A rock layer attribute classification method based on fault constraint surface

Abstract

The present invention relates to a rock formation attribute classification method based on fault constraint surface, and belongs to the field of three-dimensional geological modeling of mines. The method extracts the spatial characteristics and attribute characteristics of the fault structure from the exploration line profile and the geological plane map, converts them into the drilling data format to participate in the rock formation attribute classification, divides the data set into the upper plate data set, the lower plate data set and the data set not affected by the fault according to the label of the data set, and performs separate training according to different types of data sets to improve the accuracy of rock formation attribute classification and the characterization effect of the fault. The present invention can directly realize the characterization of the fault structure for the drilling data at the data set stage, so as to be flexibly applied to different and their learning models, and provide guarantee for the accuracy of rock formation attribute classification and modeling effect.

Description

Technical Field

[0001] The invention relates to a rock layer attribute classification method based on a fault constraint surface, and belongs to the field of three-dimensional geological modeling of mines. Background Art

[0002] At present, the classification of rock formation attributes is the basis for realizing three-dimensional geological modeling of borehole data based on machine learning. Machine learning methods can obtain feature space as much as possible when the borehole data space is sparsely distributed and the feature attributes are sparse, classify underground rock formation attributes, and finely simulate the spatial characteristics of geological objects by analyzing geological data. Using machine learning methods to classify rock formation attributes of borehole data can reduce mining sampling costs, improve modeling accuracy and efficiency, and improve the automation and intelligence of modeling.

[0003] However, in the process of machine learning for rock formation attribute classification, there is interference from fault structures on the feature extraction of rock formation attribute classification. If fault structures are developed in the strata, the faults will cause the rock formations on both sides to shift, corresponding to the abnormal distribution of rock formation attribute units. Under the condition of limited data density, it is difficult to extract the characteristic laws of the faults, resulting in poor simulation of the spatial characteristics of the geological body. Existing technologies, such as the "Fault Closure Analysis Method and Device under Structural Constraints" with application number 2019112479240, mesh the cross-section of the fault to achieve the model construction of the fault plane. In this technical solution, modeling is performed based on the collected data, and the collected data is adapted to the model, but the collected data cannot be directly migrated to other models well, and the flexibility of using data is poor when simulating the spatial characteristics of geological bodies. Summary of the invention

[0004] In order to solve the problems existing in the above-mentioned prior art, the present invention provides a rock formation attribute classification method based on fault constraint surface. The present invention can directly realize the characterization of fault structure for drilling data at the data set stage, so as to be flexibly applied to various machine learning models, thereby providing guarantee for the rock formation attribute classification accuracy and modeling effect.

[0005] To achieve the above object, the technical solution provided by the present invention is: a rock formation attribute classification method based on fault constraint surface, which is operated according to the following steps:

[0006] (1) Preprocessing of borehole data training set: Obtain the borehole data set to be processed, preprocess each borehole data point according to the depth range and corresponding three-dimensional coordinates corresponding to different rock formations, obtain the starting and ending three-dimensional coordinates of each rock formation, and assign corresponding labels to the original borehole data points according to the corresponding rock formation attributes to form a standard format data set. Resample the standard format data set, that is, equidistant encryption is performed between the starting and ending data of the same rock formation, and abnormal data in the data set is eliminated to form a borehole data training set.

[0007] (2) Constructing the fault constraint surface: combining the exploration line profile and the geological plane map, fitting the strike and shape of the fault surface in the entire space, as well as the distribution of rock layers on both sides of the fault surface on each exploration line, obtaining the coordinates of the fault line and the exploration line in the plane, and then preprocessing the data according to step (1). The spatial range of the fault surface in the classification area is used as the spatial constraint data, and the distribution of rock layers on both sides of the fault surface on each exploration line is used as the attribute constraint data. The spatial constraint data is used to divide the drilling data training set into different parts of the training set. The attribute constraint data is added to the corresponding training set as supplementary data according to the spatial constraint.

[0008] (3) Calculate the mark and divide the data set: traverse the borehole data training set, and determine whether each borehole data point in the borehole data training set in step (1) is within the range of both sides of the fault plane according to the influence range of the fault plane spatial constraint data and the attribute constraint data. If the coordinates of the borehole data point are not within the coordinate range of the fault plane, the borehole data point is less affected by the fault structure and is not marked. If the coordinates of the borehole data point are within the coordinate range of the fault plane, the plane normal vector and the distance from the point to the plane are used to calculate its relative position to the fault plane. According to the positive or negative vertical distance, combined with the positional relationship of the extracted rock layer attribute coordinate sets on both sides of the fault plane, the borehole data point is given a corresponding mark of top or mark of below.

[0009] According to the corresponding labels of the dataset, top, below and no label, the dataset is divided into the upper plate dataset, the lower plate dataset and the dataset not affected by the fault, and the extracted datasets on both sides of the fault are added to the corresponding dataset according to the labels;

[0010] (4) Training classifiers: Based on the labels of the data sets, the same machine learning algorithm is used to train different models for the hanging wall data set, the footwall data set, and the data set not affected by the fault. That is, for the hanging wall of the fault, the data labeled top is used for model training; for the footwall of the fault, the data labeled below is used for model training; for the data not affected by the fault, that is, the unlabeled data, the unlabeled data is used for model training;

[0011] (5) Output classification results: In the classification prediction process, according to the range of the data set, training classification prediction is performed in the corresponding classifier, and finally the results are merged to obtain the final test and classification results.

[0012] The method further includes, in step (1) of isometric encryption, performing isometric resampling between the starting depth coordinate and the cutoff depth coordinate of the same rock layer at a certain interval, converting the same rock layer into a rock layer sequence having three-dimensional coordinates and corresponding rock layer attributes, and arranging the corresponding data sets. Then, a maximum-minimum standardization processing method is used to reduce the influence of the dimensional differences between the features on the classification results, and a data set that can be used for rock layer attribute classification is obtained after the processing.

[0013] The method further includes: performing the following steps in step (2):

[0014] 1) Extract the plane coordinates of the fault line and the exploration line on the geological plane map: Combine the distribution of each exploration line on the fault line in the geological plane map, and obtain the X and Y coordinates of the fault line and the exploration line in the plane through formula 1. Formula 1 is as follows:

[0015] (1)

[0016] Where, X i , Y i is the X, Y coordinates of any point, X1, Y1 are the starting coordinates of the upper left corner of the geological plane, dx, dy are the coordinates X i , Y i The distance from X1 and Y1 in the figure, b is the scale of the figure;

[0017] When the fault is relatively regular, the overall coordinate range can be directly fitted according to the starting point and the end point; when the fault has an irregular shape, the equidistant coordinates on the entire fault line are fitted by encrypting the sampling points;

[0018] 2) Extract the coordinates of the fault profile and the rock attributes on both sides from the exploration line profile: Fit the fault space and attribute characteristics through the exploration line profile, and extract the coordinates of any point on the fault from it through Formula 2, where Formula 2 is as follows:

[0019] (2)

[0020] Where, X i , Y i , Z i X is the coordinate of any point on the fault in the figure, F1 , X F2 is the coordinate of the intersection of the fault and the exploration line, Z F1is the coordinate elevation of the intersection point, α is the angle between the survey line and the X axis, θ is the angle between the fault and the vertical direction, and z is the depth corresponding to the coordinate in the profile;

[0021] When the fault is relatively regular, the three-dimensional coordinates of any point on the fault line in the section can be directly calculated according to Formula 2. If it is a fault with complex morphology, the sampling points are encrypted to make the distribution regular within a smaller distance.

[0022] After the coordinates of the fault plane in the section are obtained, the two sides of the fault plane are divided into different planes, denoted as Ftop and Fbelow, respectively, to extract the different rock strata distribution characteristics of the upper and lower walls of the fault;

[0023] 3) Fault constraint surface design: According to the distribution range of the fault plane in each exploration line profile and geological plane map obtained in the above steps, the distribution of the entire fault plane in three-dimensional space is fitted; when classifying rock formation attributes, the sample data set is divided into different parts to train the classifier according to the three-dimensional coordinate range of the fault plane; the rock formation attribute characteristics corresponding to the upper and lower plates on both sides of the fault plane are added to the data set in the form of drilling data sets, and respectively participate in the corresponding rock formation attribute classification parts; the spatial constraint effect and attribute constraint effect on the fault structure are realized in the rock formation attribute classification stage.

[0024] The method also includes: in step (3), the influence range of the fault is mainly determined by whether the X, Y coordinates of the borehole data point are within the fitted fault X, Y coordinate range; if the coordinates of the borehole data point are within the fault coordinate range, it means that the borehole data point is affected by the fault; the relative position of the borehole data point to the fault plane is calculated by the plane normal vector and the distance from the point to the plane; according to the positive or negative vertical distance, combined with the positional relationship of the rock formation attribute coordinate sets on both sides of the fault plane, it is determined whether the borehole data point belongs to the upper plate or the lower plate of the fault and a corresponding mark is given; if the coordinates of the borehole data point are outside the fault coordinate range, it means that the borehole data point is not affected by the fault and no mark is given; in this process, if the borehole data point is affected by two or more fault planes at the same time, it is regarded as the same part of the data set.

[0025] The method also includes dividing the data set into different parts according to the labels of the data set in step (3), adding the extracted fault data set as supplementary data to the upper plate data and the lower plate data to characterize the attribute characteristics of the fault, and the unlabeled data is not affected by the fault and does not need to be expanded.

[0026] According to the above technical solution, the rock formation attribute classification method based on fault constraint surface provided by the present invention extracts the spatial characteristics and attribute characteristics of the fault structure from the exploration line profile and geological plane map, converts them into the drilling data format to participate in the rock formation attribute classification, divides the data set into the upper plate data set, the lower plate data set and the data set not affected by the fault according to the label of the data set, and performs separate training according to different types of data sets to improve the accuracy of rock formation attribute classification and the characterization effect of the fault. Compared with the prior art, the present invention has the following advantages:

[0027] (1) Because the technical solution adopted by the present invention combines the exploration line profile with the geological plane map and classifies the data set according to the distribution characteristics of different rock layer attributes, the present invention can directly train the data with different characteristics of the upper and lower plates separately according to the data set type, thereby achieving a good modeling effect of fault dislocation.

[0028] (2) Because the technical solution adopted by the present invention fits the spatial distribution of the fault structure and the corresponding abnormal attribute characteristics, and participates in the rock formation attribute classification in the form of a data set, the characteristics of the fault are reflected in the data processing stage, which effectively solves the problem of identifying the abnormal characteristics of the fault structure during the classification process, and improves the accuracy of rock formation attribute classification and the characterization effect of the fault. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Flowchart of rock formation attribute classification based on fault constraint surfaces;

[0030] Figure 2 Fault constraint surface construction flow chart;

[0031] Figure 3 Extraction of fault lines and exploration lines from geological plane map;

[0032] Figure 4 Extract the fault profile coordinates and the rock layer attribute diagram on both sides from the exploration line profile;

[0033] Figure 5 Schematic diagram of fault impact range calculation. DETAILED DESCRIPTION

[0034] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments, but the protection scope of the present invention is not limited to the following embodiments.

[0035] In the technical solution provided by the present invention, a rock formation attribute classification method based on a fault constraint surface is operated according to the following steps:

[0036] (1) Preprocessing of the borehole data training set: By integrating the borehole data points to be processed, the depth ranges and corresponding three-dimensional coordinates corresponding to different rock formations are obtained, and corresponding labels are assigned to the data set according to the rock formation properties. The borehole data points are resampled using an equidistant encryption method, and abnormal data in the data set are removed to form a borehole data training set. In this embodiment, the following steps are specifically included:

[0037] 1) Preprocessing of borehole data points: First, the information contained in the borehole data points is integrated to obtain an initial data sorting table that can be used for rock formation attribute classification. The sorting results are shown in Table 1.

[0038] Table 1 Drilling data point arrangement table

[0039] Drilling number X coordinate Y coordinate Z coordinate Start Depth Cut-off Depth Rock properties Geological code Lithology description

[0040] 2) Resampling of borehole data points: After obtaining the table of borehole data points, it is necessary to produce sample data used in the machine learning method, with the three-dimensional coordinates of the borehole data points as classification features and the rock formation attributes as classification labels. Although the table already includes the three-dimensional coordinates corresponding to the starting depth and cutoff depth of each rock formation, the machine learning algorithm cannot recognize the geological significance of the same attributes between the top and bottom plates of various rock formations. It is necessary to resample the borehole data points, that is, to resample the starting depth coordinates and cutoff depth coordinates of the same rock formation at a certain interval, and convert them into a rock formation sequence with three-dimensional coordinates and corresponding rock formation attributes.

[0041] 3) Standardization of drilling data points: Data standardization can map values ​​of different ranges to a unified interval. The interval is usually selected as [0, 1] or [-1, 1]. The main purpose is to eliminate the dimensional differences between different features so that the features have similar weights and scales, which can be better applied to machine learning algorithms. The data format of the standard format data set is (x, y, z, v), where x, y, and z are three-dimensional coordinates and v is the corresponding attribute. The data set obtained after standardization is shown in Table 2.

[0042] Table 2 Drilling dataset example table

[0043] X coordinate Y coordinate Z coordinate Rock Formation Properties V 0.3107566222 0.36577554355 -0.0030 1 0.3107773067 0.36577873712 -0.3530 3 0.3107644605 0.36576353285 -0.5800 5 0.3106581341 0.36578150223 -0.6511 7 0.3106570972 0.36578796291 -0.8533 9

[0044] (2) Construct fault constraint surface: Figure 2As shown in the steps, the exploration line profile is combined with the geological plane map to fit the distribution of the fault plane in the entire space and the distribution of rock layers on both sides of the fault plane on each exploration line. The data is then processed into the data format in step (1) to participate in the rock layer attribute classification process, with the spatial range of the fault plane in the classification area as the spatial constraint and the distribution of rock layers on both sides of the fault plane on each exploration line as the attribute constraint. The specific operation is as follows:

[0045] 1) Extract the plane coordinates of fault lines and exploration lines on the geological plane map: Figure 3 As shown in the figure, the geological plane map contains the coordinate range, scale and other spatial information of the entire map, and the map will show the direction and shape of the fault line in the plane. Combined with the distribution of each exploration line on the fault line, the X and Y coordinates of the fault line and the exploration line in the plane can be obtained. The specific extraction method is shown in Formula 1:

[0046] (1)

[0047] Where Xi, Yi are the X, Y coordinates of any point in the map, X1, Y1 are the starting coordinates of the upper left corner of the geological plane map, dx, dy are the distances between the required coordinates Xi, Yi and X1, Y1 in the map, and b is the map scale.

[0048] In this embodiment, if Figure 3 As shown in the figure, (X1, Y1), ..., (X4, Y4) are the XY longitude and latitude coordinates of the four vertices of the geological plane map; the red F1 and F2 are two fault lines, (X F1st ,Y F1st ),…,(X F2st ,Y F2st ) The four coordinates are the corresponding longitude and latitude coordinates of the two faults in the figure; purple K1, K2, K3 are three exploration lines, (X K1st ,Y K1st ),…,(X K2st ,Y K2st ) are the starting and ending longitude and latitude coordinates of the three survey lines in the figure; dx and dy are the distances between the two coordinate points in the x and y directions.

[0049] When the fault is relatively regular, the overall coordinate range can be directly fitted based on the starting point and the end point. When the fault has an irregular shape, the equidistant coordinates along the entire fault line are fitted by encrypting the sampling points.

[0050] 2) Extract the fault profile coordinates and rock properties on both sides from the exploration line profile: Figure 4As shown in the figure, the exploration line profile can show the distribution trend of underground fault structure and the distribution range of the coordinates within the profile and the corresponding rock attributes, fit the fault space and attribute characteristics, from which the coordinates of any point on the fault can be extracted as shown in Formula 2:

[0051] (2)

[0052] Where Xi, Yi, Zi are the coordinates of any point on the fault in the diagram, XF1 and XF2 are the coordinates of the intersection of the fault and the exploration line, ZF1 is the elevation of the intersection, α is the angle between the exploration line and the X-axis, θ is the angle between the fault and the vertical direction, and z is the depth corresponding to the coordinate in the profile.

[0053] In this embodiment, if Figure 4 As shown, A, B, and C correspond to three rock formation attributes; (X F1 ,Y F1 ) is the above Figure 3 The coordinates of the intersection of the regular fault F1 and the survey line; top , F1 below is the corresponding upper and lower plates of the fault; θ is the angle between the fault and the vertical direction; (Xi, Yi, Zi, A) (Xi, Yi, Zi, B) are the different rock attributes extracted from the upper and lower plates of the fault line when the three-dimensional coordinates are the same. The three-dimensional coordinates of any point on the fault line in the section can be calculated. If it is a complex fault, the sampling points are also encrypted to make the distribution regular within a small distance.

[0054] After the coordinates of the fault plane in the section are obtained, the corresponding attributes need to be extracted. However, the two sides of the fault plane are the corresponding upper and lower plates of the fault, which have different rock layer attribute distribution characteristics, which is also the main reason for the decrease in classification accuracy and modeling accuracy. The approach taken here is to divide the two sides of the fault plane into different planes, recorded as Ftop and Fbelow, respectively, to extract the different rock layer distribution characteristics of the upper and lower plates of the fault.

[0055] 3) Fault constraint surface design: After the above steps, the distribution range of the fault plane in each exploration line profile and geological plane map can be obtained, thereby fitting the distribution of the entire fault plane in three-dimensional space. When classifying rock formation attributes, the sample data set is divided into different parts to train the classifier according to the three-dimensional coordinate range of the fault plane. At the same time, in order to make up for the lack of spatial density of the data set and the problem of fault characterization, the rock formation attribute characteristics of the upper and lower plates on both sides of the fault plane are added to the data set in the form of drilling data sets, and participate in the corresponding rock formation attribute classification parts respectively. The spatial constraint effect and attribute constraint effect of the fault structure are realized in the rock formation attribute classification stage.

[0056] (3) Calculate the labels and divide the data set: Figure 1 As shown, the data set is traversed, and the borehole data point is determined to be within the range of both sides of the fault plane according to the influence range of the fault plane. If the coordinates of the borehole data point are not within the coordinate range of the fault plane, it means that the borehole data point is less affected by the fault structure and does not need to be marked. If the coordinates of the borehole data point are within the coordinate range of the fault plane, the plane normal vector and the distance from the point to the plane are used to calculate its relative position to the fault plane. According to the positive and negative vertical distance, combined with the positional relationship of the extracted rock layer attribute coordinate sets on both sides of the fault plane, the borehole data point is marked top or below accordingly; in this embodiment, Figure 5 As shown in the figure, (Xi, Yi) are the latitude and longitude coordinates of the corresponding drilling data points in the figure; top and below are the upper and lower plates of the fault respectively.

[0057] According to the marking of the data set, the data set is divided into an upper plate data set, a lower plate data set and a data set not affected by the fault, which are represented by the marking top, the marking below and the unmarked data set, and the extracted data sets on both sides of the fault are added to the corresponding data set according to the marking; according to the marking of the data set, the data set is divided into different parts. For the upper plate data and the lower plate data, the extracted fault data set is added as supplementary data to characterize the attribute characteristics of the fault. The unmarked data is not affected by the fault and does not need to be expanded.

[0058] In step (3), if Figure 5 As shown in the figure, the influence range of the fault is mainly determined by whether the X and Y coordinates of the borehole data point are within the range of the fitted fault X and Y coordinates. If the coordinates of the borehole data point are within the range of the fault coordinates, it means that the borehole data point is affected by the fault. The relative position of the borehole data point to the fault plane is calculated by the plane normal vector and the distance from the point to the plane. According to the positive and negative vertical distance, combined with the positional relationship of the rock layer attribute coordinate sets on both sides of the fault plane, the borehole data point is judged to belong to the upper plate or lower plate of the fault and marked accordingly. If the coordinates of the borehole data point are outside the range of the fault coordinates, it means that it is not affected by the fault and no mark is given. In this process, if the borehole data point is affected by two or more fault planes at the same time, it is regarded as the same part of the data set.

[0059] (4) Classifier training: According to the labeling of the data set, different models are trained using the same machine learning algorithm. In this embodiment, an SVM algorithm model is built and grid search is used to adjust the parameters. For the upper plate of the fault, the data labeled top is used for model training; for the lower plate of the fault, the data labeled below is used for model training; for data not affected by the fault, that is, unlabeled data, the unlabeled data is used for model training. This process takes into account that in the process of rock layer attribute classification, the movement of the fault causes a sudden change in the original rock layer distribution pattern. If the entire data set is still used for data with different characteristics of the upper and lower plates, the algorithm will learn features that do not conform to the original rules, affecting the classification effect.

[0060] (5) Output classification results: In the classification prediction process, according to the range of the data set, training classification prediction is performed in the corresponding classifier, and finally the results are merged to obtain the final test and classification results.

Claims

1. A rock formation attribute classification method based on fault constraint surface, characterized in that Proceed as follows: (1) Preprocessing of borehole data training set: Obtain the borehole data set to be processed, preprocess each borehole data point according to the depth range and corresponding three-dimensional coordinates corresponding to different rock formations, obtain the starting and ending three-dimensional coordinates of each rock formation, and assign corresponding labels to the original borehole data points according to the corresponding rock formation attributes to form a standard format data set. Resample the standard format data set, that is, equidistant encryption is performed between the starting and ending data of the same rock formation, and abnormal data in the data set is eliminated to form a borehole data training set; In isometric densification, the starting depth coordinates and the cutoff depth coordinates of the same rock layer are resampled at equal intervals according to a certain interval, and converted into a rock layer sequence with three-dimensional coordinates and corresponding rock layer attributes. After being sorted into the corresponding data set, the maximum-minimum standardization processing method is used to reduce the impact of the dimensional difference between the features on the classification results. After processing, a data set that can be used for rock layer attribute classification is obtained; (2) Constructing the fault constraint surface: combining the exploration line profile and the geological plane map, fitting the strike and shape of the fault surface in the entire space, as well as the distribution of rock layers on both sides of the fault surface on each exploration line, obtaining the coordinates of the fault line and the exploration line in the plane, and then preprocessing the data according to step (1). The spatial range of the fault surface in the classification area is used as the spatial constraint data, and the distribution of rock layers on both sides of the fault surface on each exploration line is used as the attribute constraint data. The spatial constraint data is used to divide the drilling data training set into different parts of the training set. The attribute constraint data is added to the corresponding training set as supplementary data according to the spatial constraint. Follow these steps: 1) Extract the plane coordinates of the fault line and the exploration line on the geological plane map: Combine the distribution of each exploration line on the fault line in the geological plane map, and obtain the X and Y coordinates of the fault line and the exploration line in the plane through formula 1. Formula 1 is as follows: Where, X i , Y i is the X, Y coordinates of any point, X1, Y1 are the starting coordinates of the upper left corner of the geological plane, dx, dy are the coordinates X i , Y i The distance from X1 and Y1 in the figure, b is the scale of the figure; When the fault is relatively regular, the overall coordinate range is directly fitted according to the starting point and the end point; when the fault has an irregular shape, the equidistant coordinates on the entire fault line are fitted by encrypting the sampling points; 2) Extract the coordinates of the fault profile and the rock attributes on both sides from the exploration line profile: Fit the fault space and attribute characteristics through the exploration line profile, and extract the coordinates of any point on the fault from it through Formula 2, where Formula 2 is as follows: Where, X i , Y i , Z i X is the coordinate of any point on the fault in the figure, F1 is the coordinate of the intersection of the fault and the exploration line, Z F1 is the coordinate elevation of the intersection point, α is the angle between the survey line and the X-axis, θ is the angle between the fault and the vertical direction, and z is the depth corresponding to the coordinate in the profile; When the fault is relatively regular, the three-dimensional coordinates of any point on the fault line in the section are directly calculated according to Formula 2. If it is a fault with complex morphology, the sampling points are encrypted to make the distribution regular within a small distance. After the coordinates of the fault plane in the section are obtained, the two sides of the fault plane are divided into different planes, denoted as Ftop and Fbelow, respectively, to extract the different rock strata distribution characteristics of the upper and lower walls of the fault; 3) Fault constraint surface design: According to the distribution range of the fault plane in each exploration line profile and geological plane map obtained in the above steps, the distribution of the entire fault plane in three-dimensional space is fitted; when classifying rock formation attributes, the sample data set is divided into different parts to train the classifier according to the three-dimensional coordinate range of the fault plane; the rock formation attribute characteristics corresponding to the upper and lower plates on both sides of the fault plane are added to the data set in the form of drilling data sets, and respectively participate in the corresponding rock formation attribute classification parts; the spatial constraint effect and attribute constraint effect on the fault structure are realized in the rock formation attribute classification stage; (3) Calculate the mark and divide the data set: traverse the borehole data training set, and determine whether each borehole data point in the borehole data training set in step (1) is within the range of both sides of the fault plane according to the influence range of the fault plane spatial constraint data and the attribute constraint data. If the coordinates of the borehole data point are not within the coordinate range of the fault plane, the borehole data point is less affected by the fault structure and is not marked; if the coordinates of the borehole data point are within the coordinate range of the fault plane, use the plane normal vector and the distance from the point to the plane to calculate its relative position to the fault plane. According to the positive or negative vertical distance, combined with the positional relationship of the extracted rock layer attribute coordinate sets on both sides of the fault plane, give the borehole data point a corresponding mark top or mark below; According to the corresponding labels of the dataset, top, below and no label, the dataset is divided into the upper plate dataset, the lower plate dataset and the dataset not affected by the fault, and the extracted datasets on both sides of the fault are added to the corresponding dataset according to the labels; The influence range of the fault is determined by whether the X, Y coordinates of the borehole data point are within the range of the fitted fault X, Y coordinates. If the coordinates of the borehole data point are within the range of the fault coordinates, it means that the borehole data point is affected by the fault. The relative position of the borehole data point to the fault plane is calculated by the plane normal vector and the distance from the point to the plane. According to the positive and negative vertical distance, combined with the positional relationship of the extracted rock formation attribute coordinate sets on both sides of the fault plane, it is determined whether the borehole data point belongs to the upper plate or lower plate of the fault and is marked accordingly. If the coordinates of the borehole data point are outside the range of the fault coordinates, it means that it is not affected by the fault and is not marked. In this process, if the borehole data point is affected by two or more fault planes at the same time, it is regarded as the same part of the data set. (4) Training classifiers: Based on the labels of the data sets, the same machine learning algorithm is used to train different models for the hanging wall data set, the footwall data set, and the data set not affected by the fault. That is, for the hanging wall of the fault, the data labeled top is used for model training; for the footwall of the fault, the data labeled below is used for model training; for the data not affected by the fault, that is, the unlabeled data, the unlabeled data is used for model training; Output classification results: In the process of classification prediction, according to the range of the data set, training classification prediction is performed in the corresponding classifier, and finally the results are merged to obtain the final test and classification results.

2. The rock formation attribute classification method based on fault constraint surface according to claim 1 is characterized in that: In step (3), the dataset is divided into different parts according to its labels. For the upper plate data and lower plate data, the extracted fault dataset is added as supplementary data to characterize the attribute characteristics of the fault. The unlabeled data is not affected by the fault and does not need to be expanded.

Images