CN111914943B

CN111914943B - Information vector machine method and device for comprehensively judging stability of dumping type karst dangerous rock

Info

Publication number: CN111914943B
Application number: CN202010820170.XA
Authority: CN
Inventors: 苏国韶; 李培峰; 许华杰; 张研; 罗丹旎; 黄小华; 蒋剑青; 郑志
Original assignee: Guangxi University
Current assignee: Guangxi University
Priority date: 2020-08-14
Filing date: 2020-08-14
Publication date: 2022-04-15
Anticipated expiration: 2040-08-14
Also published as: CN111914943A

Abstract

The invention discloses an information vector machine method and device for comprehensively judging stability of dumping type karst dangerous rocks, and mainly solves the problem of reasonably judging the stability of dumping type unstable collapse type dangerous rocks in a karst area. Firstly, selecting characteristic indexes according to various influence factors of the stability of the dumping karst dangerous rock; secondly, training by utilizing an engineering example sample set and establishing an information vector machine classification model with excellent automatic small sample classification performance; secondly, inputting the characteristic vector of the influence factors of the new dangerous rock mass to be judged into the information vector machine classification model on the basis that the real-time monitoring device provides dynamic information of the fracture water pressure of the main control structural plane to obtain a comprehensive stability judgment result of the new dangerous rock mass; and finally, dividing the uncertainty grade of the discrimination result, and when the uncertainty is higher or the engineering grade of the dangerous rock mass to be discriminated is higher, calculating and rechecking the discrimination result by adopting a rigid body limit balance method. The method is suitable for quickly judging the stability of the dumping dangerous rock in the karst area.

Description

Information vector machine method and device for comprehensively judging stability of dumping type karst dangerous rock

Technical Field

The invention belongs to the technical field of geological disaster prevention and control engineering, and relates to an information vector machine classification method and device for comprehensively judging dumping type dangerous rock stability in karst mountainous areas.

Background

The dangerous rock is a geologic body which is cut and separated by a plurality of groups of structural surfaces, has poor stability and can collapse in the forms of toppling, falling, sliding and the like. The dumping dangerous rock is a dangerous rock body separated from a stable parent rock by a steep dumping structural plane on a steep slope, and is easy to generate outward dumping instability under the action of factors such as gravity and the like.

The karst area of China is widely distributed, the karst belongs to soluble rock, the stability is easily influenced by water, and compared with the non-soluble rock, the karst rock on the side slope is easier to crack so as to form dangerous rock, and the collapse of the dangerous rock becomes one of the main geological disasters of the karst area. In recent decades, a great deal of research work is done at home and abroad on the aspect of judging the stability of the dumping karst dangerous rocks, people analyze the instability and collapse phenomena of the dumping karst dangerous rocks from the aspects of strength, joints, loads, natural factors, human factors and the like of dangerous rocks, and various judgment theories and methods are provided. However, compared with sliding type and falling type dangerous rocks, the stability of the dangerous rock body is not only influenced by the spatial relationship between the gravity center and the overturning point, but also influenced by other various complex factors, the collapse instability generating mechanism is more complex, and a highly complex nonlinear mapping relationship is presented between the stability influencing factor of the dumping type karst dangerous rock and the stability thereof. Under the condition that the collapse and instability mechanism of the dumping karst dangerous rock is unclear, when the stability of the dumping karst dangerous rock is judged by adopting the traditional methods such as mechanical calculation, numerical calculation and the like, the judgment error is often larger, the manpower and economic cost required to be input in the calculation is higher, and the engineering practice requirement that the judgment result needs to be obtained quickly, accurately and at low cost in the process of evaluating the stability of a large number of dangerous rock masses cannot be completely met.

An Information Vector Machine (IVM) is a new machine learning method, which is based on Bayesian statistical learning theory and kernel method, and has the advantages of super-parameter self-adaptive acquisition, high dimension and complex nonlinear problem adaptability, probability significance of prediction output and the like, and meanwhile, a method based on information entropy theory is adopted, partial samples (information vectors) with most information are selected from a large number of training samples, the same or similar effect as the original training sample set is achieved by learning the samples, and sparse kernel matrix representation is combined, so that the learning time and space complexity are remarkably reduced. In addition, the density Approximation (ADF), also called moment matching approximation method and the introduction of KL divergence (relative information entropy) are assumed, so that the method has strong approximation processing capability for non-gaussian distributed noise. The information vector machine can better take the advantages of the current mainstream machine learning methods such as an artificial neural network and a support vector machine into account, effectively avoids the limitation that the optimal hyper-parameter is difficult to determine, and has good popularization and application prospect and value. The invention introduces the method into the comprehensive judgment of the stability of the dumping karst dangerous rock, and provides an information vector machine classification method and a device for the comprehensive judgment of the stability of the dumping karst dangerous rock, so that the stability of the dumping karst dangerous rock is judged efficiently and accurately, and a foundation is laid for the safety reinforcement and prevention of the dumping karst dangerous rock.

Disclosure of Invention

The invention provides an IVM classification method and device for comprehensively distinguishing stability of dumping karst dangerous rocks, aiming at the problem that the existing method for comprehensively distinguishing stability of dumping karst dangerous rocks is poor in effect and efficiency. The method comprises the steps of establishing a sample library by widely collecting dangerous rock examples of representative karst areas at home and abroad, training an IVM classification model for stable comprehensive judgment of the dumping type karst dangerous rock by utilizing a K-fold cross validation algorithm (K-CV), establishing a nonlinear mapping relation between a plurality of leading influence factors and stability categories of the dumping type karst dangerous rock, and providing the IVM classification method and device for realizing stable comprehensive judgment of the dumping type karst dangerous rock by combining an implementation method.

The technical scheme is as follows:

an information vector machine method for comprehensively judging stability of a dumping karst dangerous rock comprises the following steps:

step S1: selecting a dumping karst dangerous rock stability characteristic index;

step S2: by widely collecting dumping type karst dangerous rock examples at home and abroad and preprocessing data, the data quality of example samples is improved, a sample library is established, and the data of the sample library is subjected to standardized processing so as to be beneficial to improving the judgment precision;

step S3: dividing a sample library into a training sample and a testing sample by adopting a cross validation strategy, training and testing an IVM classification model for the comprehensive and stable judgment of the dumping karst dangerous rock, and optimizing the hyper-parameters of the IVM classification model according to the training and testing effects to obtain a trained dumping karst dangerous rock stable and comprehensive judgment IVM classification model;

step S4: applying the trained IVM classification model for comprehensively judging stability of the dumping type karst dangerous rock, and inputting the feature vector of the dumping type karst dangerous rock sample to be judged to obtain the classification probability of the stability of the dumping type karst dangerous rock sample;

step S5: according to the classification probability, dividing the uncertainty level of the discrimination result;

step S6: and when the uncertainty level of the judgment result belongs to a high level, rechecking the accuracy of the judgment result by adopting a rigid body limit balance method. Moreover, for a new dangerous rock sample with a higher engineering grade, serious personnel and equipment loss can be caused after collapse, so that serious safety accidents can be caused, and the rechecking of the stability judgment result also needs to be carried out.

Illustratively, the invention relates to predicting and distinguishing two keywords, and it is noted that the prediction mentioned in the invention is a concept from an IVM classification model and is not a prediction on a time scale; the judgment means that the IVM classification model is applied to execute the judgment of the instability state of the dumping karst dangerous rock mass; the two terms presented in the present invention do not confuse the conflict, and can be understood as being predicted as a form, but judged for the purpose.

The specific description of each step is as follows:

step S1 specifically describes:

selecting 12 characteristic indexes which obviously influence the stability of the poured karst dangerous rock, and specifically comprising the following steps: whether the gravity center of the dangerous rock mass is outside an overturning point, the hardness of the base rock, the development degree of the karst and the weathering degree are 4 qualitative characteristic indexes, and the volume weight W of the dangerous rock mass_lHorizontal distance a from gravity center of dangerous rock mass to overturning point, characteristic period T of regional earthquake dynamic acceleration response spectrum, regional earthquake dynamic peak acceleration G and tensile strength sigma of dangerous rock mass_tThe dip angle beta of the main control structure surface, the length l of the through section of the main control structure surface and the fracture water pressure Q of the main control structure surface are 8 quantitative characteristic indexes; and according to the influence mechanism and the importance degree of each characteristic index on the dumping karst dangerous rocks, a grading rule of qualitative characteristic indexes of the dumping karst dangerous rocks is also set, and the grading rule is shown in table 1.

Step S2 specifically describes:

the step S2 includes sub-steps S2-1 and S2-2, which are explained in detail as follows.

Step S2-1: creating a sample

Through data collection, influence factor measurement and test, the measured data of 12 characteristic indexes of a plurality of dumping type karst dangerous rock engineering cases and the stable category thereof are obtained.

Establishing a sample set D ═ { x, y }, where x is_i＝[x_i1，x_i2，…，x_i12]For inputting the feature vector, the evaluation of the critical rock weight center outside the overturning point and the corresponding 4 qualitative feature indexes of the base rock hardness, the karst development degree and the weathering degreeScore, and critical rock volume weight W_lHorizontal distance a from gravity center of dangerous rock mass to overturning point, characteristic period T of regional earthquake dynamic acceleration response spectrum, regional earthquake dynamic peak acceleration G and tensile strength sigma of dangerous rock mass_tThe dip angle beta of the main control structure surface, the length l of the through section of the main control structure surface and the fracture water pressure Q of the main control structure surface are 8 quantitative characteristic indexes; y is an output target value, and corresponding unstable state and stable state are respectively-1 and 1.

TABLE 1 Scoring rule table for karst dangerous rock qualitative characteristic index

Step S2-3: data normalization

The invention applies the method that the input feature vector of each sample subtracts the mean value of all samples in each dimension, and divides by the standard deviation of all samples in each dimension, and carries out standardization processing on the sample input data, namely z-score standardization:

in the formula, x_i,jAnd x'_i,jRespectively expressed as the actual value and normalized value of the j-th dimension output characteristic of the i-th sample_jAnd sigma_jRespectively representing the mean and standard deviation of j-th dimension features of all samples. Through standardization, the input feature vectors of all samples conform to the standard normal distribution in all dimensions.

Step S3 specifically describes:

the step S3 includes substeps S3-1, S3-2 and S3-3, which are explained in detail below

The IVM binary model (instability and stability) is based on Bayesian statistical learning theory and kernel method, combines assumed density approximation ADF and a sample selection method based on information entropy theory, utilizes ADP recursion approximation to increase posterior distribution after selecting an information vector sample each time, and obtains approximate likelihood distribution, thereby ensuring the traceability and the tractability of the algorithm. The following is a brief description of some key steps of the learning and prediction process of the IVM binary model.

The meaning of the part of the symbols appearing in the formulas of the part is as follows: p represents probability distribution, q represents approximate probability distribution, N represents Gaussian distribution, X and y represent input feature vector set (matrix) and output target vector of training sample, respectively, X_*And y_*Respectively representing input eigenvectors and output targets, mu, of samples to be predicted_*And sigma_*The scores represent the predicted mean and the predicted variance, f represents the set of potential variables, m represents the likelihood surrogate variable, B or beta represents the variance of the noise distribution, mu represents the mean of the Gaussian distribution, K or sigma represents the covariance matrix of the Gaussian distribution, theta represents the hyper-parametric vector of the covariance function, and other symbolic interpretations can be obtained elsewhere in the invention.

Step S3-1: learning of IVM binary models

In the learning process of the IVM binary classification model, two sample index sets I and J are maintained, wherein I is an active set, J is a candidate set, and initially,

j ═ 1, 2,. n, and at any time,

i ═ J ═ 1, 2., N (assuming that d information vectors are filtered out of N training samples), the information vectors are obtained in a continuous, online-like learning manner: first, applying ADF approximation has I information vectors, I_iThe time-of-arrival and likelihood distributions are such that the approximate solution is consistent with the exact solution for the case of gaussian distribution.

Thereafter, the i +1 st information vector is selected as follows

The above formula represents: and selecting a sample index J capable of maximally reducing the posterior distribution information entropy in the current candidate set J, and taking a sample J as an i +1 th information vector.

In general, for a probability distribution, the information entropy H characterizes its degree of uncertainty, while statistical learning naturally wants to be able to obtain a certain (predictive) posterior distribution as much as possible. The above processes are executed in a loop until the selection of d information vectors is completed (I ═ I)_d). At this time, can obtain

In the IVM binary model, the optimal solution of the covariance function hyperparameter θ is just by maximizing the edge likelihood p (y)_IlX_I，：θ) is adaptively obtained. In particular, by taking the negative log-log (p (y)_I|X_I,:Theta)) to convert the maximization problem into the minimization problem, and then the conjugate gradient descent method is used to realize the optimal hyperparameter

Adaptive acquisition of (2).

Step S3-2, prediction of IVM binary model

The learning process realizes that the learning of the original sample set is replaced by the learning of the information vector sample specified by the active set I, the later prediction process is consistent with the method in Bayesian regression, and the corresponding variable is substituted to obtain the IVM prediction posterior distribution

For the binary problem involved in the invention, after the IVM prediction posterior distribution is obtained, the prediction of the class to which the unknown sample belongs can be obtained by the following simple transformation

In the formula, Φ (x) represents a standard normal distribution cumulative probability density function, but other response functions, such as sigmoid logic function, may be used instead.

In the IVM binary model, a covariance matrix (covariance matrix) k, also called a kernel matrix, has the same idea as a kernel function: if the original sample data is mapped to the high-order feature space using the set of basis functions phi (x), the dot product of the original sample input vector is generalized to the dot product of the basis functions by the kernel, i.e., the kernel

K_i,j≡k(x_i,x_j)＝cov(f(x_i),f(x_j))＝φ(x_i)^T∑φ (7)

The mapping relationship is nonlinear, the dimension of the feature space can be extremely high or even infinite, and phi (x) in the basis function can be infinite, so that the introduction of the covariance function (kernel method) enables the IVM to have strong nonlinear processing capability.

Step S3-3, IVM classification model feasibility test

In order to ensure that the performance of the IVM classification model for stable and comprehensive judgment of the optimal dumping type karst dangerous rock meets the requirements of learning ability and generalization ability, feasibility test is carried out on the result of the test sample output by the optimal IVM classification model. Specifically, the inspection index is the prediction accuracy of the test sample, namely the actual stability and the prediction stability of the test sample are utilized for checking, if the prediction accuracy is 100%, the performance of the established optimal IVM classification model is considered to meet the requirements, and the method has feasibility for the prediction of the karst dangerous rock stability; otherwise, retraining and modeling.

In step S3, the present invention randomly divides the training sample library into 10(K ═ 10) parts by using a typical K-fold cross validation (K-CV) method, sequentially selects 9 parts of the training sample library as training samples, and sets initial parameters such as covariance function type, noise distribution, information vector number, etc. of the IVM model for classification of stability of dump karst critical rocks, applies the IVM model to perform learning and prediction, and evaluates the learning and generalization performance of the model by using K times of calculation of average learning accuracy and prediction accuracy.

In step S3, the present invention makes an adjustment according to the cross-validation result of the IVM classification model for the comprehensive determination of stability of the dump karst crisis rock. If the performance of the cross-validated IVM classification model does not meet the requirements, the adjustment can be made by two aspects: on one hand, adjusting the initial parameter setting of the IVM classification model according to the cross validation learning and prediction results and the action effect of each initial parameter; on the other hand, considering that a plurality of dumping type karst dangerous rock examples are originated from different projects, and a certain difference may exist in the determination of the stability characteristic index of the dumping type karst dangerous rock, therefore, necessary screening needs to be carried out on training samples according to the cross validation learning and prediction results, samples incompatible with other more samples are removed, and the samples have a plurality of learning or prediction errors in the cross validation circulation. And after adjustment and cross validation training are carried out again, the process is repeatedly executed, and finally the dumping type karst dangerous rock stable comprehensive judgment IVM classification model with strong learning and generalization performance is obtained.

Step S4 specifically describes:

in step S4, for predicting the stability of the new dangerous rock mass, it is necessary to calculate an IVM classification model input feature vector for comprehensive judgment of stability of the dumping karst dangerous rock, that is, a feature vector consisting of clean and optimized characteristic indexes of stability of the dumping karst dangerous rock, and the feature vector may be obtained in step S2.

And (3) listing newly-added dangerous rock sample into a model preparation sample set P, and if the number of the preparation sample set P is increased to a threshold value, establishing a prediction extrapolation model, namely inputting the sample in the preparation sample set P into a sample set D to form a new model sample set D.

The number threshold value of the preparation sample set P is set to be based on the number of the model sample set D, the relative value of the two is used as an evaluation index, namely the update specific gravity zeta of the number m of the preparation sample set P and the number n of the model sample set D, if zeta is larger than 0.2, a new sample is considered to be added, the model prediction performance has a higher promotion space, and the IVM classification model for the stable and comprehensive judgment of the dumping type karst dangerous rock is reestablished; otherwise, the new dangerous rock sample in the preparation sample set P is retained.

Step S5 specifically describes:

in step S5, the uncertainty of the decision result is quantitatively evaluated according to the predicted classification probability outputted by the IVM classification model for the comprehensive decision of stability of the dump karst dangerous rock.

Specifically, when the discrimination result is stable, the discrimination boundary is far and near according to the classification probability, and the uncertainty level of the "stable" discrimination result is divided into three levels, namely a low level, a medium level and a high level; similarly, when the determination result is "unstability", the classification is similar, which is detailed in table 2.

TABLE 2 uncertainty level division Standard of the discrimination results

As can be seen from Table 2, the uncertainty of the discrimination result can be known from the value range of the classification probability corresponding to each discrimination result of the IVM classification model, and the discrimination result y of the sample can be judged according to the uncertainty_*Whether the accuracy needs to be rechecked; when the uncertainty level belongs to a low or medium level, the uncertainty degree of the discrimination result is considered to be low, and rechecking can not be carried out; when the uncertainty level belongs to a high level, the judgment result has high uncertainty degree, and the judgment result needs to be checked again. Therefore, the uncertainty level provides a scientific basis which can be conveniently obtained for the credibility decision of the dangerous rock stability judgment result.

Step S6 specifically describes:

in step S6, when the uncertainty level of the model discrimination result of the new dangerous rock mass in step S5 is a high level, the accuracy of the discrimination result should be rechecked by using a rigid body limit balance method. If the project and the like to which the new dangerous rock mass belongs are high in grade, the collapse of the new dangerous rock mass can cause serious safety accidents of personnel and equipment loss, and therefore the judgment result needs to be rechecked.

The invention adopts a limit balance mechanics analysis method to carry out recheck on the dumping karst dangerous rock, and the worst load combination is as follows: dead weight, fracture water pressure (rainstorm working condition) and earthquake force.

The invention discloses a solving and calculating method for solving a stability coefficient of a dump karst dangerous rock by using a rigid body limit balance method, which comprises the following steps:

(1) when the gravity center of the dumping dangerous rock mass is at the inner side of the overturning point, the stability coefficient K calculation method comprises the following steps:

(2) when the gravity center of the toppled dangerous rock mass is outside the overturning point, the stability coefficient K calculation method comprises the following steps:

wherein W is dead weight of the dangerous rock mass, P is horizontal seismic force, Q is fracture water pressure acting on the karst dangerous rock main control structure surface, f_lkIs a standard value of the tensile strength of the dangerous rock mass, f_0kIs a standard value of tensile strength between the dangerous rock mass and the base, and f is a standard value of tensile strength between the dangerous rock mass and the base when the base is a rock mass_0k＝f_1kWhen the base is soft rock, acquiring an actual standard value of tensile strength; beta is the angle of inclination of the main control structural plane, gamma_wThe water is the severe of the fracture water, H is the height of the main control structural surface, e is the vertical height of the through section of the main control structural surface, l is the length of the karst dangerous rock mass along the trend direction of the scarp, l is the depth of the rock mass along the trend direction of the scarp_bThe distance from the top end of the main control structure surface at the bottom of the dangerous rock body to the overturning point, and a is the horizontal distance from the center of gravity of the dangerous rock body to the overturning pointK is the calculated stability factor value of the dumped karst crisis, see figure 1.

The stability of the poured karst dangerous rock is evaluated by applying the stability coefficient K. Referring to domestic and foreign documents and past engineering experience, when K is less than or equal to 1, the dumping dangerous rock is considered to be in a destabilizing state, when K is less than 1 and less than 1.35, the dumping dangerous rock is considered to be in a low stable state, and when K is more than or equal to 1.35, the dumping dangerous rock is considered to be in a high stable state.

The invention provides an information vector machine device for comprehensively judging stability of a dumping karst dangerous rock, which comprises:

an input module: the device is used for receiving the collected dumping karst dangerous rock example original data and obtaining an initial model original sample set;

a processing module: the IVM classification model sample set feature vector is used for analyzing, optimizing, extracting and integrating the initial original sample set to obtain the pouring type karst dangerous rock stability comprehensive discrimination IVM classification model sample set feature vector;

IVM classification model training module: the method comprises the steps of constructing a sample set of the IVM classification model according to a feature vector of the stability of the dumping karst dangerous rock, training and testing the sample set, and obtaining the IVM classification model with good performance for comprehensively judging the stability of the dumping karst dangerous rock according to the accuracy of the training and testing;

the IVM classification model prediction module: the method comprises the steps of inputting a sample of a new dangerous rock body to be predicted into an IVM classification model for comprehensive judgment of stability of the dumping karst dangerous rock so as to judge the stability category of the new dangerous rock body;

a rechecking module: the system is used for rechecking the model discrimination result according to the uncertainty of the discrimination result of the stability of the new dangerous rock mass output by the IVM classification model and the engineering grade of the new dangerous rock mass;

an output module: and the method is used for outputting the model discrimination result and the rechecking result of the new dangerous rock mass for the user to check.

Illustratively, the processing module includes:

and the preprocessing unit is used for eliminating or perfecting the data with missing, invalid and wrong formats in the original data set, acquiring more complete, effective and reasonable data and improving the quality of the original data.

The extraction unit is used for extracting and establishing a characteristic vector sample of the stability of each dumping type karst dangerous rock from the processed data set according to the characteristic index of the stability of the dumping type karst dangerous rock;

and the standardization unit is used for avoiding adverse effects on the machine learning model judgment accuracy caused by overlarge numerical value difference of different characteristic parameters or overlarge discreteness of the same characteristic parameter, and carrying out standardization processing on the input characteristic vectors of all the established samples. I.e. z-score normalization:

Illustratively, the preprocessing unit includes:

the effectiveness calculation subunit is used for carrying out effectiveness retrieval on the collected original dumping type karst dangerous rock data so as to determine the range of the original dumping type karst dangerous rock data, remove unnecessary fields, fill missing contents and re-value operation cleaning data;

the logic calculation subunit is used for performing logic retrieval on the collected original dumping karst dangerous rock data so as to remove weight, remove unreasonable values and correct contradictory content operation cleaning data;

and the necessity calculating subunit is used for performing necessity retrieval on the collected original dumping karst crisis data so as to delete unnecessary redundant data and operate cleaning data.

Illustratively, the extraction unit includes:

the similarity calculation subunit is used for comparing and calculating the preprocessed sample set and the characteristic indexes of the stability of the dump karst dangerous rock selected by the invention to obtain the similarity between all data information and the characteristic indexes of different samples in the data set, eliminating data with low similarity according to the calculated data similarity, and establishing a first characteristic vector data set for the sample with high similarity;

and the eigenvector reprogramming subunit is used for matching the coding information calculated according to each data in the first eigenvector set with the selected calibration eigenvector, reordering the eigenvectors and establishing a second eigenvector set, namely the final eigenvector group.

Illustratively, the IVM classification model training module comprises:

the model training and testing unit is used for training and testing the model sample set and establishing an optimal dumping type IVM classification model for comprehensive judgment of karst dangerous rock stability;

and the model checking unit is used for carrying out feasibility checking on the optimal IVM classification model, judging whether the model performance meets the requirements or not, and mainly checking the prediction accuracy of the test sample according to the checking index.

Illustratively, the model training and testing unit includes:

the model initial training subunit is used for inputting the established training sample into the model and training the IVM classification model for the comprehensive judgment of the stability of the dumping karst dangerous rocks;

and the model training and adjusting subunit is used for readjusting the parameters of the model and the sample library according to the accuracy of the cross validation strategy training and prediction so as to obtain the optimal parameters of the IVM classification model and the training set and test set division of the sample library.

Illustratively, the IVM classification model prediction module comprises:

the characteristic vector operator unit is used for inputting data of a sample to be predicted into the processing module so as to obtain characteristic vectors required by the IVM classification model for stable and comprehensive judgment of the dumping karst dangerous rock;

the prediction operator unit is used for inputting the characteristic vector into an IVM classification model for comprehensive judgment of stability of the dumping karst dangerous rocks to obtain the classification probability of the new dangerous rock sample;

the uncertainty evaluation operator unit is used for carrying out uncertainty grade division on the discrimination result according to the classification probability of the discrimination result of the new sample, and listing the new dangerous rock sample with uncertainty belonging to a high grade into a first rechecking group;

and the extrapolation model evaluation operator unit is used for carrying out model prediction performance improvement space evaluation on the new dangerous rock mass and judging whether the existing prediction model sample library needs to be updated or not according to the update proportion between the number of the new samples and the number of the IVM classification model sample libraries.

Illustratively, the review module includes:

a prediction reinspection unit: the system comprises a sample set, a second rechecking group and a third rechecking group, wherein the sample set is used for determining samples with high engineering levels in the sample set according to engineering level labels to which input new samples belong and listing the samples in the second rechecking group;

the merging processing unit is used for carrying out similarity calculation on the first and second reinspection groups to eliminate the duplicate samples and establish a final reinspection group;

and the final rechecking unit is used for calculating the actual value of the stability coefficient of the dangerous rock mass by a rigid body limit balancing method according to the rechecking group data and comparing the actual value with the judgment result output by the IVM classification model prediction module.

Compared with the prior art, the invention has the beneficial effects that:

(1) aiming at the dumping karst dangerous rock in the karst area, the invention perfects the characteristic indexes influencing the dumping karst dangerous rock, and specifically comprises the following steps: whether the gravity center of the dangerous rock mass is outside an overturning point, the hardness of the base rock, the development degree of the karst and the weathering degree are 4 qualitative characteristic indexes, and the volume weight W of the dangerous rock mass_lHorizontal distance a from gravity center of dangerous rock mass to overturning point, characteristic period T of regional earthquake dynamic acceleration response spectrum, regional earthquake dynamic peak acceleration G and tensile strength sigma of dangerous rock mass_t8 quantitative characteristic indexes including the inclination angle beta of the main control structure surface, the length l of the through section of the main control structure surface and the fracture water pressure Q of the main control structure surface are set up, a multi-index dumping type karst dangerous rock stability comprehensive judgment method is established, the characteristic indexes complement each other, and the problem that the common dangerous rock indexes cannot well reflect the dumping type karst dangerous rock stability generation mechanism is solvedThe accuracy of the judgment result is low;

(2) according to the invention, a large number of dumping type karst dangerous rock examples at home and abroad are widely collected, a rich karst dangerous rock database is established, and meanwhile, the IVM machine learning method with superior statistical model prediction performance is used for training and predicting, so that the complex work of derivation of a karst dangerous rock stability prediction empirical formula and the over-idealization of a calculation model are effectively avoided, and the reliability of the judgment result of the invention is ensured;

(3) the IVM is based on Bayesian statistical learning theory and a kernel method, combines assumed density approximation ADF and a sample selection method based on an information entropy theory, has the advantages of complete and strict theory, self-adaptive acquisition of hyper-parameters, simple and efficient implementation process, strong adaptability to complex nonlinear problems, probability significance of prediction output and the like, and has obvious advantages compared with the support vector machine and the artificial neural network which are widely applied at present;

(4) applying a cross validation strategy to train and evaluate an information vector machine classification model for comprehensive judgment of stability of the dumping type karst dangerous rock, adjusting an initial parameter value according to a training result, dividing a sample library (a training sample and a test sample), and adjusting the training sample library and the test sample library to be in an optimal state to ensure that an IVM model obtained finally is optimal; selectively extrapolating to construct a new IVM classification model according to the number of the newly added karst dangerous rocks, and further improving the accuracy of the stable comprehensive judgment of the dump karst dangerous rocks;

(5) the method not only provides an IVM classification model for the stability and comprehensive judgment of the dumping type karst dangerous rock, but also provides an uncertainty degree of the judgment result of the quantitative evaluation model based on the classification probability of the judgment result of the IVM classification model on the basis, and rechecks the samples with higher uncertainty level and the samples with higher engineering level.

(6) The IVM classification model for the comprehensive judgment of the stability of the dump karst dangerous rock has the advantages of strong learning capacity of small samples, self-adaptive acquisition of the optimal parameters of the prediction model, good generalization capacity of the prediction model and the like, overcomes the defects that the optimal network topology and the hyper-parameters are not easy to determine in the conventional widely applied artificial neural network method, and has strong applicability to the nonlinear mapping prediction problem between the characteristic index of the karst dangerous rock and the stability of the karst dangerous rock;

(7) the model is applied to the comprehensive judgment of the stability of the dumping karst dangerous rock, engineering personnel are not required to deeply know the collapse and instability mechanism of the dumping karst dangerous rock, the requirement on professional knowledge level and engineering experience is not high, complex and time-consuming mechanical calculation or numerical simulation analysis is not required, and the stable state of the dangerous rock can be obtained only by inputting the obtained evaluation characteristic indexes corresponding to the karst dangerous rock sample to be judged into the IVM classification model for the comprehensive judgment of the stability of the dumping karst dangerous rock, so that the method has the advantages of economy, practicability, simplicity and high efficiency.

Drawings

FIG. 1 is a schematic diagram of the dump type stress of dangerous rock provided by the invention;

FIG. 2 is a flowchart of an IVM classification method for comprehensive determination of stability of a dump karst crisis according to an embodiment of the present invention;

fig. 3 is a schematic diagram of a cloud server apparatus according to an embodiment of the present invention;

fig. 4 is a schematic structural diagram of an IVM classification device for comprehensive determination of stability of a dump karst crisis according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of an input device according to an embodiment of the present invention;

FIG. 6 is a schematic diagram of a processing apparatus according to an embodiment of the present invention;

FIG. 7 is a schematic diagram of a preprocessing sub-apparatus according to an embodiment of the present invention;

FIG. 8 is a schematic diagram of an extraction sub-apparatus according to an embodiment of the present invention;

FIG. 9 is a schematic diagram of an IVM classification model training apparatus according to an embodiment of the present invention;

FIG. 10 is a schematic diagram of a model training and testing sub-apparatus according to an embodiment of the present invention;

FIG. 11 is a schematic diagram of an IVM classification model prediction apparatus according to an embodiment of the present invention;

fig. 12 is a schematic diagram of a review apparatus according to an embodiment of the present invention.

Detailed Description

The following further describes embodiments of the present invention with reference to the drawings and examples. It is noted that the drawings show only some of the relevant aspects of the invention and not all of the results. And the specific examples are to be construed as merely illustrative, and not limitative of the remainder of the disclosure in any way whatsoever.

Example 1

Referring to fig. 2, an example of the present invention provides an IVM classification method for comprehensive determination of stability of a dump karst crisis. The method is suitable for comprehensively judging the dumping type dangerous rock stability in the karst region, and specifically comprises the following steps:

in step S1, referring to domestic and foreign documents and engineering experiences, 4 qualitative characteristic indexes of whether the gravity center of the dangerous rock mass is outside the overturning point, the hardness of the base rock, the development degree of the karst and the weathering degree are selected, and the volume weight W of the dangerous rock mass_lHorizontal distance a from gravity center of dangerous rock mass to overturning point, characteristic period T of regional earthquake dynamic acceleration response spectrum, regional earthquake dynamic peak acceleration G and tensile strength sigma of dangerous rock mass_tThe dip angle beta of the main control structure surface, the length l of the through section of the main control structure surface and the fracture water pressure Q of the main control structure surface are 8 quantitative characteristic indexes, and the qualitative characteristic indexes are converted into grading grades according to a grading rule table 1.

In step S2, a dump karst crisis characteristic index is selected in accordance with step S1, and 107 representative dump karst crisis cases at home and abroad are widely collected.

In step S2-1, the characteristic indexes and stability of the dumping karst dangerous rock are respectively used as the input characteristic vector and the output target of the sample, and the method can be used for widely collecting representative examples of the dumping karst dangerous rock at home and abroad and establishing a rich training sample library. Sample data of dumping karst dangerous rockIn the format of (x)_i，y_i). Wherein x_i＝[x_i1，x_i2，…，x_i12]For inputting the feature vector, the elements are sequentially the grade corresponding to 4 qualitative feature indexes of whether the weight center of the dangerous rock is outside the overturning point, the hardness of the base rock, the development degree of the karst and the weathering degree, and the volume weight W of the dangerous rock_lHorizontal distance a from gravity center of dangerous rock mass to overturning point, characteristic period T of regional earthquake dynamic acceleration response spectrum, regional earthquake dynamic peak acceleration G and tensile strength sigma of dangerous rock mass_tThe dip angle beta of the main control structure surface, the length l of the through section of the main control structure surface and the fracture water pressure Q of the main control structure surface are 8 quantitative characteristic indexes; y is an output target value, and corresponding unstable state and stable state are respectively-1 and 1.

In step S2-3, the present invention applies a method of subtracting the mean of all samples in each dimension from the input feature vector of each sample, and dividing by the standard deviation of all samples in each dimension, to perform normalization on the sample input data, i.e. z-score normalization:

in the formula, x_i,jAnd x'_i,jRespectively expressed as the actual value and normalized value of the j-th dimension output characteristic of the i-th sample_jAnd sigma_jRespectively representing the mean and standard deviation of j-th dimension features of all samples. Through the normalization, the feature vectors of all sample inputs will meet the standard normal score, x ', in each dimension'_：，jN (0, 1), the model sample set is shown in Table 3.

In step 3, the embodiment of the invention adopts an IVM classification model for cross validation training pouring type karst dangerous rock stability comprehensive judgment, specifically adopts a typical K (K takes 10) times cross validation (K-fold cross validation, K-CV) method to divide a sample library into 10 parts at random, sequentially selects 9 parts as training samples, and takes 1 part as a test sample; referring to the explanation and the prior experience of use of the classification problem in the IVM toolbox, the initial parameters of the IVM classification model for the preliminary dump karst crisis stable comprehensive judgment include that 'probit' distribution noise is adopted, the information vector number d is set to be 60 (according to the 10-fold cross validation strategy, the training sample number N is 9/10 of all sample numbers, namely 107 × 9/10, about 96), and the Radial substrate variance Function 'RBF' (Radial Basis Function) is selected.

Its hyper-parameter theta ═ l, sigma_f，σ_n]Self-adaptive acquisition is carried out in learning, and other initial parameters are set by default; then, the IVM model is applied to learn and predict training and testing samples, and the learning and generalization performance of the model is evaluated by using the average learning accuracy and the testing accuracy calculated for 10 times.

After repeated adjustment and repeated cross validation training, the process is repeatedly executed to finally obtain the IVM classification model for stable and comprehensive judgment of the dumping type karst dangerous rocks with strong learning and generalization performance, and the initial parameters of the IVM classification model for stable and comprehensive judgment of the dumping type karst dangerous rocks are set as follows: the noise is distributed by adopting 'probit', the information vector number d is 60, and the covariance function adopts an RBF covariance function.

In order to ensure that the performance of the IVM classification model for the stable and comprehensive judgment of the optimal dumping type karst dangerous rock meets the requirements of learning ability and generalization ability, the embodiment of the invention carries out feasibility test on the result of the test sample output by the optimal IVM classification model. The actual stability and the prediction stability of the test sample are used for checking, and the prediction result table of the model test sample in the table 4 shows that the prediction result of the IVM classification model test sample has no prediction error condition, the performance of the established optimal IVM classification model meets the requirements, and the method has feasibility for comprehensive judgment of stability of the dumping karst dangerous rocks.

TABLE 3 IVM classification model sample set for comprehensive judgment of stability of dumping type karst dangerous rocks

TABLE 4 test sample discrimination results based on the present invention

Numbering	Classification probability (%)	Actual state	Determining the state
				2	0.89	1	1
12	0.21	-1	-1
				15	0.37	-1	-1
31	0.87	1	1
				38	0.91	1	1
47	0.16	-1	-1
				61	0.39	-1	-1
74	0.94	1	1
				85	0.85	1	1
92	0.39	-1	-1
				106	0.95	1	1

In step S4, 10 new dump karst dangerous rock sample examples are collected, which are from karst dangerous rock protection engineering in Guilin city of Zhuang autonomous region of Guangxi province, and feature vector sample set D of dump karst dangerous rock is obtained after the collection is processed in step S2, and is input to the dump karst dangerous rock sample set DIn the IVM classification model for stable comprehensive judgment, the stability probability and the stability judgment result are output, namely the classification probability of each sample in the example

And the corresponding stability determination result

See table 5.

In step S5, the uncertainty of the new dangerous rock mass is quantitatively determined according to the new dangerous rock mass determination result obtained based on the IVM classification model of the present invention and the corresponding classification probability, which is detailed in table 5.

TABLE 5 New dangerous rock mass discrimination results and uncertainty level of the discrimination results based on the present invention

As can be seen from the results of the discrimination of the uncertain size of the 10 new dangerous rock sample discrimination results in table 5, the sample discrimination results with numbers 3 and 4 in the example samples obtained based on the IVM classification model of the present invention have high uncertainty level evaluation, and the other sample discrimination results have low uncertainty levels.

In step S6, the embodiment of the present invention rechecks the accuracy of the discrimination result for 2 samples with numbers 3 and 4 having high uncertainty levels of the discrimination result for the new dangerous rock sample and 2 samples with numbers 8 and 10 having higher engineering grades.

According to the combination mode of self weight, fracture water pressure (rainstorm working condition) and seismic load, the actual stable states of the 4 samples are calculated through a rigid body limit balance method, the accuracy of the judgment result of the IVM classification model is rechecked, and the comparison result is shown in a table 6.

TABLE 6 accuracy rechecking result of new dangerous rock mass discrimination result based on the present invention

As can be seen from tables 5 and 6, the information vector machine classification model for the comprehensive judgment of the stability of the dump karst dangerous rock, which is established by the embodiment of the invention, has better performance. Specifically, in 4 samples in the recheck sample set, the actual stable state of only 1 sample does not accord with the discrimination stable state of the IVM classification model, and the uncertainty level of the discrimination result of the sample is high. From the above results, the IVM classification model for the stable and comprehensive judgment of the dump karst dangerous rock has high prediction certainty, can meet the requirements of engineering application, and can quantitatively evaluate the uncertainty of the judgment result according to the classification probability corresponding to the model output judgment result, namely, the closer the classification probability is to the judgment limit, the greater the uncertainty of the judgment result.

Example 2

Referring to fig. 3, the present invention is directed to a cloud server device 100 comprising one or more processors 100-1, one or more storage devices 100-2, an input device 100-3, and an output device 100-4, which are interconnected via a bus system 100-5 and/or other type of connection mechanism. It should be noted that the components and structure of the cloud server apparatus 100 shown in fig. 3 are merely exemplary and not limiting, and the cloud server apparatus may have other components and structures as needed.

The processor 100-1 may be a Central Processing Unit (CPU) or other form of processing unit having data processing capabilities and/or instruction execution capabilities, and may control other components in the cloud server apparatus 100 to perform desired functions.

Illustratively, the processor 100-1 may perform the steps of preprocessing the original sample set, extracting the stability characteristic index of the poured karst dangerous rock sample, training the IVM classification model, determining the stability of the new dangerous rock mass, and rechecking the uncertainty of the determination result in the method of the present invention (S2-S6).

Storage 100-2 may include one or more computer program products that may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. The volatile memory may include, for example, Random Access Memory (RAM), cache memory (cache), and/or the like. The non-volatile memory may include, for example, Read Only Memory (ROM), hard disk, flash memory, etc. On which one or more computer program instructions may be stored that may be executed by processor 100-1 to implement the computer functions (implemented by the processor) of the embodiments of the invention described below and/or other desired functions. Various applications and various data, such as various data used and/or generated by the applications, may also be stored in the computer-readable storage medium.

The input device 100-3 may be a device for receiving commands input by a user and collecting data, and the input mode thereof adopts a combination of wireless and wired transmission.

The output device 100-4 may output various information such as text data, images or sounds to the outside (e.g., a user), and may include one or more of a display, a speaker, etc., and the application of the present invention is mainly directed to text data output.

Fig. 4 is a schematic diagram of an IVM device 200 for comprehensive determination of karst crisis stability in a dumping type karst area, according to an embodiment of the present invention, for determining the karst area crisis stability, the device being used for executing the method according to the above embodiment of the present invention, wherein the IVM device 200 includes:

the input module 200-1 is used for receiving the collected dumping type karst dangerous rock example original data and obtaining an initial model original sample set; the processing module 200-2 may be implemented by the processor 100-1 in the cloud server apparatus 100 shown in fig. 3 running the program instructions stored in the storage apparatus 100-2, and may execute a corresponding part of step S2 of the information vector machine classification method for comprehensive determination of stability of dump karst crisis provided in the embodiment of the present invention;

and the processing module 200-2 is used for analyzing, optimizing, extracting and integrating the initial original sample set to obtain the characteristic vector of the IVM classification model sample set for comprehensive judgment of stability of the dump karst dangerous rock. The processing module 200-2 may be implemented by the processor 100-1 in the cloud server apparatus 100 shown in fig. 3 running the program instructions stored in the storage apparatus 100-2, and may execute a corresponding part of step S2 of the information vector machine classification method for comprehensive determination of stability of dump karst crisis provided in the embodiment of the present invention;

and the IVM classification model training module 200-3 is used for constructing a sample set of the IVM classification model according to the characteristic vector of the stability of the dumping karst dangerous rock, training and testing the sample set, and obtaining the IVM classification model with good performance for comprehensively judging the stability of the dumping karst dangerous rock according to the accuracy of training and prediction. The IVM classification model training module 200-3 may be implemented by the processor 100-1 in the cloud server device 100 shown in fig. 3 running the program instructions stored in the storage device 100-2, and may execute step S3 of the information vector machine classification method for comprehensive determination of stability of dump karst crisis provided in the embodiment of the present invention;

the IVM classification model prediction module 200-4 is used for inputting the new dangerous rock sample to be predicted into the IVM classification model for the dumping type karst dangerous rock stability comprehensive judgment so as to predict the prediction mean value and the prediction variance of the new dangerous rock stability. The IVM classification model prediction module 200-4 may be implemented by the processor 100-1 in the cloud server device 100 shown in fig. 3 running the program instructions stored in the storage device 100-2, and may execute steps S4 and S5 of the information vector machine classification method for comprehensive determination of stability of dump karst rocks in the embodiment of the present invention;

and the rechecking module 200-5 is used for rechecking the judgment result of the sample with larger uncertainty of the judgment result of the new dangerous rock mass output by the IVM classification model or higher engineering grade of the new dangerous rock mass. The rechecking module 200-5 may be implemented by the processor 100-1 in the cloud server apparatus 100 shown in fig. 3 running the program instructions stored in the storage apparatus 100-2, and may execute step S6 of the information vector machine classification method for comprehensive determination of stability of dump karst crisis provided in the embodiment of the present invention;

output module 200-6: and the method is used for outputting the model discrimination result and the rechecking result of the new dangerous rock mass for the user to check.

Illustratively, referring to fig. 5, the input module 200-1 includes:

the input receiving unit 200-1-1 is used for receiving collected dumping type karst dangerous rock example original data, wherein the data comprises karst characteristics of dangerous rocks, rock mechanical parameters, lithology of rocks, hydrological conditions and the like;

the fracture water pressure acquisition unit 2-1-2 is used for dynamically acquiring the fracture water pressure of the main control structure surface of the sample according to the collected example sample, and the reasons are as follows: the fracture water pressure is a dynamic parameter, and according to the characteristics of non-continuity, non-uniform distribution, dynamic change and the like of the fracture water pressure in a rock body, the fracture water pressure needs to be acquired in real time in order to improve the accuracy of dangerous rock stability judgment.

Illustratively, referring to fig. 6, the processing module 200-2 includes:

the preprocessing unit 200-2-1 is used for removing or perfecting data with missing, invalid and wrong formats in the original data set, acquiring more complete, effective and reasonable data and improving the quality of the original data;

the extraction unit 200-2-2 is used for extracting and establishing a feature vector sample of the stability of the dump type karst dangerous rock from the processed data set according to the characteristic index of the stability of the dump type dangerous rock;

and the standardization unit 200-2-3 is used for avoiding adverse effects on the machine learning model judgment accuracy caused by overlarge differences of different characteristic parameter values or overlarge discreteness of the same characteristic parameter, and carrying out standardization processing on the input characteristic vectors of all the samples. I.e. z-score normalization:

Illustratively, referring to FIG. 7, the pre-processing unit 200-2-1-1 includes:

the effectiveness calculating subunit 200-2-1-1 is used for performing effectiveness retrieval on the collected original dumping type karst dangerous rock data to determine the range of the data, remove unnecessary fields, fill missing contents and re-value operation cleaning data;

the logical calculation subunit 200-2-1-2 is used for performing logical retrieval on the collected original dumping karst dangerous rock data so as to remove the weight, remove the unreasonable value and correct the contradictory content operation cleaning data;

a necessity calculation subunit 200-2-1-3 for performing a necessity search on the collected raw dump karst crisis data to delete unnecessary surplus data and manipulate the cleaning data.

Illustratively, the extraction unit 200-2-2 is specifically configured to compare each sample raw data set with a selected characteristic index of karst dangerous rock stability based on a distance measurement principle, that is, based on a characteristic variable, and if the similarity exceeds a threshold, perform data extraction to the characteristic variable, and establish a characteristic variable group of a certain sample as a characteristic vector.

Illustratively, referring to fig. 8, the extraction unit 200-2-2 includes:

the similarity calculation operator unit 200-2-2-1 is used for calculating the similarity according to the calibrated feature vector and the original data set, if the similarity s exceeds a set threshold k, the similarity is listed as a first feature vector group, and the rest are residual data groups, and a Pearson error algorithm is adopted:

and the eigenvector reprogramming subunit 200-2-2-2 is used for matching the coding information calculated according to each data in the first eigenvector set with the selected calibration eigenvector, reordering the eigenvectors, and establishing a second eigenvector set, namely the final eigenvector group.

Illustratively, referring to fig. 9, the IVM classification model training module 200-3 includes:

the model training and testing unit 200-3-1 is used for performing model training and testing by utilizing the model sample set so as to establish an optimal dumping type IVM classification model for karst dangerous rock stability comprehensive judgment;

the model checking unit 200-3-2 is used for carrying out feasibility checking on the optimal IVM classification model, and checking indexes of model performance mainly comprise prediction accuracy and generalization capability;

illustratively, referring to FIG. 10, the model training and testing unit 200-3-1 includes:

the model initial training subunit 200-3-1-1 is used for inputting the established training sample into the model and training the dumping type IVM classification model for karst dangerous rock stability comprehensive judgment;

and the model training and adjusting subunit 200-3-1-2 is used for readjusting the parameters of the model and the sample base according to the accuracy of cross validation strategy training and model prediction so as to obtain the optimal parameters of the IVM classification model and the optimal division of the training set and the test set of the sample base.

Illustratively, referring to fig. 11, the IVM classification model prediction module 200-4 includes:

the characteristic vector operator unit 200-4-1 is used for inputting data of a sample to be predicted into the processing module so as to obtain characteristic vectors required by the IVM classification model for the stable and comprehensive judgment of the dumping karst dangerous rocks;

the prediction operator unit 200-4-2 is used for inputting the feature vectors into an IVM classification model for comprehensive judgment of stability of the dump karst dangerous rocks to obtain the classification probability of the new dangerous rock sample;

uncertainty evaluation operator unit 200-4-3: the system is used for carrying out uncertainty grade division on the discrimination result according to the classification probability of the discrimination result of the new sample, and listing the new dangerous rock sample with uncertainty belonging to a high grade into a first rechecking group;

and the extrapolation model evaluation operator unit 200-4-4 is used for performing model prediction performance improvement space evaluation on the new dangerous rock mass and judging whether the existing prediction model sample library needs to be updated or not according to the number of new samples and the update proportion between the number of the IVM classification model sample libraries.

Illustratively, referring to fig. 12, the review module 200-5 includes:

the prediction rechecking unit 200-5-1 is used for determining samples with high engineering levels in the sample set according to the engineering level labels to which the input new samples belong and listing the samples into a second rechecking group;

the merging processing unit 200-5-2 is used for carrying out similarity calculation on the first and second reinspection groups to eliminate the duplicate samples and establish a final reinspection group;

and the final rechecking unit 200-5-3 is used for calculating the actual value and the stable state of the stability coefficient of the dangerous rock mass by a rigid body limit balancing method according to the rechecking group data and comparing the actual value and the stable state with the judgment result output by the IVM classification model prediction module 200-4.

For example, the apparatuses mentioned in the present invention may be implemented by the processor 100-1 in the cloud server apparatus 100 in fig. 3 running program instructions stored in the storage apparatus 100-2, and modules and units not described are not meant to refer to this apparatus processing.

Exemplarily, in the device provided by the present invention, only the input module 200-1, the IVM classification model prediction module 200-4 and the review module 200-5 relate to the input device 100-3 and the output device 100-4 in the cloud server device 100 in fig. 3, and the input and output of the other devices in the information vector machine classification device for the stable and comprehensive judgment of the dump karst dangerous rock are all performed inside the cloud server 100 device, that is, only the input, prediction and review modules perform the interaction among users, so that the inconvenience problem brought to the users by frequent interaction is reduced, and the prediction device provided by the present invention is closer to intellectualization.

The IVM classification device of stable comprehensive judgement of formula karst dangerous rock is emptyd to this application embodiment. Firstly, acquiring dangerous rock original data by using a data structure module, carrying out preprocessing cleaning operation on the original data according to the integrity, effectiveness, consistency and the like of the data, carrying out one-to-one corresponding calculation on a data set and a dumping karst dangerous rock characteristic variable label, and establishing a model characteristic vector by using original sample data according to the calculation similarity; then, training and testing the IVM classification model by adopting a cross validation algorithm, and checking the learning ability and generalization performance of the IVM classification model to obtain an optimal model; further, carrying out quantization operation on the new dangerous rock sample, inputting the characteristic vector of the stability of the new dangerous rock sample into a discrimination model to obtain a stability discrimination result, and quantitatively evaluating the uncertainty of the discrimination result according to the classification probability and the discrimination type; and finally, rechecking the judgment result according to the uncertainty degree and the engineering grade level. Therefore, the comprehensive distinguishing process of the stability of the dumping type karst dangerous rocks is realized, a model is established by a simple and efficient method without requiring higher professional level requirements, the outstanding sample is rechecked by combining a traditional high-reliability calculation method, and the stability of the dumping type karst dangerous rocks is verified again.

It should be understood that portions of the present application may be implemented in hardware, software, firmware, or a combination thereof.

It will be understood by those skilled in the art that all or part of the steps carried by the method for implementing the above embodiments may be implemented by hardware related to instructions of a program, which may be stored in a computer readable storage medium, and when the program is executed, the program includes one or a combination of the steps of the method embodiments.

In addition, each unit/sub-unit in the embodiments of the present application may be integrated into one module, or each sub-unit may exist alone physically, or two or more units/sub-units may be integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. The integrated module, if implemented in the form of a software functional module and sold or used as a stand-alone product, may also be stored in a computer readable storage medium.

It is noted that the above examples are disclosed for the purpose of aiding a further understanding of the invention, but those skilled in the art will appreciate that: numerous obvious variations, changes and substitutions may be made without departing from the scope of the present invention. Therefore, the invention is not limited to the disclosure of the examples, but rather the scope of the invention is defined by the appended claims.

Claims

1. The information vector machine method for comprehensively judging the stability of the dumping karst dangerous rock is characterized by comprising the following steps of:

step S1: selecting characteristic indexes which obviously influence the stability of the poured karst dangerous rock, and specifically comprising the following steps: whether the gravity center of the dangerous rock mass is outside an overturning point, the hardness of the base rock, the development degree of the karst and the weathering degree are 4 qualitative characteristic indexes, and the volume weight W of the dangerous rock mass_lHorizontal distance a from gravity center of dangerous rock mass to overturning point, characteristic period T of regional earthquake dynamic acceleration response spectrum, regional earthquake dynamic peak acceleration G and tensile strength sigma of dangerous rock mass_tThe dip angle beta of the main control structure surface, the length l of the through section of the main control structure surface and the fracture water pressure Q of the main control structure surface are 8 quantitative characteristic indexes;

step S2: collecting the 12 characteristic indexes and stable categories of the multiple examples of the dumping karst dangerous rock masses, scoring the qualitative indexes according to the formulated scoring rules to convert the qualitative indexes into quantitative indexes, preprocessing and standardizing the example data of the multiple examples of the dumping karst dangerous rock masses, and establishing a machine learning sample set;

step S3: dividing a sample set into a training sample set and a testing sample set, training an IVM classification method with optimal parameter self-adaptive acquisition, small sample processing and strong adaptability to a high-dimensional nonlinear classification problem by adopting a cross validation strategy, and obtaining an IVM classification model with good extrapolation discrimination performance for comprehensively discriminating the stability of the dumping karst dangerous rock, thereby establishing a reliable nonlinear mapping relation between the stability of the dumping karst dangerous rock and various influence factors of the dumping karst dangerous rock;

step S4: for the newly dumped karst dangerous rock mass to be distinguished, 12 characteristic indexes are obtained through data collection and engineering tests, wherein qualitative indexes are graded according to a grading rule table and converted into quantitative indexes, 12-dimensional characteristic index vectors are constructed, the characteristic index vectors are input into a trained IVM classification model, and the classification probability of the stability of the characteristic index vectors is obtained through calculation, wherein the classification probability is more than 50% of the attribute stability, and the classification probability is less than 50% of the attribute instability;

step S5: according to the classification probability, quantifying the uncertainty of the newly-dumped karst dangerous rock mass judgment result, and dividing the uncertainty grade;

step S6: and (3) according to the uncertainty level of the newly dumped karst dangerous rock mass stability judgment result and the engineering level of the dangerous rock mass, mechanically calculating a stability coefficient by adopting a rigid body limit balance method for the dangerous rock mass with higher uncertainty level, and rechecking the accuracy of the judgment result.

2. The information vector machine method for the comprehensive judgment of the stability of the poured karst dangerous rocks according to claim 1, wherein the step S5 comprises the steps of evaluating the uncertainty of the judgment result by applying classification probability, dividing the uncertainty levels of the judgment result of 'stable' and 'unstable' into 3 levels respectively according to the classification probability, wherein the uncertainty levels provide scientific basis for the credibility decision of the judgment result of the stability of the dangerous rocks, and the scientific basis can be obtained conveniently.

3. The information vector machine method for comprehensive judgment of stability of dumping karst dangerous rocks according to claim 1, wherein when uncertainty of a judgment result of the new dangerous rocks in the step S5 is high level or the engineering level to which the new dangerous rocks belong is high, a rigid limit balancing method is adopted to perform recheck calculation on the stability of the new dangerous rocks, and according to a stress mechanism and an evolution process of the dumping karst dangerous rocks, the worst load combination is considered: calculating a stability coefficient of the gravity, fracture water pressure and seismic force, judging the stability coefficient to be stable when the stability coefficient value is more than 1, and judging the stability to be unstable if the stability coefficient value is not more than 1, so that the stability coefficient is used as a rechecking reference of an IVM classification model judgment result; wherein the fracture water pressure is the fracture water pressure under the rainstorm working condition.

4. Stable information vector machine device of differentiating of empting formula karst dangerous rock, its characterized in that includes:

an input module: the system is used for receiving the collected dumping karst dangerous rock example original data;

the IVM classification model prediction module: the method comprises the steps of inputting a sample of a new dangerous rock mass to be predicted into an IVM classification model for comprehensive judgment of stability of the poured karst dangerous rock mass to obtain the classification probability of the new dangerous rock mass;

5. The information vector machine device for the comprehensive judgment of the stability of the poured karst crisis according to claim 4, wherein the processing module comprises:

the preprocessing unit is used for eliminating or perfecting data with missing, invalid and wrong formats in the original data set, acquiring more complete, effective and reasonable data and improving the quality of the original data;

and the standardization unit is used for avoiding adverse effects on the machine learning model judgment accuracy caused by overlarge numerical value difference of different characteristic parameters or overlarge discreteness of the same characteristic parameter, and carrying out standardization processing on the input characteristic vectors of all the established samples.

6. The information vector machine device for the comprehensive judgment of the stability of the poured karst crisis according to claim 4, wherein the IVM classification model training module comprises:

and the model checking unit is used for carrying out feasibility checking on the optimal IVM classification model, judging whether the model performance meets the requirements or not, and checking the prediction accuracy of the test sample as the checking index.

7. The information vector machine device for the comprehensive judgment of the stability of the poured karst crisis according to claim 4, wherein the IVM classification model prediction module comprises:

the characteristic vector operator unit is used for inputting data of a sample to be judged to the processing module so as to obtain characteristic vectors required by the IVM classification model for stable and comprehensive judgment of the dumping karst dangerous rocks;

8. The information vector machine device for the comprehensive judgment of the stability of the poured karst crisis according to claim 4, wherein the rechecking module comprises:

a prediction reinspection unit: the system comprises a sample set, a second rechecking group and a third rechecking group, wherein the sample set is used for determining samples with higher engineering levels in the sample set according to engineering level labels to which input new samples belong and listing the samples in the second rechecking group;