CN113326661A - RS-XGboost-based tunnel advanced drilling quantitative interpretation method and device - Google Patents
RS-XGboost-based tunnel advanced drilling quantitative interpretation method and device Download PDFInfo
- Publication number
- CN113326661A CN113326661A CN202110674747.5A CN202110674747A CN113326661A CN 113326661 A CN113326661 A CN 113326661A CN 202110674747 A CN202110674747 A CN 202110674747A CN 113326661 A CN113326661 A CN 113326661A
- Authority
- CN
- China
- Prior art keywords
- xgboost
- data
- drilling
- quantitative
- model
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
Abstract
The invention relates to the field of tunnel engineering, in particular to a tunnel advanced drilling quantitative interpretation method and device based on RS-XGboost. The tunnel to be excavated is randomly sampled, drilling data of the tunnel to be excavated is acquired and is subjected to preliminary processing, the drilling data is input into a pre-established RS-XGboost model for quantitative interpretation, and a quantitative interpretation result is output. By combining strong nonlinear data analysis performance of the XGboost machine learning model and high-efficiency super-parameter optimization capability of RS random search, the difficulty of model establishment is greatly reduced on the premise of ensuring the accuracy of identification and classification of tunnel poor geologic bodies, and the adverse effect of manual parameter adjustment is avoided; meanwhile, the method provides that the types of the poor geologic bodies are used as machine learning model interpretation results, several types of poor geologic bodies which are common to tunnels and have strong harmfulness are used as quantitative intelligent interpretation results, and the excavation mode and the support measures are adjusted in time according to the interpretation results, so that the field construction of the tunnels is guided.
Description
Technical Field
The invention relates to the field of tunnel engineering, in particular to a tunnel advanced drilling quantitative interpretation method and device based on RS-XGboost.
Background
Since the 21 st century, along with the high-speed development of the transportation industry in China, the construction scale of the road tunnel is increasingly huge. According to data statistics, as far as 2020, the national road tunnels 21316 and 2199.93 kilometers, wherein the extra-long tunnel 1394 and 623.55 kilometers, and the long tunnel 5541 and 963.32 kilometers, become the countries with the largest scale, the largest number and the fastest development speed of the road tunnels in the world at present. In the process that the whole tunnel construction gradually changes to the direction of large burial depth and long tunnel line, the characteristics of far site selection, high stress, strong karst, high water pressure, complex structure and the like are gradually highlighted, and various bad geological bodies cause frequent disasters in the current tunnel construction stage.
Advanced geological forecast is an important technical means for exploring geological conditions of tunnels and further ensuring tunnel construction safety, and is always the research focus of tunnel construction development. The conventional advanced geological prediction method mainly comprises various geophysical prospecting methods such as geological radar, seismic wave and infrared detection and an advanced drilling method, and the various methods have various advantages in the aspects of convenience in operation, occupied tunnel face time, prediction accuracy and the like, and are specifically shown in table 1:
TABLE 1 common advance geological forecast method for tunnel
As can be seen from table 1, in terms of identification of tunnel advance poor geologic bodies, the conventional geophysical prospecting method has certain limitations, and the advance drilling method can most intuitively reflect real geologic information in front of a tunnel face by drilling surrounding rocks in front of the tunnel face (namely, sampling a tunnel to be excavated through a tunnel puncher). However, the interpretation work of the drilling data researched at present mainly depends on technicians to carry out the interpretation work by combining with the actual drilling situation on site, and although the measurement while drilling system carried by the drilling machine can record and provide various drilling parameters in real time, the measurement while drilling system is only used as an interpretation reference and is not fully utilized. The interpretation mode does not fall off the scope of empirical judgment, is rough and is a 'pseudo-quantitative' interpretation.
With the development of informatization technologies such as big data and computer technology, the idea of machine learning gradually permeates into various fields, and a new idea is provided for data analysis. In recent years, researchers have introduced machine learning methods into tunnel advance geological forecast for quantitative interpretation, and certain results have been obtained in geological radars and TSPs respectively.
However, two problems still exist in the existing research: firstly, the surrounding rock grade or stratum attribute is mostly used as a prediction result, because the currently common surrounding rock grade is a relatively large concept, different unfavorable geologic bodies and different stratum lithologies can be the same surrounding rock grade, and the surrounding rock grade is basically determined at the tunnel design stage, so that the situation is often inconsistent with the actual excavation situation on site, although the prediction accuracy is high, the guiding significance for tunnel construction is limited; secondly, the traditional machine learning model has less hyper-parameters, and the manual parameter adjustment can basically meet the requirements, but the accuracy is low; the conventional XGboost model is excellent in theoretical performance and high in accuracy, but the number of super parameters needing to be adjusted is large, and the model performance cannot be fully exerted by manual parameter adjustment.
Disclosure of Invention
The invention aims to overcome the defects in the prior art and provide a tunnel advanced drilling quantitative interpretation method based on RS-XGboost.
In order to achieve the above purpose, the invention provides the following technical scheme:
a tunnel advanced drilling quantitative interpretation method based on RS-XGboost comprises the following steps:
s1: randomly sampling a tunnel to be excavated, acquiring drilling data of the tunnel to be excavated and performing primary processing; the drilling data comprises rate of penetration, thrust, torque, and rotational speed;
s2: inputting the preliminarily processed drilling data into a pre-established RS-XGboost model for quantitative interpretation, and outputting quantitative interpretation results, wherein the quantitative interpretation results comprise more complete to more broken, broken to extremely broken and soft mud filling;
the RS-XGboost model is obtained by training and optimizing the XGboost model through an RS algorithm. According to the method, the strong nonlinear data analysis performance of the XGboost machine learning model and the high-efficiency super-parameter optimizing capability of RS random search are combined to construct the RS-XGboost tunnel advanced drilling bad geologic body quantitative interpretation model, and the super-parameter combination capable of fully exerting the performance of the XGboost model is efficiently searched out, so that the difficulty of model establishment is greatly reduced on the premise of ensuring the accuracy of tunnel bad geologic body identification and classification, and the adverse effect of manual parameter adjustment is avoided; meanwhile, the method provides that the types of the bad geological bodies are used as machine learning model interpretation results, three types of bad geological bodies (more complete to more broken, broken to extremely broken and soft mud filled) which are common and have strong harmfulness to the tunnel are used as quantitative intelligent interpretation results, and the excavation mode and the support measures are adjusted in time according to the interpretation results, so that the field construction of the tunnel is guided. Namely, the type of the poor geologic body related to the invention is a definite concept, is more suitable for field construction, and has stronger guidance effect.
As a preferred embodiment of the present invention, when the quantitative interpretation result is a soft mud filling, further division is made:
the adjacent interpretation units interpret the soft mud filling and output the soft mud filling karst cave as mud;
outputting the soft interlayer when the adjacent interpretation units do not interpret the soft mud filling;
wherein the interpretation unit is a basic unit in quantitative interpretation. According to the invention, the soft mud filling is further divided into the soft interlayer and the argillaceous filling karst cave, so that the site constructor is informed of the existence of the soft interlayer or the karst cave in the rock mass filled with the soft mud in front, the site constructor is more suitable for site construction, and the guiding effect is stronger.
As a preferable scheme of the invention, the establishment of the RS-XGboost model comprises the following steps:
s21: inputting sample data and marking to form marked sample data; the label is a quantitative interpretation result corresponding to the sample data;
the sample data comprises a plurality of geological data, each geological data comprises a plurality of quantitative indexes, and the quantitative indexes comprise depth, drilling speed, pressure stabilization, cutting force, propelling force, torque and/or rotating speed;
s22: preprocessing the labeled sample data;
s23: inputting the preprocessed labeled sample data into an XGboost model, performing model training on the XGboost model through an RS algorithm, and outputting an RS-XGboost model.
As a preferred embodiment of the present invention, the model training in step S23 specifically includes the following steps:
s231: setting a value range of an over-parameter in the XGboost model;
s232: inputting preprocessed labeled sample data into an XGboost model, performing super-parameter optimization on the XGboost model through a random search algorithm in the value range, and acquiring a model performance evaluation index value and a corresponding super-parameter;
s233: when the number of times of optimization is less than the preset value, returning to the step S231; when the number of times of optimization is greater than or equal to the preset value, go to step S234;
s234: and selecting a value with the highest model performance evaluation index value from the model performance evaluation index values, and taking a super parameter corresponding to the value with the highest model performance evaluation index value as a preferred super parameter of the XGboost model. According to the method, the RS algorithm is used for automatically optimizing, and after the number of times of optimizing is preset, the model with the highest model performance evaluation index value is selected as output, so that the difficulty of model training and adjustment is reduced, and the accuracy of the model is greatly guaranteed.
As a preferred aspect of the present invention, the hyper-parameters in step S231 include the number of weak evaluators, the maximum depth, the learning rate, the sample weight, and the ratio of randomly sampled samples.
As a preferred embodiment of the present invention, in the step S21, performing correlation analysis on the quantitative index to obtain a preferred quantitative index; the preferred quantitative indicators include rate of penetration, thrust, torque, and rotational speed. According to the method, the quantitative indexes with higher relevance in the sample data are removed by performing relevance analysis on the data of the quantitative indexes, the calculated amount of model training is greatly reduced on the premise of not influencing the accuracy of the model, and the time cost of the model training is further greatly reduced.
As a preferable aspect of the present invention, the preprocessing in step S22 includes the steps of:
a: performing data noise reduction by deleting the rise data in the labeled sample data, wherein the rise data is acquired when the drilling machine for advanced drilling does not reach a stable state;
b: traversing the missing values of the labeled sample data after denoising, and filling the missing values through the mean values of the index data corresponding to the missing values to obtain denoised and complemented data;
c: the data after noise reduction and gap filling are equidistantly divided into a plurality of sections according to a preset division interval;
d: calculating the secondary indexes of the optimized quantitative indexes in each paragraph after the equidistant segmentation; wherein the secondary indicators comprise the mean and variance of each preferred quantitative indicator;
e: and carrying out data standardization on the secondary indexes by adopting a standard deviation method. According to the invention, the rising section data without interpretation value is removed and the missing data in the stable section is filled in the preprocessing, so that the data quality is improved; through the data segmentation step, on the premise of ensuring the prediction accuracy, the situation that the real result of the tunnel segment is estimated by using the tunnel face interpretation result is avoided, so that the subsequent output result is more visual and reliable; meanwhile, the invention accurately reflects the data characteristics of various bad geologic bodies in each paragraph by calculating the secondary indexes, and provides a good data base for subsequent calculation.
As a preferable embodiment of the present invention, the division pitch preset in the step c is [0.5m,1.5m ]. According to the invention, the segmentation interval is set to be 0.5m or 1.5m, so that the prediction precision of the method is improved as much as possible on the premise of ensuring the prediction accuracy.
As a preferred embodiment of the present invention, the secondary indexes with low correlation are removed, and the obtained preferred secondary indexes are: drilling rate mean, drilling rate variance, propulsion mean, torque variance, and rotational speed mean.
An RS-XGboost-based tunnel advanced drilling quantitative interpretation device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of the above.
Compared with the prior art, the invention has the beneficial effects that:
1. according to the method, the strong nonlinear data analysis performance of the XGboost machine learning model and the high-efficiency super-parameter optimizing capability of RS random search are combined to construct the RS-XGboost tunnel advanced drilling bad geologic body quantitative interpretation model, and the super-parameter combination capable of fully exerting the performance of the XGboost model is efficiently searched out, so that the difficulty of model establishment is greatly reduced on the premise of ensuring the accuracy of tunnel bad geologic body identification and classification, and the adverse effect of manual parameter adjustment is avoided; three types of poor geologic bodies (more complete to more broken, broken to extremely broken and soft mud filled) which are common to tunnels and have stronger harmfulness are used as quantitative intelligent interpretation results, and the excavation mode and the supporting measures are timely adjusted according to the interpretation results, so that the field construction of the tunnels is guided. Namely, the type of the poor geologic body related to the invention is a definite concept, is more suitable for field construction, and has stronger guidance effect.
2. According to the invention, the soft mud filling is further divided into the soft interlayer and the argillaceous filling karst cave, so that the site constructor is informed of the existence of the soft interlayer or the karst cave in the rock mass filled with the soft mud in front, the site constructor is more suitable for site construction, and the guiding effect is stronger.
3. According to the method, the RS algorithm is used for automatically optimizing, and after the number of times of optimizing is preset, the model with the highest model performance evaluation index value is selected as output, so that the difficulty of model training and adjustment is reduced, and the accuracy of the model is greatly guaranteed.
4. According to the method, the quantitative indexes with higher relevance in the sample data are removed by performing relevance analysis on the data of the quantitative indexes, the calculated amount of model training is greatly reduced on the premise of not influencing the accuracy of the model, and the time cost of the model training is further greatly reduced.
5. According to the invention, the rising section data without interpretation value is removed and the missing data in the stable section is filled in the preprocessing, so that the data quality is improved; through the data segmentation step, on the premise of ensuring the prediction accuracy, the situation that the real result of the tunnel segment is estimated by using the tunnel face interpretation result is avoided, so that the subsequent output result is more visual and reliable; meanwhile, the invention accurately reflects the data characteristics of various bad geologic bodies in each paragraph by calculating the secondary indexes, and provides a good data base for subsequent calculation.
6. According to the invention, the partition distance is set to be 0.5m, so that the prediction precision of the method is improved as much as possible on the premise of ensuring the prediction accuracy.
Drawings
FIG. 1 is a schematic flow chart of a tunnel advanced drilling quantitative interpretation method based on RS-XGboost in embodiment 1 of the present invention;
fig. 2 is a flow chart of geological prediction of advanced drilling of a tunnel advanced drilling quantitative interpretation method based on RS-XGBoost according to embodiment 1 of the present invention;
FIG. 3 is a schematic diagram of four types of poor geologic bodies in a tunnel advanced drilling quantitative interpretation method based on RS-XGboost according to embodiment 1 of the present invention;
FIG. 4 is a thermodynamic diagram related to drilling quantitative indicators of the RS-XGboost-based tunnel advanced drilling quantitative interpretation method in embodiment 1 of the present invention;
FIG. 5 is a distribution scatter diagram of the mean drilling speed in the RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to embodiment 1 of the present invention;
FIG. 6 is a distribution scatter diagram of the mean torque value in the RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to embodiment 1 of the present invention;
FIG. 7 is a distribution scatter diagram of drilling speed variance in the RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to embodiment 1 of the present invention;
FIG. 8 is a distribution scatter diagram of the rotation speed variance in the RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to embodiment 1 of the present invention;
FIG. 9 is a distribution scatter diagram of the mean thrust force in the RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to embodiment 1 of the present invention;
FIG. 10 is a distribution scatter diagram of the mean rotation speed in the RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to embodiment 1 of the present invention;
FIG. 11 is a distribution scatter diagram of the propulsion variance in the RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to embodiment 1 of the present invention;
FIG. 12 is a distribution scatter diagram of the torque variance in the RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to embodiment 1 of the present invention;
fig. 13 is a schematic diagram illustrating classification and prediction of an XGBoost model in the RS-XGBoost-based tunnel advanced drilling quantitative interpretation method according to embodiment 1 of the present invention;
FIG. 14 is a schematic diagram of GS and RS reference finding in the RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to embodiment 1 of the present invention;
FIG. 15 is a flow chart of an RS-XGboost poor geologic body prediction model in the RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to embodiment 1 of the present invention;
FIG. 16 is a predicted set predicted tag distribution diagram in the RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to embodiment 1 of the present invention;
FIG. 17 is a graph of drilling data YK73+ 506-YK 73+491 in a RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to embodiment 2 of the present invention;
FIG. 18 is a chart of drilling data from ZK73+ 570-ZK 73+560 in the RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to embodiment 2 of the present invention;
fig. 19 is a schematic structural diagram of a tunnel advanced drilling quantitative interpretation device based on RS-XGBoost in embodiment 3 of the present invention.
Detailed Description
The present invention will be described in further detail with reference to test examples and specific embodiments. It should be understood that the scope of the above-described subject matter is not limited to the following examples, and any techniques implemented based on the disclosure of the present invention are within the scope of the present invention.
Example 1
A tunnel advanced drilling quantitative interpretation method based on RS-XGboost comprises the following steps:
s1: randomly sampling the tunnel to be excavated through an advanced drilling technology, acquiring drilling data of the tunnel to be excavated and performing primary processing; the drilling data comprises four quantitative indexes of drilling speed, propelling force, torque and rotating speed; the preliminary treatment comprises the following steps:
a: performing data noise reduction on input data by deleting rise data, wherein the data collected when the drilling machine for the advanced drilling in the rise data does not reach a stable state is preferably data of 0-0.5 m;
b: traversing missing values in input data, and filling the missing values through the mean values of the index data corresponding to the missing values;
c: equally dividing the input data subjected to noise reduction and filling into a plurality of paragraphs at preset division intervals;
d: calculating the secondary indexes of each quantitative index in each paragraph after the equidistant segmentation; the secondary indexes comprise the mean value and the variance of each quantitative index;
e: and carrying out data standardization on the secondary indexes by adopting a standard deviation method.
S2: inputting the preliminarily processed drilling data into a pre-established RS-XGboost model for quantitative interpretation, and outputting quantitative interpretation results, wherein the quantitative interpretation results comprise more complete to more broken, broken to extremely broken and soft mud filling;
when the quantitative interpretation result is soft mud filling, further division is performed:
the adjacent interpretation units interpret the soft mud filling and output the soft mud filling karst cave as mud; outputting the soft interlayer when the adjacent interpretation units do not interpret the soft mud filling; the interpretation unit is a basic unit in quantitative interpretation.
After four bad geologic bodies are interpreted, the construction method can carry out surrounding rock grade division on each bad geologic body section according to an interpretation report, and on the basis of excavation and supporting measures made in the early stage of each grade of surrounding rock, the following supporting and strengthening measures can be prepared in advance:
(1) when the quantitative interpretation result is more complete-more broken and broken-extremely broken, the preferable construction steps after interpretation are as follows:
firstly, constructing a small guide pipe for advancing and radial grouting: before excavation, small guide pipes are radially driven into the stratum at a certain angle at the periphery of the face and the excavated surrounding rock, and slurry is made to permeate through the small guide pipes by means of the pressure of a grouting pump and is diffused to the end of a stratum gap or crack, so that the crushing degree of the rock mass is improved, and the physical and mechanical properties of the rock mass are improved.
Secondly, initial spraying in time: after the broken rock mass is excavated, the broken rock mass is easy to fall off and collapse, and the initial spraying is required to be carried out in time.
(2) When the quantitative interpretation result is a weak interlayer, the preferable construction steps after interpretation are as follows:
firstly, short footage excavation: the exposed area of the surrounding rock is reduced, the exposed practice of the surrounding rock is shortened, collapse is avoided, and short footage is adopted for excavation.
Reducing the burst strength: the blast hole loading is strictly controlled, smooth blasting is adopted, the distance between peripheral holes is controlled to be 30-40 cm, and the thickness of a smooth layer is controlled to be about 65-70 cm.
(3) When the quantitative interpretation result is that the karst cave is filled with mud, the preferable construction steps after interpretation are as follows:
adopting subsection excavation: when the karst cave appears on one side of the tunnel, the side is excavated firstly, and the other side is excavated after primary support is finished.
Secondly, the excavation cycle length is strictly controlled, more holes are drilled and shallow holes are drilled in each cycle of blast hole drilling, and blasting vibration is controlled.
Thirdly, after the karst cave is revealed, further surveying the information of the scale of the karst cave, the size of the karst cave and the like, and taking the next step of measures according to the field situation.
The RS-XGboost model comprises the following building process:
s21: inputting labeled sample data;
s211: data source
The model building adopts sample data of a certain tunnel, the tunnel is a separated tunnel, the whole tunnel sequentially passes through a north-south valley and a east-west karst peak from north to south, the left line extends to mileage ZK72+ 040-ZK 73+660, and the tunnel length is 1620 m; the right line extends to the mileage YK72+ 060-YK 73+660, the length of the tunnel is 1600m, the designed road surface elevation is 250-280 m, and the tunnel is a long tunnel. The geological conditions of tunnel engineering are complex, the grade of surrounding rock is mainly IV-V grade, the surrounding rock is formed by medium weathering sandstone and strong weathering sandstone, a karst cave develops, and most of the surrounding rock is filled with argillaceous substances.
Advanced drilling work is carried out on site by using a multifunctional crawler-type drilling machine of model C6-2 of Casagrand, and geological forecast is carried out based on drilling data and site conditions, as shown in figure 2. Collecting sample data according to the accumulated advanced drilling geological forecast data in the early stage, collecting 8893 geological data, accumulating about 160 meters of tunnel length, wherein the related data labels comprise relatively complete to relatively broken rock masses, broken to extremely broken rock masses and soft mud filling (relatively complete to relatively broken, broken to extremely broken are qualitative evaluation indexes of rock integrity degree, and the corresponding quantitative index is Kv＝(vpm/vpr)2Wherein v ispmIs the longitudinal wave velocity, v, of the rock massprIs the longitudinal wave velocity of the rock, KvThe range is 0-1, wherein the complete-to-break ratio is 0.35-0.75, and the break-to-maximum ratio is 0-0.35; soft mud filling is a geological concept, which means that soft mud layers are sandwiched between continuous rocks and can be divided into soft and weak interlayers and argillaceous cavern filling according to the filling range), and the soft mud filling range can be divided into the soft and weak interlayers and the argillaceous cavern filling in the forecasting process, namely, quantitative interpretation and research work can be carried out on four types of unfavorable geologic bodies (as shown in figure 3) based on the sample data.
S212: drilling data structures and features
In the process of advanced drilling, the system carries out random sampling along with the change of footage, about 50 sample data are collected per meter, the sample data comprise a plurality of geological data, each geological data comprises seven quantitative indexes, namely Depth (Depth), drilling rate (drilingrate), pressure stability (stability), cutting force (Toolforce), propelling force (Thrust pressure), Torque (Torque) and Rotation speed (Rotation). By carrying out structural analysis on sample data, the method mainly has the following three characteristics:
(1) sampling stage: the sampling process has stronger continuity as a whole, and presents obvious stage division, an ascending section at the beginning of drilling sampling and a stable section in the sampling process, wherein the ascending section is usually concentrated in a footage range of 0-0.5 m, the concrete parameter significance relates to air drilling and initial spraying concrete drilling, and the quantitative interpretation of bad geologic bodies has no reference significance.
(2) Data non-linearity: the quantitative index shows obvious nonlinear correlation.
(3) The dispersion degree is large: the specific sampling parameters of the drilling speed, the torque and the rotating speed indexes show large discreteness, and the specific discreteness degree is closely related to the drilling of different unfavorable geologic bodies.
S213: first order index correlation analysis
Besides the qualitative characteristic analysis of the data, quantitative index correlation analysis is also necessary. In order to improve the data quality of the training set, reduce the data analysis dimension and improve the model accuracy, correlation analysis needs to be performed on quantitative indexes involved in the drilling process so as to find out the correlation among the indexes and remove indexes with higher correlation.
According to the method, by using the pandas and matplotlib libraries in Scikit-leern, the seven primary indexes are subjected to correlation analysis after the collected original sample data is imported, and the result is shown in the following figure 4.
From the content of fig. 4, it can be understood that the Depth (Depth) has a high correlation with the steady pressure (stabilizing pressure), the drilling rate (drilingrate) and the cutting force (Tool force), and the correlation coefficients are 0.89 (positive correlation) and-0.54 (negative correlation), respectively. Meanwhile, considering that the depth value is meaningless to interpret the categories of the bad geologic body, finally, in order to reduce the time cost of model training, the depth, the pressure stabilization and the cutting force are removed by referring to relevant documents, and the four items of drilling speed, propelling force, torque and rotating speed are used as first-level indexes of drilling quantitative interpretation.
S22: preprocessing the labeled sample data;
s221: data denoising: according to the operating condition and quantitative data characteristics of the drilling machine, a complete circulation section for drilling can be divided into an ascending section and a stable section, the drilling data of the ascending section is usually regarded as meaningless because the drilling machine does not reach a stable state, and the depth range of the section is usually 0-0.5 m. Therefore, the method needs to remove the data of the ascending section before the data is divided, and achieves the purpose of data noise reduction by removing the data of the ascending section (0-0.5 m) in the labeled sample data.
S222: missing value filling: when a drilling system collects quantitative index data, a small amount of data of individual indexes is lost sometimes due to operation of a manipulator, aiming at the condition of data loss, a missing value is filled by adopting an input module in Scikit-left, namely a mean value of the index data in a training set is filled by adopting a parameter of 'mean' input in the 'training' so as to ensure the integrity of the marking sample data of the training set.
S223: data equidistant segmentation: in order to avoid errors in the estimation of the real situation in a section of the tunnel by using the tunnel face interpretation result, the training set marking sample data is segmented into a plurality of sections by adopting a data equidistant segmentation mode, and the mining and calculation of the secondary indexes are carried out by taking the segmented sections as units.
Data equidistant segmentation: after data noise reduction and missing value filling are completed, marking equidistant segmentation of sample data is carried out, and the data segmentation interval d is set to be 0.5m, mainly because of the following two points:
when the segmentation distance d is less than 0.5m and is too small, sample data is inevitably higher or lower than actual data and cannot be removed through noise reduction due to the mechanical system and the operation of a manipulator in the operation process of the drilling machine, and the consideration weight of the abnormal data is increased when the segmentation distance is too small, so that the prediction result is inconsistent with the actual data;
and secondly, when the segmentation distance d is larger than 0.5m, the segmentation distance is too large, and in a tunnel with complicated engineering geological conditions, the situation of the surrounding rock is often changed within the range of 1m or more, for example, the situation is changed from crushing to crushing, even extremely crushing, and the too large segmentation distance can cause that a bad geologic body with the proportion lower than 50% in the segmentation distance is ignored, so that the prediction result is inconsistent with the reality.
S224: calculating secondary indexes:
after the data segmentation is completed, in order to deeply mine the data characteristics and rules of various poor geologic bodies corresponding to the drilling data, the data in the segmentation sections of the indexes are subjected to secondary calculation to form a second-level index which is used as an index system of a machine learning model training set. By analyzing the characteristics of the labeled sample data, the invention determines to select the mean value and the variance as secondary indexes:
(1) mean value: the digital drilling data of different unfavorable geologic bodies have a certain value range, the average value is important embodiment of the value range, and the influence of abnormal data in the segmentation interval on the whole real data can be reduced by the means of taking the average value, so that the accuracy of the prediction result is improved.
Mean value
The calculation was performed as follows:
wherein n is the number of samples, xnIs the nth sample.
(2) Variance: the drilling data are influenced by objective factors such as surrounding rocks and machinery, the amplitude of different degrees, namely the discrete degree, appears in the sampling process, the discrete degree is particularly obvious in various unfavorable geologic bodies, if the discrete degree of sample data of complete surrounding rocks is small compared with the mean value, the discrete degree of sample data of broken surrounding rocks is large compared with the mean value, and the discrete degree of sample data of various unfavorable geologic bodies can be reflected scientifically and reasonably by taking the variance, so that the accuracy of the prediction result is improved.
Variance (variance)
The calculation was performed as follows (2).
In the process of drilling a tunnel, 8893 collected original drilling sample data are subjected to equidistant segmentation and secondary index calculation, and a machine learning model training set formed by 324 data is total, wherein 116 data are more complete and more broken, 35.80% are used, 107 data are used for breaking and extremely breaking, 33.03% is used, 101 soft mud is filled, 31.17% is used, and the ratio of three types of bad geologic bodies is basically balanced. Meanwhile, in order to make the machine learning model accurately interpret and distinguish various unfavorable geologic bodies, the data set classification labels need to be set and encoded. The labels are coded as "0", "1" and "2" in order of more complete to more broken, broken to extremely broken and soft mud filling. Specific training set data are shown in table 2.
TABLE 2 advanced drilling data training set
S225: data normalization
The great difference of the data value range in the machine model training set often affects the model prediction effect, and in order to avoid the situation, data standardization is usually required. When the data (x) is centered on the mean (μ) and then scaled by the standard deviation (σ), the data follows a normal distribution (i.e., a standard normal distribution) with a mean of 0 and a variance of 1, which is called data normalization, and is shown in equation (3):
x*＝(x-μ)/σ (3)
in Scikit-lern, preprocessing. Meanwhile, in order to verify the scientificity and rationality of 0.5m as the equidistant segmentation interval and further screen and reduce the dimension of the data set indexes, 100 pieces of data of each of three types of poor geologic bodies in the training set data are selected to be drawn into a scatter diagram for analysis and explanation, and the results are shown in fig. 5-12.
As can be seen from FIGS. 5-12, the mean values of the four indexes of the three types of poor geologic bodies show different degrees of clustering phenomena, wherein the mean values of the drilling speed, the propulsion and the torque are the most obvious. In the aspect of the variance of the four indexes, the drilling speed variance and the torque variance are distinguished to a certain extent on the broken-extremely broken surrounding rock and the relatively complete-broken surrounding rock respectively, which shows that two secondary indexes of the mean value and the variance are calculated after 0.5m equidistant segmentation, so that three types of bad geologic bodies needing to be interpreted are effectively classified. Meanwhile, the superposition degree of the propulsion variance and the rotation speed variance is high, so that the indexes of the two training sets of the propulsion variance and the rotation speed variance are removed in order to optimize the data quality of the training sets, improve the prediction accuracy of the machine model on various bad geological bodies and achieve the purpose of eliminating the propulsion variance and the rotation speed variance.
The following table shows the Accuracy (ACC) of the prediction model obtained by multiple experiments when training with different segmentation intervals and index systems:
TABLE 3 segmentation Pitch optimization
| Division type (m) | Index system (index number) | ACC(％) |
| 0 | First level index (7) | 75.2 |
| 0 | Screening the latter index (4) | 77.6 |
| 0.25 | Second level index (8) | 93.3 |
| 0.5 | Second level index (8) | 96.1 |
| 1 | Second level index (8) | 94.2 |
| 1.5 | Second level index (8) | 93.8 |
| 2 | Second level index (8) | 86.7 |
| 0.25 | Second level index after screening (6) | 93.9 |
| 0.5 | Second level index after screening (6) | 97.8 |
| 1 | Second level index after screening (6) | 96.5 |
| 1.5 | Second level index after screening (6) | 95.7 |
| 2 | Second level index after screening (6) | 88.9 |
ACC (accuracy) — (TP + TN)/(TP + TN + FP + FN),
in the formula: TP is the number of correct positive case predictions, FP is the number of negative case prediction errors, TN is the number of correct negative case predictions, and FN is the number of positive case prediction errors.
As can be seen from Table 3, the accuracy is highest when 0.5m is used as the segmentation pitch and the second-level index (6) after screening is used as the command system, so the scheme is selected for calculation.
The preprocessing step of the invention is a data processing means aiming at the characteristics of advanced drilling data, and aims to effectively distinguish different poor geologic body types. After the advanced drilling sample data is visualized, the data of different unfavorable geologic bodies have obvious discreteness, and if the data are directly imported into a machine learning model by taking an original acquisition point (section) as a unit, the classification accuracy is inevitably reduced. Therefore, the original primary index and the original sample collection unit are not applicable, and a new sample unit and a secondary index created according to the primary index should be formed. To achieve the purpose, the sample data of different bad geologic bodies, which are composed of four primary indexes and subjected to noise reduction, are equally divided by 0.5m, and a section is taken as a sample unit to replace a section (acquisition point) which is taken as a sample unit before. Then, the mean and variance (the reason is described in the previous data) of each index within 0.5m are calculated as the secondary indexes, and all the calculated secondary indexes are collected to form a training set. In order to illustrate the effect of the method, 8 scatter diagrams are drawn, so that the data processing method can effectively distinguish different types of the unfavorable geologic bodies, a data set with extremely high quality is provided for training and learning of the machine learning model, and conditions are created for efficiently distinguishing various unfavorable geologic bodies.
S23: building an RS-XGboost model:
summary and principle of XGboost model:
the XGboost model is essentially an improved algorithm based on a gradient lifting decision tree, and can effectively construct an enhanced treeAnd the model runs in parallel, and has the characteristics of low calculation complexity, high running speed, high accuracy and the like. Wherein the objective function Obj(t)The XGboost model is an important index for measuring the quality of the XGboost model, and the smaller the minimum value of the XGboost model is, the better the model is considered to perform.
The objective function is shown in equation (4):
in the formula: n represents the total amount of data imported into the kth tree, the first term represents the conventional loss function, and the true label y is measurediAnd the predicted value
The second term represents the complexity of the model, and is expressed by using some kind of transformation omega of the tree model, and the change represents a formula for measuring the complexity of the tree model from the structure of the tree, and the development is shown in formula (5).
Ω(fk)＝γT+λ||ω||2/2 (5)
In the formula: gamma and lambda represent the coefficient of the complexity of the model, and T represents the number of leaf nodes of the decision tree of the model.
To solve the objective function, equation (4) can be operated using Taylor expansion, resulting in equation (6):
meanwhile, formula (7) is defined:
substituting the formula (7) into the formula (6) to obtain Obj(t)Formula (8):
based on the above formula, a schematic diagram of classification prediction of the XGBoost model is shown in fig. 13.
Fast optimization of RS hyper-parameter
Adjusting the value of the model hyper-parameters to improve the model performance is an important part of machine learning, and the optimal hyper-parameter combination is difficult to find out along with the increase of the number of the hyper-parameters by artificial parameter adjustment, so that an automatic hyper-parameter optimization tool is needed. Currently, the automatic parameter adjustment commonly used mainly includes two kinds of Grid Search (GS, CV) and Random Search (RS). Both of which are shown in simplified schematic form in fig. 14.
In the figure, the green curve is g (x), the yellow curve is h (y), the objective function is f (x, y) ═ g (x) + h (y), f (x, y) is solvedmax. Wherein f (x, y) ═ g (x) + h (y) ≈ g (x) and x) because g (x) is numerically significantly larger than h (y), i.e., f (x, y) is solvedmaxIn the process (e), g (x) is an important hyperparameter, and h (y) is a non-important hyperparameter.
The left graph is a grid search, the principle is that traversal is carried out in a parameter space until the optimal hyper-parameter combination is found out, the right graph is a random search, and both graphs are searched for 9 times in a defined parameter space. From f (x, y) ═ g (x) + h (y) ≈ g (x), the left image actually explores 3 points, f (x, y)maxA and the right picture actually explores 9 points, f (x, y)maxB. It is clear that the right graph is more likely to find the maximum of the objective function, i.e. introducing a random factor may in some cases improve the optimization efficiency.
In random search and grid search, all parameters influencing the model are searched simultaneously, and the parameters and the grid search explore the identical parameter space, and the result of parameter setting is very similar. However, the method is limited to the operation mechanism of grid search, when the number of the hyper-parameters is more than three and the search precision is high, the operation cost is increased, and at the moment, the random search can be a better choice by integrating the double factors of the search effect and the operation time.
Performance evaluation of the RS-XGboost model:
by integrating an integrated algorithm, a weak evaluator and other operation processes, the number of the hyper-parameters which can be adjusted by the XGboost model exceeds 20, and n _ estimators (the number of the weak evaluators), max _ depth (the maximum depth), learning _ rate (learning rate), min _ child _ weight and subsample (the proportion of randomly sampled samples) are selected according to the importance degree. Firstly, according to the rule of a random search algorithm, combining the characteristics of the XGboost algorithm, a grid search space is defined, namely the value range and the search precision of each hyper-parameter. In the process of searching the super-parameter, the grid search space is continuously adjusted by taking Accuray (accuracy rate) as an evaluation index, and finally the optimal parameter combination is obtained and is brought into the XGboost algorithm model to improve the prediction performance.
The RS-XGBoost model flow is schematically illustrated in fig. 15, the idea of adjusting the optimization of the RS hyper-parameter combination is shown, the adjusted flow is to manually adjust the search grid during optimization by using the RS, the optimization times are circulated until the tenth time (the optimal hyper-parameter combination in each grid space can be found basically for the tenth time, the optimization times are set to be ten at this time), and the circulation is terminated after the tenth time. Since each time the RS is optimized, a model performance evaluation index gard.
When the XGboost model is used for optimizing by using the RS, an ending condition is set for the optimizing process according to the requirement so as to achieve the preset accuracy. The scheme can be applied to the situations with lower requirements on accuracy and needing less model training, and can also be applied to the situations with clear requirements on accuracy.
According to the characteristic of the value of the hyper-parameter of the XGboost model, selecting five hyper-parameters (n _ estimators, max _ depth, learning _ rate, min _ child _ weight and subsample) commonly used in the XGboost model to select a hyper-parameter combination optimal selection scheme for operation, wherein the accuracy of each optimal selection scheme is shown in Table 4.
TABLE 4 preferred combination of superparametric
In order to improve the accuracy, the invention selects the scheme with the highest accuracy, namely selects the scheme with five super-parameters of n _ estimators, max _ depth, learning _ rate, min _ child _ weight and subsample, and the constructed grid search space sequentially comprises (10, 100, 1), (5, 10, 1), (0.1, 1, 0.1) and (0.1, 1, 0.1), wherein the first two values in the brackets are the search range of the super-parameters, and the third value is the search accuracy. In addition, 5-fold cross validation is also set after each search to ensure the authenticity and validity of the selected hyper-parameter combination. The training set is divided into 70% of the learning set and 30% of the prediction set, the accuracy of the final model is 97.8%, and the prediction condition is shown in fig. 16 (the division ratio of the training set is set according to actual needs, the learning set is used for training the model, and the prediction set is used for checking the accuracy of the model).
In order to evaluate the performance of the XGboost model, the performance of the XGboost model is compared with that of other models in a default over-parameter value-taking state. The final results are shown in Table 5.
TABLE 5 XGboost model and other models effect comparison table (model default no parameter adjustment)
Wherein: DT is Decision Tree (Decision Tree); RF is Random Forest (Random Forest); the SVM is a support vector Machine (Supportvector Machine); the ANN is an Artificial Neural Network (Artificial Neural Network).
Accuracy (Accuracy) — (TP + TN)/(TP + TN + FP + FN);
precision TP/(TP + FP);
recall (Recall) ═ TP/(TP + FN);
F1-Score (F value) ═ 2 × Precision × reduce)/(Precision + reduce); the TP, TN, FP, and FN specifications are shown in Table 6:
TABLE 6 TP, TN, FP and FN description
| Prediction as positive samples | Predicting as negative sample | |
| The label is a positive sample | TP (true Positive sample of TruePositive couple) | FN (false negative sample) |
| Label is negative sample | FP (false positive sample) | TN (negative sample of true negative pair) |
Meanwhile, in order to evaluate the performance of the RS-XGboost model, after a grid search space is constructed for random search optimization, the optimization result of the GS in the search space is compared with the performance of the model in the XGboost default over-parameter value state. The final results are shown in table 7.
TABLE 7 Performance evaluation Table of RS-XGboost model
| Serial number | Model (model) | Value of hyper-parameter | Run time(s) | Accuracy (%) |
| 1 | XGBoost | (10，6，1，1，1) | 3 | 92.5 |
| 2 | GS-XGBoost | (70，5，0.4，0.6，0.2) | 1347 | 96.6 |
| 3 | RS-XGBoost | (68，7，0.75，0.3，0.55) | 11 | 97.8 |
Note: the model operating environment is a Win 10 system, Intel (R) i7-6700HQ CPU @2.6GHz and 16 GB RAM, the editor is JupyterLab based on Python 3.8, the hyper-parameter value of the XGboost model is a model default value, and the hyper-parameter values of the GS-XGboost model and the RS-XGboost model are values optimized by a GS algorithm and an RS algorithm respectively.
As can be seen from Table 3, the RS-XGboost model has the highest accuracy rate of 97.8%, the order of the superparameter combination at this time according to n _ estimators, max _ depth, learning _ rate, min _ child _ weight and subsample is (68, 7, 0.75, 0.3, 0.55), the accuracy rate of GS-XGboost is closer to that of the GS-XGboost model and is 96.6%, and the accuracy rate of the XGboost model is the lowest under the condition of default superparameter value and is only 92%. Meanwhile, in the aspect of running time, the running time of the GS-XGboost reaches 1347 seconds, and other two models are greatly redundant. The GS-XGboost model has the best performance by integrating the accuracy and the running time.
Example 2
This embodiment is an example of performing actual prediction by using the model established in embodiment 1. In order to test the actual interpretation effect of the RS-XGboost tunnel poor geologic body prediction model in the advanced drilling geological prediction, two representative prediction examples in a certain tunnel of a project are selected and used for explanation, and the rationality and the practicability of the model are verified by comparing a manual interpretation conclusion and an actual excavation result in the prediction. It should be noted that, in the interpretation process, the label "2" (soft mud filling) is independent between "0" (relatively intact to relatively broken) and "1" (broken to and broken), and can be interpreted as "weak interlayer", and when continuously appeared, can be interpreted as "soft mud filling type karst cave".
Engineering example A
And selecting 15 meters of YK73+ 506-YK 73+491 sections of a certain tunnel as a verification sample. The interpretation result of the section in the advanced drilling geological forecast report is as follows: the suspected soft mud filling type cavern with the diameter of 5-6 m is complete and broken with the diameter of 6-14 m, and the suspected soft mud filling type cavern with the diameter of 14-20 m is shown in a drilling image in a figure 17.
The interpretation results of the RS-XGboost model are shown in Table 8 below.
TABLE 8 YK73+507 YK73+491 advanced drilling RS-XGboost interpretation results
| Depth (m) | Interpreting tags | Interpretation of the |
| 5～5.5 | 2 | Mud filled karst cave |
| 5.5～13 | 0 | Relatively complete to relatively broken |
| 13～13.5 | 1 | Crushing-extreme crushing |
| 13.5～14 | 2 | |
| 14～15.5 | 1 | Crushing-extreme crushing |
| 15.5～20 | 2 | Mud filled karst cave |
Engineering example B
And selecting 10 meters of a tunnel ZK73+ 570-ZK 73+560 section as a verification sample. The interpretation result of the section in the advanced drilling geological forecast report is as follows: filling the karst cave with suspected mud of 0-2 m, completing and crushing the surrounding rock of 2-8 m with a weak interlayer, and filling the karst cave with soft mud of 8-10 m. The drilling image is shown in figure 18.
The interpretation results of the RS-XGboost model are shown in Table 9 below.
TABLE 9 YK73+507 YK73+491 advanced drilling RS-XGboost interpretation results
| Depth (m) | Interpreting tags | Interpretation of the |
| 0～2 | 2 | Mud filled |
| 2～3 | 0 | Relatively complete to relatively broken |
| 3～3.5 | 2 | Soft sandwich |
| 3.5～5 | 0 | Relatively complete to relatively broken |
| 5～5.5 | 1 | Crushing-extreme crushing |
| 5.5～6 | 2 | Soft sandwich |
| 6～7.5 | 1 | Crushing-extreme crushing |
| 7.5～10 | 2 | Mud filled karst cave |
Analysis of model interpretation results:
comparing the reported manual interpretation results of the engineering case A and the engineering case B with the quantitative interpretation results of the RS-XGboost model, it can be known that the quantitative interpretation results and the manual interpretation results keep higher consistency by virtue of the excellent performance of the RS-XGboost model on a training set under the premise of practical application and no excessive noise reduction on sample data, and particularly have higher accuracy in the aspect of prediction of more complete to more broken surrounding rocks and mud filled karst caves, and generally meet the requirements of advanced geological forecast engineering application.
Fourthly, experimental conclusion:
(1) the invention aims at the quantitative interpretation problem of tunnel advanced Drilling data, Drilling sample data is qualitatively and quantitatively analyzed, a Drilling speed (Drilling rate), a Thrust (Thrust), a Torque (Torque) and a Rotation speed (Rotation) are used as a primary index system for classification and prediction of the unfavorable geologic body, the quality of a training set is improved by data preprocessing methods such as equidistant data segmentation and secondary index calculation on the basis, meanwhile, an RS-XGboost tunnel advanced Drilling unfavorable geologic body quantitative interpretation model is constructed by combining the strong nonlinear data analysis performance of an XGboost machine learning model and the high-efficiency multiparameter optimizing capability of RS random search, the performance of the model on the prediction set is excellent, and the accuracy of the prediction rate is up to 97.8%.
(2) In order to evaluate the performance of the RS-XGboost model, the model is compared with the XGboost default model and the GS-XGboost model based on grid search by taking Accuray as an evaluation index. The comparison result shows that the RS-XGboost performance is optimal by integrating the model running time and the model accuracy.
(3) The RS-XGboost model is applied to the advanced drilling geological forecast of the actual tunnel engineering, and the result shows that the RS-XGboost model can provide a finer drilling data interpretation result for technical staff to refer to, the interpretation result basically meets the forecast requirement of the tunnel on poor geological bodies, and the tunnel construction can be guided to a great extent.
Example 3
As shown in fig. 19, an RS-XGBoost based tunnel advanced drilling quantitative interpretation apparatus comprises at least one processor, and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a method of RS-XGBoost based quantitative interpretation of tunnel boring as described in previous embodiments. The input and output interface can comprise a display, a keyboard, a mouse and a USB interface and is used for inputting and outputting data; the power supply is used for supplying electric energy to the electronic equipment.
Those skilled in the art will understand that: all or part of the steps for realizing the method embodiments can be completed by hardware related to program instructions, the program can be stored in a computer readable storage medium, and the program executes the steps comprising the method embodiments when executed; and the aforementioned storage medium includes: various media that can store program codes, such as a removable memory device, a Read Only Memory (ROM), a magnetic disk, or an optical disk.
When the integrated unit of the present invention is implemented in the form of a software functional unit and sold or used as a separate product, it may also be stored in a computer-readable storage medium. Based on such understanding, the technical solutions of the embodiments of the present invention may be essentially implemented or a part contributing to the prior art may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the methods described in the embodiments of the present invention. And the aforementioned storage medium includes: a removable storage device, a ROM, a magnetic or optical disk, or other various media that can store program code.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (10)
1. A tunnel advanced drilling quantitative interpretation method based on RS-XGboost is characterized by comprising the following steps:
s1: randomly sampling a tunnel to be excavated, acquiring drilling data of the tunnel to be excavated and performing primary processing; the drilling data comprises rate of penetration, thrust, torque, and rotational speed;
s2: inputting the preliminarily processed drilling data into a pre-established RS-XGboost model for quantitative interpretation, and outputting quantitative interpretation results, wherein the quantitative interpretation results comprise more complete to more broken, broken to extremely broken and soft mud filling;
the RS-XGboost model is obtained by training and optimizing the XGboost model through an RS algorithm.
2. The RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to claim 1, characterized in that when the quantitative interpretation result is soft mud filling, further division is performed:
the adjacent interpretation units interpret the soft mud filling and output the soft mud filling karst cave as mud;
outputting the soft interlayer when the adjacent interpretation units do not interpret the soft mud filling;
wherein the interpretation unit is a basic unit in quantitative interpretation.
3. The RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to claim 2, wherein the construction of the RS-XGboost model comprises the following steps:
s21: inputting sample data and marking to form marked sample data; the label is a quantitative interpretation result corresponding to the sample data;
the sample data comprises a plurality of geological data, each geological data comprises a plurality of quantitative indexes, and the quantitative indexes comprise depth, drilling speed, pressure stabilization, cutting force, propelling force, torque and/or rotating speed;
s22: preprocessing the labeled sample data;
s23: inputting the preprocessed labeled sample data into an XGboost model, performing model training on the XGboost model through an RS algorithm, and outputting an RS-XGboost model.
4. The RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to claim 3, wherein the model training in the step S23 specifically comprises the following steps:
s231: setting a value range of an over-parameter in the XGboost model;
s232: inputting preprocessed labeled sample data into an XGboost model, performing super-parameter optimization on the XGboost model through a random search algorithm in the value range, and acquiring a model performance evaluation index value and a corresponding super-parameter;
s233: when the number of times of optimization is less than the preset value, returning to the step S231; when the number of times of optimization is greater than or equal to the preset value, go to step S234;
s234: and selecting a value with the highest model performance evaluation index value from the model performance evaluation index values, and taking a super parameter corresponding to the value with the highest model performance evaluation index value as a preferred super parameter of the XGboost model.
5. The RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to claim 4, wherein the hyper-parameters in step S231 comprise weak evaluator number, maximum depth, learning rate, sample weight and random sampling sample ratio.
6. The RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to any one of claims 3 to 5, wherein the quantitative index is subjected to correlation analysis in the step S21 to obtain a preferred quantitative index; the preferred quantitative indicators include rate of penetration, thrust, torque, and rotational speed.
7. The RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to claim 6, wherein the preprocessing in the step S22 comprises the following steps:
a: performing data noise reduction by deleting the rise data in the labeled sample data, wherein the rise data is acquired when the drilling machine for advanced drilling does not reach a stable state;
b: traversing the missing values of the labeled sample data after denoising, and filling the missing values through the mean values of the index data corresponding to the missing values to obtain denoised and complemented data;
c: the data after noise reduction and gap filling are equidistantly divided into a plurality of sections according to a preset division interval;
d: calculating the secondary indexes of the optimized quantitative indexes in each paragraph after the equidistant segmentation; wherein the secondary indicators comprise the mean and variance of each preferred quantitative indicator;
e: and carrying out data standardization on the secondary indexes by adopting a standard deviation method.
8. The RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to claim 7, wherein the segmentation pitch preset in the step c is [0.5m,1.5m ].
9. The RS-XGboost-based tunnel advanced drilling quantitative interpretation method according to claim 7, characterized in that the secondary indexes with low correlation are removed, and the preferred secondary indexes are: drilling rate mean, drilling rate variance, propulsion mean, torque variance, and rotational speed mean.
10. An RS-XGboost-based tunnel advanced drilling quantitative interpretation device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110674747.5A CN113326661B (en) | 2021-06-17 | 2021-06-17 | RS-XGboost-based tunnel advanced drilling quantitative interpretation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110674747.5A CN113326661B (en) | 2021-06-17 | 2021-06-17 | RS-XGboost-based tunnel advanced drilling quantitative interpretation method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113326661A true CN113326661A (en) | 2021-08-31 |
| CN113326661B CN113326661B (en) | 2022-02-15 |
Family
ID=77423743
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110674747.5A Active CN113326661B (en) | 2021-06-17 | 2021-06-17 | RS-XGboost-based tunnel advanced drilling quantitative interpretation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113326661B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113779880A (en) * | 2021-09-06 | 2021-12-10 | 广西路桥工程集团有限公司 | Tunnel surrounding rock two-dimensional quality evaluation method based on advanced drilling data |
| CN114439500A (en) * | 2021-12-16 | 2022-05-06 | 山东大学 | TBM (tunnel boring machine) through unfavorable geology intelligent tunneling system and method based on while-drilling test |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103513293A (en) * | 2013-10-12 | 2014-01-15 | 广西大学 | Tunnel geology comprehensive advanced forecasting expert system and implementation method thereof |
| CN103713335A (en) * | 2014-01-07 | 2014-04-09 | 山东大学 | Comprehensive advance geological detection system carried by tunnel boring machine |
| CN107367772A (en) * | 2017-08-29 | 2017-11-21 | 西南石油大学 | The advanced prediction method of unfavorable geology in front of a kind of prediction tunnel |
| CN109034582A (en) * | 2018-07-16 | 2018-12-18 | 山东大学 | Tunnel Passing inrush through faults based on cloud model and combination weighting are dashed forward mud risk evaluating method |
| CN110244344A (en) * | 2019-06-05 | 2019-09-17 | 中南大学 | A kind of TBM method for forecasting advanced geology based on deep learning |
| CN111538073A (en) * | 2020-06-04 | 2020-08-14 | 中铁西南科学研究院有限公司 | Advanced geological forecast detection simulation device and method for tunnel poor geologic body |
| CN111721603A (en) * | 2020-07-31 | 2020-09-29 | 中国电建集团成都勘测设计研究院有限公司 | Assembled tunnel geological forecast geophysical prospecting test model test device and method |
| CN112287595A (en) * | 2020-09-16 | 2021-01-29 | 同济大学 | Method for predicting shield tunnel wall back grouting thickness based on ground penetrating radar detection and machine learning |
| KR102211421B1 (en) * | 2020-06-17 | 2021-02-02 | 에스케이건설 주식회사 | Method and system for determining tbm control parameters based on prediction geological condition ahead of tunnel face |
| CN112819195A (en) * | 2020-12-25 | 2021-05-18 | 广西大学 | Tunnel advanced drilling geology refined forecasting method and application |
-
2021
- 2021-06-17 CN CN202110674747.5A patent/CN113326661B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103513293A (en) * | 2013-10-12 | 2014-01-15 | 广西大学 | Tunnel geology comprehensive advanced forecasting expert system and implementation method thereof |
| CN103713335A (en) * | 2014-01-07 | 2014-04-09 | 山东大学 | Comprehensive advance geological detection system carried by tunnel boring machine |
| CN107367772A (en) * | 2017-08-29 | 2017-11-21 | 西南石油大学 | The advanced prediction method of unfavorable geology in front of a kind of prediction tunnel |
| CN109034582A (en) * | 2018-07-16 | 2018-12-18 | 山东大学 | Tunnel Passing inrush through faults based on cloud model and combination weighting are dashed forward mud risk evaluating method |
| CN110244344A (en) * | 2019-06-05 | 2019-09-17 | 中南大学 | A kind of TBM method for forecasting advanced geology based on deep learning |
| CN111538073A (en) * | 2020-06-04 | 2020-08-14 | 中铁西南科学研究院有限公司 | Advanced geological forecast detection simulation device and method for tunnel poor geologic body |
| KR102211421B1 (en) * | 2020-06-17 | 2021-02-02 | 에스케이건설 주식회사 | Method and system for determining tbm control parameters based on prediction geological condition ahead of tunnel face |
| CN111721603A (en) * | 2020-07-31 | 2020-09-29 | 中国电建集团成都勘测设计研究院有限公司 | Assembled tunnel geological forecast geophysical prospecting test model test device and method |
| CN112287595A (en) * | 2020-09-16 | 2021-01-29 | 同济大学 | Method for predicting shield tunnel wall back grouting thickness based on ground penetrating radar detection and machine learning |
| CN112819195A (en) * | 2020-12-25 | 2021-05-18 | 广西大学 | Tunnel advanced drilling geology refined forecasting method and application |
Non-Patent Citations (4)
| Title |
|---|
| XIAO WANG等: "Application of Artificial Neural Network in Tunnel Engineering: A Systematic Review", 《IEEE ACCESS》 * |
| 冯春萌等: "水平钻探在山区隧道超前地质预报中的应用研究", 《公路》 * |
| 房昱纬等: "基于超前钻探测试的隧道地层智能识别方法", 《岩土力学》 * |
| 谭立等: "提高复杂地质条件下隧道施工的超前预报精度", 《四川建筑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113779880A (en) * | 2021-09-06 | 2021-12-10 | 广西路桥工程集团有限公司 | Tunnel surrounding rock two-dimensional quality evaluation method based on advanced drilling data |
| CN113779880B (en) * | 2021-09-06 | 2024-04-12 | 广西路桥工程集团有限公司 | Tunnel surrounding rock two-dimensional quality evaluation method based on advanced drilling data |
| CN114439500A (en) * | 2021-12-16 | 2022-05-06 | 山东大学 | TBM (tunnel boring machine) through unfavorable geology intelligent tunneling system and method based on while-drilling test |
| CN114439500B (en) * | 2021-12-16 | 2023-09-05 | 山东大学 | TBM (Tunnel boring machine) tunneling system and method for crossing unfavorable geology based on while-drilling test |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113326661B (en) | 2022-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113326661B (en) | RS-XGboost-based tunnel advanced drilling quantitative interpretation method and device | |
| CN111160571B (en) | Method for driving inversion of geology based on earth pressure balance shield machine parameter data | |
| CN111999765B (en) | Micro-seismic multi-precursor method and device for early warning of tension-cracking falling type karst dangerous rock instability | |
| CN113779880B (en) | Tunnel surrounding rock two-dimensional quality evaluation method based on advanced drilling data | |
| CN106407493A (en) | Multi-dimensional Gaussian cloud model-based rock burst grade evaluation method | |
| Wang et al. | An integrated parameter prediction framework for intelligent TBM excavation in hard rock | |
| CN108760305B (en) | Bearing fault detection method, device and equipment | |
| CN112948932A (en) | Surrounding rock grade prediction method based on TSP forecast data and XGboost algorithm | |
| CN112901183B (en) | Method for determining geological features of shield construction process based on machine learning | |
| CN111022119B (en) | Shield tunnel structure anti-seismic safety evaluation method based on pattern recognition | |
| CN113431635B (en) | Semi-supervised shield tunnel face geological type estimation method and system | |
| CN113640380B (en) | Multistage classification method and system for rail damage detection | |
| Jiang et al. | Development of a core feature identification application based on the Faster R-CNN algorithm | |
| CN115688251A (en) | Earthquake multi-occurrence-zone tunnel risk decision method and system based on deep learning | |
| Qiu et al. | TBM tunnel surrounding rock classification method and real-time identification model based on tunneling performance | |
| Ma et al. | Complex texture contour feature extraction of cracks in timber structures of ancient architecture based on YOLO algorithm | |
| Li et al. | Cross-project utilisation of tunnel boring machine (TBM) construction data: A case study using big data from Yin-Song diversion project in China | |
| CN109543268A (en) | The recognition methods of TBM propulsive force major influence factors based on kriging model | |
| CN109886421B (en) | Swarm intelligence coal-winning machine cutting pattern recognition system based on ensemble learning | |
| CN115759351A (en) | Slurry shield tunneling comprehensive early warning method and system and storage medium | |
| CN114818493A (en) | Method for quantitatively evaluating integrity degree of tunnel rock mass | |
| CN114114390B (en) | Oil and gas exploration method, device, equipment and storage medium | |
| Shuai et al. | Collapse column prediction model based on random forest algorithm | |
| CN114663453B (en) | Reservoir multi-scale fracture network quantitative description method and device | |
| Wu et al. | Automatic Borehole Lithology Stratification Study Based on Computer Vision |
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 |