Method and device for efficient determination of tunnel face failure probability based on dynamic multi-classifier system

Abstract

The application provides a tunnel face failure probability efficient determination method and device based on a dynamic multi-classifier system, which can greatly improve the calculation efficiency while ensuring the calculation precision and the accuracy of analysis and prediction results. The method comprises the following steps: step 1, selecting A samples as initial test sample points to obtain A initial critical points through numerical simulation; step 2, taking the A initial critical points as classification basis to obtain A initial classifiers, which jointly divide the parameter space into a joint stable domain Ω + and a joint unstable domain Ω ‑ ; step 3, classifying and determining the to-be-classified samples through the joint domains; step 4, updating the joint domain range through the undetermined samples; step 5, classifying and determining the remaining undetermined samples through the updated joint domain range, so that the number of undetermined samples in the undetermined set decreases; step 6, repeating steps 4 and 5 until the number of undetermined samples in the undetermined set is zero, and then stopping iteration and completing classification; and step 7, calculating the tunnel face failure probability.

Description

Technical Field

[0001] The present invention belongs to the technical field of tunnel engineering design and construction, and particularly relates to a method and device for efficiently determining the failure probability of a tunnel face based on a dynamic multi-classifier system. Background Art

[0002] In tunnels constructed in weak surrounding rock, especially soil, stress release occurs in the longitudinal direction of the face after excavation and gradually decreases to zero, which can easily lead to large extrusion deformation. If extrusion deformation is not promptly and effectively controlled, the face may become unstable and collapse, seriously threatening the lives of construction workers and the safety of mechanical construction equipment. Therefore, it is necessary to assess the safety of the face before tunnel construction. If the face is stable, normal excavation should be carried out. Otherwise, appropriate reinforcement measures should be taken to prevent the face from collapsing. Since tunnel face stability is a complex three-dimensional nonlinear problem, it is difficult to analyze it using empirical formulas or generally applicable analytical methods. Therefore, existing studies often use three-dimensional numerical calculation methods to evaluate face stability.

[0003] However, in practical engineering, some parameters that influence tunnel face stability, such as the cohesion and internal friction angle of the surrounding soil, are often uncertain. Therefore, tunnel face instability is a probabilistic problem. Predicting the probability of tunnel face failure based on the statistical characteristics of soil parameters is an urgent problem that needs to be solved.

[0004] Common probabilistic analysis methods include analytical methods and Monte Carlo methods. Monte Carlo methods offer the highest accuracy by performing deterministic analysis on a large number of samples, directly counting the number of failures in these samples and using this to calculate the failure probability. Deterministic analysis can be performed using three-dimensional numerical calculations. However, probabilistic analysis requires a large number of samples, potentially reaching tens of thousands or even hundreds of thousands. Performing three-dimensional numerical calculations on all these samples would require prohibitively long computation times, making it virtually impossible to perform using conventional computers.

[0005] Therefore, if Figure 1 As shown, existing techniques generally first select a small number of test samples and then determine an approximate limit state plane using methods such as polynomials, support vector machines, and neural networks. This limit state plane then divides the parameter space into two subspaces: the stability domain and the failure domain for prediction. However, due to the complexity of the model and strata, the true limit state plane of the tunnel face cannot be determined. Furthermore, the approximate limit state plane cannot completely match the true limit state plane, inevitably resulting in an error domain. Consequently, samples falling within this error domain will receive erroneous predictions, ultimately affecting the accuracy of the probabilistic analysis results. Summary of the Invention

[0006] The present invention is made to solve the above-mentioned problems, and its purpose is to provide a method and device for efficiently determining the failure probability of a tunnel face based on a dynamic multi-classifier system, which can greatly improve the calculation efficiency while ensuring the accuracy of the calculation accuracy and analysis and prediction results.

[0007] In order to achieve the above purpose, the present invention adopts the following scheme:

[0008] <Method>

[0009] like Figure 2 As shown (taking A=5 as an example), the present invention provides a method for efficiently determining the failure probability of a tunnel face based on a dynamic multi-classifier system, which is characterized by comprising the following steps:

[0010] Step 1: Select A samples as initial test sample points on the tunnel face using the central composite sampling method. Perform strength reduction calculations on these test sample points and convert the resulting safety factors to obtain A initial critical points on the limit state plane. (Here, only A points on the limit state plane, i.e., critical points, are determined, not the limit state plane.)

[0011] Step 2, such as Figure 3 As shown in (a) to (e) (corresponding to A = 5), A initial critical points are used as the classification basis, and A initial classifiers are obtained; Figure 3 Each of the figures (a) to (e) represents the division of an initial classifier. Each initial classifier is located at the corresponding initial critical point and the area to the upper right of the point in the parameter space (for example, Figure 3 The first initial critical point and the upper right black area in (a) are all divided into stable domains, and the lower left area of ​​the corresponding initial critical point is all divided into unstable domains (for example, Figure 3 (a) the lower left gray area of ​​the first initial critical point); Figure 3 As shown in (f), A initial classifiers jointly divide the parameter space into: the joint stable domain Ω obtained by the union of the stable domains of A initial classifiers + , the joint unstable domain Ω obtained by the union of the unstable domains of A initial classifiers - , and neither belongs to Ω + It does not belong to Ω - The undetermined domain;

[0012] Step 3: According to the probability density function of soil cohesion and internal friction angle, a predetermined number of samples to be classified (for example, several hundred to several thousand) are extracted by Latin sampling; then, the joint stability domain Ω divided in step 2 is used. + , joint unstable region Ω - And the undetermined domain is used to classify the samples to be classified: it will fall into the joint stable domain Ω +The samples to be classified are determined to be stable samples; those falling in the joint unstable domain Ω - The samples to be classified that fall into the undetermined domain are determined as unstable samples, and the samples to be classified that fall into the undetermined domain are determined as undetermined samples, forming an undetermined set.

[0013] Step 4, update the joint domain range through the undetermined samples;

[0014] like Figure 3 As shown in any of the figures (g) to (i), an undetermined sample is obtained from the undetermined set as a sample point, and its state is determined by three-dimensional strength reduction calculation (different from the initial test sample point to form the initial classifier, when the undetermined sample is used to form a new classifier, it is not necessary to calculate the safety factor and critical point, only the state needs to be determined). If it is a stable state, the sample point is used as the classification basis for the new stable domain to form a new classifier, so that the upper right area of ​​the parameter space surrounded by the horizontal and vertical coordinates of the sample point (for example Figure 3 The point pointed by the black arrow in (i) and the upper right area enclosed by the extension lines of the horizontal and vertical coordinates of the point are merged into the joint stability domain Ω + If it is in an unstable state, the sample point is used as the classification basis for the newly added unstable domain to form a new classifier, so that the lower left area of ​​the parameter space surrounded by the horizontal and vertical coordinates of the critical point (for example Figure 3 The point pointed by the black arrow in (g) or (h) and the lower left area enclosed by the extension lines of the horizontal and vertical coordinates of the point are merged into the joint instability domain Ω - middle;

[0015] Step 5: Use the updated joint stability region Ω from step 4 + and the joint unstable region Ω - Classify and judge the remaining undetermined samples so that the number of undetermined samples in the undetermined set decreases;

[0016] Step 6: Repeat steps 4 and 5 until the number of undetermined samples in the undetermined set reaches zero, and the classification is completed.

[0017] Step 7: After the sample classification is completed, count the number of unstable samples N - , if there is N - ≥100, the failure probability of the tunnel face can be calculated:

[0018] P f =N - / (N - +N + ),

[0019] Where, P f represents the failure probability of the tunnel face; N + Represents the number of all stable samples; the total number of samples is N = N - +N+ = N - / P f ≥ 100 / P f ;

[0020] If N - < 100, repeat steps 3 to 7 until N - ≥ 100, and the tunnel face failure probability is calculated.

[0021] Preferably, the tunnel face failure probability efficient determination method based on the dynamic multi-classifier system provided by the present application can also have the following features: the time required for comparing the sample cohesion and the internal friction angle is much lower than the numerical calculation, so the larger the area of the combined stable domain Ω + and the combined unstable domain Ω - , the higher the calculation efficiency of the method; in three-dimensional strength reduction calculation, if only the state (stable or unstable) is determined, it can be completed within a few minutes, and if the safety factor is to be determined, the calculation time is relatively longer, about 5-10 times of the stable state determination, in order to reduce the calculation consumption of the test sample as much as possible, in step 1, the value range is recommended to be 3 ≤ A ≤ 11, and A is an odd number; the optimal value is A = 5, and the five initial test sample points are respectively: μ c and respectively represent the average value of the cohesion and the internal friction angle of the soil body, σ c and respectively represent the standard deviation of the cohesion and the internal friction angle of the soil body; the strength reduction calculation is performed on the five initial test sample points, and the safety factor obtained is converted, so that the five initial critical points on the limit state plane can be obtained.

[0022] Preferably, the tunnel face failure probability efficient determination method based on the dynamic multi-classifier system provided by the present application can also have the following features: the initial critical point coordinates are set as c cr , the cohesion of the critical point, , the internal friction angle of the critical point, and the coordinates of an arbitrary point in the parameter space are c * , the cohesion of the sample, , the internal friction angle of the sample, so in step 2, the initial classifier divides the parameter space in the right upper region of the corresponding initial critical point into a stable domain, which means that all points satisfying c * ≥ c cr and form a stable domain; the initial classifier divides the parameter space in the left lower region of the corresponding initial critical point into an unstable domain, which means that all points satisfying c * < ccr and All points of If there is cohesion c * ≥ c cr and internal friction angle It is explained that the sample The corresponding working condition is stable state; if there is cohesion c * < c cr and internal friction angle It is explained that the sample The corresponding working condition is unstable state.

[0023] Preferably, the tunnel face failure probability efficient determination method based on dynamic multi-classifier system provided by the present application can also have the following features: for a sample The safety factor is obtained by strength reduction calculation:

[0024]

[0025] In the formula, The critical point on the limit state plane is determined by the initial value of cohesion and internal friction angle and the strength reduction factor FS determined by calculation:

[0026]

[0027] [Device]

[0028] Further, the present application also provides a tunnel face failure probability efficient determination device based on dynamic multi-classifier system, characterized in that it comprises:

[0029] An initial critical point acquisition unit selects A samples as initial test sample points by using central composite sampling on the tunnel face; the test sample points are subjected to strength reduction calculation, and the obtained safety factor is converted to obtain A initial critical points;

[0030] A domain range division unit obtains A initial classifiers based on the A initial critical points; each initial classifier divides the right upper region of the corresponding initial critical point and the parameter space into a stable domain, and divides the left lower region of the corresponding initial critical point into an unstable domain; the A initial classifiers together divide the parameter space into a joint stable domain Ω + obtained by the stable domains of the A initial classifiers, a joint unstable domain Ω - obtained by the unstable domains of the A initial classifiers, and an undetermined domain which does not belong to Ω + nor Ω - ; and

[0031] The classification unit classifies the to-be-classified samples according to the joint stable domain Ω + , the joint unstable domain Ω - and the undetermined domain according to the probability density functions of the cohesion and internal friction angle of the soil mass + : the to-be-classified samples falling in the joint stable domain Ω - are determined as stable samples; the to-be-classified samples falling in the joint unstable domain Ω - are determined as unstable samples; and the to-be-classified samples falling in the undetermined domain are determined as undetermined samples, forming an undetermined set

[0032] The domain range updating unit updates the domain range by using the undetermined samples: an undetermined sample is obtained from the undetermined set as a sample point, and the state of the sample point is determined by using the three-dimensional strength reduction calculation; if the sample point is in a stable state, the sample point is used as a classification basis for a newly added stable domain, a new classifier is formed, and the right upper region of the parameter space surrounded by the horizontal and vertical coordinates of the sample point is merged (union) into the joint stable domain Ω + ; if the sample point is in an unstable state, the sample point is used as a classification basis for a newly added unstable domain, a new classifier is formed, and the left lower region of the parameter space surrounded by the horizontal and vertical coordinates of the sample point is merged into the joint unstable domain Ω - ;

[0033] The undetermined sample classification unit classifies the remaining undetermined samples according to the joint stable domain Ω + and the joint unstable domain Ω - updated by the domain range updating unit, so that the number of undetermined samples in the undetermined set is reduced

[0034] The iteration unit repeats the processing of the domain range updating unit and the undetermined sample classification unit until the number of undetermined samples in the undetermined set is zero, and then the iteration is stopped, and the classification is completed

[0035] The failure probability determination unit determines the failure probability of the tunnel face after the classification of the samples is completed, and the number of unstable samples N - is counted; if N - ≥ 100, the failure probability of the tunnel face can be calculated as follows:

[0036] P f = N - / (N - +N + ),

[0037] wherein P f represents the failure probability of the tunnel face; N + represents the number of all stable samples; and the total number of samples is N = N - +N + ;

[0038] If N - <100, repeat steps 3 to 7 until N - ≥100, the program is terminated, and the tunnel face failure probability is calculated;

[0039] The control unit is in communication with the initial critical point acquisition unit, the domain range division unit, the classification unit, the domain range update unit, the undetermined sample classification unit, the iteration unit, and the failure probability determination unit, and controls the operation of them.

[0040] Preferably, the tunnel face failure probability efficient determination device based on the dynamic multi-classifier system provided by the application can further comprise an input display unit in communication with the initial critical point acquisition unit, the domain range division unit, the classification unit, the domain range update unit, the undetermined sample classification unit, the iteration unit, the failure probability determination unit, and the control unit, for inputting operation instructions and displaying accordingly.

[0041] Preferably, the tunnel face failure probability efficient determination device based on the dynamic multi-classifier system provided by the application can further have the following features: in the initial critical point acquisition unit, 3≤A≤11, and A is an odd number.

[0042] Preferably, the tunnel face failure probability efficient determination device based on the dynamic multi-classifier system provided by the application can further have the following features: in the initial critical point acquisition unit, A=5, and the 5 initial test sample points are respectively:

[0043]

[0044] μ c and respectively represent the average values of the soil cohesion and the internal friction angle, and σ c and respectively represent the standard deviations of the soil cohesion and the internal friction angle.

[0045] The 5 initial test sample points are subjected to strength reduction calculation, and the safety factor obtained is converted, so that the upper 5 initial critical points on the limit state plane can be obtained.

[0046] Preferably, the tunnel face failure probability efficient determination device based on the dynamic multi-classifier system provided by the application can further have the following features: the initial critical point coordinates are set as The coordinates of any point on the parameter space are set as In the domain range division unit, the initial classifier divides the parameter space into stable domains, which are located at the corresponding initial critical points and the upper right regions of the points, and the stable domains are defined as follows: c * ≥c cr and All points of the stable domain; the initial classifier divides the left lower area of the corresponding initial critical point in the parameter space into the stable domain, which means that c * <c cr and All points of the unstable domain.

[0047] Effects of the application

[0048] The tunnel face failure probability efficient determination method and device based on the dynamic multi-classifier system provided by the application first selects initial test sample points through central composite sampling, then calculates the tunnel face safety factor by using the intensity reduction method, and determines the initial critical point on the limit state plane corresponding to each initial test sample point, which is used as an initial classifier to form a joint stable domain and a joint unstable domain, and to construct a multi-classifier system. On this basis, a batch of prediction samples are obtained through Latin sampling, and the multi-classifier system is used to judge the stable state of the samples. For the undetermined samples falling outside the joint stable domain and the joint unstable domain, the numerical model is introduced for calculation to determine the state of the samples, and the samples are used as new classifiers to predict the states of the remaining unclassified samples, and then the new classifiers are added to the original classifier system to update the joint domain range. The process is repeated until the total number of failure samples reaches 100 or more, and the face failure probability is calculated. Based on this, the multi-classifier system is dynamically constructed through numerical calculation of a small number of samples, the generation of the error domain is avoided, the states of a large number of samples of the tunnel face are judged by using the classifier system, the number of numerical calculations required can be greatly reduced, and the calculation efficiency is greatly improved.

[0049] In summary, the application can ensure the calculation accuracy and the accuracy of the analysis and prediction results, effectively improve the processing efficiency, and provide a new way for efficient and accurate determination of the face stability and rapid, safe and reliable tunnel construction. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 The test sample and limit state plane schematic diagram involved in the background art;

[0051] Figure 2 The flowchart of the tunnel face failure probability efficient determination method based on the dynamic multi-classifier system involved in the embodiment of the application;

[0052] Figure 3 The method principle diagram (the curves in the diagram are irrelevant to the principle of the method, and are only used to distinguish the diagram, and the method is not calculated or used in the actual processing process) involved in the application;

[0053] Figure 4 The tunnel numerical model diagram involved in the embodiment of the application;

[0054] Figure 5 The calculation result schematic diagram of the 30kPa support pressure classifier determined in the embodiment one of the present application;

[0055] Figure 6 The partition schematic diagram of the combined stable region, combined unstable region and undetermined region in the embodiment one of the present application;

[0056] Figure 7 The classification result schematic diagram of the combined region classified by the 5 initial classifiers under the 30kPa support pressure in the embodiment one of the present application;

[0057] Figure 8 The classification result schematic diagram of the combined region classified after the range of the combined region is updated under the 30kPa support pressure in the embodiment one of the present application;

[0058] Figure 9 The calculation result schematic diagram of the 50kPa support pressure classifier determined in the embodiment two of the present application;

[0059] Figure 10 The classification result schematic diagram of the combined region classified by the 5 initial classifiers under the 50kPa support pressure in the embodiment two of the present application;

[0060] Figure 11 The classification result schematic diagram of the combined region classified after the range of the combined region is updated under the 50kPa support pressure in the embodiment two of the present application. DETAILED DESCRIPTION

[0061] The specific implementation of the tunnel face failure probability efficient determination method and device based on the dynamic multi-classifier system will be described in detail in combination with the drawings. In the following embodiments, the analysis software is FLAC 3D , and the steps and methods are all conventional methods unless otherwise specified.

[0062] <Embodiment one>

[0063] A circular tunnel with a diameter and a buried depth of 10m is selected, and the cohesion and internal friction angle of the soil in the stratum are random variables, with average values of μ c =7kPa and standard deviations of σ c =1.4kPa and Other variables are all ordinary variables, the soil bulk density is γ=18kN / m 3 , the elastic modulus is E=240MPa, and the Poisson's ratio is v=0.3. The numerical analysis model is established as Figure 4As shown, due to symmetry, only half of the numerical model is built, with the model having dimensions of 35 m, 70 m and 50 m in the x, y and z directions respectively. The bottom of the model is fully constrained, the four outer faces perpendicular to the x and y coordinate axes are normal constraints, and the top is a free surface.

[0064] In this embodiment, it is assumed that the support pressure applied to the tunnel face is 30 kPa, and the support pressure applied to the face when excavating to 3 times the hole diameter is 30 kPa.

[0065] Central composite sampling is used, and five initial test sample points are selected as follows: (4.2 kPa, 20.4 0 ), (5.6 kPa, 18.7 0 ), (7 kPa, 17 0 ), (8.4 kPa, 15.3 0 ), and (9.8 kPa, 13.6 0 ).

[0066] When the support pressure is 30 kPa, strength reduction calculations are performed for each initial test sample point to obtain safety factors, and five initial critical points on the limit state plane are determined, with the converted data shown in Table 1. The distribution of the initial test sample points (test sample points) and the corresponding initial critical points (critical sample points) is shown in Figure 5 .

[0067] Table 1 Test sample points and critical sample points under the action of a support pressure of 30 kPa

[0068]

[0069] As shown in Figure 6 , according to the five initial classifiers determined, the parameter space can be divided into three parts: a joint stable domain (the upper right region enclosed by the solid line), a joint unstable domain (the lower left region enclosed by the dashed line), and an undetermined domain (the remaining region enclosed by the dashed and solid boundary lines).

[0070] According to the probability density functions of the soil cohesion and internal friction angle, 200 samples are extracted using the Latin sampling method, and after the process shown in Figure 2 , the total number of unstable samples obtained is less than 100. 200 samples are again extracted for judgment and calculation, and the number of unstable samples obtained is N - = 123. Thus, the failure probability is calculated as:

[0071] P f = N - / (N - +N + ) = 123 / 400 = 0.3075,

[0072] As shown in Figure 7As shown, there are 218 stable samples, 87 unstable samples, and 95 indefinite samples (the indefinite set) predicted by the five initial classifiers. Among the 95 indefinite samples, 31 stable samples, 26 unstable samples are determined by numerical calculation, and the joint domain range is updated as Figure 8 As shown, based on the updated joint domain range, 28 stable samples and 10 unstable samples are determined from the other indefinite samples in the indefinite set. In this way, the number of samples requiring three-dimensional numerical calculation is reduced from 400 to 57, with a reduction of 85.7%, greatly improving the calculation efficiency.

[0073] <Embodiment Two>

[0074] The tunnel face failure probability in Embodiment One is 0.3075, which has a relatively large risk of collapse. In Embodiment Two, the face support pressure is changed to 50 kPa, and the face failure probability is determined according to the method of the present application. The five initial classifiers (critical sample points) determined under the action of the 50 kPa support pressure are as shown in Figure 9

[0075] The calculation results obtained according to the calculation process of the present application are as shown in Figure 10 and 11 After multiple iterations, a total of 34100 samples are collected, among which 33861 stable samples, 62 unstable samples, and 177 indefinite samples are predicted by the five initial classifiers. These indefinite samples are determined by numerical simulation to be 117 stable samples and 32 unstable samples, and the joint domain range is updated accordingly. According to the updated joint domain range, further classification of the remaining indefinite samples is performed, and 22 stable samples and 6 unstable samples are determined. The number of unstable samples obtained is N - = 100, and the failure probability is calculated as:

[0076] P f = N - / (N - +N + ) = 100 / 34100 = 0.0029,

[0077] In this way, the total number of simulation samples is reduced from 34100 to 149, with a reduction of 99.6%, greatly improving the calculation efficiency.

[0078] From the results of the above two embodiments, it can be seen that the smaller the tunnel face failure probability, the more obvious the advantage of the present application in calculation efficiency.

[0079] <Embodiment Three>

[0080] ​The embodiment three provides a tunnel face failure probability efficient determination device based on a dynamic multi-classifier system, which can automatically implement the above method. The device comprises an initial critical point acquisition part, a domain range division part, a classification part, a domain range updating part, an undetermined sample classification part, an iteration part, a failure probability determination part, an input display part and a control part.

[0081] The initial critical point acquisition part selects A samples as initial test sample points by using central composite sampling on the tunnel face; performs strength reduction calculation on the test sample points, converts the obtained safety factors to obtain A initial critical points.

[0082] The domain range division part obtains A initial classifiers based on the A initial critical points; each initial classifier divides the right upper region of the corresponding initial critical point and the parameter space into a stable domain, and divides the left lower region of the corresponding initial critical point into an unstable domain; the A initial classifiers collectively divide the parameter space into a joint stable domain Ω + obtained by the stable domains of the A initial classifiers, a joint unstable domain Ω - obtained by the unstable domains of the A initial classifiers, and an undetermined domain which neither belongs to Ω + nor belongs to Ω - .

[0083] The classification part extracts a predetermined number of samples to be classified by using Latin sampling according to the probability density function of the soil cohesion and internal friction angle; then, the joint stable domain Ω + , the joint unstable domain Ω - and the undetermined domain divided by the domain range division part are used to classify the samples to be classified: the samples to be classified falling in the joint stable domain Ω + are determined as stable samples; the samples to be classified falling in the joint unstable domain Ω - are determined as unstable samples; the samples to be classified neither falling in the stable domain Ω + nor falling in the unstable domain Ω - are determined as undetermined samples, forming an undetermined set.

[0084] The domain range updating part updates the domain range through the undetermined samples: an undetermined sample is obtained from the undetermined set as a sample point, and the state of the sample point is determined by using three-dimensional strength reduction calculation; if the sample point is in a stable state, the sample point is used as classification basis of a newly added stable domain to form a newly added classifier, and the right upper region of the parameter space surrounded by the horizontal and vertical coordinates of the sample point is merged into the joint stable domain Ω + ; if the sample point is in an unstable state, the sample point is used as classification basis of a newly added unstable domain to form a newly added classifier, and the left lower region of the parameter space surrounded by the horizontal and vertical coordinates of the sample point is merged into the joint unstable domain Ω - .

[0085] The undetermined sample classification part uses the joint stable domain Ω updated by the domain range update part + and the joint unstable region Ω - The remaining undetermined samples are classified and judged, so that the number of undetermined samples in the undetermined set is reduced.

[0086] The iteration part repeats the processing of the domain range updating part and the undetermined sample classification part until the number of undetermined samples in the undetermined set reaches zero, and the iteration is stopped, thus completing the classification.

[0087] After the sample classification is completed, the failure probability determination unit counts the number of unstable samples N - , if there is N - ≥100, the failure probability of the tunnel face can be calculated:

[0088] P f =N - / (N - +N + ),

[0089] Where, P f represents the failure probability of the tunnel face; N + Represents the number of all stable samples; the total number of samples is N = N - +N + ;

[0090] If N - <100, repeat steps 3 to 7 until N is satisfied - The program is terminated when ≥100, and the failure probability of the tunnel face is calculated.

[0091] The input display unit is communicatively connected to the initial critical point acquisition unit, the domain range division unit, the classification unit, the domain range update unit, the undetermined sample classification unit, the iteration unit, and the failure probability determination unit, and is used to allow the user to input operation instructions and display them accordingly.

[0092] The control unit is connected to the initial critical point acquisition unit, the domain range division unit, the classification unit, the domain range update unit, the undetermined sample classification unit, the iteration unit, the failure probability determination unit, and the input display unit to control their operations.

[0093] The above embodiments are merely illustrative of the technical solutions of the present invention. The method and apparatus for efficiently determining tunnel face failure probability based on a dynamic multi-classifier system, as described herein, are not limited solely to the content described in the above embodiments but are subject to the scope defined by the claims. Any modifications, supplements, or equivalent substitutions made by persons skilled in the art based on these embodiments are considered within the scope of protection claimed in the claims.

Claims

1. A method for efficient determination of the probability of face failure of a tunnel based on a dynamic multi-classifier system, characterized in that, The following steps are involved: Step 1: A samples are selected as initial test sample points on the tunnel face using the central composite sampling method; Perform strength reduction calculation on these test sample points and convert them using the obtained safety factor to obtain A initial critical points; Step 2, taking A initial critical points as classification basis, obtaining A initial classifiers; each initial classifier divides the parameter space into stable domain and unstable domain, wherein the stable domain is located in the right upper region of the corresponding initial critical point, and the unstable domain is located in the left lower region of the corresponding initial critical point; A initial classifiers jointly divide the parameter space into: the joint stable domain Ω + obtained by the union of the stable domains of A initial classifiers, and the joint unstable domain Ω - obtained by the union of the unstable domains of A initial classifiers; Step 3: According to the probability density function of soil cohesion and internal friction angle, a predetermined number of samples to be classified are extracted by Latin sampling method; then, the joint stability domain Ω divided in step 2 is obtained. + and the joint unstable region Ω - Classification judgment of the samples to be classified: will fall into the joint stable domain Ω + The samples to be classified are determined to be stable samples; those falling in the joint unstable domain Ω - The samples to be classified are judged as unstable samples; they do not fall into the stable domain Ω + Nor does it fall into the unstable region Ω - The samples to be classified are determined as undetermined samples, forming an undetermined set; Step 4, update the joint domain range through the undetermined samples; An undetermined sample is obtained from the undetermined set as a sample point, and its state is determined by using three-dimensional intensity reduction calculation; if it is in stable state, the sample point is taken as a new stable domain classification basis to form a new classifier, so that the right upper region of the parameter space surrounded by the horizontal and vertical coordinates of the sample point is merged into the joint stable domain Ω + If it is in unstable state, the sample point is taken as a new unstable domain classification basis to form a new classifier, so that the left lower region of the parameter space surrounded by the horizontal and vertical coordinates of the sample point is merged into the joint unstable domain Ω - ​ Step 5, update the joint stable region Ω with the result of step 4 + and the joint unstable region Ω - Classify the remaining undecided samples so that the number of undecided samples in the undecided set is reduced; Step 6: Repeat steps 4 and 5 until the number of undetermined samples in the undetermined set reaches zero, and the classification is completed. Step 7, after the sample classification is completed, the number N of unstable samples is counted - If N - ≥ 100, the failure probability of the tunnel face can be calculated: P f = N - / (N - + N + ), In the formula, P f represents the tunnel face failure probability; N + represents the number of all stable samples; the total number of samples is N=N - +N + ; If N - <100, repeat steps 3 to 7 until N - ≥100, the program is terminated, and the tunnel face failure probability is calculated.

2. The method for efficiently determining tunnel face failure probability based on a dynamic multi-classifier system according to claim 1 is characterized by: wherein In step 1, 3≤A≤11, and A is an odd number.

3. The method for efficiently determining tunnel face failure probability based on a dynamic multi-classifier system according to claim 1 is characterized by: wherein In step 1, A = 5, and the 5 initial trial sample points are: μ c and respectively represent the mean value of soil cohesion and internal friction angle, σ c and respectively represent the standard deviation of soil cohesion and internal friction angle; The five initial test sample points are subjected to strength reduction calculations and converted using the obtained safety factors to obtain the five initial critical points.

4. The method for efficiently determining tunnel face failure probability based on a dynamic multi-classifier system according to claim 1 is characterized by: wherein, let the initial critical point coordinates are c cr the cohesion of the critical point, the internal friction angle of the critical point, the coordinates of an arbitrary point in the parameter space are c * the cohesion of the sample, the internal friction angle of the sample, then in step 2, the initial classifier divides the parameter space into stable domains and unstable domains, wherein the initial critical point and the upper right region of the point are located in the parameter space, which means that all points satisfying c * ≥c cr and form a stable domain; the initial classifier divides the parameter space into stable domains and unstable domains, wherein the initial critical point and the lower left region of the point are located in the parameter space, which means that all points satisfying c * <c cr and form an unstable domain.

5. The method for efficiently determining tunnel face failure probability based on a dynamic multi-classifier system according to claim 1 is characterized by: wherein, For one sample Safety factor calculated by strength reduction: wherein denotes the critical point on the limit state plane, whose value is determined by the initial values of the cohesion and the internal friction angle and the strength reduction factor FS determined by the calculation:

6. Apparatus for efficient determination of the probability of face failure of a tunnel face based on a dynamic multiple classifier system, characterized in that include: The initial critical point acquisition part selects A samples as the initial test sample points on the tunnel face using the central composite sampling method; Perform strength reduction calculation on these test sample points and convert them using the obtained safety factor to obtain A initial critical points; The domain range division part obtains A initial classifiers with A initial critical points as classification basis; each initial classifier divides the right upper region of the corresponding initial critical point and the parameter space into a stable domain, and divides the left lower region of the corresponding initial critical point into an unstable domain; the A initial classifiers jointly divide the parameter space into: a joint stable domain Ω obtained by the union of the stable domains of the A initial classifiers + , and a joint unstable domain Ω - obtained by the union of the unstable domains of the A initial classifiers. The classification unit extracts a predetermined number of samples to be classified by using Latin sampling method according to the probability density function of the soil cohesion and internal friction angle; then the joint stable domain Ω + , and the joint unstable domain Ω - are divided by the domain range division unit The samples to be classified are classified and determined: the samples to be classified falling in the joint stable domain Ω + are determined as stable samples; the samples to be classified falling in the joint unstable domain Ω - are determined as unstable samples; the samples to be classified neither falling in the stable domain Ω + nor falling in the unstable domain Ω - are determined as undetermined samples, forming an undetermined set; The domain range updating unit updates the domain range by an undetermined sample: an undetermined sample is obtained from the undetermined set as a sample point, the state of the sample point is judged by using three-dimensional intensity reduction calculation, if the sample point is in stable state, the sample point is taken as a new stable domain classification basis to form a new classifier, and the right upper region of the parameter space surrounded by the horizontal and vertical coordinates of the sample point is merged into the joint stable domain Ω + If the sample point is in unstable state, the sample point is taken as a new unstable domain classification basis to form a new classifier, and the left lower region of the parameter space surrounded by the horizontal and vertical coordinates of the sample point is merged into the joint unstable domain Ω - ​ The undetermined sample classification unit classifies the undetermined samples in the undetermined set according to the updated joint stable domain Ω + and the joint unstable domain Ω - The undetermined sample classification unit classifies the undetermined samples in the undetermined set according to the updated joint stable domain Ω The iteration part repeats the processing of the domain range update part and the undetermined sample classification part until the number of undetermined samples in the undetermined set reaches zero, and the iteration stops and the classification is completed; The failure probability determination unit counts the number N of unstable samples after the sample classification is completed - If N - ≥ 100, the failure probability of the tunnel face can be calculated: P f = N - / (N - + N + ), In the formula, P f represents the tunnel face failure probability; N + represents the number of all stable samples; the total number of samples is N=N - +N + ; If N - <100, repeat steps 3 to 7 until N - ≥100, the program is terminated, and the tunnel face failure probability is calculated. The control unit is connected to the initial critical point acquisition unit, the domain range division unit, the classification unit, the domain range update unit, the undetermined sample classification unit, the iteration unit, and the failure probability determination unit to control their operations.

7. The dynamic multi-classifier system based tunnel face failure probability efficient determination apparatus according to claim 6, wherein, Also includes: The input display unit is communicatively connected with the initial critical point acquisition unit, the domain range division unit, the classification unit, the domain range update unit, the undetermined sample classification unit, the iteration unit, the failure probability determination unit, and the control unit, and is used to allow the user to input operation instructions and display them accordingly.

8. The device for efficiently determining tunnel face failure probability based on a dynamic multi-classifier system according to claim 6 is characterized by: wherein, In the initial critical point acquisition section, 3≤A≤11, and A is an odd number.

9. The device for efficiently determining tunnel face failure probability based on a dynamic multi-classifier system according to claim 6, characterized in that: wherein, In the initial critical point acquisition unit, A = 5, and the 5 initial test sample points are respectively: μ c and respectively represent the average value of soil cohesion and internal friction angle, σ c and respectively represent the standard deviation of soil cohesion and internal friction angle; The five initial test sample points are subjected to strength reduction calculations and converted using the obtained safety factors to obtain the five initial critical points.

10. The device for efficiently determining tunnel face failure probability based on a dynamic multi-classifier system according to claim 6, characterized in that: wherein, let The coordinates of the initial critical point are The coordinates of an arbitrary point on the parameter space are In the domain range division section, the initial classifier divides the parameter space located in the corresponding initial critical point and the right upper region of the point into stable domains, which means that c * ≥ c cr and all points of form a stable domain; the initial classifier divides the parameter space located in the corresponding initial critical point and the left lower region of the point into unstable domains, which means that c * < c cr and all points of form an unstable domain.

Images