CN-121997297-A - Machine learning-based drilling rock mass structure quantitative prediction method and system

CN121997297ACN 121997297 ACN121997297 ACN 121997297ACN-121997297-A

Abstract

The invention relates to the technical field of machine learning, and provides a method and a system for predicting the quantization of a drilling rock mass structure based on machine learning, which are implemented by constructing an associated data set taking an actual rock mass structure disclosed by drilling television images and/or exploration caves as a reference, taking an actual rock mass structure quantized value as a reference, and calculating an integral deviation correction coefficient of the rock mass integrity coefficient and the rock quality index based on the associated data set, correcting the deviation of the original data by using the correction coefficient, and realizing automatic mapping from double index input to rock mass structure quantization predicted value output by a machine learning model. The decision coefficient of the prediction model on the test set reaches more than 0.9, the average absolute error is not more than 0.3, the single-group data prediction time is not more than 1.5 seconds, and the prediction precision and efficiency are far higher than those of the traditional empirical method. The invention supports local offline deployment, adapts to a field network-free environment, builds a data closed-loop mechanism to realize continuous iteration of the model, and provides reliable basis for dam foundation safety evaluation.

Inventors

SUN NING
LIANG LEI
ZHAO YIHAN
YU ZHENGXING
PENG SHULIANG
ZHANG YUNHAN
WANG WEI
Tang Guanxiong

Assignees

中国电建集团中南勘测设计研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20260407

Claims (10)

1. The quantitative prediction method for the drilling rock mass structure based on machine learning is characterized by comprising the following steps of: The method comprises the steps of S1, constructing an associated data set taking an actual rock mass structure as a reference, wherein the associated data set comprises a plurality of data records, each data record is associated with a rock mass integrity coefficient, a rock quality index and an actual rock mass structure quantization value of the same drilling depth or a corresponding exploration area, the actual rock mass structure quantization value is determined based on a drilling television image and/or a rock mass structure disclosed by an exploration flat hole, the rock mass structure is divided into a plurality of categories, and each category corresponds to a unique quantization value; S2, respectively calculating the rock mass integrity coefficient and the integral deviation correction coefficient of the rock quality index based on the associated data set; s3, acquiring original rock mass integrity coefficient data and original rock quality index data of the drill hole to be predicted, and respectively correcting the original rock mass integrity coefficient data and the original rock quality index data by utilizing the integral deviation correction coefficient to obtain corrected rock mass integrity coefficient data and corrected rock quality index data; And S4, inputting the corrected rock mass integrity coefficient data and the corrected rock mass quality index data into a pre-trained machine learning model, and outputting a rock mass structure quantized predicted value, wherein the machine learning model is trained based on a training data set, and the training data set comprises corrected data obtained by correcting the rock mass integrity coefficient data and the rock mass quality index data in the associated data set through the integral deviation correction coefficient and an actual rock mass structure quantized value in the associated data set.
2. The machine learning based quantitative prediction method for a drilled rock mass structure according to claim 1, wherein the step S1 comprises the steps of: Collecting a borehole television image, wherein the resolution of the borehole television image is not lower than 720P, capturing a frame of image every 0.5 m along the depth direction of the borehole, and recording the depth of the borehole, the number of cracks, the width of the cracks and the type of filler corresponding to each frame of image; Acquiring rock mass structure data disclosed by an exploration flat hole, carrying out full-section mapping on the exploration flat hole wall, and recording the rock mass structure type, crack development parameters and rock mass weathering degree; And (3) aligning the drilling television image with rock structure data disclosed by the exploration cave according to depth, comparing the rock structure type reflected by the image in the same depth interval with the rock structure type disclosed by the cave wall, determining a quantized value corresponding to the rock structure type as an actual rock structure quantized value of the depth interval if the rock structure type reflected by the drilling television image and the rock structure type reflected by the exploration cave are consistent, and comprehensively judging by combining rock core data and a geological survey report if the rock structure type reflected by the drilling television image and the rock structure type reflected by the exploration cave are inconsistent, and determining the actual rock structure quantized value of the depth interval.
3. The machine learning based drilling rock mass structure quantification prediction method according to claim 2, wherein the step S2 comprises the steps of: S21, grouping rock integrity coefficient data and rock quality index data in the associated data set according to the actual rock structure quantized value in the associated data set and rock structure categories; S22, calculating the average value of all rock integrity coefficient data in each rock structure type as the actual measurement average value of the rock integrity coefficient of the rock structure type, and calculating the average value of all rock quality index data in the rock structure type as the actual measurement average value of the rock quality index of the rock structure type; S23, obtaining a standard rock mass integrity coefficient value and a standard rock mass quality index value corresponding to each rock mass structure type determined according to engineering rock mass grading standards; S24, respectively calculating a rock integrity coefficient deviation value and a rock quality index deviation value of each rock structural category according to each rock structural category; S25, calculating the proportion of the number of samples of each rock mass structure type to the total number of samples of the associated data, and taking the proportion as a weight coefficient of the rock mass structure type; And S26, multiplying the rock mass integrity coefficient deviation values of the rock mass structural categories by the corresponding weight coefficients respectively, and then summing to obtain the overall deviation correction coefficients of the rock mass integrity coefficients, and multiplying the rock mass index deviation values of the rock mass structural categories by the corresponding weight coefficients respectively, and then summing to obtain the overall deviation correction coefficients of the rock mass indexes.
4. The machine learning based drilling rock mass structure quantification prediction method of claim 1, wherein the step S3 comprises the steps of: s31, obtaining original rock mass integrity coefficient data and original rock quality index data of a to-be-predicted drilling hole; S32, carrying out data cleaning on the original rock mass integrity coefficient data and the original rock quality index data, removing abnormal values exceeding a preset reasonable range, and filling the missing values to obtain cleaned rock mass integrity coefficient data and cleaned rock quality index data; S33, carrying out standardization processing on the cleaned rock mass integrity coefficient data and the cleaned rock mass quality index data to convert the data into standardized data with zero mean value and one standard deviation, thereby obtaining standardized rock mass integrity coefficient data and standardized rock mass quality index data; S34, correcting the standardized rock mass integrity coefficient data and the standardized rock mass index data by utilizing the integral deviation correction coefficient, subtracting the integral deviation correction coefficient of the rock mass integrity coefficient from the standardized rock mass integrity coefficient data to obtain corrected rock mass integrity coefficient data, and subtracting the integral deviation correction coefficient of the rock mass index from the standardized rock mass index data to obtain corrected rock mass index data.
5. The machine learning based quantitative prediction method for a drilled rock mass structure according to claim 4, wherein the machine learning model in step S4 is obtained by: s41, constructing a training data set, wherein the training data set comprises corrected rock mass integrity coefficient data, corrected rock quality index data and an actual rock mass structure quantized value; s42, dividing the training data set into a training set and a testing set; S43, determining a search space of super parameters, wherein the super parameters comprise maximum tree depth, learning rate, number of weak learners, sample sampling proportion and characteristic sampling proportion; S44, traversing each super-parameter combination in the search space of the super-parameters by adopting a grid search and cross verification mode, taking the average absolute error as an evaluation index, and selecting the super-parameter combination with the minimum average absolute error as the optimal super-parameter; s45, training a machine learning model on the training set by using the optimal super parameters to obtain a trained machine learning model.
6. The method according to claim 5, wherein in the step S42, when the training data set is divided into a training set and a test set, a hierarchical sampling mode is adopted to make the sample ratio of the actual rock structure quantized value corresponding to each rock structure category in the training set and the test set consistent with the sample ratio of the actual rock structure quantized value corresponding to each rock structure category in the training data set, wherein the number of samples of the training set is seventy percent of the total number of samples of the training data, and the number of samples of the test set is thirty percent of the total number of samples of the training data.
7. The method for predicting the structure of a drilled rock mass based on machine learning according to claim 5, wherein in the step S44, each super-parameter combination is traversed in a search space of the super-parameters by combining grid search with 3-fold cross validation, and a super-parameter combination with the smallest average absolute error is selected as an optimal super-parameter by taking the average absolute error as an evaluation index, wherein the value of the average absolute error is less than or equal to zero point three, and the decision coefficient of the machine learning model on the test set is greater than or equal to zero point seven.
8. The machine learning based drilling rock mass structure quantization prediction method of claim 5, wherein the search range of the maximum tree depth is 2 to 4, the search range of the learning rate is 0.05 to 0.2, the search range of the weak learner number is 100 to 200, the search range of the sample sampling ratio is 0.9 to 1.0, and the search range of the feature sampling ratio is 0.9 to 1.0.
9. The machine learning based drilling rock mass structure quantification prediction method according to claim 1, wherein the step S4 is followed by the further steps of: Selecting at least one verification drilling hole, acquiring corrected rock mass integrity coefficient data and corrected rock mass quality index data of the verification drilling hole, and inputting a trained machine learning model to obtain a rock mass structure quantitative predicted value; and calculating the average absolute error of the two, supplementing the data of the verified drilling hole to the associated data set when the average absolute error exceeds zero, recalculating the integral deviation correction coefficient and retraining the machine learning model.
10. A machine learning based drilling rock mass structure quantification prediction system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the machine learning based drilling rock mass structure quantification prediction method of any of claims 1 to 9 when the computer program is executed.

Description

Machine learning-based drilling rock mass structure quantitative prediction method and system Technical Field The invention relates to the technical field of machine learning, in particular to a method and a system for quantitatively predicting a drilling rock mass structure based on machine learning. Background The accurate division of the rock mass structure of the riverbed dam foundation is a core premise of water conservancy and hydropower engineering design, construction and safety evaluation, and directly determines the dam foundation treatment scheme, engineering cost and operation safety. At present, the exploration means of the riverbed dam foundation part is limited by the terrain and hydrologic conditions, is mainly limited to drilling exploration, and lacks comprehensive and visual rock mass exposure means, so that drilling related indexes become core basis for rock mass structure division. In the prior art, a rock mass structure is divided by commonly used single index, the rock mass integrity coefficient is calculated by commonly used drilling longitudinal wave velocity at present, and then the rock mass structure type is divided by combining the rock mass integrity coefficient, or the rock mass index is obtained by drilling the rock core, and the rock mass structure is divided according to the rock mass index value. However, in practice, it has been found that these two indexes have inherent drawbacks: The wave velocity of the drilling longitudinal wave reflects the integrity characteristics of the rock mass in situ, and is influenced by factors such as rock mass crack filling, underground water, test depth and the like, so that systematic deviation exists in rock mass structure types divided based on rock mass integrity coefficients; If the rock quality index is acquired based on a drilling rock core, disturbance is inevitably generated in the rock core taking process, and the rock quality index is influenced by the rock core taking rate, the crack development direction and the like, so that systematic deviation exists in rock structure types divided based on the rock quality index; More importantly, the two division results often have contradiction, cannot be directly fused and judged, and are required to be manually corrected depending on the experience of engineers, so that the subjectivity is strong, the efficiency is low, the correction precision is difficult to guarantee, and the requirements of large hydraulic engineering on the accuracy and standardization of rock mass structure division cannot be met. In view of the foregoing, there is a need for a machine learning-based quantitative prediction method and system for a drilled rock mass structure that addresses or at least alleviates the above-mentioned drawbacks. Disclosure of Invention The invention mainly aims to provide a quantitative prediction method and a quantitative prediction system for a drilling rock mass structure based on machine learning, so as to solve the technical problem. In order to achieve the above purpose, the invention provides a quantitative prediction method for a drilling rock mass structure based on machine learning, which comprises the following steps: The method comprises the steps of S1, constructing an associated data set taking an actual rock mass structure as a reference, wherein the associated data set comprises a plurality of data records, each data record is associated with a rock mass integrity coefficient, a rock quality index and an actual rock mass structure quantization value of the same drilling depth or a corresponding exploration area, the actual rock mass structure quantization value is determined based on a drilling television image and/or a rock mass structure disclosed by an exploration flat hole, the rock mass structure is divided into a plurality of categories, and each category corresponds to a unique quantization value; S2, respectively calculating the rock mass integrity coefficient and the integral deviation correction coefficient of the rock quality index based on the associated data set; s3, acquiring original rock mass integrity coefficient data and original rock quality index data of the drill hole to be predicted, and respectively correcting the original rock mass integrity coefficient data and the original rock quality index data by utilizing the integral deviation correction coefficient to obtain corrected rock mass integrity coefficient data and corrected rock quality index data; And S4, inputting the corrected rock mass integrity coefficient data and the corrected rock mass quality index data into a pre-trained machine learning model, and outputting a rock mass structure quantized predicted value, wherein the machine learning model is trained based on a training data set, and the training data set comprises corrected data obtained by correcting the rock mass integrity coefficient data and the rock mass quality index data in the associated data set through the integral de