CN115840921B - Rock mass quality grading method based on machine learning - Google Patents

Rock mass quality grading method based on machine learning Download PDF

Info

Publication number
CN115840921B
CN115840921B CN202310159304.1A CN202310159304A CN115840921B CN 115840921 B CN115840921 B CN 115840921B CN 202310159304 A CN202310159304 A CN 202310159304A CN 115840921 B CN115840921 B CN 115840921B
Authority
CN
China
Prior art keywords
rock mass
quality grading
mass quality
index
grading
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202310159304.1A
Other languages
Chinese (zh)
Other versions
CN115840921A (en
Inventor
李地元
罗平框
赵君杰
马金银
茹文凯
夏志远
吕欣欣
周奥辉
蒋京泰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Central South University
Original Assignee
Central South University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Central South University filed Critical Central South University
Priority to CN202310159304.1A priority Critical patent/CN115840921B/en
Publication of CN115840921A publication Critical patent/CN115840921A/en
Application granted granted Critical
Publication of CN115840921B publication Critical patent/CN115840921B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02PCLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30Computing systems specially adapted for manufacturing

Landscapes

  • Earth Drilling (AREA)
  • Investigation Of Foundation Soil And Reinforcement Of Foundation Soil By Compacting Or Drainage (AREA)

Abstract

The invention relates to a rock mass quality grading method based on machine learning, which comprises the following steps: (1) Acquiring drilling information, and establishing a drilling image database according to the drilling information; (2) Setting minimum measurement calculation units, scoring and summing the rock mass quality grading indexes of each minimum measurement calculation unit to obtain a rock mass quality grading score; (3) Establishing a rock mass quality grading database, and recording rock mass quality grading indexes and rock mass quality grading scores of each drilling hole; (4) Processing the rock mass quality grading database based on a machine learning method, and obtaining a machine learning prediction model through training; (5) And inputting the rock mass quality grading index data into a machine learning prediction model to obtain a rock mass quality grading result. The rock mass quality grading method based on machine learning can improve the utilization rate of drilling and can obtain rock mass quality grading results more quickly.

Description

Rock mass quality grading method based on machine learning
Technical Field
The invention relates to a rock mass quality evaluation method for mining engineering and underground rock engineering, in particular to a rock mass quality grading method based on machine learning.
Background
In the construction and exploitation process of underground metal mines, the quality of rock mass is an important basis and precondition for evaluating the engineering characteristics and stability of the rock mass.
In the method, various mechanical indexes of a drilling rock core are required to be obtained as basic basis of rock mass classification, and the joint fracture occurrence information of the rock is one of important indexes for evaluating the rock mass classification.
However, core dislocation often occurs in the coring and core placement processes, and real surrounding rock joint fracture occurrence information cannot be obtained, so that a more scientific and effective method is needed to measure and count the joint fracture of a rock body, at present, when the quality of underground metal mine rock mass is evaluated in China, a large amount of time and effort are needed to be spent on carrying out a large amount of field work by researchers based on field joint fracture investigation results of a roadway or stope typical of a certain middle section, however, the investigation results often have limitations, and only rock mass quality evaluation results of local areas can be reflected, and a large amount of geological drilling holes are needed to be constructed in the underground metal mine construction and exploitation processes to reach the results of exploratory and geological investigation.
In view of the above problems, the present invention provides a rock mass quality grading method based on machine learning.
Disclosure of Invention
The rock mass quality grading method based on machine learning can improve the utilization rate of drilling, reduce the time of on-site investigation and measurement of workers, and can obtain rock mass quality grading results more quickly through machine learning.
In order to solve the technical problems, the invention provides a rock mass quality grading method based on machine learning, which is characterized by comprising the following steps:
(1) Acquiring drilling information, wherein the drilling information comprises drilling images, drilling depth, azimuth, inclination angle and serial numbers of the drilling holes and serial numbers of cores corresponding to the drilling holes, and establishing a drilling image database according to the drilling information;
(2) Setting minimum measurement calculation units, scoring and summing rock mass quality grading indexes of each minimum measurement calculation unit based on RMR rock mass quality grading standards, and obtaining rock mass quality grading scores of each minimum measurement calculation unit;
(3) Establishing a rock mass quality grading database, and recording the rock mass quality grading index and the rock mass quality grading score of each drilling hole into the rock mass quality grading database;
(4) Processing the rock mass quality grading database based on at least one machine learning method, and obtaining a machine learning prediction model through training;
(5) And inputting various rock mass quality grading index data acquired from the rock mass in the drilling depth range into a machine learning prediction model, wherein the model prediction result is the rock mass quality grading result.
In particular, the borehole image, borehole depth, azimuth and inclination of the borehole are obtained by a borehole television.
Preferably, the number of the drill hole corresponds to the number of the core corresponding to the drill hole one by one.
Specifically, the minimum measurement calculation unit is a single drilling hole, and the rock mass quality grading index comprises a rock mass joint interval index, a rock mass joint state index, a complete rock strength index, a rock mass index, a groundwater state index and a ground stress index.
Further, according to the drilling image of the single drilling hole, counting and recording the number of the joint cracks in the drilling hole, obtaining the total number of the joint cracks of the drilling hole passing through the rock mass in the depth range of the single drilling hole, and calculating the rock mass joint interval index:
the joint spacing index is equal to the total number of joint cracks of the borehole passing through the rock mass within the borehole depth/the borehole depth range of the single borehole.
Further, according to the drilling image, counting the joint state of a single drilling hole passing through the rock mass joint surface, and obtaining the joint state index of the rock mass in the single drilling depth range, wherein the joint state comprises the roughness degree, the corrosion condition and the closure degree of the drilling hole passing through the rock mass joint surface.
Preferably, the machine learning method comprises a random forest regression algorithm and an Adaboost regression algorithm, and the rock mass quality classification database is subjected to machine learning by using the random forest regression algorithm and the Adaboost regression algorithm.
Specifically, the step of processing the rock mass quality grading database by using the random forest regression algorithm comprises the following steps:
a1 Selecting an optimal segmentation variable j and a segmentation point S for the rock mass quality classification database based on a segmentation selection formula, wherein the segmentation variable j and the segmentation point S divide the rock mass quality classification database into a data set S1 and a data set S2, and the segmentation selection formula is as follows:
Figure SMS_1
wherein :
Figure SMS_2
output means for the samples of data set S1, < >>
Figure SMS_3
For the sample output average value of the S2 data set, traversing the segmentation variable j, traversing the segmentation point S for the fixed segmentation variable j, and solving a segmentation pair (j, S) enabling the segmentation selection formula to reach the minimum value;
a2 For a selected said segmentation pair (j, s), dividing the region and determining the corresponding output value:
Figure SMS_4
Figure SMS_5
wherein x represents all rock mass quality grading indexes;
Figure SMS_8
a hierarchical database of quality of the rock mass for selection of the cut; />
Figure SMS_12
Representing a region where the value of the corresponding rock mass quality grading index to be segmented in the rock mass quality grading database is smaller than or equal to the segmentation point s, +.>
Figure SMS_16
Representing a region in the rock mass quality grading database, corresponding to the rock mass quality grading index to be segmented, having a value greater than the segmentation point s, < >>
Figure SMS_7
Representation->
Figure SMS_13
Or->
Figure SMS_17
Number of samples in area, +.>
Figure SMS_20
Representing the divided sub-regions->
Figure SMS_6
and />
Figure SMS_10
,/>
Figure SMS_14
Representation->
Figure SMS_18
and />
Figure SMS_9
In a corresponding borehole rock mass quality grading result, < >>
Figure SMS_11
Representation->
Figure SMS_15
and />
Figure SMS_19
The average value of the output;
a3 Dividing the rock mass quality grading database into M sub-regions
Figure SMS_21
Establishing a final prediction model:
Figure SMS_22
wherein I represents an identity matrix.
Specifically, the step of processing the rock mass quality grading database by using the Adaboost regression algorithm comprises the following steps:
b1 Initializing the rock mass quality grading index weight and grading all the rock mass quality grading indexes
Figure SMS_23
Is initialized to 1/N:
Figure SMS_24
Figure SMS_25
Figure SMS_26
representing the number of loop iterations;
b2 Performing loop iteration on the weighted rock mass quality grading index, wherein the steps of the loop iteration comprise:
b1 Sample distribution of the rock mass quality grading index is as follows
Figure SMS_27
On the basis of (1) training a weak classifier using a training set +.>
Figure SMS_28
;
b2 Calculating a weak classifier
Figure SMS_29
Maximum error on training set +.>
Figure SMS_30
Figure SMS_31
wherein ,
Figure SMS_32
representing weak classifier->
Figure SMS_33
Sample set of quality grading indicators for said rock mass>
Figure SMS_34
Is predicted by->
Figure SMS_35
Representing sample set +.>
Figure SMS_36
A corresponding target value;
b3 Calculating the weak classifier
Figure SMS_37
Sample of quality grading index for each of said rock masses +.>
Figure SMS_38
Relative error of (2):
Figure SMS_39
b4 According to the relative error of the sample
Figure SMS_40
Calculating the current weak classifier +.>
Figure SMS_41
Error rate of (c):
Figure SMS_42
wherein ,
Figure SMS_43
representing the weight value corresponding to each variable;
b5 Updating the current weak classifier
Figure SMS_44
Weight of (2):
Figure SMS_45
wherein ,
Figure SMS_46
representing weak classifier->
Figure SMS_47
Weight coefficient of>
Figure SMS_48
Weight representing last updated sample point, +.>
Figure SMS_49
Weight of sample point representing the current update, +.>
Figure SMS_50
Representing the normalization factor;
b3 End of (d)
Figure SMS_51
And (5) performing round iteration to finally obtain a strong regression:
Figure SMS_52
wherein ,
Figure SMS_53
all->
Figure SMS_54
Middle of (a) i.e. the weighted output results of all weak learners, +.>
Figure SMS_55
Preferably, based on the random forest regression algorithm and the Adaboost regression algorithm, a random forest model and an Adaboost regression algorithm model are obtained, and the average value of the prediction results of the random forest model and the Adaboost regression algorithm model is taken as a final rock mass quality classification result.
Through the technical scheme, the invention has the following beneficial effects:
according to the invention, the drilling information containing the drilling image is scored by utilizing the RMR (rock mass grade) rock mass grading standard, and rock mass grading indexes of the minimum measurement calculation units are summed to obtain the RMR (rock mass grade) rock mass grading score, so that the grading result is more reliable; the rock mass grading indexes of each drilled hole and the rock mass grading scores are established into a rock mass grading database and are processed by a machine learning algorithm to obtain a machine learning prediction model, the rock mass grading indexes of a single minimum measurement and calculation unit are input into the machine learning prediction model to obtain more accurate grading results, the utilization rate of geological drilling is improved, rock mass grading can be performed more accurately and efficiently, transparent geology is facilitated to be realized in the exploitation process of underground metal mines, corresponding countermeasures are adopted, and the safety production efficiency of the mines is improved.
Other advantages and technical effects of the preferred embodiments of the present invention will be further described in the following detailed description.
Drawings
FIG. 1 is a method flow diagram of a machine learning based rock mass quality grading method of the present invention;
FIG. 2 is a flow chart of rock mass quality classification results obtained by a rock mass quality classification prediction model in the rock mass quality classification method based on machine learning of the present invention;
FIG. 3 is a schematic illustration of joint cracks in a machine learning based rock mass quality grading method of the present invention;
FIG. 4 is a schematic illustration of joint fracture states in a machine learning based rock mass quality grading method of the present invention;
fig. 5 is a schematic diagram of point load intensity test and equivalent diameter calculation in the machine learning based rock mass grading method of the present invention.
Detailed Description
The following describes specific embodiments of the present invention in detail with reference to the drawings. It should be understood that the detailed description and specific examples, while indicating and illustrating the invention, are not intended to limit the invention.
In addition, the technical features of the different embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
As shown in fig. 1, as an embodiment of the rock mass quality grading method based on machine learning provided by the invention, the method comprises the following steps:
(1) Acquiring drilling information, wherein the drilling information comprises drilling images, drilling depth, azimuth, inclination angle and serial numbers of drilling holes and serial numbers of cores corresponding to each drilling hole, and establishing a drilling image database according to the drilling information;
(2) Setting minimum measurement calculation units, scoring and summing rock mass quality grading indexes of each minimum measurement calculation unit based on RMR (rock mass grade) rock mass quality grading standards to obtain rock mass quality grading scores of each minimum measurement calculation unit;
(3) Establishing a rock mass quality grading database, and recording rock mass quality grading indexes and rock mass quality grading scores of each drilling hole into the rock mass quality grading database;
(4) Processing the rock mass quality grading database based on at least one machine learning method, and obtaining a rock mass quality grading prediction model through training;
(5) And inputting various rock mass quality grading index data acquired from the rock mass in the drilling depth range into a machine learning prediction model, wherein the model prediction result is the rock mass quality grading result.
Specifically, as shown in fig. 3 and fig. 4, the borehole television is slowly pushed into the borehole, the borehole television can acquire image information of the wall of the borehole and the depth, azimuth and inclination of the borehole, and the borehole television can send the acquired information to the computer, so that a user can view the information acquired by the borehole television on the computer through corresponding software, and the information is facilitated to be checked and stored.
Further, in order to facilitate statistics and summary arrangement of drill holes and hole cores obtained through geological drilling, the drill holes and cores need to be numbered, and the numbers of the drill holes and the cores corresponding to the drill holes are in one-to-one correspondence, so that the number and statistics of the drill holes and the cores are facilitated.
Specifically, a single drill hole is used as a minimum measurement calculation unit to improve accuracy of a grading result, drill hole depth of the single drill hole is measured through a drill hole television, rock mass in a drill hole depth range of the single drill hole is used as the minimum measurement calculation unit, scores of the minimum measurement calculation unit are obtained through calculation of rock mass joint interval indexes, rock mass joint state indexes, complete rock strength indexes, rock quality indexes, underground water state indexes and ground stress indexes of the minimum measurement calculation unit, and scores of the rock mass joint interval indexes, the rock mass joint state indexes, the complete rock strength indexes, the rock quality indexes, the underground water state indexes and the ground stress indexes of the minimum measurement calculation unit are summed to obtain a RMR (rock mass grade) rock mass grading score.
Further, as shown in fig. 3, the number of the joint cracks in the drill hole is recorded according to the drill hole image of the single drill hole, so as to obtain the total number of the joint cracks of the drill hole passing through the rock body in the depth range of the single drill hole, and the rock body joint interval index is calculated:
the joint spacing index is equal to the total number of joint cracks drilled through the rock mass over a single borehole depth/single borehole depth range.
Further, as shown in fig. 4, according to the drilling image, counting the joint state of a single drilling hole passing through the joint surface of the rock mass, wherein the joint state comprises the roughness degree, the corrosion condition and the closure degree of the drilling hole passing through the joint surface of the rock mass, and according to the joint state of the drilling hole passing through the joint surface, obtaining the joint state index of the rock mass within the depth range of the single drilling hole.
Further, in a single borehole, the cores corresponding to the boreholes are sampled, the number of samples sampled by each borehole is not less than 10, the samples are numbered and recorded, in a specific embodiment, the samples can be sampled by taking standard uniaxial compression samples (the size phi 50 x 100 mm) for uniaxial compression test and also can be sampled by taking irregular samples for point load test, and in order to simplify the description, the description is mainly made by taking the irregular samples for point load test, of course, the standard uniaxial compression samples can also be sampled for uniaxial compression test, and on the basis of knowing the technical scheme of the invention, the standard uniaxial compression samples can be adopted for uniaxial compression test by a person skilled in the art, so that the description is omitted.
In a specific embodiment, as shown in fig. 5, the sample size of the point load test needs to meet the requirement of the irregular rock point load test, so as to develop the irregular rock point load test and obtain the load when the rock breaks; calculating the equivalent diameter of the sample according to the size of the sample, calculating the point load intensity of the sample in combination with the load when the sample breaks, correcting the point load intensity, and calculating to obtain a corrected point load intensity index, thereby obtaining a complete rock intensity index, wherein the point load intensity correction method comprises the following steps:
a. unmodified point load intensity index calculation
Figure SMS_56
Figure SMS_57
Wherein Is the point load intensity index before correction, P Is the load of the sample during damage, de Is the equivalent diameter of the sample, W Is the average width of the smallest section passing through two loading points, and D Is the loading point distance;
b. point load intensity index correction
Figure SMS_58
Figure SMS_59
wherein Is50 F is a correction coefficient, m is a correction coefficient index, and 0.5 is taken as the corrected point load intensity index;
and obtaining the corrected point load intensity index, namely the rock intensity index through calculation.
Further, measuring the actual length of a rock core with the length of more than or equal to 10cm in a single hole, summing, and obtaining rock quality indexes of a rock body in a single drilling depth range according to drilling information, wherein the calculation formula is as follows:
Figure SMS_60
further, according to the drilling image information and the on-site investigation result, the development condition of the underground water in the single drilling hole is counted, and the underground water state of all rock bodies in the depth range of the single drilling hole can be judged through the water inflow amount of the drilling hole in unit time, and the underground water state index of the rock bodies in the drilling hole is obtained.
Further, the common measurement methods of the ground stress include a drilling pressure relief method, a hydraulic fracturing method and an acoustic emission method, and the drilling pressure relief test can be carried out by using a geological drill to drill holes in an underground metal mine so as to obtain the ground stress distribution condition of the rock mass and obtain the ground stress index.
And recording the calculated rock mass joint interval index, rock mass joint state index, complete rock strength index, rock quality index, underground water state index and ground stress index and the sum of the rock mass quality grading indexes in an established rock mass quality grading database so as to process the rock mass quality grading indexes by at least one machine learning method and obtain a rock mass quality grading prediction model through training.
Preferably, the rock mass quality classification database is learned by a random forest regression algorithm and an Adaboost regression algorithm.
Specifically, the step of processing the rock mass quality grading database by using a random forest regression algorithm comprises the following steps:
a1 Selecting an optimal segmentation variable j and a segmentation point S on the basis of a segmentation selection formula for the rock mass quality classification database, dividing the rock mass quality classification database into a data set S1 and a data set S2 by the segmentation variable j and the segmentation point S, wherein the segmentation selection formula is as follows:
Figure SMS_61
wherein :
Figure SMS_62
output means for the samples of data set S1, < >>
Figure SMS_63
For the sample output average value of the S2 data set, traversing the segmentation variable j, traversing the segmentation point S for the fixed segmentation variable j, and solving a segmentation pair (j, S) enabling the segmentation selection formula to reach the minimum value;
a2 For the selected segmentation pair (j, s), dividing the region and determining the corresponding output value:
Figure SMS_64
Figure SMS_65
wherein x represents all rock mass quality grading indexes;
Figure SMS_68
a hierarchical database of rock mass quality for selection of the cut;
Figure SMS_72
representing a region in the rock mass quality grading database where the value of the corresponding rock mass quality grading index to be sliced is less than or equal to the slicing point s,/for>
Figure SMS_76
Representing a region in the rock mass quality grading database corresponding to the rock mass quality grading index to be sliced, the value of which is greater than the slicing point s, +.>
Figure SMS_67
Representation->
Figure SMS_71
Or->
Figure SMS_75
Number of samples in area, +.>
Figure SMS_79
Representing the divided sub-regions
Figure SMS_66
and />
Figure SMS_70
,/>
Figure SMS_74
Representation->
Figure SMS_78
and />
Figure SMS_69
Corresponding drill rock mass quality grading score,/->
Figure SMS_73
Representation->
Figure SMS_77
and />
Figure SMS_80
The average value of the output;
a3 Dividing a rock mass quality grading database into M sub-regions
Figure SMS_81
Establishing a final prediction model:
Figure SMS_82
wherein I represents an identity matrix.
Specifically, the step of processing the rock mass quality grading database by using an Adaboost regression algorithm comprises the following steps:
b1 Initializing the weight of rock mass quality grading index, and grading all rock mass quality grading indexes
Figure SMS_83
Is initialized to 1/N:
Figure SMS_84
Figure SMS_85
Figure SMS_86
representing the number of loop iterations;
b2 The weighted rock mass quality grading index is subjected to cyclic iteration, and the steps of the cyclic iteration comprise:
b1 Sample distribution of rock mass quality grading index
Figure SMS_87
On the basis of (1) training a weak classifier using a training set +.>
Figure SMS_88
;
b2)Computing weak classifiers
Figure SMS_89
Maximum error on training set +.>
Figure SMS_90
Figure SMS_91
wherein ,
Figure SMS_92
representing weak classifier->
Figure SMS_93
Sample set of index for grading quality of all rock masses>
Figure SMS_94
Is predicted by->
Figure SMS_95
Representing sample set +.>
Figure SMS_96
A corresponding target value; />
b3 Calculating a weak classifier
Figure SMS_97
Sample of quality grading index for each rock mass +.>
Figure SMS_98
Relative error of (2):
Figure SMS_99
b4 According to the relative error of the sample
Figure SMS_100
Calculate the current weak classifier +.>
Figure SMS_101
Error rate of (c):
Figure SMS_102
wherein ,
Figure SMS_103
representing the weight value corresponding to each variable;
b5 Updating the current weak classifier
Figure SMS_104
Weight of (2):
Figure SMS_105
wherein ,
Figure SMS_106
representing weak classifier->
Figure SMS_107
Weight coefficient of>
Figure SMS_108
The weight of the sample point that represents the last update,
Figure SMS_109
weight of sample point representing the current update, +.>
Figure SMS_110
Representing the normalization factor;
b3 End of (d)
Figure SMS_111
And (5) performing round iteration to finally obtain a strong regression:
Figure SMS_112
wherein ,
Figure SMS_113
all->
Figure SMS_114
Middle of (a) i.e. the weighted output results of all weak learners, +.>
Figure SMS_115
Preferably, the rock mass quality grading index is processed based on a random forest regression algorithm and an Adaboost regression algorithm to obtain a random forest model and an Adaboost regression algorithm model, each rock mass quality grading index data acquired from the rock mass within the drilling depth range is input into the random forest model and the Adaboost regression algorithm model, and the average value of the prediction results of the random forest model and the Adaboost regression algorithm model is taken as the final rock mass quality grading result.
According to the rock mass quality grading method based on machine learning, provided by the invention, the drilling television is slowly pushed into the drilling hole to obtain drilling hole images and more accurate drilling hole information, so that the rock mass quality grading standard of a single minimum measurement calculation unit is utilized to grade rock mass joint interval indexes, rock mass joint state indexes, complete rock strength indexes, rock quality indexes, underground water state indexes and ground stress indexes based on the drilling hole information, and the rock mass quality grading indexes of the minimum measurement calculation unit are summed to obtain RMR (rock mass grade) rock mass quality grading grade, and the drilling hole information obtained based on the drilling hole television is utilized to carry out RMR (rock mass grade) rock mass quality grading grade, and the single drilling hole is used as the minimum measurement calculation unit, so that the grading result is more reliable; the rock mass grading indexes and the rock mass grading scores of each drilled hole are established into a rock mass grading database and are processed by a machine learning algorithm to obtain a random forest model and an Adaboost regression algorithm model, the rock mass grading indexes of a single minimum measurement calculation unit are input into the random forest model and the Adaboost regression algorithm model, the average value of the prediction results of the random forest model and the Adaboost regression algorithm model is taken as the final rock mass grading result, and a more accurate grading result is obtained.
The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the specific details of the above embodiments, and various simple modifications can be made to the technical solution of the present invention within the scope of the technical concept of the present invention, and all the simple modifications belong to the protection scope of the present invention.
In addition, the specific features described in the foregoing embodiments may be combined in any suitable manner, and in order to avoid unnecessary repetition, the present invention is not limited to the various possible combinations, so long as the concept of the present invention is not violated.

Claims (7)

1. The rock mass quality grading method based on machine learning is characterized by comprising the following steps:
(1) Acquiring drilling information, wherein the drilling information comprises drilling images, drilling depth, azimuth, inclination angle and serial numbers of the drilling holes and serial numbers of cores corresponding to the drilling holes, and establishing a drilling image database according to the drilling information;
(2) Taking a single drilling hole as a minimum measurement calculation unit, scoring and summing rock mass grading indexes of each minimum measurement calculation unit based on RMR rock mass grading standards to obtain a rock mass grading score of each minimum measurement calculation unit, wherein the rock mass grading indexes comprise a rock mass joint interval index, a rock mass joint state index, a complete rock strength index, a rock mass index, a groundwater state index and a ground stress index;
(3) Establishing a rock mass quality grading database, and recording the rock mass quality grading index and the rock mass quality grading score of each drilling hole into the rock mass quality grading database;
(4) Processing the rock mass quality grading database based on a random forest regression algorithm and an Adaboost regression algorithm respectively, and obtaining two machine learning prediction models through training;
(5) And respectively inputting various rock mass quality grading index data acquired from the rock mass within the drilling depth range into the two machine learning prediction models, wherein the average value of the output results of the two machine learning prediction models is the rock mass quality grading result.
2. The machine learning based rock mass quality grading method according to claim 1, wherein the borehole image, borehole depth, azimuth and inclination of the borehole are obtained by a borehole television.
3. The machine learning based rock mass quality grading method according to claim 1, wherein the number of the drill holes corresponds to the number of the cores corresponding to the drill holes one by one.
4. A machine learning based rock mass quality grading method according to claim 3, wherein the number of joint cracks in a single borehole is statistically recorded based on the borehole image of the single borehole to obtain the total number of joint cracks of the borehole passing through the rock mass within the depth range of the single borehole, and a rock mass joint interval index is calculated:
the joint spacing index is equal to the total number of joint cracks of the borehole passing through the rock mass within the borehole depth/the borehole depth range of the single borehole.
5. A machine learning based rock mass quality grading method according to claim 3, wherein the joint state of a single drill hole passing through the rock mass joint surface is counted according to the drill hole image, and the joint state index of the rock mass in the single drill hole depth range is obtained, wherein the joint state index comprises the roughness, corrosion condition and closure degree of the drill hole passing through the rock mass joint surface.
6. The machine learning based rock mass quality grading method of claim 1, wherein the step of processing the rock mass quality grading database using the random forest regression algorithm comprises:
a1 Selecting an optimal segmentation variable j and a segmentation point S for the rock mass quality classification database based on a segmentation selection formula, wherein the segmentation variable j and the segmentation point S divide the rock mass quality classification database into a data set S1 and a data set S2, and the segmentation selection formula is as follows:
Figure QLYQS_1
wherein :
Figure QLYQS_2
output means for the samples of data set S1, < >>
Figure QLYQS_3
For the sample output average value of the S2 data set, traversing the segmentation variable j, traversing the segmentation point S for the fixed segmentation variable j, and solving a segmentation pair (j, S) enabling the segmentation selection formula to reach the minimum value; />
A2 For a selected said segmentation pair (j, s), dividing the region and determining the corresponding output value:
Figure QLYQS_4
Figure QLYQS_5
wherein x represents allThe rock mass quality grading index;
Figure QLYQS_8
a hierarchical database of quality of the rock mass for selection of the cut; />
Figure QLYQS_12
Representing a region where the value of the corresponding rock mass quality grading index to be segmented in the rock mass quality grading database is smaller than or equal to the segmentation point s, +.>
Figure QLYQS_15
Representing a region in the rock mass quality grading database, corresponding to the rock mass quality grading index to be segmented, having a value greater than the segmentation point s, < >>
Figure QLYQS_9
Representation->
Figure QLYQS_13
Or->
Figure QLYQS_16
The number of samples in the region,
Figure QLYQS_19
representing the divided sub-regions->
Figure QLYQS_6
and />
Figure QLYQS_10
,/>
Figure QLYQS_14
Representation->
Figure QLYQS_18
and />
Figure QLYQS_7
In a corresponding borehole rock mass quality grading result, < >>
Figure QLYQS_11
Representation->
Figure QLYQS_17
and />
Figure QLYQS_20
The average value of the output;
a3 Dividing the rock mass quality grading database into M sub-regions
Figure QLYQS_21
Establishing a final prediction model:
Figure QLYQS_22
wherein I represents an identity matrix.
7. The machine learning based rock mass quality grading method according to claim 1, wherein the step of processing the rock mass quality grading database using the Adaboost regression algorithm comprises:
b1 Initializing the rock mass quality grading index weight and grading all the rock mass quality grading indexes
Figure QLYQS_23
Is initialized to 1/N:
Figure QLYQS_24
Figure QLYQS_25
Figure QLYQS_26
representing the number of loop iterations;
b2 Performing loop iteration on the weighted rock mass quality grading index, wherein the steps of the loop iteration comprise:
b1 Sample distribution of the rock mass quality grading index is as follows
Figure QLYQS_27
On the basis of (1) training a weak classifier using a training set +.>
Figure QLYQS_28
;
b2 Calculating a weak classifier
Figure QLYQS_29
Maximum error on training set +.>
Figure QLYQS_30
Figure QLYQS_31
wherein ,
Figure QLYQS_32
representing weak classifier->
Figure QLYQS_33
Sample set of quality grading indicators for all said rock masses +.>
Figure QLYQS_34
Is predicted by->
Figure QLYQS_35
Representing sample set +.>
Figure QLYQS_36
A corresponding target value;
b3 Calculating the weak classifier
Figure QLYQS_37
Sample of quality grading index for each of said rock masses +.>
Figure QLYQS_38
Relative error of (2):
Figure QLYQS_39
b4 According to the relative error of the sample
Figure QLYQS_40
Calculating the current weak classifier +.>
Figure QLYQS_41
Error rate of (c): />
Figure QLYQS_42
wherein ,
Figure QLYQS_43
representing the weight value corresponding to each variable;
b5 Updating the current weak classifier
Figure QLYQS_44
Weight of (2):
Figure QLYQS_45
Figure QLYQS_46
Figure QLYQS_47
wherein ,
Figure QLYQS_48
representing weak classifier->
Figure QLYQS_49
Weight coefficient of>
Figure QLYQS_50
The weight of the sample point that represents the last update,
Figure QLYQS_51
weight of sample point representing the current update, +.>
Figure QLYQS_52
Representing the normalization factor;
b3 End of (d)
Figure QLYQS_53
And (5) performing round iteration to finally obtain a strong regression:
Figure QLYQS_54
wherein ,
Figure QLYQS_55
all->
Figure QLYQS_56
I.e., the median of the weighted output results of all weak learners,
Figure QLYQS_57
。/>
CN202310159304.1A 2023-02-24 2023-02-24 Rock mass quality grading method based on machine learning Active CN115840921B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310159304.1A CN115840921B (en) 2023-02-24 2023-02-24 Rock mass quality grading method based on machine learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310159304.1A CN115840921B (en) 2023-02-24 2023-02-24 Rock mass quality grading method based on machine learning

Publications (2)

Publication Number Publication Date
CN115840921A CN115840921A (en) 2023-03-24
CN115840921B true CN115840921B (en) 2023-05-16

Family

ID=85580105

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310159304.1A Active CN115840921B (en) 2023-02-24 2023-02-24 Rock mass quality grading method based on machine learning

Country Status (1)

Country Link
CN (1) CN115840921B (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115017791A (en) * 2021-12-18 2022-09-06 中国铁道科学研究院集团有限公司电子计算技术研究所 Tunnel surrounding rock grade identification method and device

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109635461B (en) * 2018-12-18 2022-04-29 中国铁建重工集团股份有限公司 Method and system for automatically identifying surrounding rock grade by using while-drilling parameters
CN110410027A (en) * 2019-07-16 2019-11-05 山东黄金矿业科技有限公司深井开采实验室分公司 It is a kind of to carry out the continuous evaluation method of rock-mass quality and system using drilling core
CN112378753A (en) * 2020-10-27 2021-02-19 西北矿冶研究院 Method for evaluating quality of surface mine slope rock mass
CN114818493A (en) * 2022-04-24 2022-07-29 广西路桥工程集团有限公司 Method for quantitatively evaluating integrity degree of tunnel rock mass
CN115630257B (en) * 2022-12-19 2023-04-21 中南大学 Blasting hopper volume prediction method

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115017791A (en) * 2021-12-18 2022-09-06 中国铁道科学研究院集团有限公司电子计算技术研究所 Tunnel surrounding rock grade identification method and device

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
岩体质量分类的PCA-RF模型及应用;刘强等;黄金科学技术(第01期);全文 *

Also Published As

Publication number Publication date
CN115840921A (en) 2023-03-24

Similar Documents

Publication Publication Date Title
US10546072B2 (en) Obtaining micro- and macro-rock properties with a calibrated rock deformation simulation
CA3019107C (en) Obtaining micro-and macro-rock properties with a calibrated rock deformation simulation
CN109839493B (en) Underground engineering rock quality evaluation method and device, storage medium and electronic equipment
CN110006568B (en) Method and system for acquiring three-dimensional ground stress by using rock core
WO2020125682A1 (en) Method and system for calculating rock strength using logging-while-drilling data
CN113222347B (en) Method for evaluating gray system of safety risk of open-air blasting
CN112948932A (en) Surrounding rock grade prediction method based on TSP forecast data and XGboost algorithm
CN110889440A (en) Rockburst grade prediction method and system based on principal component analysis and BP neural network
CN116522692B (en) Underground engineering surrounding rock structural feature in-situ detection and classification method
CN110847969A (en) Method for determining deformation grading early warning index of underground cavern group under rock mass condition
CN115017833A (en) High ground stress soft rock body ground stress calculation method based on deep neural network
CN113779880A (en) Tunnel surrounding rock two-dimensional quality evaluation method based on advanced drilling data
CN110648082B (en) Quick table lookup method for rock burst grade evaluation of deep-buried hard rock tunnel
CN116229354A (en) Face image surrounding rock grade identification method based on characteristic parameter automatic extraction
CN107506556A (en) A kind of short-cut method for determining fresh intact rock sound wave velocity of longitudinal wave value
CN117292148B (en) Tunnel surrounding rock level assessment method based on directional drilling and test data
CN115840921B (en) Rock mass quality grading method based on machine learning
CN110568495A (en) Rayleigh wave multi-mode dispersion curve inversion method based on generalized objective function
CN109543268A (en) The recognition methods of TBM propulsive force major influence factors based on kriging model
JP2015001100A (en) Method for evaluating base rock
CN112329255A (en) Rock burst prediction method based on tendency degree and uncertain measure
CN116973550B (en) Explosion parameter determining method based on advanced geological drilling
Kluckner et al. Estimation of the in situ block size in jointed rock masses using three-dimensional block simulations and discontinuity measurements
CN108614947B (en) Method for discriminating weathering and unloading value of rock mass
CN112598061A (en) Tunnel surrounding rock clustering and grading method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant