Identification and Classification Method of Different Cut Patterns in Blasting of Long-Span and Small-Net-Spacing Tunnels

Abstract

The present invention relates to a method for identifying and classifying different cut patterns in the blasting of large-span small clear distance tunnels, comprising the following steps: P1, collecting the blasting vibration signals of the middle rock mass under n known cut patterns; P2, jointly denoising the blasting vibration signals of the middle rock mass under any collected cut pattern by using multi-synchronous compression transform MSST and wavelet threshold denoising WTD; P3, obtaining the multi-fractal (α-f(α)) spectrum by using multi-fractal detrended fluctuation analysis MF-DFA for the denoised signal data, extracting multiple eigenvalues, and establishing a multi-dimensional feature vector; P4, repeating steps P2 and P3, establishing a number of multi-dimensional feature vectors as a training set and inputting them into a support vector machine SVM for training to obtain a well-trained SVM classifier; P5, inputting the blasting vibration signals of the middle rock mass to be identified into the well-trained SVM classifier to obtain the category of the cut pattern to which the blasting vibration signals of the middle rock mass belong. The present invention realizes the scientific discrimination and effective classification of cut patterns according to the above method.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel blasting, and more particularly, to a method for identifying and classifying different cut patterns in the blasting of large-span and small-spacing tunnels. Background Art

[0002] During the tunneling process of large-span and small-spacing tunnels by the drill-and-blast method, the influence of blasting-induced vibration on the stability of the middle rock stratum cannot be ignored. Under the condition of a single free face, the strong vibration in tunnel blasting often comes from the initiation of the cut section. Therefore, the optimization of the cut pattern is crucial for controlling blasting secondary disasters. At present, the vibration monitoring method for the middle rock stratum in the blasting of large-span and small-spacing tunnels has defects such as poor test accuracy, cumbersome procedures, and high labor intensity. There is still a lack of quantitative indicators for the vibration characteristics of the middle rock stratum under different cut patterns in the tunnel. The identification and classification of cut patterns through blasting signal characteristics have rarely been reported and applied, which cannot effectively guide production practice and far from meet the urgent requirements of engineering technicians and managers for the identification and classification of cut patterns. Summary of the Invention

[0003] Based on the technical problems of the lack of quantitative indicators and the failure to effectively identify and classify the cut patterns of large-span and small-spacing tunnels proposed in the background art, the present invention proposes a method for fractal characteristics and identification and classification of the vibration signals of the middle rock stratum under different cut patterns in the blasting of large-span and small-spacing tunnels.

[0004] A method for identifying and classifying different cut patterns in the blasting of large-span and small-spacing tunnels includes the following steps:

[0005] P1, collecting the blasting vibration signals of the middle rock stratum under n known cut patterns. For each cut pattern, a number of groups of blasting vibration signals of the middle rock stratum that meet the analysis requirements are collected;

[0006] P2, performing joint denoising of multi-synchronous compression transform MSST and wavelet threshold denoising WTD on the collected blasting vibration signals of the middle rock stratum under any cut pattern condition to obtain the denoised signal data;

[0007] P3, using multi-fractal detrended fluctuation analysis MF-DFA on the denoised signal data to obtain the multi-fractal ( - ) spectrum, extracting multiple characteristic values from the multi-fractal ( - ) spectrum, and establishing a multi-dimensional feature vector;

[0008] P4, repeating steps P2 and P3, for the several blasting vibration signals of the middle rock stratum under n cut patterns collected, establishing the corresponding number of multi-dimensional feature vectors, and taking the established multiple multi-dimensional feature vectors as a training set to input into a support vector machine SVM for calculation and training to obtain a trained SVM classifier;

[0009] P5. Input the blasting vibration signals of the middle rock pillar under any one or more of the cut patterns to be recognized into the trained SVM classifier for the recognition and classification of the cut patterns, and obtain the category of the cut pattern to which each blasting vibration signal of the middle rock pillar belongs.

[0010] As an improvement to the method for recognizing and classifying different cut patterns in the blasting of large-span and small-spacing tunnels of the present invention, the n known cut patterns include parallel hole cut, single-wedge cut, and double-wedge cut.

[0011] As an improvement to the method for recognizing and classifying different cut patterns in the blasting of large-span and small-spacing tunnels of the present invention, in step P3, extract the spectral width - , dimension difference , maximum value , asymmetry index , and spectral coverage area A, five characteristic values , and establish a five-dimensional feature vector . .

[0012] As an improvement to the method for recognizing and classifying different cut patterns in the blasting of large-span and small-spacing tunnels of the present invention, step P3 is specifically as follows:

[0013] For a blasting signal sequence { x k} of length N, k = 1, 2, …, N , the MF-DFA calculation steps are as follows:

[0014] (1) Calculate the average value of the sequence sample { x k}:

[0015]

[0016] (2) Determine the cumulative deviation of the signal sample:

[0017]

[0018] where i = 1, 2, …, N ;

[0019] (3) Divide the cumulative deviation sequence obtained in step (2) into N s small intervals, ; if N cannot be divided evenly by s, then starting from the tail of the deviation sequence , The tail is divided into N s small intervals, and finally 2N s equal-length small intervals are obtained;

[0020] (4) Perform m-order polynomial fitting of the least squares method on s points in each equal-length small interval obtained by the division in step (3):

[0021]

[0022] where i = 1, 2, …, s ;

[0023] (5) Calculate the mean square error. Let the interval be v = 1, 2, …, 2 N s , calculate the mean square error :

[0024]

[0025] When v = N s + 1, N s + 2, …, 2 N s , calculate the mean square error :

[0026]

[0027] (6) Take the average value of the detrended , then the q fluctuation function can be obtained:

[0028]

[0029] where q is an arbitrary non-zero real number, increases with the increase of s in a power-law relationship, that is , then for each s, there is a corresponding function value , and the slope in the ln[Fq(s)] - lns function relationship diagram is the generalized Hurst exponent h(q);

[0030] (7) Calculate the quality index:

[0031] (8) Calculate the generalized dimension:

[0032] (9) Plot the multifractal spectrum - Spectrum, calculate the spectral width , dimension difference , maximum value , asymmetry index , spectral coverage area A as five characteristic parameters of the multifractal spectrum; among them, the spectral width ; dimension difference ; , L is the distance from the left pole of the multifractal spectrum to the perpendicular line of the maximum value, R is the distance from the right pole to the perpendicular line of the maximum value; the spectral coverage area A is obtained by numerical integration, and the upper and lower limits of the integral are respectively , .

[0033] As an improvement to the identification and classification method of different cuttings in the large-span and small clear distance tunnel blasting of the present invention, in step P4, steps P2 and P3 are repeated, and for a number of intermediate rock blasting vibration signals of n types of cuttings collected, multiple five-dimensional feature vectors are established, and the obtained multiple five-dimensional feature vectors are used as the training set and input into the support vector machine SVM for calculation and training, and the signal category attribution label is determined according to the number of votes ( , , ), and a trained SVM classifier is obtained, where is the straight hole cutting category label, is the single wedge cutting category label, is the double wedge blasting category label.

[0034] As an improvement to the identification and classification method of different cuttings in the large-span and small clear distance tunnel blasting of the present invention, n = 3, and the n known cuttings are straight hole cutting, single wedge cutting and double wedge cutting; the support vector machine SVM is a three-class support vector machine. In step P4, the training set is input into the support vector machine SVM for calculation and training to obtain a trained SVM classifier, which specifically includes: for the training set category attribution label ( , , ), a decision function is constructed by using a three-class support vector machine to separate these three types of sample points. For the input F of a five-dimensional feature vector, a voting method is used to judge which class it belongs to. When , F belongs to class; when , F belongs to class; when , F belongs to class; finally, the number of votes is counted, and F belongs to the class with the most votes; all the data in the training set are classified and trained repeatedly.

[0035] As an improvement to the identification and classification method of different cutting modes in large-span and small-clearance tunnel blasting of the present invention, the classifier is repeatedly trained using multiple sets of rock-intercalated blasting vibration data until the training classification error is less than 10 -3 until.

[0036] As an improvement to the method for identifying and classifying different cutting modes in large-span and small-clearance tunnel blasting of the present invention, the specific process of step P2 is as follows:

[0037] (1) First, center the rock blasting vibration signal Perform short-time Fourier transform to obtain the spectral coefficients ;

[0038] (2) Spectral coefficients Perform synchronous compression transformation to obtain synchronous compression transformation coefficients ;

[0039] (3) Using wavelet threshold to analyze the signal Denoise and compress transform coefficients simultaneously Perform filtering;

[0040] (4) Set the number of iterations of MSST to 2, and perform synchronous compression transformation on the coefficients Perform another synchronous compression transformation to obtain the secondary synchronous compression transformation coefficient ,use Perform inverse transform to obtain effective signal;

[0041] (5) The denoising effect is evaluated using four evaluation indicators: signal-to-noise ratio (SNR), mean square error (MSE), kurtosis signal-to-noise ratio (PSNR), and smoothness (SM). If the indicators are not met, the number of iterations is adjusted until the analysis accuracy requirements are met.

[0042] As an improvement to the method for identifying and classifying different cutting modes in large-span and small-clearance tunnel blasting of the present invention, in step P1, collecting rock-interposed blasting vibration signals under n known cutting modes specifically includes:

[0043] S1: Select the stable rock section in the middle of a large-span, small-spacing tunnel and use an integrated blasting vibration monitoring method to arrange measurement points on the rock wall.

[0044] S2: Determine the blasthole layout and related charging parameters for the upper step of the tunnel, drill different types of slot holes, and charge and detonate them. The total charge of the various slot holes must be consistent.

[0045] S3, monitoring the vibration signals of the rock blasting under different cutting modes, storing and outputting the waveforms to the computer terminal.

[0046] As an improvement to the identification and classification method of different cut patterns in the blasting of long-span and small clear distance tunnels of the present invention, in step S1, the integrated blasting vibration monitoring method relies on an integrated monitoring system to execute. The integrated monitoring system includes: a customized large-capacity lithium battery power supply module for power supply, and a step-down board is set to reduce the battery voltage to the working voltage of the vibration measuring instrument; a protective hose is set on the surface of the wire of the vibration measuring instrument; a customized special instrument protective box is provided, and the protective box has locking and warning functions; the inner backboard of the protective box is fixed on the surface of the middle rock pillar by expansion bolts, the sensor probe is fixed on the angle iron with a prefabricated hole on the surface of the middle rock pillar and a lock washer is set to keep the connection firm, and the gap with the rock wall is filled and bonded with lime powder stirred into a paste; before testing, the main body of the vibration measuring instrument, the step-down board, the power supply module and the wire protection hose are integrated in the protection box, and the protection box is fixed on the explosion-facing side of the sensor, and the relevant wires are protected by the hose, forming the integrated monitoring system.

[0047] The present invention extracts the fractal features of the vibration signals of the middle rock pillar under different cut patterns in the long-span and small clear distance tunnels, and realizes the scientific discrimination and effective classification of the cut patterns in engineering practice according to the above features, and realizes the adaptive discrimination, identification and classification of different cut patterns in the long-span and small clear distance tunnels. Further, the present invention also realizes the prediction and classification of three cut patterns, namely parallel hole cut, wedge cut and double wedge cut, according to the fractal features of the collected blasting signals, and provides a quantitative index for the selection of the tunnel cut pattern and the evaluation and classification of the blasting effect. It provides a method for discriminating and classifying tunnel cut patterns that adapts to future digitalization, intelligentization and refinement, solves the problems of blindness and poor accuracy of manual supervision and discrimination, overcomes the defects of traditional manual recording and a large amount of data sorting and discrimination, and provides a convenient, efficient, intelligent prediction, discrimination and classification method for the safety and refined design and construction of tunnel structures and the optimization and selection of cut patterns. Description of the Drawings

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained according to these drawings.

[0049] Figure 1 is the flow chart of the identification and classification method of different cut patterns in the blasting of long-span and small clear distance tunnels of the present invention.

[0050] Figure 2 is the typical blast hole layout diagram of the bench method for the face of the subsequent tunnel.

[0051] Figure 3It is the layout drawing of the straight-hole cut holes in the upper bench of a long-span and small clear-span tunnel.

[0052] Figure 4 It is the layout drawing of the single-wedge cut holes in the upper bench of a long-span and small clear-span tunnel.

[0053] Figure 5 It is the layout drawing of the double-wedge cut holes in the upper bench of a long-span and small clear-span tunnel.

[0054] Figure 6 It is the time history curve diagram of the blasting signal for the straight-hole cut method in a long-span and small clear-span tunnel.

[0055] Figure 7 It is the time history curve diagram of the blasting signal for the single-wedge cut method in a long-span and small clear-span tunnel.

[0056] Figure 8 It is the time history curve diagram of the blasting signal for the double-wedge cut method in a long-span and small clear-span tunnel.

[0057] Figure 9 It is the multifractal correlation spectrum curve of the blasting signal under three different cut methods ( - ) diagram.

[0058] Figure 10 It is the schematic diagram of the multifractal - spectral statistical characteristics (spectral width, dimension difference, maximum value, asymmetry index) of the tunnel blasting vibration signal.

[0059] Figure 11 It is the schematic diagram of the multifractal - spectral statistical characteristic (spectral coverage area) of the tunnel blasting vibration signal. Specific implementation manner

[0060] Please refer to Figure 1 , a fractal feature and identification classification method for the vibration response of the middle rock stratum under different cut methods in the blasting of a long-span and small clear-span tunnel, including the following specific steps S1 to S8:

[0061] S1: Select a section with stable lithology of the middle rock stratum in a long-span and small clear-span tunnel, and arrange measuring points at the rock wall to be concerned by using the integrated blasting vibration monitoring method.

[0062] For the blasting of large-span and small-spacing tunnels, the construction environment is complex and the safety risks are extremely high. Therefore, extremely high requirements are imposed on the vibration control of the middle rock stratum induced by tunnel blasting construction. Practice has shown that the strong vibrations generated by tunnel blasting often come from the cut section. Therefore, the selection of the cut method is crucial for the stability of the middle rock stratum. How to adopt a reasonable monitoring method to objectively monitor the vibration response of the middle rock stratum and scientifically distinguish and classify different cut methods through signal characteristics, so as to provide a quantitative basis for the determination of different cut methods, is a key problem faced by engineering and technical personnel.

[0063] The influence of the vibrations generated by the blasting of large-span and small-spacing tunnels on the stability of the middle rock stratum cannot be ignored. Blasting vibration is an important index for the adjustment of blasting parameters and the evaluation of blasting effects. Therefore, carrying out blasting vibration monitoring has positive practical significance. Aiming at the defects of the traditional "monitoring while moving" blasting vibration monitoring method, such as frequent layout of measuring points and cumbersome procedures, step S1 of this embodiment proposes an integrated vibration monitoring method for the middle rock stratum in tunnel blasting. Considering the limited capacity of the built-in battery of the vibration meter, a large-capacity lithium battery power supply module is customized for power supply, and a step-down board is set to reduce the battery voltage to the working voltage of the vibration meter. A protective hose is arranged on the surface of the vibration meter wire to prevent the damage of the wire by blasting flying rocks. A special instrument protective box is customized, which has locking and warning functions. The inner backboard of the protective box is fixed on the surface of the middle rock stratum by expansion bolts. The sensor probe is fixed on the angle iron with a prefabricated hole on the surface of the middle rock stratum and a lock washer is set to keep the connection firm. The gap with the rock wall is filled and bonded with lime powder stirred into a paste. Before the test, the vibration meter host, step-down board, power supply module and wire protection hose are integrated in the protective box, and the protective box is fixed on the blasting side of the sensor. The relevant wires are protected by the hose, which plays a certain role in protecting the sensor probe to a certain extent, forming an integrated monitoring system. The integrated tunnel monitoring scheme overcomes the problem that it is difficult to guarantee the test accuracy caused by the frequent layout of sensors in the traditional test method, reduces the interference of the test to normal construction, and avoids the defects of the frequent layout and recovery of wires and instruments in the previous "monitoring while moving" mobile test method. The long-term uninterrupted monitoring of the middle rock stratum can be realized by simply replacing the lithium battery pack, ensuring the accuracy and integrity of the test data, and providing a guarantee for the scientific nature of the vibration response monitoring and identification classification of the middle rock stratum under the above different cut blasting forms.

[0064] S2: Determine the blast hole layout and related charging parameters of the upper bench of the tunnel and drill and charge different forms of cut holes for initiation, and keep the total charge of different forms of cut holes consistent.

[0065] Basic principles for the arrangement of cut holes: First, under the condition that other blasting parameters remain unchanged, only the cut method is changed; second, the number of drilled holes and the total amount of charge for each cut method are kept as consistent as possible, only the distribution of cut holes is different; third, the spread area of each cut method on the heading face is approximately the same; finally, the batches and quantities of blasting consumables such as detonators used for different cut methods are kept consistent. On this basis, there is a basis for identifying and classifying the waveform characteristics of each cut method, avoiding the influence of other blasting parameter factors on the cut method and the cut effect. Rock blasting itself has great discreteness, which is jointly determined by various factors, such as the physical and mechanical properties of rocks, explosive characteristics, wave impedance, occurrence environment, construction level, blasting parameter design and other factors. Therefore, in this embodiment, for the three cut methods of parallel cut, wedge cut and double wedge cut to be tested, blasting tests and vibration monitoring are carried out in the same geological section. Each cut is repeated for multiple tests, and the footage is 2m for each. It should be noted that the three cut methods are subject to prerequisite factors such as drilling depth, drilling angle, number of holes and charge amount, and they are not the most optimized methods in each cut method. The present invention focuses on effectively identifying and classifying the fractal characteristics of blasting vibration waveforms under different cut methods.

[0066] The specific arrangements of the three types of cut holes are as follows:

[0067] (1) Parallel cut

[0068] The parallel cut is designed with 6 holes. The 6 cut holes are arranged in 2 columns at equal intervals in the cut area. The distance between the two cut columns is 2m. The detonator segments used for the 6 cut holes are the same, all being the first series of millisecond-delay plastic detonator tubes, MS1 segment. The drilling angle of the blast holes is perpendicular to the heading face at 90°, and the blast holes are drilled to meet the requirements of being flat, accurate and straight.

[0069] (2) Single wedge cut

[0070] The single wedge cut is arranged with 6 holes. The 6 wedge cut holes are arranged in a wedge shape in two left and right columns, with 3 cut holes in each column. The 6 wedge cut holes are inclined holes, with an angle of 80° with the plane of the heading face. The distance between the orifices of the two columns of cut holes is 1.98m. The vertical distance of the cut holes is 200mm deeper than that of other blast holes. The bottoms of the cut holes must fall on the same plane behind the heading face, which is beneficial to removing the protruding rock ledges at the bottom of the wedge cut cavity. The 6 wedge cut holes are all charge holes, and the detonator segments used are the same, all being the first series of millisecond-delay plastic detonator tubes, MS1 segment. The distance between the bottoms of the two columns of cut holes falling on the bottom of the blast holes is 442mm, which is more beneficial to the formation of the blasting cavity, preventing the bottoms of the blast holes from intersecting due to too small an angle and avoiding the blast holes from being drilled through.

[0071] (3) Double wedge cut

[0072] The number of double-wedge cut holes is 10. All 10 wedge cut holes are inclined holes, arranged in a plum blossom shape in four columns. Among them, the two central columns are the first wedge cut holes, with 3 blast holes in each column, a total of 6 blast holes; the other two columns are the second wedge cut holes, with 2 blast holes in each column, a total of 4 blast holes. The drilling angles all face the tunnel center. The detonator segments used for the first wedge cut holes are the same, all being No. 1 detonators, ensuring that they can be detonated simultaneously. The second wedge cut holes skip to No. 3 detonators, detonating slightly later than the first wedge cut holes, forming a layered and graded blasting. The cavity volume spaces formed by the sequential detonation of the compound wedge cut holes are superimposed in sequence to achieve a good cut effect. The bottoms of the first wedge cut holes must be kept on the same plane behind the face. The projected length of the hole depth perpendicular to the face is 1.2 m, and the distance between the bottoms of the blast holes is 424 mm; the bottoms of the second wedge cut holes must be kept on the same plane behind the face. The projected length of the hole depth perpendicular to the face is 2.2 m, which is 10% deeper than other holes, and the distance between the bottoms of the blast holes is 648 mm.

[0073] The blasting part parameters of the above three cut methods are summarized in Table 1 below.

[0074] Table 1 Blasting parameter table of three cut methods

[0075]

[0076] As Figure 2 shown, it is a typical blast hole layout diagram of the bench method for the face of the subsequent tunnel. When the above different cut methods are detonated, the blasting sequence and detonator segments follow the order of detonating from the inside to the outside, making the detonation of the blast holes have a certain level and time interval. The charge amount of the auxiliary holes is arranged according to the principle that the charge amount in the inner circle is large and the charge amount in the outer circle gradually decreases. The detonation of the blast holes in the same circle must be simultaneous detonation to ensure good blasting effect. Except for special geological conditions (except in the case of gas and combustible gases), all cut holes are detonated by reverse charging method to achieve the most ideal blasting effect.

[0077] S3: Effectively monitor the blasting vibration signals of the middle rock pillar under different cut methods, store and output the waveforms to the computer terminal.

[0078] During the bench method blasting construction of large-span and small clear distance tunnels, the strong vibration generated by blasting usually comes from the upper bench with a single free face. Therefore, the vibration responses generated by the upper bench blasting are selected for signal acquisition under the three cut methods. As Figures 3 to 5 shown, they are respectively the straight hole cut hole layout, single wedge cut hole layout and double wedge cut hole layout of the upper bench of the large-span and small clear distance tunnel. The cut parts of different forms are all arranged in the middle and lower parts of the upper bench of the tunnel. The number and depth of the cut holes are basically the same, especially the charge amount of the cut holes is the same, and the specific charge of each blasting design is basically the same.

[0079] During the process of selecting vibration test data under different cut methods, it is necessary to select measuring points at the same distance from the tunnel face to ensure the accuracy of model training. In order to achieve the optimal blasting construction plan, the total charge of a single cycle at the tunnel face remains unchanged under different cut methods, so as to obtain the most scientific blasting hole pattern parameters and improve the data prediction and classification effect.

[0080] As Figures 6 to 8 shown, they are respectively the time history curve of blasting signals for the parallel cut method, the single-wedge cut method, and the double-wedge cut method in a large-span and small-spacing tunnel. Blasting vibration signals have the characteristics of short duration (generally at the millisecond level), rapid mutation (rapid signal transition), and poor smoothness (violent conversion between wave peaks and wave valleys). Many signal characteristics of blasting vibration are contained in their multi-fractal spectra. It is precisely due to the differences in the above characteristics that different response characteristics are generated in engineering structures. This is also the main basis for identifying and classifying cuts through the differences in multi-fractal spectra.

[0081] S4: Perform joint denoising of multi-synchronous compression + wavelet threshold (MSST+WTD) on the effectively collected signals under different cut methods to obtain the denoised signal data.

[0082] In the field of signal analysis, the essence of multi-synchrosqueezing transform (MSST) is to iterate multiple synchrosqueezing transform (SST) operations. Multiple compressions are carried out in the frequency direction, and theoretically, perfect signal reconstruction can be achieved. Wavelet threshold de-noising (WTD) decomposes the signal into wavelets at each scale and retains the largest scale. That is, the decomposition values at high resolutions are set with thresholds for processing, and then the effective signal coefficients are extracted. In actual analysis, the wavelet threshold is estimated by calculating the median absolute deviation (MAD) of the noise variance , and use to estimate the optimal threshold : . Filter the spectrum coefficients of the synchronous compression transform of tunnel blasting signals using the wavelet threshold de-noising method, and then perform inverse transformation to achieve the reconstruction of blasting signals. The specific process of using the combined algorithm of multi-synchronous compression transform and wavelet threshold to denoise the tunnel blasting signals collected on-site is as follows:

[0083] 1. First, perform short-time Fourier transform on the blasting vibration signal of the middle rock pillar to obtain the spectrum coefficients ;

[0084] 2. For the spectrum coefficients Perform synchronous compression transformation to obtain synchronous compression transformation coefficients ;

[0085] 3. Use wavelet threshold to denoise the signal and filter the synchronous compression transformation coefficients ;

[0086] 4. When MSST (number of iterations = 2), perform another synchronous compression transformation on the SST synchronous compression transformation coefficients to obtain secondary synchronous compression transformation coefficients , and use to perform inverse transformation to obtain the effective signal

[0087] 5. Use four evaluation indicators, namely signal-to-noise ratio (SNR), mean square error (MSE), peak signal-to-noise ratio (PSNR), and smoothness (SM), to evaluate the denoising effect. If not satisfied, adjust the value of the number of iterations until the analysis accuracy requirement is met

[0088] S5: Use multifractal detrended fluctuation analysis (MF-DFA) on the denoised signal to obtain the multifractal - spectrum, and extract five eigenvalues, namely spectral width , dimension difference , maximum value , asymmetry index , and spectral coverage area A , to establish a five-dimensional feature vector

[0089] MF-DFA is an analysis method that can be used to analyze the multi-scale characteristics of signals. Through this method, the multifractal spectrum of the signal sample, that is, the functional relationship between the generalized Hurst exponent h q and the generalized dimension D q , can be obtained

[0090] For the burst signal sequence of length N{ x k}, k = 1, 2, …, N , the MF-DFA calculation steps are as follows

[0091] 1. Calculate the average value of the sequence sample{ x k}:

[0092]

[0093] 2. Determine the cumulative deviation of the signal sample

[0094]

[0095] Among them, i = 1, 2, …, N ;

[0096] 3. Divide the cumulative deviation sequence obtained in step 2 into N s small intervals, ; If N cannot be divided evenly by s, then starting from the tail of the deviation sequence , divide 's tail into N s small intervals, and finally obtain 2N s equal-length small intervals;

[0097] 4. Perform m-order polynomial fitting of the least squares method on the s points in each equal-length small interval obtained by the division in step 3:

[0098]

[0099] Among them i = 1, 2, …, s ;

[0100] 5. Calculate the mean square error. Let the interval be v = 1, 2, …, 2 N s , calculate the mean square error :

[0101]

[0102] When v = N s + 1, N s + 2, …, 2 N s , calculate the mean square error :

[0103]

[0104] 6. Take the average value of the detrended , then the q fluctuation function can be obtained:

[0105]

[0106] Among them, q is an arbitrary non-zero real number, ass increases in a power-law relationship, that is . Then for each s , there is a corresponding function value , for ln Fq ( s )] — ln s the slope in the function relationship diagram is the generalized Hurst exponent h( q ).

[0107] 7. Calculate the quality index:

[0108] 8. Calculate the generalized dimension:

[0109] 9. As Figures 9 to 10 , plot the multifractal spectrum - spectrum, calculate the multifractal spectrum width , dimension difference , maximum value , asymmetry index , multifractal spectrum coverage area A as five characteristic parameters of the multifractal spectrum, as Figure 11 shown. Among them, the fractal spectrum width ; dimension difference ; , L is the distance from the left pole of the multifractal spectrum to the perpendicular line of the maximum value, R is the distance from the right pole to the perpendicular line of the maximum value; the multifractal spectrum coverage area A is obtained by numerical integration, and the upper and lower limits of the integral are , . In one embodiment, as shown in Table 2, they are the multifractal spectrum related parameters of the middle rock burst vibration signals under three cut forms.

[0110] Table 2 Multifractal spectrum related parameters of the middle rock vibration signals under three cut forms

[0111]

[0112] S6: Respectively collect 270 groups of middle rock burst vibration signals in total of 90 groups under each of the three cut methods of straight hole cut, single wedge cut and double wedge cut, to form a training category label set ( , , ) and the final test category label set C = ( c 1, c 2, c3), corresponding to the straight-hole cut blasting, single-wedge cut blasting, and double-wedge blasting target matrices respectively, adopt the above analysis process and obtain the five-dimensional feature vectors under different blasting schemes , and each signal fractal feature attribute is composed of a five-dimensional feature vector, corresponding to the spectral width , dimensional difference , maximum value , asymmetry index , and the fractal spectrum coverage area A， to obtain the multi-dimensional information characteristics of the signal.

[0113] S7: Divide the test data into a training set and a test set according to a ratio of 2:1. Randomly select 60 groups of signals under each of the three cut blasting methods, a total of 180 groups of signals, and input them into the support vector machines (SVM) classification model. Use the five fractal feature values of each group of signals to calculate, train, and determine the signal training class membership label according to the number of votes ( , , ).

[0114] The basic principle of SVM is:

[0115] Support vector machine is a supervised learning algorithm. As a multi-class support vector machine, the three-class support vector machine is also a classification algorithm developed on the basis of the binary SVM. For the training set , is the classification data, is the data class label. To achieve data classification, define the feature space as E , and the data in this feature space is generated by the kernel function through the inner product. φ is a non-linear mapping function, which realizes mapping the input space data to a high-dimensional feature space. By constructing a hyperplane E in the feature space , this hyperplane divides the feature data into three categories, and searches for the hyperplane decision function , so that the output of the hyperplane decision function satisfies the formula:

[0116]

[0117] In the formula: F is the five-dimensional feature vector of the test signal, ; l is the total number of sample data; l 1 is the total number of sample data with the output value of the decision function being +1; l 2 is the total number of sample data with the output value of the decision function being -1; l-l 1 -​l 2 is the total number of the third - type sample data for which the output value of the decision function is 0.

[0118] To enable the hyperplane to have a correct and good data classification function, it is necessary to ensure that the distance from the optimal classification hyperplane to the nearest sample point is maximized. Such a problem is ultimately transformed into an optimization problem in the form of quadratic programming under specific constraints, that is:

[0119]

[0120] The constraints are:

[0121]

[0122] In the formula: slack variable , , ; l 12 = l 1 + l 2; , b are the hyperplane coefficients; C > 0, D > 0, are the weighted parameters for balancing classification errors. To avoid overlap between classes when the slack variable is zero, usually take .

[0123] Using the above - mentioned analysis process and obtaining the five - dimensional multi - fractal feature vectors of signals under different cut - hole patterns , each signal feature attribute is composed of a five - dimensional feature vector, corresponding to the fractal spectrum width , dimension difference , maximum value , asymmetry index and spectral coverage area value A respectively, to obtain the multi - dimensional digital information characteristics of the signal. During the data training process, for any three - class training attribution sets ( , , ), 180 groups of signals, 60 groups for each of the three cut - hole patterns, are selected to construct a decision function to separate these three - class training sample points. For a new input of a signal multi - dimensional feature vector F, the voting method is also used to determine which class F belongs to. When , F belongs to class; when , F belongs to class; when , F belongs to category. Finally, count the votes, and F belongs to the category with the most votes. Set the number of training times to 3000 times, and use multiple groups of vibration data to repeatedly train the classifier until the training classification error is less than 10 -3 until;

[0124] S8: Input the feature vectors of the remaining 30 groups of signals (a total of 90 groups) under different blasting cut methods into the trained SVM classifier, and determine the voting status of the given different categories ( c 1, c 2, c 3) as the final result of the blasting cut scheme recognition classification, and realize the discriminant classification of three different cut methods: parallel hole cut, single-wedge cut, and double-wedge cut.

[0125] The most intuitive and authoritative evaluation criterion for the performance of the classification recognition result is the classification accuracy rate, that is, the recognition rate. Its mathematical expression is:

[0126]

[0127] In the formula, is the classification accuracy rate, is the number of correctly classified samples, is the total number of test samples. Usually, if the classification accuracy rate is higher than 90%, the classification result can be considered valid.

[0128] The above repeatedly trains the support vector machine SVM through three cut methods, and the accuracy rate results of repeatedly identifying and testing the cut methods using the trained SVM are shown in Table 3.

[0129] Table 3 Comparison of the accuracy rates of the training set and test set of the support vector machine SVM prediction classification model

[0130]

[0131] It can be seen from Table 3 that the recognition classification method of different cut methods in the large-span small clear distance tunnel blasting of the present invention has an accuracy rate of more than 94% for the discriminant classification of three different cut methods: parallel hole cut, single-wedge cut, and double-wedge cut. The present invention provides a convenient, efficient, intelligent prediction, discrimination and classification method for tunnel structure safety, refined design and construction, and optimization of cut scheme selection.

[0132] The above is only the preferred embodiment of the present invention and is not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for identifying and classifying different cut patterns in the blasting of large-span and small-spacing tunnels, characterized in that It includes the following steps: P1. Collect the middle rock burst vibration signals under n known cut - hole patterns. For each cut - hole pattern, collect several groups of middle rock burst vibration signals that meet the analysis requirements; P2. Perform joint denoising on the collected middle rock burst vibration signals under any one of the cut - hole patterns by using multi - synchronous compressive transform (MSST) and wavelet threshold denoising (WTD) to obtain the denoised signal data; P3, the multi-fractal detrended fluctuation analysis (MF-DFA) is used for the denoised signal data to obtain a multi-fractal ( - ) spectrum, multiple eigenvalues in the multi-fractal ( - ) spectrum are extracted, and a multi-dimensional feature vector is established; P4. Repeat steps P2 and P3. For the several middle rock burst vibration signals of the n cut - hole patterns collected, establish the corresponding number of multi - dimensional feature vectors. Input the established multiple multi - dimensional feature vectors as a training set into the support vector machine (SVM) for calculation and training to obtain a well - trained SVM classifier; P5. Input the middle rock burst vibration signals under any one or more cut - hole patterns to be identified into the well - trained SVM classifier for identifying and classifying the cut - hole patterns, and obtain the category of the cut - hole pattern to which each middle rock burst vibration signal belongs. Among them, for the middle rock burst vibration signals under any one or more cut - hole patterns to be identified, after establishing the corresponding number of multi - dimensional feature vectors by using the above - mentioned analysis process, input them into the well - trained SVM classifier.

2. The recognition and classification method for different cut patterns in the blasting of long-span and small clear distance tunnels according to claim 1, characterized in that: The n known cut - hole patterns include parallel - hole cut, single - wedge cut, and double - wedge cut.

3. The method for identifying and classifying different cut patterns in the blasting of long-span and small clear-distance tunnels according to claim 1 or 2, characterized in that: In step P3, extract the multifractal ( - ) spectral width in the spectrum , dimensional difference , maximum value , asymmetric index , spectrum coverage area A five characteristic values , establish a five-dimensional feature vector .

4. The recognition and classification method for different cut patterns in the blasting of long-span and small clear distance tunnels according to claim 3, characterized in that, Step P3 is specifically as follows: For a burst signal sequence { x k} of length N, k = 1, 2, …, N , the MF-DFA calculation steps are as follows: (1) Calculate the average value of the sequence sample { x k}: (2) Determine the cumulative deviation of the signal samples: Among them, i = 1, 2, …, N ; (3) Divide the cumulative deviation sequence obtained in step (2) into N s sub - intervals. ; If N cannot be divided evenly by s, then starting from the tail of the deviation sequence , divide the tail into N s sub - intervals, and finally obtain 2N s equal - length sub - intervals; s represents the number of points divided within each equal - length sub - interval. (4) Perform m - order polynomial fitting of the least - squares method on s points in each equal - length small interval obtained by dividing in step (3): wherein i = 1, 2, …, s ; (5) Calculate the mean square error, and assume the interval is v = 1, 2, …, 2 N s , calculate the mean square error : When v = N s + 1, N s + 2,…,2 N s , calculate the mean square error : (6) After detrending the taking the average value, the q fluctuation function can be obtained: Among them, q is any non-zero real number, As s increases, it increases in a power-law relationship, that is , then for each s, there is a corresponding function value , for ln F q (s) - the slope in the function relationship graph of lns is the generalized Hurst exponent h ( q ); (7) Calculate the quality index: (8) Calculate the generalized dimension: (9) Plot the multifractal spectrum - Spectrum, calculate the spectral width 、Dimension difference 、Maximum value 、Asymmetry index 、Spectral coverage area A as five characteristic parameters of the multifractal spectrum; among them, the spectral width ; Dimension difference ; , L is the distance from the left pole of the multifractal spectrum to the perpendicular line of the maximum value, R is the distance from the right pole to the perpendicular line of the maximum value; the spectral coverage area A is obtained by numerical integration, and the upper and lower limits of the integral are respectively 、 .

5. The identification and classification method of different cut patterns in the blasting of large-span and small clear-distance tunnels according to claim 3, characterized in that: In step P4, steps P2 and P3 are repeated to establish multiple five-dimensional feature vectors for a number of intermediate rock blasting vibration signals of n cuttings collected , and the multiple five-dimensional feature vectors obtained are used as the training set and input into the support vector machine SVM for calculation and training, and the signal category attribution label is determined according to the number of votes ( , , ), and a trained SVM classifier is obtained. Among them, is the straight-hole cutting category label, is the single-wedge cutting category label, is the double-wedge blasting category label.

6. The identification and classification method for different cut methods in the blasting of large-span and small clear distance tunnels according to claim 5, characterized in that n = 3. The n known cut - hole patterns are parallel - hole cut, single - wedge cut, and double - wedge cut; the support vector machine (SVM) is a three - class support vector machine; In step P4, the training set is input into the support vector machine (SVM) for calculation and training to obtain a well-trained SVM classifier. Specifically, for the class membership labels of the training set ( , , ), a decision function is constructed using a three-class support vector machine to separate these three types of sample points. For the input F of a five-dimensional feature vector, a voting method is used to determine which class it belongs to. When , F belongs to class; when , F belongs to class; when , F belongs to class; finally, the votes are counted, and F belongs to the class with the most votes. All the data in the training set are classified and trained repeatedly.

7. The recognition and classification method for different cut patterns in the blasting of long-span and small clear distance tunnels according to claim 6, characterized in that The classifier was trained repeatedly using multiple sets of rock blasting vibration data until the training classification error was less than 10 -3 until.

8. The recognition and classification method for different cut patterns in the blasting of large-span and small-spacing tunnels according to claim 1, characterized in that, The specific process of step P2 is as follows: (1)First, perform a short-time Fourier transform on the intermediate rock burst vibration signal to obtain the spectral coefficients ; (2) Perform synchrosqueezing transform on the spectral coefficients to obtain synchrosqueezing transform coefficients ; (3) Use wavelet threshold to denoise the signal and synchronously compress the transform coefficients for filtering; (4) Set the number of iterations of MSST to 2, and perform the synchrosqueezing transform on the synchrosqueezing transform coefficients once again to obtain the second synchrosqueezing transform coefficients , and use to perform the inverse transform to obtain the effective signal; (5) Evaluate the denoising effect by using four evaluation indexes: signal - to - noise ratio (SNR), mean square error (MSE), peak signal - to - noise ratio (PSNR), and smoothness (SM). If it does not meet the requirements, adjust the value of the iteration times until the analysis accuracy requirements are met.

9. The recognition and classification method for different cut patterns in the blasting of long-span and small clear distance tunnels according to claim 1, characterized in that, In step P1, collecting the middle rock burst vibration signals under n known cut - hole patterns specifically includes: S1. Select the stable section of the middle rock lithology in the large - span and small - clear - distance tunnel, and arrange measuring points on the rock wall by using the integrated blasting vibration monitoring method; S2. Determine the blast - hole layout and related charging parameters on the upper bench of the tunnel, drill different forms of cut - holes and charge and detonate. Among them, the total charge of multiple different forms of cut - holes remains the same; S3. Monitor the middle rock burst vibration signals under different cut - hole patterns, store them and output the waveforms to the computer terminal.

10. The recognition and classification method for different cut patterns in the blasting of long-span and small clear distance tunnels according to claim 9, characterized in that, In step S1, the integrated blasting vibration monitoring method depends on the integrated monitoring system to execute; The integrated monitoring system includes: a customized large-capacity lithium battery power supply module for power supply, and a step-down board is set to reduce the battery voltage to the working voltage of the vibration measuring instrument; a protective hose is set on the surface of the wire of the vibration measuring instrument; a customized special instrument protective box with locking and warning functions; the inner backboard of the protective box is fixed on the surface of the middle rock stratum by expansion bolts, the sensor probe is fixed on the angle iron with prefabricated holes on the surface of the middle rock stratum and a lock washer is set to keep the connection reliable, and the gap with the rock wall is filled and bonded with lime powder stirred into a paste; before the test, the main unit of the vibration measuring instrument, the step-down board, the power supply module and the wire protection hose are integrated in the protection box, and the protection box is fixed on the explosion-facing side of the sensor, and the relevant wires are protected by the hose, thus forming the integrated monitoring system.

Images