates to the field of petroleum exploration technology, and more specifically, to an effective crack identification method, device, equipment and medium based on random forest. Background Art

[0002] Tight sandstone reservoirs are widely distributed in continental basins. They have large oil and gas reservoir areas and rich actual geological reserves, making them an important area for increasing production and reserves. The structural pattern of the tight sandstone reservoirs of the Xujiahe Formation in the Sichuan Basin is complex, and the law of fracture development is unclear. The development and connectivity of fractures is one of the key factors for the high yield of tight sandstone reservoirs. It is particularly important to obtain accurate fracture distribution laws and development degrees. The fractures in the area have the characteristics of complex formation process, diverse types, large scale differences, and strong heterogeneity. Therefore, the precise prediction of fractures is a major difficulty in oil and gas exploration and development in this area. Usually, fracture prediction methods are roughly divided into two aspects. One is to make full use of outcrop, core and imaging logging data to evaluate single well fractures. This method can accurately determine the fracture development and fracture type, and is the basis for conducting logging fracture effectiveness evaluation. However, the logging method only represents single-point information, and it is difficult to evaluate the spatial distribution of fracture development in the area. Second, using single seismic attributes such as coherence, curvature, ant body or anisotropic inversion to predict fractures can evaluate the development of fractures in the entire area, but due to the limitations of the method itself and the quality of seismic data, fracture prediction often produces multiple solutions and limited accuracy. In summary, it is difficult to accurately characterize the development degree and spatial distribution of effective fractures using only conventional seismic attribute technology, which restricts the exploration and evaluation of reservoirs in this area.

[0003] At present, there is still a need to develop an effective crack identification method, device, equipment and medium based on random forest.

[0004] The information disclosed in the background technology section of the present invention is only intended to deepen the understanding of the general background technology of the present invention, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art already known to those skilled in the art. Summary of the invention

[0005] The present invention proposes an effective fracture identification method, device, equipment and medium based on random forest. Based on the evaluation of fracture effectiveness by logging and supported by seismic multi-attribute analysis, an improved random forest model is constructed to realize automatic identification of effective fractures, thus providing a reference for the evaluation of effective fractures in reservoirs.

[0006] In a first aspect, an embodiment of the present disclosure provides an effective crack identification method based on random forest, comprising:

[0007] Based on the logging data, the fracture effectiveness evaluation standard is established and the fracture effectiveness index is determined;

[0008] A fracture prediction data set is established based on a plurality of seismic attributes and inversion results characterizing fractures of different scales, and a label is annotated with the fracture effectiveness index;

[0009] Training a random forest effective crack prediction model to predict the unknown crack effectiveness index;

[0010] Based on the trained random forest effective crack prediction model, effective cracks in the entire three-dimensional area are predicted.

[0011] Preferably, based on the well logging data, the fracture effectiveness evaluation criteria are established including:

[0012] Based on the analysis of fracture logging response characteristics, a fracture effectiveness index is established to evaluate fracture effectiveness:

[0013] FIND=a×RDB_FI+b×ACSH_FI+C×RDS_FI (1)

[0014] Among them, FIND is the fracture effectiveness index, RDB_FI is the absolute value decrease of resistivity, ACSH_FI is the increase of acoustic time difference, RDS_FI is the resistivity invasion amplitude difference, and a, b, and c are regional empirical parameters determined by the imaging logging scale.

[0015] Preferably, the random forest effective crack prediction model uses multiple CART decision trees as basic units, and multiple decision trees are constructed by random repeated sampling technology and node random splitting technology. The key parameters are obtained by using the sparrow search algorithm, and the predicted crack effectiveness index is output. The key parameters include the number of decision trees and the maximum number of features.

[0016] Preferably, the CART decision tree determines the segmentation scheme based on the minimum joint impurity J value of the binary space samples. s,t For J:

[0017] J(R s,t )=P(x∈X S )·I(X S )+P(x∈X T )·I(X T ) (2)

[0018] Here, I is the impurity measure, expressed as the variance of the predictor variables in X:

[0019]

[0020] Among them, E(y|x∈X) represents the conditional expectation that the input is y when the input x is given.

[0021] Preferably, after the decision tree recursive optimization growth is completed, the entire training set is divided into M mutually exclusive subsets X1, X2...X M , that is, the decision tree has M leaf nodes;

[0022] When the sample x* is brought into this decision tree, a leaf node X is obtained according to the logical rules of the branch nodes in the tree. l , at this point, the conditional expectation of y on x* is calculated as the result of the decision tree regression prediction:

[0023]

[0024] Where N represents the number of samples in the sample set, I(x i ∈X l ) is a conditional function, when x i ∈X l If true, I takes the value 1, otherwise it takes the value 0.

[0025] Preferably, the sum of the errors between the training set and the test set of the random forest effective crack prediction model is used as the objective function F, and the model parameters that minimize F can be iteratively found using the sparrow search algorithm. Training the random forest effective crack prediction model includes:

[0026] ① Initialize the population, determine the number of iterations, and the ratio of discoverers to joiners;

[0027] ② Randomly select m sample points from the training sample set S to obtain multiple training subsets;

[0028] ③ Training the CRAT decision tree based on the training subset;

[0029] ④ Repeat steps ②-③ to generate multiple independent CART decision tree models, each CART decision tree obtains a prediction result, and the result of the random forest effective crack prediction model is the mean of the prediction results of all CART decision trees;

[0030] ⑤ Calculate the fitness of the random forest effective crack prediction model, sort it, and determine the current optimal and worst positions;

[0031] ⑥ Obtain the discoverer, joiner and warning positions and feed them back to the random forest effective crack prediction model, calculate the fitness, and update the position if the new position is better than the previous one;

[0032] ⑦ Determine whether the stopping condition is met. If it is met, exit and output the final prediction result of the effective crack of the random forest. Otherwise, repeat steps ②-⑥.

[0033] Preferably, it also includes:

[0034] The well point imaging data is used as quality control, and the prediction result is the fracture effective factor FIND three-dimensional data body. The prediction result is superimposed on the seismic waveform, and the seismic structural morphology is compared to verify the accuracy of the identification result.

[0035] In a second aspect, the embodiment of the present disclosure further provides an effective crack identification device based on random forest, comprising:

[0036] The calibration module establishes the fracture effectiveness evaluation standard and determines the fracture effectiveness index based on the well logging data;

[0037] A labeling module, which establishes a fracture prediction data set based on a plurality of seismic attributes and inversion results characterizing fractures of different scales, and labels the data with the fracture effectiveness index;

[0038] A training module is used to train a random forest effective crack prediction model to predict the unknown crack effectiveness index;

[0039] The prediction module predicts effective cracks in the entire three-dimensional area based on the trained random forest effective crack prediction model.

[0040] Preferably, based on the well logging data, the fracture effectiveness evaluation criteria are established including:

[0041] Based on the analysis of fracture logging response characteristics, a fracture effectiveness index is established to evaluate fracture effectiveness:

[0042] FIND=a×RDB_FI+b×ACSH_FI+C×RDS_FI (1)

[0043] Among them, FIND is the fracture effectiveness index, RDB_FI is the absolute value decrease of resistivity, ACSH_FI is the increase of acoustic time difference, RDS_FI is the resistivity invasion amplitude difference, and a, b, and c are regional empirical parameters determined by the imaging logging scale.

[0044] Preferably, the random forest effective crack prediction model uses multiple CART decision trees as basic units, and multiple decision trees are constructed by random repeated sampling technology and node random splitting technology. The key parameters are obtained by using the sparrow search algorithm, and the predicted crack effectiveness index is output. The key parameters include the number of decision trees and the maximum number of features.

[0045] Preferably, the CART decision tree determines the segmentation scheme based on the minimum joint impurity J value of the binary space samples. s,t For J:

[0046] J(R s,t )=P(x∈X S )·I(X S )+P(x∈X T)·I(X T ) (2)

[0047] Here, I is the impurity measure, expressed as the variance of the predictor variables in X:

[0048]

[0049] Among them, E(y|x∈X) represents the conditional expectation that the input is y when the input x is given.

[0050] Preferably, after the decision tree recursive optimization growth is completed, the entire training set is divided into M mutually exclusive subsets X1, X2...X M , that is, the decision tree has M leaf nodes;

[0051] When the sample x* is brought into this decision tree, a leaf node X is obtained according to the logical rules of the branch nodes in the tree. l , at this point, the conditional expectation of y on x* is calculated as the result of the decision tree regression prediction:

[0052]

[0053] Where N represents the number of samples in the sample set, I(x i ∈X l ) is a conditional function, when x i ∈X l If true, I takes the value 1, otherwise it takes the value 0.

[0054] Preferably, the sum of the errors between the training set and the test set of the random forest effective crack prediction model is used as the objective function F, and the model parameters that minimize F can be iteratively found using the sparrow search algorithm. Training the random forest effective crack prediction model includes:

[0055] ① Initialize the population, determine the number of iterations, and the ratio of discoverers to joiners;

[0056] ② Randomly select m sample points from the training sample set S to obtain multiple training subsets;

[0057] ③ Training the CRAT decision tree based on the training subset;

[0058] ④ Repeat steps ②-③ to generate multiple independent CART decision tree models, each CART decision tree obtains a prediction result, and the result of the random forest effective crack prediction model is the mean of the prediction results of all CART decision trees;

[0059] ⑤ Calculate the fitness of the random forest effective crack prediction model, sort it, and determine the current optimal and worst positions;

[0060] ⑥ Obtain the discoverer, joiner and warning positions and feed them back to the random forest effective crack prediction model, calculate the fitness, and update the position if the new position is better than the previous one;

[0061] ⑦ Determine whether the stopping condition is met. If it is met, exit and output the final prediction result of the effective crack of the random forest. Otherwise, repeat steps ②-⑥.

[0062] Preferably, it also includes:

[0063] The well point imaging data is used as quality control, and the prediction result is the fracture effective factor FIND three-dimensional data body. The prediction result is superimposed on the seismic waveform, and the seismic structural morphology is compared to verify the accuracy of the identification result.

[0064] In a third aspect, an embodiment of the present disclosure further provides an electronic device, the electronic device comprising:

[0065] A memory storing executable instructions;

[0066] A processor runs the executable instructions in the memory to implement the effective crack identification method based on random forest.

[0067] In a fourth aspect, an embodiment of the present disclosure further provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the effective crack identification method based on random forest is implemented.

[0068] Its beneficial effects are:

[0069] The present invention is based on the characterization of the spatial distribution characteristics and development degree of fractures of different scales by multiple seismic attributes, and uses single well coring and imaging logging data for quality control. It uses the random forest technology improved by the sparrow search algorithm to finely characterize the effective fractures in tight sandstone, forming an efficient automatic fracture identification technology, reducing the prediction multi-solution, and providing technical support for the exploration of terrestrial tight sandstone reservoirs.

[0070] The methods and apparatus of the present invention have other features and advantages that will be apparent from, or will be described in detail in, the accompanying drawings and subsequent detailed descriptions incorporated herein, which together serve to explain the specific principles of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The above and other objects, features and advantages of the present invention will become more apparent through a more detailed description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present invention.

[0072] Figure 1 A flow chart showing the steps of an effective crack identification method based on random forest according to an embodiment of the present invention.

[0073] Figure 2 A schematic diagram showing single-well effective fracture factor FIND values ​​of some wells according to an embodiment of the present invention is shown.

[0074] Figure 3a , Figure 3b , Figure 3c A schematic plan view of target layer ant tracking attributes, structural deformation intensity attributes, and pre-stack longitudinal wave anisotropic fracture prediction according to an embodiment of the present invention is shown respectively.

[0075] Figure 4 A schematic diagram showing a random forest model structure according to an embodiment of the present invention.

[0076] Figure 5 A random forest effective crack prediction flow chart according to an embodiment of the present invention is shown.

[0077] Figure 6a , Figure 6b , Figure 6c Schematic diagrams of the B1 well logging data set, the well bypass seismic attribute data set, and the FIND prediction results of the two data samples according to an embodiment of the present invention are respectively shown.

[0078] Figure 7a A schematic diagram showing a random forest algorithm FIND prediction plane diagram according to an embodiment of the present invention is shown; Figure 7b A schematic diagram showing a comparison diagram of prediction results of the B2 well and imaging logging according to an embodiment of the present invention is shown.

[0079] Figure 8 A schematic diagram of an effective fracture prediction cross-section through wells A8 and B2 according to an embodiment of the present invention is shown.

[0080] Fig. 9 A block diagram of an effective crack identification device based on random forest according to an embodiment of the present invention is shown.

[0081] Description of reference numerals:

[0082] 201, calibration module; 202, labeling module; 203, training module; 204, prediction module. DETAILED DESCRIPTION

[0083] The preferred embodiments of the present invention will be described in more detail below. Although the preferred embodiments of the present invention are described below, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0084] The present invention provides an effective crack identification method based on random forest, comprising:

[0085] Based on the logging data, the fracture effectiveness evaluation standard is established and the fracture effectiveness index is determined;

[0086] A fracture prediction dataset is established based on a variety of seismic attributes and inversion results that characterize fractures of different scales, and the labels are annotated with fracture effectiveness indexes;

[0087] Train the random forest effective crack prediction model to predict the unknown crack effectiveness index;

[0088] Based on the trained random forest effective crack prediction model, effective cracks in the entire three-dimensional area are predicted.

[0089] In one example, based on well logging data, fracture effectiveness evaluation criteria are established including:

[0090] Based on the analysis of fracture logging response characteristics, a fracture effectiveness index is established to evaluate fracture effectiveness:

[0091] FIND=a×RDB_FI+b×ACSH_FI+C×RDS_FI (1)

[0092] Among them, FIND is the fracture effectiveness index, RDB_FI is the absolute value decrease of resistivity, ACSH_FI is the increase of acoustic time difference, RDS_FI is the resistivity invasion amplitude difference, and a, b, and c are regional empirical parameters determined by the imaging logging scale.

[0093] In one example, the random forest effective crack prediction model uses multiple CART decision trees as basic units. Multiple decision trees are constructed by random repeated sampling technology and node random splitting technology. The sparrow search algorithm is used to obtain key parameters and output the predicted crack effectiveness index. The key parameters include the number of decision trees and the maximum number of features.

[0094] In one example, the CART decision tree determines the segmentation scheme based on the minimum joint impurity J value of the binary space samples. For the segmentation scheme R s,t For J:

[0095] J(R s,t )=P(x∈X S )·I(X S )+P(x∈X T )·I(X T ) (2)

[0096] Here, I is the impurity measure, expressed as the variance of the predictor variables in X:

[0097]

[0098] Among them, E(y|x∈X) represents the conditional expectation that the input is y when the input x is given.

[0099] In one example, after the decision tree recursive optimization growth is completed, the entire training set is divided into M mutually exclusive subsets X1, X2, ...X M , that is, the decision tree has M leaf nodes;

[0100] When the sample x* is brought into this decision tree, a leaf node X is obtained according to the logical rules of the branch nodes in the tree. l , at this point, the conditional expectation of y on x* is calculated as the result of the decision tree regression prediction:

[0101]

[0102] Where N represents the number of samples in the sample set, I(x i ∈X l ) is a conditional function, when x i ∈X l If true, I takes the value 1, otherwise it takes the value 0.

[0103] In one example, the sum of the errors between the training set and the test set of the random forest effective crack prediction model is used as the objective function F, and the sparrow search algorithm can be used to iteratively find the model parameters that minimize F. Training the random forest effective crack prediction model includes:

[0104] ① Initialize the population, determine the number of iterations, and the ratio of discoverers to joiners;

[0105] ② Randomly select m sample points from the training sample set S to obtain multiple training subsets;

[0106] ③Train the CRAT decision tree based on the training subset;

[0107] ④ Repeat steps ②-③ to generate multiple independent CART decision tree models. Each CART decision tree obtains a prediction result. The result of the random forest effective crack prediction model is the mean of the prediction results of all CART decision trees.

[0108] ⑤ Calculate the fitness of the random forest effective crack prediction model, sort it, and determine the current optimal and worst positions;

[0109] ⑥ Obtain the discoverer, joiner and warning positions and feed them back to the random forest effective crack prediction model, calculate the fitness, and update the position if the new position is better than the previous one;

[0110] ⑦ Determine whether the stopping condition is met. If it is met, exit and output the final prediction result of the effective crack of the random forest. Otherwise, repeat steps ②-⑥.

[0111] In one example, it also includes:

[0112] The well point imaging data is used as quality control, and the prediction result is the fracture effective factor FIND three-dimensional data body. The prediction result is superimposed with the seismic waveform, and the seismic structural morphology is compared to verify the accuracy of the identification result.

[0113] Specifically, based on the well logging data, the fracture effectiveness evaluation criteria are established;

[0114] In order to predict effective fractures, it is necessary to first establish a fracture effectiveness evaluation standard that is universal in the region. Well logging technology provides conditions for evaluating fracture effectiveness. Conventional and imaging well logging are used to detect the development of wellbore fractures, and dipole acoustic waves are used in combination with seismic data to detect the spatial distribution of fractures. The response characteristics of fractures on conventional well logging can be divided into three categories: a decrease in the absolute value of resistivity, an increase in the acoustic time difference, and a large difference in the resistivity invasion amplitude. Based on the analysis of fracture logging response characteristics, a fracture effectiveness index (FIND) is established to evaluate fracture effectiveness, as shown in formula (1).

[0115] A variety of seismic attributes and inversion results that characterize fractures of different scales are selected to establish a fracture prediction dataset, and the dataset is labeled with a fracture effectiveness index.

[0116] The key to using machine learning to obtain accurate prediction results is to establish a large number of reliable data sets. In seismic data, the degree of fracture development can change the dynamic characteristics of seismic waves, and the seismic phase axis morphology is also different. Seismic attributes such as coherence, curvature, ant body, structural deformation intensity, and pre-stack longitudinal wave anisotropy inversion can be used to characterize fractures of different scales, reflecting the changes in seismic data caused by fractures from multiple angles such as seismic trace correlation, geometric morphology, and anisotropy. Therefore, based on seismic multi-attribute analysis and anisotropic inversion, based on the degree of fracture development of different scales reflected by multiple attributes and inversion results, wellside traces are extracted and annotated in combination with the fracture effectiveness index FIND evaluated by logging, to obtain a real and reliable label set.

[0117] Train a random forest effective crack prediction model;

[0118] The random forest model constructed by the present invention can generate an effective fracture index FIND prediction result for each sampling point of seismic data. For this regression problem, the training model is a plurality of decision trees formed by random repeated sampling (Bootstrap) technology and node random splitting technology with several CART decision trees as basic units, and the key parameters such as the number of decision trees are obtained by using the Sparrow Search Algorithm (SSA) with strong optimization ability and fast convergence. The appropriate decision tree prediction results are combined by ensemble learning and output as a whole to obtain the FIND prediction.

[0119] The CART decision tree is constructed by recursive segmentation, splitting nodes from top to bottom until a leaf node is reached. When each internal node is split, the most appropriate segmentation scheme needs to be selected. The CART decision tree selects the optimal segmentation scheme based on the minimum joint impurity J value of the binary space samples. For the segmentation scheme R s,t J can be defined as formula (2), where I is the impurity measure, expressed as the variance of the predictor variable in X, that is, formula (3).

[0120] After the decision tree recursive optimization growth is completed, the entire training set is divided into M mutually exclusive subsets X1, X2…X M , that is, the decision tree has M leaf nodes. When the sample x* is brought into this decision tree, a leaf node X is obtained according to the logical rules of the branch nodes in the tree. l At this time, the conditional expectation of x* can be estimated as the result of the decision tree regression prediction as formula (4).

[0121] The sparrow search algorithm is an optimization algorithm. Sparrow individuals are divided into two categories: discoverers and joiners. Discoverers will give priority to obtaining food during the search process and have a larger search range than joiners. In each iteration, the positions of discoverers and joiners will be updated. The current position is determined to be optimal based on the calculated fitness value. Multiple iterations are performed until the optimal position is obtained or the maximum number of iterations is reached. The sparrow search algorithm can avoid overfitting caused by improper setting of key parameters. The number of decision trees will greatly affect the prediction results of random forests, especially when processing noisy data such as seismic data. SSA is an optimization algorithm based on the biological characteristics of sparrows for iterative optimization. In the algorithm, some sparrows act as discoverers and play a role in searching for food for the population. More individuals join as joiners and join according to the foraging direction provided by the discoverers. The identities of the two change dynamically. Finally, the remaining individuals act as vigilants to observe whether there is any danger in the surrounding environment. Once danger is found, they immediately send a signal, and all sparrows take anti-predation behavior. In each iteration, the positions of discoverers, joiners, and vigilants are updated until the iteration ends.

[0122] The Bagging ensemble framework is a randomized ensemble learning algorithm framework. If N decision trees need to be built based on the training data set S, then N training subsets must be extracted from the set S. The extraction principle is to use unweighted sampling (Bagging), that is, random repeated sampling. In the random forest prediction stage, the mean of the prediction results of these N decision trees is integrated as the final prediction result.

[0123] In summary, the process of establishing a random forest model optimized by the sparrow algorithm is as follows:

[0124] ① Initialize the population, determine the number of iterations, and the ratio of discoverers to joiners.

[0125] ② Randomly select m sample points from the training sample set S to obtain a new training subset S1…Sn.

[0126] ③ Use the training subset to train a CRAT decision tree. During the training process, the splitting rule for each node is to randomly select one feature from all features, and then select the splitting scheme with the smallest sample joint impurity J value from this feature for binary division.

[0127] ④ According to the above method, multiple independent CART decision tree models are generated, and each CART decision tree will get a prediction result. The random forest model result is the mean of all CART decision tree prediction results.

[0128] ⑤Calculate the fitness of the current random forest model and sort it to find the current optimal and worst positions.

[0129] ⑥ Update the discoverer, joiner and alert positions and feed them back to the random forest model, calculate the fitness, and update the position if the new position is better than the previous one.

[0130] ⑦ Determine whether the stopping condition is met. If it is met, exit and finally predict the result. Otherwise, repeat steps ② to ⑥.

[0131] Prediction of effective fractures in the whole area;

[0132] Based on a real and reliable data set, a high-performance random forest effective crack prediction model is obtained. By inputting three-dimensional target data, the three-dimensional effective crack spatial distribution and development degree results can be obtained.

[0133] Well point imaging data was used as quality control to verify the feasibility of the technology and the accuracy of the prediction results.

[0134] The final prediction result is the fracture effective factor FIND three-dimensional data body. By superimposing the prediction result with the seismic waveform and comparing the seismic structural morphology, the feasibility of the method and the accuracy of the identification result can be verified as a whole. In addition, by comparing the real drilling imaging logging and core data that did not participate in the training, the prediction of effective fractures can be further verified.

[0135] The present invention also provides an effective crack identification device based on random forest, comprising:

[0136] The calibration module establishes the fracture effectiveness evaluation standard and determines the fracture effectiveness index based on the well logging data;

[0137] The labeling module builds a fracture prediction dataset based on a variety of seismic attributes and inversion results that characterize fractures of different scales, and labels them with fracture effectiveness indices;

[0138] Training module, training the random forest effective crack prediction model to predict unknown crack effectiveness index;

[0139] The prediction module predicts effective cracks in the entire three-dimensional area based on the trained random forest effective crack prediction model.

[0140] In one example, based on well logging data, fracture effectiveness evaluation criteria are established including:

[0141] Based on the analysis of fracture logging response characteristics, a fracture effectiveness index is established to evaluate fracture effectiveness:

[0142] FIND=a×RDB_FI+b×ACSH_FI+C×RDS_FI (1)

[0143] Among them, FIND is the fracture effectiveness index, RDB_FI is the absolute value decrease of resistivity, ACSH_FI is the increase of acoustic time difference, RDS_FI is the resistivity invasion amplitude difference, and a, b, and c are regional empirical parameters determined by the imaging logging scale.

[0144] In one example, the random forest effective crack prediction model uses multiple CART decision trees as basic units. Multiple decision trees are constructed by random repeated sampling technology and node random splitting technology. The sparrow search algorithm is used to obtain key parameters and output the predicted crack effectiveness index. The key parameters include the number of decision trees and the maximum number of features.

[0145] In one example, the CART decision tree determines the segmentation scheme based on the minimum joint impurity J value of the binary space samples. For the segmentation scheme R s,t For J:

[0146] J(R s,t )=P(x∈X S )·I(XS )+P(x∈X T )·I(X T ) (2)

[0147] Here, I is the impurity measure, expressed as the variance of the predictor variables in X:

[0148]

[0149] Among them, E(y|x∈X) represents the conditional expectation that the input is y when the input x is given.

[0150] In one example, after the decision tree recursive optimization growth is completed, the entire training set is divided into M mutually exclusive subsets X1, X2, ...X M , that is, the decision tree has M leaf nodes;

[0151] When the sample x* is brought into this decision tree, a leaf node X is obtained according to the logical rules of the branch nodes in the tree. l , at this point, the conditional expectation of y on x* is calculated as the result of the decision tree regression prediction:

[0152]

[0153] Where N represents the number of samples in the sample set, I(x i ∈X l ) is a conditional function, when x i ∈X l If true, I takes the value 1, otherwise it takes the value 0.

[0154] In one example, the sum of the errors between the training set and the test set of the random forest effective crack prediction model is used as the objective function F, and the sparrow search algorithm can be used to iteratively find the model parameters that minimize F. Training the random forest effective crack prediction model includes:

[0155] ① Initialize the population, determine the number of iterations, and the ratio of discoverers to joiners;

[0156] ② Randomly select m sample points from the training sample set S to obtain multiple training subsets;

[0157] ③Train the CRAT decision tree based on the training subset;

[0158] ④ Repeat steps ②-③ to generate multiple independent CART decision tree models. Each CART decision tree obtains a prediction result. The result of the random forest effective crack prediction model is the mean of the prediction results of all CART decision trees.

[0159] ⑤ Calculate the fitness of the random forest effective crack prediction model, sort it, and determine the current optimal and worst positions;

[0160] ⑥ Obtain the discoverer, joiner and warning positions and feed them back to the random forest effective crack prediction model, calculate the fitness, and update the position if the new position is better than the previous one;

[0161] ⑦ Determine whether the stopping condition is met. If it is met, exit and output the final prediction result of the effective crack of the random forest. Otherwise, repeat steps ②-⑥.

[0162] In one example, it also includes:

[0163] The well point imaging data is used as quality control, and the prediction result is the fracture effective factor FIND three-dimensional data body. The prediction result is superimposed with the seismic waveform, and the seismic structural morphology is compared to verify the accuracy of the identification result.

[0164] Specifically, based on the well logging data, the fracture effectiveness evaluation criteria are established;

[0165] In order to predict effective fractures, it is necessary to first establish a fracture effectiveness evaluation standard that is universal in the region. Well logging technology provides conditions for evaluating fracture effectiveness. Conventional and imaging well logging are used to detect the development of wellbore fractures, and dipole acoustic waves are used in combination with seismic data to detect the spatial distribution of fractures. The response characteristics of fractures on conventional well logging can be divided into three categories: a decrease in the absolute value of resistivity, an increase in the acoustic time difference, and a large difference in the resistivity invasion amplitude. Based on the analysis of fracture logging response characteristics, a fracture effectiveness index (FIND) is established to evaluate fracture effectiveness, as shown in formula (1).

[0166] A variety of seismic attributes and inversion results that characterize fractures of different scales are selected to establish a fracture prediction dataset, and the dataset is labeled with a fracture effectiveness index.

[0167] The key to using machine learning to obtain accurate prediction results is to establish a large number of reliable data sets. In seismic data, the degree of fracture development can change the dynamic characteristics of seismic waves, and the seismic phase axis morphology is also different. Seismic attributes such as coherence, curvature, ant body, structural deformation intensity, and pre-stack longitudinal wave anisotropy inversion can be used to characterize fractures of different scales, reflecting the changes in seismic data caused by fractures from multiple angles such as seismic trace correlation, geometric morphology, and anisotropy. Therefore, based on seismic multi-attribute analysis and anisotropic inversion, based on the degree of fracture development of different scales reflected by multiple attributes and inversion results, wellside traces are extracted and annotated in combination with the fracture effectiveness index FIND evaluated by logging, to obtain a real and reliable label set.

[0168] Train a random forest effective crack prediction model;

[0169] The random forest model constructed by the present invention can generate an effective fracture index FIND prediction result for each sampling point of seismic data. For this regression problem, the training model is a plurality of decision trees formed by random repeated sampling (Bootstrap) technology and node random splitting technology with several CART decision trees as basic units, and the key parameters such as the number of decision trees are obtained by using the Sparrow Search Algorithm (SSA) with strong optimization ability and fast convergence. The appropriate decision tree prediction results are combined by ensemble learning and output as a whole to obtain the FIND prediction.

[0170] The CART decision tree is constructed by recursive segmentation, splitting nodes from top to bottom until a leaf node is reached. When each internal node is split, the most appropriate segmentation scheme needs to be selected. The CART decision tree selects the optimal segmentation scheme based on the minimum joint impurity J value of the binary space samples. For the segmentation scheme R s,t J can be defined as formula (2), where I is the impurity measure, expressed as the variance of the predictor variable in X, that is, formula (3).

[0171] After the decision tree recursive optimization growth is completed, the entire training set is divided into M mutually exclusive subsets X1, X2…X M , that is, the decision tree has M leaf nodes. When the sample x* is brought into this decision tree, a leaf node X is obtained according to the logical rules of the branch nodes in the tree. l At this time, the conditional expectation of x* can be estimated as the result of the decision tree regression prediction as formula (4).

[0172] The sparrow search algorithm can avoid overfitting caused by improper setting of key parameters. The number of decision trees will greatly affect the prediction results of random forests, especially when processing noisy data such as seismic data. SSA is an optimization algorithm based on the biological characteristics of sparrows for iterative optimization. In the algorithm, some sparrows act as discoverers, playing the role of exploring the way for the population to search for food. More individuals join as joiners according to the foraging direction provided by the discoverer, and the identities of the two change dynamically. Finally, the remaining individuals act as sentinels to observe whether there is any danger in the surrounding environment. Once danger is found, a signal is immediately sent, and all sparrows take anti-predation actions. In each iteration, the positions of the discoverer, joiner, and sentinel are updated until the iteration ends.

[0173] The Bagging ensemble framework is a randomized ensemble learning algorithm framework. If N decision trees need to be built based on the training data set S, then N training subsets must be extracted from the set S. The extraction principle is to use unweighted sampling (Bagging), that is, random repeated sampling. In the random forest prediction stage, the mean of the prediction results of these N decision trees is integrated as the final prediction result.

[0174] In summary, the process of establishing a random forest model optimized by the sparrow algorithm is as follows:

[0175] ① Initialize the population, determine the number of iterations, and the ratio of discoverers to joiners.

[0176] ② Randomly select m sample points from the training sample set S to obtain a new training subset S1…Sn.

[0177] ③ Use the training subset to train a CRAT decision tree. During the training process, the splitting rule for each node is to randomly select one feature from all features, and then select the splitting scheme with the smallest sample joint impurity J value from this feature for binary division.

[0178] ④ According to the above method, multiple independent CART decision tree models are generated, and each CART decision tree will get a prediction result. The random forest model result is the mean of all CART decision tree prediction results.

[0179] ⑤Calculate the fitness of the current random forest model and sort it to find the current optimal and worst positions.

[0180] ⑥ Update the discoverer, joiner and alert positions and feed them back to the random forest model, calculate the fitness, and update the position if the new position is better than the previous one.

[0181] ⑦ Determine whether the stopping condition is met. If it is met, exit and finally predict the result. Otherwise, repeat steps ② to ⑥.

[0182] Prediction of effective fractures in the whole area;

[0183] Based on a real and reliable data set, a high-performance random forest effective crack prediction model is obtained. By inputting three-dimensional target data, the three-dimensional effective crack spatial distribution and development degree results can be obtained.

[0184] Well point imaging data was used as quality control to verify the feasibility of the technology and the accuracy of the prediction results.

[0185] The final prediction result is the fracture effective factor FIND three-dimensional data body. By superimposing the prediction result with the seismic waveform and comparing the seismic structural morphology, the feasibility of the method and the accuracy of the identification result can be verified as a whole. In addition, by comparing the real drilling imaging logging and core data that did not participate in the training, the prediction of effective fractures can be further verified.

[0186] The present invention also provides an electronic device, which includes: a memory storing executable instructions; and a processor, which runs the executable instructions in the memory to implement the above-mentioned effective crack identification method based on random forest.

[0187] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned effective crack identification method based on random forest is implemented.

[0188] To facilitate understanding of the solutions and effects of the embodiments of the present invention, four specific application examples are given below. Those skilled in the art should understand that the examples are only for facilitating understanding of the present invention, and any specific details thereof are not intended to limit the present invention in any way.

[0189] Example 1

[0190] Figure 1 A flow chart showing the steps of an effective crack identification method based on random forest according to an embodiment of the present invention.

[0191] like Figure 1 As shown, the effective fracture identification method based on random forest includes: step 101, establishing fracture effectiveness evaluation criteria based on well logging data and determining fracture effectiveness index; step 102, establishing fracture prediction data set according to multiple seismic attributes and inversion results that characterize fractures of different scales, and marking labels with fracture effectiveness index; step 103, training random forest effective fracture prediction model to predict unknown fracture effectiveness index; step 104, predicting effective fractures in the entire three-dimensional area based on the trained random forest effective fracture prediction model.

[0192] The continental tight sandstone area in northeastern Sichuan Basin was selected as a prediction example. The method of the present invention was used to comprehensively evaluate the effectiveness of fractures in the study area, providing a reliable reference for the optimization of favorable targets for exploration and development.

[0193] Figure 2 A schematic diagram of the single-well effective fracture factor FIND values ​​of some wells according to an embodiment of the present invention is shown. In this example, conventional logging and imaging logging data of 8 drilled wells are first used to analyze the factors affecting fracture effectiveness under the guidance of geological background, and the fracture response pattern in the region is summarized. Then, an effective fracture evaluation method with regional universality is established, and finally, the single-well effective fracture factor FIND is obtained using formula (1) and used as an effective fracture label.

[0194] Figure 3a , Figure 3b , Figure 3c A schematic plan view of target layer ant tracking attributes, structural deformation intensity attributes, and pre-stack longitudinal wave anisotropic fracture prediction according to an embodiment of the present invention is shown respectively.

[0195] In this embodiment, the ant tracking technology of the bionic optimization algorithm is used to retrieve the discontinuity between seismic data, so as to complete the tracking and identification of large-scale fractures. The structural deformation intensity attribute reflects the local deformation degree of structures such as anticlines and synclines by calculating the distribution of the direction vector between the current data point and the adjacent data point, and characterizes the development of medium-scale fractures. Prestack amplitude anisotropy is used to predict interlayer small-scale fractures. Since the reflection amplitude and other factors of seismic waves are related to the propagation azimuth when the seismic wave propagates, the fracture density can be predicted by extracting prestack seismic attributes related to the azimuth and the angle of incidence. The above three seismic attributes can characterize the spatial distribution and development degree of fractures of different scales, and the three seismic attribute data bodies in the study area are calculated as the data set for training the random forest effective fracture model.

[0196] Figure 4 A schematic diagram of a random forest model structure according to an embodiment of the present invention is shown. The random forest algorithm model is a tree structure as a whole. After testing, the random forest model in the example is composed of 600 decision trees. First, the training samples are randomly divided into several subsets by random sampling without replacement. Each decision tree is divided into two according to a randomly selected feature according to a prescribed segmentation scheme to generate a prediction result. Finally, the prediction result of the entire random forest model is obtained by means of mean integration.

[0197] Figure 5 A random forest effective crack prediction flow chart according to an embodiment of the present invention is shown.

[0198] The prediction of effective fractures by random forest can be divided into three parts. First, data preprocessing is performed, including the calculation of FIND values ​​of label data based on core, conventional logging and imaging logging data, and three well bypass data that characterize the seismic attributes of fractures of different scales. Through fine single well calibration and time-depth conversion, seismic attributes and FIND labels are matched one by one to form a training sample set. The training sample set of the example contains 10,000 data points from 10 wells in the study area. The second part is model establishment. The random forest algorithm is used to extract the features of the nonlinear relationship between seismic multi-attributes and effective fracture FIND values ​​and optimize the parameters to establish a random forest model. The third part is prediction. The model is used to perform regression prediction on the samples to be predicted to obtain effective fracture results. The example selects data from two wells that do not participate in model training as verification.

[0199] Figure 6a , Figure 6b , Figure 6c Schematic diagrams of the B1 well logging data set, the well bypass seismic attribute data set, and the FIND prediction results of the two data samples according to an embodiment of the present invention are respectively shown.

[0200] The prediction effect of the random forest algorithm was tested by a single well. First, two types of training samples were established, and 7 types of well logging data related to the FIND value were selected as training sample 1; three types of seismic attributes such as curvature, ant body, and structural deformation intensity were selected to make training sample 2. When making samples, it is necessary to match the seismic data (time domain) with the well logging data (depth domain) in time and depth, and integrate the two into the same scale domain. On the basis of well-seismic calibration, resampling is used to increase the seismic attribute data sample points according to the number of FIND value samples interpreted by well logging to ensure the consistency of model input and output. As can be seen from the figure, the FIND value of well B1 is predicted using two types of data samples, and the trend of the prediction result is the same as the true value. The random forest algorithm based on seismic attribute samples can be used to predict FIND, and the accuracy can meet the needs of actual tasks, which proves the feasibility of the method used.

[0201] Figure 7a The schematic diagram of the FIND prediction plane map of the random forest algorithm according to an embodiment of the present invention is shown; the white part represents the low fracture effectiveness area with FIND less than 0.02. The yellow and red parts represent the high fracture effectiveness area with FIND greater than 0.06, and the test production is also relatively higher. The prediction results can better reflect the distribution characteristics and development of effective fractures in the area. The FIND values ​​predicted by the random forest of wells B1 and B2 used for verification are highly consistent with the FIND values ​​calculated by actual well logging.

[0202] Figure 7b A schematic diagram of a comparison diagram of the prediction results and imaging logging of Well B2 according to an embodiment of the present invention is shown. In the figure, large-scale fractures are developed in the upper subsection of Xu 4 in Well B2, and the prediction results match the imaging logging data well. The effect comparison shows that the effect of predicting effective fractures using the random forest algorithm is highly consistent with the actual results, showing the great potential of this method in fracture prediction.

[0203] Figure 8 The schematic diagram of the effective fracture prediction profile of the well-connected A8 and B2 wells according to an embodiment of the present invention is shown. The prediction results of the A8 and B2 wells in the figure are well consistent with the true values, which verifies the accuracy of the technology of the present invention. In addition, the effective fracture distribution characteristics reflected in the well-connected section are more in line with the actual geological background, which has a certain guiding significance for the evaluation of the favorable fracture zone.

[0204] Example 2

[0205] Fig. 9 A block diagram of an effective crack identification device based on random forest according to an embodiment of the present invention is shown.

[0206] like Fig. 9 As shown, the effective crack identification device based on random forest includes:

[0207] The calibration module 201 establishes a fracture effectiveness evaluation standard based on the well logging data and determines the fracture effectiveness index;

[0208] A labeling module 202 is used to establish a fracture prediction data set based on a plurality of seismic attributes and inversion results that characterize fractures of different scales, and label the data set with a fracture effectiveness index;

[0209] A training module 203 is used to train a random forest effective crack prediction model to predict an unknown crack effectiveness index;

[0210] The prediction module 204 predicts effective fractures in the entire three-dimensional area based on the trained random forest effective fracture prediction model.

[0211] As an optional solution, based on well logging data, the fracture effectiveness evaluation criteria include:

[0212] Based on the analysis of fracture logging response characteristics, a fracture effectiveness index is established to evaluate fracture effectiveness:

[0213] FIND=a×RDB_FI+b×ACSH_FI+C×RDS_FI (1)

[0214] Among them, FIND is the fracture effectiveness index, RDB_FI is the absolute value decrease of resistivity, ACSH_FI is the increase of acoustic time difference, RDS_FI is the resistivity invasion amplitude difference, and a, b, and c are regional empirical parameters determined by the imaging logging scale.

[0215] As an optional solution, the random forest effective crack prediction model uses multiple CART decision trees as basic units. Multiple decision trees are constructed by random repeated sampling technology and node random splitting technology. The sparrow search algorithm is used to obtain key parameters and output the predicted crack effectiveness index. The key parameters include the number of decision trees and the maximum number of features.

[0216] As an alternative, the CART decision tree determines the segmentation scheme based on the minimum joint impurity J value of the binary space samples. s,t For J:

[0217] J(R s,t )=P(x∈X S )·I(X S )+P(x∈X T )·I(X T ) (2)

[0218] Here, I is the impurity measure, expressed as the variance of the predictor variables in X:

[0219]

[0220] Among them, E(y|x∈X) represents the conditional expectation that the input is y when the input x is given.

[0221] As an alternative, after the decision tree recursive optimization growth is completed, the entire training set is divided into M mutually exclusive subsets X1, X2…X M , that is, the decision tree has M leaf nodes;

[0222] When the sample x* is brought into this decision tree, a leaf node X is obtained according to the logical rules of the branch nodes in the tree. l , at this point, the conditional expectation of y on x* is calculated as the result of the decision tree regression prediction:

[0223]

[0224] Where N represents the number of samples in the sample set, I(x i ∈X l ) is a conditional function, when x i ∈X l If true, I takes the value 1, otherwise it takes the value 0.

[0225] As an optional solution, the sum of the errors between the training set and the test set of the random forest effective crack prediction model is used as the objective function F. The sparrow search algorithm can be used to iteratively find the model parameters that minimize F. Training the random forest effective crack prediction model includes:

[0226] ① Initialize the population, determine the number of iterations, and the ratio of discoverers to joiners;

[0227] ② Randomly select m sample points from the training sample set S to obtain multiple training subsets;

[0228] ③Train the CRAT decision tree based on the training subset;

[0229] ④ Repeat steps ②-③ to generate multiple independent CART decision tree models. Each CART decision tree obtains a prediction result. The result of the random forest effective crack prediction model is the mean of the prediction results of all CART decision trees.

[0230] ⑤ Calculate the fitness of the random forest effective crack prediction model, sort it, and determine the current optimal and worst positions;

[0231] ⑥ Obtain the discoverer, joiner and warning positions and feed them back to the random forest effective crack prediction model, calculate the fitness, and update the position if the new position is better than the previous one;

[0232] ⑦ Determine whether the stopping condition is met. If it is met, exit and output the final prediction result of the effective crack of the random forest. Otherwise, repeat steps ②-⑥.

[0233] As an option, it also includes:

[0234] The well point imaging data is used as quality control, and the prediction result is the fracture effective factor FIND three-dimensional data body. The prediction result is superimposed with the seismic waveform, and the seismic structural morphology is compared to verify the accuracy of the identification result.

[0235] Example 3

[0236] The present disclosure provides an electronic device, which includes: a memory storing executable instructions; and a processor running the executable instructions in the memory to implement the above-mentioned effective crack identification method based on random forest.

[0237] An electronic device according to an embodiment of the present disclosure includes a memory and a processor.

[0238] The memory is used to store non-temporary computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0239] The processor may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In one embodiment of the present disclosure, the processor is used to run the computer-readable instructions stored in the memory.

[0240] Those skilled in the art should be able to understand that in order to solve the technical problem of how to obtain a good user experience, the present embodiment may also include well-known structures such as a communication bus and an interface, and these well-known structures should also be included in the protection scope of the present disclosure.

[0241] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0242] Example 4

[0243] An embodiment of the present disclosure provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the effective crack identification method based on random forest is implemented.

[0244] According to the computer-readable storage medium of the embodiment of the present disclosure, non-transitory computer-readable instructions are stored thereon. When the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the above-mentioned methods of each embodiment of the present disclosure are executed.

[0245] The above-mentioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or mobile hard disk), media with built-in rewritable non-volatile memory (e.g., memory card) and media with built-in ROM (e.g., ROM box).

[0246] Those skilled in the art should understand that the purpose of the above description of the embodiments of the present invention is only to exemplarily illustrate the beneficial effects of the embodiments of the present invention, and is not intended to limit the embodiments of the present invention to any given examples.

[0247] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. An effective crack identification method based on random forest, characterized in that: include: Based on the logging data, the fracture effectiveness evaluation standard is established and the fracture effectiveness index is determined; A fracture prediction data set is established based on a plurality of seismic attributes and inversion results characterizing fractures of different scales, and a label is annotated with the fracture effectiveness index; Training a random forest effective crack prediction model to predict the unknown crack effectiveness index; Based on the trained random forest effective crack prediction model, effective cracks in the entire three-dimensional area are predicted.

2. The effective crack identification method based on random forest according to claim 1, wherein: Based on the logging data, the fracture effectiveness evaluation criteria are established, including: Based on the analysis of fracture logging response characteristics, a fracture effectiveness index is established to evaluate fracture effectiveness: FIND=a×RDB_FI+b×ACSH_FI+C×RDS_FI (1) Among them, FIND is the fracture effectiveness index, RDB_FI is the absolute value decrease of resistivity, ACSH_FI is the increase of acoustic time difference, RDS_FI is the resistivity invasion amplitude difference, and a, b, and c are regional empirical parameters determined by the imaging logging scale.

3. The effective crack identification method based on random forest according to claim 1, wherein: The random forest effective crack prediction model uses multiple CART decision trees as basic units, constructs multiple decision trees by random repeated sampling technology and node random splitting technology, optimizes key parameters using the sparrow search algorithm, and outputs the predicted crack effectiveness index, wherein the key parameters include the number of decision trees and the maximum number of features.

4. The effective crack identification method based on random forest according to claim 3, wherein: The CART decision tree determines the segmentation scheme based on the minimum joint impurity J value of the binary space samples. For the segmentation scheme R s,t For J: J(R s,t )=P(x∈X S )·I(X S )+P(x∈X T )·I(X T ) (2) Here, I is the impurity measure, expressed as the variance of the predictor variables in X: Among them, E(y|x∈X) represents the conditional expectation that the input is y when the input x is given.

5. The effective crack identification method based on random forest according to claim 4, wherein: After the decision tree recursive optimization growth is completed, the entire training set is divided into M mutually exclusive subsets X1, X2…X M , that is, the decision tree has M leaf nodes; When the sample x* is brought into this decision tree, a leaf node X is obtained according to the logical rules of the branch nodes in the tree. l , at this point, the conditional expectation of y on x* is calculated as the result of the decision tree regression prediction: Where N represents the number of samples in the sample set, I(x i ∈X l ) is a conditional function, when x i ∈X l If true, I takes the value 1, otherwise it takes the value 0.

6. The effective crack identification method based on random forest according to claim 3, wherein: The sum of the errors between the training set and the test set of the random forest effective crack prediction model is used as the objective function F. The sparrow search algorithm can be used to iteratively find the model parameters that minimize F. Training the random forest effective crack prediction model includes: ① Initialize the population, determine the number of iterations, and the ratio of discoverers to joiners; ② Randomly select m sample points from the training sample set S to obtain multiple training subsets; ③ Training the CRAT decision tree based on the training subset; ④ Repeat steps ②-③ to generate multiple independent CART decision tree models, each CART decision tree obtains a prediction result, and the result of the random forest effective crack prediction model is the mean of the prediction results of all CART decision trees; ⑤ Calculate the fitness of the random forest effective crack prediction model, sort it, and determine the current optimal and worst positions; ⑥ Obtain the discoverer, joiner and warning positions and feed them back to the random forest effective crack prediction model, calculate the fitness, and update the position if the new position is better than the previous one; ⑦ Determine whether the stopping condition is met. If it is met, exit and output the final prediction result of the effective crack of the random forest. Otherwise, repeat steps ②-⑥.

7. The effective crack identification method based on random forest according to claim 1, wherein: Also includes: The well point imaging data is used as quality control, and the prediction result is the fracture effective factor FIND three-dimensional data body. The prediction result is superimposed on the seismic waveform, and the seismic structural morphology is compared to verify the accuracy of the identification result.

8. An effective crack identification device based on random forest, characterized in that: include: The calibration module establishes the fracture effectiveness evaluation standard and determines the fracture effectiveness index based on the well logging data; A labeling module, which establishes a fracture prediction data set based on a plurality of seismic attributes and inversion results characterizing fractures of different scales, and labels the data with the fracture effectiveness index; A training module is used to train a random forest effective crack prediction model to predict the unknown crack effectiveness index; The prediction module predicts effective cracks in the entire three-dimensional area based on the trained random forest effective crack prediction model.

9. An electronic device, characterized in that: The electronic device comprises: A memory storing executable instructions; A processor, wherein the processor runs the executable instructions in the memory to implement the effective crack identification method based on random forest according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the effective crack identification method based on random forest according to any one of claims 1 to 7 is implemented.

Images