CN112966425B - Slope stability prediction and evaluation method - Google Patents

Slope stability prediction and evaluation method Download PDF

Info

Publication number
CN112966425B
CN112966425B CN202110376378.1A CN202110376378A CN112966425B CN 112966425 B CN112966425 B CN 112966425B CN 202110376378 A CN202110376378 A CN 202110376378A CN 112966425 B CN112966425 B CN 112966425B
Authority
CN
China
Prior art keywords
slope
stability
prediction
slope stability
samples
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110376378.1A
Other languages
Chinese (zh)
Other versions
CN112966425A (en
Inventor
张科
张凯
保瑞
刘享华
李娜
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Kunming University of Science and Technology
Original Assignee
Kunming University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Kunming University of Science and Technology filed Critical Kunming University of Science and Technology
Priority to CN202110376378.1A priority Critical patent/CN112966425B/en
Publication of CN112966425A publication Critical patent/CN112966425A/en
Application granted granted Critical
Publication of CN112966425B publication Critical patent/CN112966425B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/23—Design optimisation, verification or simulation using finite element methods [FEM] or finite difference methods [FDM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/08—Probabilistic or stochastic CAD
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
    • Y02A10/23—Dune restoration or creation; Cliff stabilisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Geometry (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Hardware Design (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Civil Engineering (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Architecture (AREA)
  • Data Mining & Analysis (AREA)
  • Structural Engineering (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a method for predicting and evaluating slope stability, which comprises the following steps: 1, selecting a slope stability judgment index and determining an allowable safety factor [ K ]; 2, constructing a slope model according to a slope to be predicted, and acquiring a data set of a judgment index and a corresponding stability result by utilizing engineering simulation software; 3, carrying out maximum value and minimum value normalization processing on the data set; 4, constructing a slope stability prediction model based on an integrated learning algorithm according to the data set after normalization processing; and 5, acquiring actual judgment index data of the slope to be detected during evaluation, and bringing the actual judgment index data into a slope stability prediction model to obtain a slope stability prediction result. The method and the device can realize the stability safety prediction of the side slope on the basis of not damaging the soil body structure, and have the advantages of high prediction reliability and good accuracy.

Description

Slope stability prediction and evaluation method
Technical Field
The invention relates to the technical field of slope soil safety monitoring, in particular to a slope stability prediction and evaluation method.
Background
The slope stability refers to the stability degree of the rock and soil bodies of the slope under the conditions of certain slope height and slope angle. According to the cause, the side slopes are divided into natural slopes and artificial slopes, and the latter are divided into excavation side slopes, dam side slopes and the like. Unstable natural slopes and artificial slopes with too large design slope angles are frequently damaged by sliding or collapsing under the action of gravity, water pressure, vibration force and other external forces of rocks and soil bodies. Large-scale damage of the rock and soil bodies on the side slopes can cause traffic interruption, building collapse, river blockage and reservoir silting, and bring huge losses to the lives and properties of people. Therefore, the method has important significance in researching the prediction and monitoring of slope instability.
The existing slope stability monitoring method generally realizes monitoring by burying a sensor in a slope soil body or arranging a camera above a slope. For example, CN207335617U discloses a monitoring structure for stability of an existing roadbed and a slope, and CN112432661A discloses a slope stability monitoring system based on a BIM platform. All belong to this technology. The mode of burying the sensor underground in the side slope easily causes the influence to the structure and the stability of side slope self. And the mode of camera control, the reliability is relatively poor.
In the prior art, there are some methods for predicting slope stability, for example, a slope stability monitoring and instability prediction method based on rock-soil mass strain state mutation disclosed in CN 101718876B; CN103163563B discloses a three-dimensional slope stability prediction method. But still has the defects of great influence on the slope structure, less consideration, low prediction precision and the like.
Therefore, how to provide a slope stability monitoring method with more comprehensive consideration and better reliability becomes a problem to be considered by the technical personnel in the field.
Disclosure of Invention
Aiming at the defects of the prior art, the technical problems to be solved by the invention are as follows: the method for predicting and evaluating the stability of the slope can realize the safety prediction of the stability of the slope on the basis of not damaging the soil structure and has high prediction reliability.
In order to solve the technical problems, the invention adopts the following technical scheme:
a slope stability prediction and evaluation method is characterized by comprising the following steps:
1 selecting slope stability judgment index and determining allowable safety factor K]The selected determination indexes are: volume weight gamma, cohesion c, angle of friction
Figure GDA0003611245220000011
Side slope angle phi, side slope height H and pore pressure ratio ru；
2, constructing a slope model according to a slope to be predicted, and calculating to obtain a judgment index and a corresponding data set of a safety coefficient by utilizing engineering simulation software; then, judging the stability, if the slope safety coefficient is greater than or equal to the allowable safety coefficient, determining the slope safety coefficient as stable, and if the slope safety coefficient is less than the allowable safety coefficient, determining the slope safety coefficient as unstable, and further obtaining a judgment index and a data set corresponding to a stability result;
3, carrying out maximum value and minimum value normalization processing on the data set;
4, constructing a slope stability prediction model based on an integrated learning algorithm according to the data set after normalization processing;
and 5, acquiring actual judgment index data of the slope to be detected during evaluation, and bringing the data into a slope stability prediction model to obtain a slope stability prediction result.
Therefore, in the method, the selected judgment indexes have the internal relevance with the slope stability, but are in a nonlinear relation with each other, so that the prediction precision can be more comprehensively improved according to the judgment indexes. Specifically, the side slope angle phi and the side slope height H belong to geometric factors, which means that the higher the side slope height of the side slope body is, i.e. the larger the side slope angle is, the lower the safety coefficient of the side slope is, and the more easy the instability is; conversely, the smaller the slope height, i.e., the smaller the slope angle, the higher the safety factor of the slope, and at this time, the more stable the slope. Volume weight gamma, cohesion c, angle of friction
Figure GDA0003611245220000021
And pore pressure ratio ruThe soil mechanical indexes of the slope body are all the greater the volume weight of the soil body is, the greater the cohesive force is, the greater the friction angle is, and the greater the stability coefficient of the slope isHigh, namely the greater the safety factor, the more stable the slope body is; on the contrary, the smaller the volume weight of the soil body, the smaller the cohesive force, the smaller the friction angle and the lower the stability coefficient of the side slope, i.e. the lower the safety coefficient, the instability of the slope body is easy to generate. Meanwhile, the 6 influencing factors and the slope stability are in a complex nonlinear relation, and the traditional method cannot accurately describe the slope stability. Meanwhile, in the method, the engineering simulation software is utilized to obtain the judgment index and the data set corresponding to the safety coefficient, so that the strong simulation capability of the engineering simulation software is utilized, the data does not need to be acquired on site, the defect of data shortage is overcome, and the prediction precision can be better improved.
Further, the slope allowable safety factor [ K ] is set to 1.3, and a slope safety factor of 1.3 or more is regarded as stable, and a slope safety factor of less than 1.3 is regarded as unstable.
The allowable safety coefficient of the side slope is generally between 1.1 and 1.35, and a larger safety coefficient is preferably selected in the scheme, so that the safety can be better ensured.
Further, step 2 specifically includes the following steps:
2.1 randomly generating not less than 100 groups of index parameters in Matlab software to obtain a data set related to the judgment index
Figure GDA0003611245220000022
Wherein the ratio of gamma, c,
Figure GDA0003611245220000023
Φ,H,rurespectively a one-dimensional column vector;
2.2, constructing a slope model according to the slope to be predicted, inputting randomly generated index parameters (other required parameters are selected to be default) by utilizing engineering simulation software, and solving the safety coefficient K corresponding to each group of index parameters by adopting a strength reduction method;
2.3 the calculated safety factor K and the allowable safety factor K]By comparison, if K is not less than [ K ]]Considered stable, represented by 0, K < [ K ]]Considered destabilizing, indicated by 1; obtaining a data set of decision metrics and corresponding stability results
Figure GDA0003611245220000031
Figure GDA0003611245220000032
R is ∈ {0,1}, and is a 1-dimensional column vector.
The Matlab software is the existing mathematical software for data analysis, and index parameters are randomly generated by the software, so that the mean values of the index parameters can be better close to the real index parameters, and the reference value is improved. The engineering simulation software can be realized by adopting finite elements, discrete elements, finite differences and other software, is mature engineering simulation software, has strong simulation capability, and can be used for simulation and performance calculation of building or geological engineering units. The specific simulation process is the prior art for realizing the functions of the software, and is not detailed here. The safety coefficient is solved by adopting the intensity reduction method, so that the reliability of solving the safety coefficient can be better ensured.
Further, in step 2.2, a model of the slope to be detected is established in the CAD software, and is output as a format file (such as a × dxf format) that can be identified by the engineering simulation software, and then the format file is imported into the engineering simulation software; then, grid division is firstly carried out in engineering simulation software, and boundary conditions are determined, wherein the left side and the right side of the boundary conditions are generally horizontally constrained, the lower part of the boundary conditions is fixed, and the upper part of the boundary conditions is a free boundary; the initial ground stress is selected as a dead weight ground stress field; then, randomly generated index parameters, namely gamma, c,
Figure GDA0003611245220000033
Φ,H,ru(other parameters are selected as defaults), and solving the safety coefficient K by adopting an intensity reduction method.
Therefore, the method has better operability and is convenient and quick to operate.
Further, the engineering simulation software is implemented by using OptunG 2 software.
The software is the existing rock-soil analysis software for finite element analysis, and the data set of the slope safety coefficient is obtained by adopting the software, so that the method is more accurate and reliable. Of course, other existing finite element analysis software, discrete element analysis software, or finite difference software may be used.
Further, the step 2 specifically includes the following steps: 2.4, carrying out Pearson correlation analysis on the data set T1 to obtain a correlation coefficient matrix, and if the absolute value of the correlation is less than or equal to a preset value, indicating that the selected index and the generated data are reasonable; otherwise the data should be regenerated.
Therefore, random generated data which are not reasonable enough can be eliminated, the rationality of data adopted by subsequent training is better ensured, and the reliability of the evaluation model is improved. The preset value can be usually 0.5, and according to research, when the absolute value R of the correlation coefficient is 1, complete correlation is indicated; when R is more than or equal to 0.8 and less than 1, the correlation is high; when R is greater than or equal to 0.5 and less than 0.8, significant correlation is indicated, and when R is less than 0.5, low correlation or no correlation is indicated. Therefore 0.5 can be set as a critical threshold as our preset value. When the correlation is smaller, the generated data have better independence, and a more complex nonlinear relation exists, so that the method can be directly used for the model; if the correlation is large, the independence between the generated data is poor, and at this time, the effectiveness of the selected index is reduced and is not in accordance with the reality.
Further, step 3 specifically includes: carrying out maximum and minimum normalization processing on the data set to reduce the influence of the dimension on the prediction result, wherein the mapping interval is [0,1 ]; the concrete formula is as follows:
Figure GDA0003611245220000041
in the formula: z is the original characteristic value, zmaxAnd zminRespectively the maximum and minimum values of the class characteristic, z*And taking the value of the feature after normalization.
After the data are normalized, the influence of the dimension on the prediction result can be reduced, and the reliability of a subsequent training model is ensured.
Further, step 4 specifically includes the following steps:
4.0 normalization ofData set T1Dividing the training set into a training set A and a test set B, wherein the length of the training set A is generally larger than that of the test set B;
4.1, constructing a slope stability tendency prediction model based on the XGboost ensemble learning algorithm by using the samples of the training set A:
4.2, optimizing main parameters of a slope stability tendency prediction model based on the XGboost ensemble learning algorithm by adopting a grid search algorithm and 5-fold cross validation:
and 4.3, testing the prediction result of the model with the optimized parameters by using the samples in the test set B, and if the error rate is lower than a threshold value (generally, the error rate is less than or equal to 0.1, the test is qualified, otherwise, the parameters in the model are optimized again until the requirements are met), determining that the test is passed.
In this way, the reliability of the model can be better ensured.
Further, in step 4.1, the XGBoost ensemble learning algorithm-based slope stability inclination prediction model has the expression:
Figure GDA0003611245220000042
in the formula: y isrIs the predicted value of the r-th sample in the model, fkIs the basis function of the kth classification regression tree, K is the total number of classification regression trees, xrIs the r-th input sample, and F is the hypothesis space;
wherein, for each classification regression tree, the objective function L is represented as:
Figure GDA0003611245220000043
in the formula: m is the total number of samples, l (-) represents the loss function, yiAnd
Figure GDA0003611245220000046
actual values and predicted values are respectively, and omega (-) is a regular term;
wherein, the edges based on XGboost ensemble learning algorithmInputting a slope stability tendency prediction model: in A
Figure GDA0003611245220000044
Figure GDA0003611245220000045
The method comprises the following steps of outputting a slope stability tendency prediction model based on an XGboost ensemble learning algorithm: utilizing softmax in the XGboost algorithm as an objective function, and finally returning the prediction category, namely judging whether the prediction category is 0 or 1;
the method comprises the following steps of A, obtaining a loss function of a slope stability tendency prediction model based on an XGboost ensemble learning algorithm: a default binary error rate (error) is used.
The model can be used for realizing prediction more accurately and reliably.
Further, step 4.2 may comprise the steps of:
in the first step, the required control parameters are determined, where 8 parameters are selected that have a greater influence on the model, namely: the method comprises the steps of classifying the number of regression trees, randomly extracting a sample proportion, classifying the maximum depth of the regression trees, learning rate, randomly adopting a characteristic proportion, a complexity penalty term, an L2 regular term, leaf node weight and a minimum value;
secondly, determining the value range of the adjusting parameter, and setting the optimization standard to be that error is less than or equal to 0.01 as shown in table 2;
TABLE 2 parameter settings ranges
Figure GDA0003611245220000051
Thirdly, preliminarily constructing a grid according to different value ranges of each parameter, setting a proper step length, calculating the average value of two-classification error rates after 5 times of iterative computation of each point in the grid, and recording data;
fourthly, judging the size relation between the recorded value and 0.01, and if the recorded value is less than or equal to 0.01, determining the recorded value as an optimal parameter combination; if the ratio is more than 0.01, carrying out the fifth step;
fifthly, setting a smaller step length by taking the point with the minimum error value in the third step as a center and taking the adjacent points as boundaries to obtain finer grid division, calculating the two-classification error rate error of each point in the grid after 5 times of iterative computation, and recording data;
and repeating the fourth step until the requirements are met.
The method is adopted to realize the optimization of the model parameters, the XGboost algorithm model is optimized by using the grid search method, the tuning process is simple, the overall optimal parameters can be obtained, and the accuracy of the model is further improved; meanwhile, the use of 5-fold cross validation can effectively avoid accidental errors of the test to a certain extent.
Further, step 4 also includes step 4.4: constructing a confusion matrix to judge the generalization ability of the model; the specific process is as follows:
the confusion matrix is established as shown in table 3, and the number of actually stable samples is TP and the number of actually unstable samples is FP in the samples predicted to be stable; in the samples predicted to be unstable, the number of the samples actually stable is FN, and the number of the samples actually unstable is TN;
TABLE 3 confusion matrix
Figure GDA0003611245220000061
And 4 types of indexes are provided for comprehensively evaluating the prediction result:
the accuracy is as follows:
Figure GDA0003611245220000062
the precision ratio is as follows:
Figure GDA0003611245220000063
the recall ratio is as follows:
Figure GDA0003611245220000064
harmonic mean of precision and recall:
Figure GDA0003611245220000065
the calculated values of the 4 indexes are larger than a preset value (generally, Acc, Pre, Rec and F1-Score are more than or equal to 0.90 and meet the requirement, but the larger the index value is, the more accurate the prediction month of the model is, the stronger the generalization ability is, and the more accurate the prediction ability of the method is), the generalization ability is considered to meet the requirement.
Therefore, the reliability and the accuracy of model prediction can be better ensured through the generalization capability judgment.
In the step 5, the acquisition modes of the actual values of the 6 indexes are mature prior art, and specifically, the side slope angle phi and the side slope height H can be acquired through equipment such as a total station and the like; for volume weight gamma, cohesion c, angle of friction
Figure GDA0003611245220000066
And pore pressure ratio ruThe method can be used for sampling the researched slope soil body, then transporting the sample to a laboratory for carrying out corresponding soil mechanics experiments, and further obtaining related parameters, wherein the obtaining process is not a place where the method contributes to the prior art, and is not detailed here.
To sum up, this application can realize the stability safety prediction of side slope on the basis of not destroying the soil body structure, has the prediction reliability height, advantage that the accuracy is good.
Drawings
FIG. 1 is a schematic diagram of a slope model to be detected established in cad software.
FIG. 2 is a schematic diagram of mesh partitioning and boundary conditions performed in the engineering simulation software.
FIG. 3 is a schematic flow chart of the method in practice.
Fig. 4 is a schematic flow chart of step 4.2 of the method in practice.
Detailed Description
The present invention will be described in further detail with reference to specific embodiments.
The specific implementation mode is as follows: referring to fig. 1-4, a slope stability prediction evaluation method includes the following steps (see fig. 3):
1 selecting slope stability judgment index and determining allowable safety factor K]The selected determination indexes are: volume weight gamma, cohesion c, angle of friction
Figure GDA0003611245220000071
Side slope angle phi, side slope height H and pore pressure ratio ru；
2, constructing a slope model according to a slope to be predicted, and calculating to obtain a data set of a judgment index and a corresponding safety coefficient by utilizing engineering simulation software; then, judging the stability, if the slope safety coefficient is greater than or equal to the allowable safety coefficient, determining the slope safety coefficient as stable, and if the slope safety coefficient is less than the allowable safety coefficient, determining the slope safety coefficient as unstable, and further obtaining a judgment index and a data set corresponding to a stability result;
3, carrying out maximum value and minimum value normalization processing on the data set;
4, constructing a slope stability prediction model based on an integrated learning algorithm according to the data set after normalization processing;
and 5, acquiring actual judgment index data of the slope to be detected during evaluation, and bringing the data into a slope stability prediction model to obtain a slope stability prediction result.
Therefore, in the method, the selected judgment indexes have the internal relevance with the slope stability, but are in a nonlinear relation with each other, so that the prediction precision can be more comprehensively improved according to the judgment indexes. Specifically, the side slope angle phi and the side slope height H belong to geometric factors, which means that the higher the side slope height of the side slope body is, i.e. the larger the side slope angle is, the lower the safety coefficient of the side slope is, and the more easy the instability is; conversely, the smaller the slope height, i.e., the smaller the slope angle, the higher the safety factor of the slope, and at this time, the more stable the slope. Volume weight gamma, cohesion c, angle of friction
Figure GDA0003611245220000072
And pore pressure ratio ruAll the soil mechanical indexes of the slope body, the larger the volume weight of the soil body is, the cohesionThe larger the force is, the larger the friction angle is, the higher the stability coefficient of the side slope is, namely, the higher the safety factor is, the more stable the slope body is; on the contrary, the smaller the volume weight of the soil body, the smaller the cohesive force, the smaller the friction angle and the lower the stability coefficient of the side slope, i.e. the lower the safety coefficient, the instability of the slope body is easy to generate. Meanwhile, the 6 influencing factors and the slope stability are in a complex nonlinear relation, and the traditional method cannot accurately describe the slope stability. Meanwhile, in the method, the engineering simulation software is utilized to obtain the judgment index and the data set corresponding to the safety coefficient, so that the strong simulation capability of the engineering simulation software is utilized, the data does not need to be acquired on site, the defect of data shortage is overcome, and the prediction precision can be better improved.
In specific implementation, the slope allowable safety factor [ K ] is set to be 1.3, the slope safety factor is considered to be stable when the slope allowable safety factor [ K ] is greater than or equal to 1.3, and the slope allowable safety factor [ K ] is considered to be unstable when the slope allowable safety factor [ K ] is less than 1.3.
The allowable safety coefficient of the side slope is generally between 1.1 and 1.35, and a larger safety coefficient is preferably selected in the scheme, so that the safety can be better ensured.
In specific implementation, the step 2 specifically comprises the following steps:
2.1 randomly generating not less than 100 groups of index parameters in Matlab software to obtain a data set related to the judgment index
Figure GDA0003611245220000081
Wherein the ratio of gamma, c,
Figure GDA0003611245220000082
Φ,H,rurespectively a one-dimensional column vector; the respective mean values of the two methods are optimized by approximate real index parameters;
2.2, constructing a slope model according to the slope to be predicted, inputting randomly generated index parameters (other required parameters are selected to be default) by utilizing engineering simulation software, and solving the safety coefficient K corresponding to each group of index parameters by adopting a strength reduction method;
2.3 the calculated safety factor K and the allowable safety factor K]By comparison, if K is not less than [ K ]]Considered stable, represented by 0, K < [ K ]]Considered destabilizing, indicated by 1; obtaining a determination index and corresponding stabilityResulting data set
Figure GDA0003611245220000083
Figure GDA0003611245220000084
R is ∈ {0,1}, and is a 1-dimensional column vector.
Table 1 shows the partial data set obtained using the inopumg 2 module.
TABLE 1 data set
Figure GDA0003611245220000085
The Matlab software is the existing mathematical software for data analysis, and index parameters are randomly generated by adopting the software, so that the respective mean values of the index parameters can be better close to the real index parameters, and the reference value is improved. The engineering simulation software can be realized by adopting finite elements, discrete elements, finite differences and other software, is mature engineering simulation software, has strong simulation capability, and can be used for simulation and performance calculation of building or geological engineering units. The specific simulation process is the prior art for realizing the functions of the software, and is not detailed here. The safety coefficient is solved by adopting the intensity reduction method, so that the reliability of solving the safety coefficient can be better ensured.
In step 2.2, a model of the slope to be detected (as shown in fig. 1) is established in CAD software, a format file (such as a × dxf format) which can be identified by the engineering simulation software is output, and then the format file is imported into the engineering simulation software; then, firstly, grid division is carried out in engineering simulation software (as shown in figure 2), boundary conditions are determined, the left side and the right side of the boundary conditions are generally horizontally restricted, the lower part of the boundary conditions is fixed, and the upper part of the boundary conditions is a free boundary; the initial ground stress is selected as a dead weight ground stress field; then, randomly generated index parameters, namely gamma, c,
Figure GDA0003611245220000086
Φ,H,ru(other parameter selection defaults) byAnd solving the safety coefficient K by the intensity reduction method.
Therefore, the method has better operability and is convenient and quick to operate.
Wherein, the engineering simulation software is realized by OptunG 2 software.
The software is the existing rock-soil analysis software for finite element analysis, and the data set of the slope safety coefficient is obtained by adopting the software, so that the method is more accurate and reliable. Of course, other existing finite element analysis software, discrete element analysis software, or finite difference software may be used.
In practice, the step 2 further comprises the following steps: 2.4 pairs of data sets T1Performing Pearson correlation analysis to obtain a correlation coefficient matrix, and if the correlation is less than or equal to a preset value (in the embodiment, the preset value is 0.5), indicating that the selected index and the generated data are reasonable; otherwise the data should be regenerated.
Therefore, random generated data which are not reasonable enough can be eliminated, the rationality of data adopted by subsequent training is better ensured, and the reliability of the evaluation model is improved. The preset value can be usually 0.5, and according to research, when the absolute value R of the correlation coefficient is 1, complete correlation is indicated; when R is more than or equal to 0.8 and less than 1, the correlation is high; when R is greater than or equal to 0.5 and less than 0.8, significant correlation is indicated, and when R is less than 0.5, low correlation or no correlation is indicated. Therefore 0.5 can be set as a critical threshold as our preset value. When the correlation is smaller, the generated data have better independence, and a more complex nonlinear relation exists, so that the method can be directly used for the model; if the correlation is large, the independence between the generated data is poor, and at this time, the effectiveness of the selected index is reduced and is not in accordance with the reality.
In practice, step 3 specifically comprises: carrying out maximum and minimum normalization processing on the data set to reduce the influence of the dimension on the prediction result, wherein the mapping interval is [0,1 ]; the specific formula is as follows:
Figure GDA0003611245220000091
in the formula: z is the original characteristic value, zmaxAnd zminRespectively the maximum and minimum values of the class characteristic, z*And taking the value of the feature after normalization.
After the data are normalized, the influence of the dimension on the prediction result can be reduced, and the reliability of a subsequent training model is ensured.
When implemented, the step 4 specifically comprises the following steps:
4.0 normalization of the processed data set T1Dividing the training set into a training set A and a test set B, wherein the length of the training set A is generally larger than that of the test set B;
4.1, constructing a slope stability tendency prediction model based on the XGboost ensemble learning algorithm by using the samples of the training set A:
4.2, optimizing main parameters of a slope stability tendency prediction model based on the XGboost ensemble learning algorithm by adopting a grid search algorithm and 5-fold cross validation:
4.3, testing the prediction result of the model with the optimized parameters by using the samples in the test set B, if the error rate is less than or equal to 0.1, the test is qualified, otherwise, the parameters in the model are optimized again until the requirements are met.
In this way, the reliability of the model can be better ensured.
In the implementation, in step 4.1, the expression of the slope stability tendency prediction model based on the XGBoost ensemble learning algorithm is as follows:
Figure GDA0003611245220000101
in the formula: y isrIs the predicted value of the r-th sample in the model, fkIs the basis function of the kth classification regression tree, K is the total number of classification regression trees, xrIs the r-th input sample, and F is the hypothesis space;
wherein, for each classification regression tree, the objective function L is represented as:
Figure GDA0003611245220000102
in the formula: m is the total number of samples, l (-) represents the loss function, yiAnd
Figure GDA0003611245220000106
actual values and predicted values are respectively, and omega (-) is a regular term;
the method comprises the following steps of inputting a slope stability tendency prediction model based on an XGboost ensemble learning algorithm: in A
Figure GDA0003611245220000103
Figure GDA0003611245220000104
The method comprises the following steps of outputting a slope stability tendency prediction model based on an XGboost ensemble learning algorithm: utilizing softmax in the XGboost algorithm as an objective function, and finally returning the prediction category, namely judging whether the prediction category is 0 or 1;
the method comprises the following steps of A, obtaining a loss function of a slope stability tendency prediction model based on an XGboost ensemble learning algorithm: a default binary error rate (error) is used.
The model can be used for realizing prediction more accurately and reliably.
In practice, referring to fig. 4, step 4.2 may comprise the following steps: (in practice, when the parameter value range is referred to in the present application, the value is usually obtained based on an empirical value in a conventional manner when corresponding software operates, if not described otherwise)
In a first step, the required control parameters are determined, wherein 8 parameters that have a greater influence on the model are selected, namely: the method comprises the steps of classifying the number of regression trees, randomly extracting a sample proportion, classifying the maximum depth of the regression trees, learning rate, randomly adopting a characteristic proportion, a complexity penalty term, an L2 regular term, leaf node weight and a minimum value;
secondly, determining the value range of the adjusting parameter, and setting the optimization standard to be that error is less than or equal to 0.01 as shown in table 2;
TABLE 2 parameter settings ranges
Figure GDA0003611245220000105
Thirdly, preliminarily constructing a grid according to different value ranges of each parameter, setting a proper step length, calculating the error rate error mean value of each point in the grid after 5 times of iterative computation, and recording data;
fourthly, judging the size relation between the recorded value and 0.01, and if the recorded value is less than or equal to 0.01, determining the recorded value as the optimal parameter combination; if the ratio is more than 0.01, carrying out the fifth step;
fifthly, setting a smaller step length by taking the point with the minimum error value in the third step as a center and taking the adjacent points as boundaries to obtain finer grid division, calculating the two-classification error rate error of each point in the grid after 5 times of iterative computation, and recording data;
and repeating the fourth step until the requirements are met.
The method is adopted to realize the optimization of the model parameters, the XGboost algorithm model is optimized by using the grid search method, the tuning process is simple, the overall optimal parameters can be obtained, and the accuracy of the model is further improved; meanwhile, the use of 5-fold cross validation can effectively avoid accidental errors of the test to a certain extent.
Wherein, step 4 also includes step 4.4: constructing a confusion matrix to judge the generalization ability of the model; the specific process is as follows:
the confusion matrix is established as shown in table 3, and the number of actually stable samples is TP and the number of actually unstable samples is FP in the samples predicted to be stable; in the samples predicted to be unstable, the number of the samples actually stable is FN, and the number of the samples actually unstable is TN;
TABLE 3 confusion matrix
Figure GDA0003611245220000111
And 4 types of indexes are provided for comprehensively evaluating the prediction result:
the accuracy is as follows:
Figure GDA0003611245220000112
the precision ratio is as follows:
Figure GDA0003611245220000113
the recall ratio is as follows:
Figure GDA0003611245220000114
harmonic mean of precision and recall:
Figure GDA0003611245220000115
the calculated values of the 4 indexes are larger than a preset value (generally, Acc, Pre, Rec and F1-Score are more than or equal to 0.90 and meet the requirement, but the larger the index value is, the more accurate the prediction month of the model is, the stronger the generalization ability is, and the more accurate the prediction ability of the method is), the generalization ability is considered to meet the requirement.
Therefore, the reliability and the accuracy of model prediction can be better ensured through the generalization capability judgment.
In the step 5, the acquisition modes of the actual values of the 6 indexes are mature prior art, and specifically, the side slope angle phi and the side slope height H can be acquired through equipment such as a total station and the like; for volume weight gamma, cohesion c, angle of friction
Figure GDA0003611245220000121
And pore pressure ratio ruThe method can be used for sampling the researched slope soil body, then transporting the sample to a laboratory for carrying out corresponding soil mechanics experiments, and further obtaining related parameters, wherein the obtaining process is not a place where the method contributes to the prior art, and is not detailed here.

Claims (6)

1. A slope stability prediction and evaluation method is characterized by comprising the following steps:
1 selecting and determining slope stability judgment indexesPermissible safety factor [ K]The selected judgment indexes are respectively: volume weight gamma, cohesion c, angle of friction
Figure FDA0003622068520000011
Side slope angle phi, side slope height H and pore pressure ratio ru；
2, constructing a slope model according to a slope to be predicted, and calculating to obtain a judgment index and a corresponding data set of a safety coefficient by utilizing engineering simulation software; then, judging the stability, if the slope safety coefficient is greater than or equal to the allowable safety coefficient, determining the slope safety coefficient as stable, and if the slope safety coefficient is less than the allowable safety coefficient, determining the slope safety coefficient as unstable, and further obtaining a judgment index and a data set corresponding to a stability result;
3, carrying out maximum value and minimum value normalization processing on the data set;
4, constructing a slope stability prediction model based on an integrated learning algorithm according to the data set after normalization processing;
5, acquiring actual judgment index data of the slope to be detected during evaluation, and bringing the data into a slope stability prediction model to obtain a slope stability prediction result;
the step 2 specifically comprises the following steps:
2.1 randomly generating not less than 100 groups of index parameters in Matlab software to obtain a data set related to the judgment index
Figure FDA0003622068520000012
Wherein the ratio of gamma, c,
Figure FDA0003622068520000013
Φ,H,rurespectively a one-dimensional column vector;
2.2 constructing a slope model according to the slope to be predicted, inputting randomly generated index parameters by utilizing engineering simulation software, and solving the safety coefficient K corresponding to each group of index parameters by adopting a strength reduction method;
2.3 the calculated safety factor K and the allowable safety factor K]By comparison, if K is not less than [ K ]]Considered stable, represented by 0, K < [ K ]]Considered destabilizing, indicated by 1; obtaining the judgment index and the corresponding stability nodeData set of fruits
Figure FDA0003622068520000014
Figure FDA0003622068520000015
R belongs to {0,1}, and is a 1-dimensional column vector;
step 2.2 specifically, a model of the slope to be detected is established in CAD software, a format file which can be identified by engineering simulation software is output, and then the format file is imported into the engineering simulation software; then, firstly, carrying out grid division in engineering simulation software, determining boundary conditions, wherein the left side and the right side of the boundary conditions are horizontally constrained, the lower part of the boundary conditions is fixed, and the upper part of the boundary conditions is a free boundary; the initial ground stress is selected as a dead weight ground stress field; then, randomly generated index parameters, namely gamma, c,
Figure FDA0003622068520000016
Φ,H,rusolving the safety coefficient K by adopting an intensity reduction method;
the engineering simulation software is realized by OptunG 2 software;
the step 2 specifically comprises the following steps: 2.4 pairs of data sets T1Performing Pearson correlation analysis to obtain a correlation coefficient matrix, and if the correlation is smaller than or equal to a preset value, indicating that the selected index and the generated data are reasonable; otherwise the data should be regenerated.
2. The slope stability prediction evaluation method of claim 1, wherein the slope allowable safety factor [ K ] is set to 1.3, and a slope safety factor greater than or equal to 1.3 is considered stable, and a slope safety factor less than 1.3 is considered unstable.
3. The slope stability prediction assessment method according to claim 1, wherein step 3 specifically comprises: carrying out maximum and minimum normalization processing on the data set to reduce the influence of the dimension on the prediction result, wherein the mapping interval is [0,1 ]; the specific formula is as follows:
Figure FDA0003622068520000021
in the formula: z is the original characteristic value, zmaxAnd zminRespectively the characteristic maximum and minimum, z*And taking the value of the feature after normalization.
4. The slope stability prediction assessment method of claim 1, characterized in that: the step 4 specifically comprises the following steps:
4.0 normalization of the processed data set T1Dividing the training set into a training set A and a testing set B, wherein the length of the training set A is larger than that of the testing set B;
4.1, constructing a slope stability tendency prediction model based on the XGboost ensemble learning algorithm by using the samples of the training set A:
4.2, optimizing main parameters of a slope stability tendency prediction model based on the XGboost ensemble learning algorithm by adopting a grid search algorithm and 5-fold cross validation:
4.3 using the samples in the test set B to test the prediction result of the model after the parameters are optimized, and if the error rate is lower than the threshold value, the test is regarded as passing.
5. The method for predicting and evaluating slope stability according to claim 4, wherein in step 4.1, the expression of the slope stability tendency prediction model based on the XGboost ensemble learning algorithm is as follows:
Figure FDA0003622068520000022
in the formula: y isrIs the predicted value of the r-th sample in the model, fkIs the basis function of the kth classification regression tree, K is the total number of classification regression trees, xrIs the r-th input sample, and F is the hypothesis space;
wherein, for each classification regression tree, the objective function L is represented as:
Figure FDA0003622068520000023
in the formula: m is the total number of samples, l (-) represents the loss function, yiAnd
Figure FDA0003622068520000024
actual values and predicted values are respectively, and omega (-) is a regular term;
the method comprises the following steps of inputting a slope stability tendency prediction model based on an XGboost ensemble learning algorithm: in A
Figure FDA0003622068520000025
Figure FDA0003622068520000026
The method comprises the following steps of outputting a slope stability tendency prediction model based on an XGboost ensemble learning algorithm: utilizing softmax in the XGboost algorithm as a target function, and finally returning the predicted category, namely judging whether the predicted category is 0 or 1;
the method comprises the following steps of A, obtaining a loss function of a slope stability tendency prediction model based on an XGboost ensemble learning algorithm: a default binary error rate is employed.
6. The slope stability prediction assessment method of claim 4, wherein step 4 further comprises step 4.4: constructing a confusion matrix to judge the generalization ability of the model; the specific process is as follows:
the confusion matrix is established as shown in table 3, and the number of actually stable samples is TP and the number of actually unstable samples is FP in the samples predicted to be stable; in the samples predicted to be unstable, the number of the samples actually stable is FN, and the number of the samples actually unstable is TN;
TABLE 3 confusion matrix
Figure FDA0003622068520000031
And 4 types of indexes are provided for comprehensively evaluating the prediction result:
the accuracy is as follows:
Figure FDA0003622068520000032
the precision ratio is as follows:
Figure FDA0003622068520000033
the recall ratio is as follows:
Figure FDA0003622068520000034
harmonic mean of precision and recall:
Figure FDA0003622068520000035
and if the calculated values of the 4 indexes are larger than a preset threshold value, the generalization capability is considered to meet the requirement.
CN202110376378.1A 2021-04-06 2021-04-06 Slope stability prediction and evaluation method Active CN112966425B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110376378.1A CN112966425B (en) 2021-04-06 2021-04-06 Slope stability prediction and evaluation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110376378.1A CN112966425B (en) 2021-04-06 2021-04-06 Slope stability prediction and evaluation method

Publications (2)

Publication Number Publication Date
CN112966425A CN112966425A (en) 2021-06-15
CN112966425B true CN112966425B (en) 2022-06-07

Family

ID=76279984

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110376378.1A Active CN112966425B (en) 2021-04-06 2021-04-06 Slope stability prediction and evaluation method

Country Status (1)

Country Link
CN (1) CN112966425B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113465680A (en) * 2021-07-19 2021-10-01 中铁十九局集团矿业投资有限公司 DINI 03-based refuse dump stability monitoring method and system
CN114266088B (en) * 2021-11-24 2024-05-31 中航勘察设计研究院有限公司 Slope stability prediction method
CN114330168B (en) * 2021-12-30 2022-06-21 中国科学院力学研究所 Method for dynamically evaluating slope safety
CN114357912B (en) * 2022-01-11 2022-09-16 湖南工程学院 Stability analysis system of river bank slope
CN114896548B (en) * 2022-05-20 2023-04-07 西南交通大学 Slope stability judging method, device and equipment and readable storage medium
CN117435891B (en) * 2023-12-20 2024-02-27 成都嘉新科技集团有限公司 Soil slope stability simulation evaluation method and system

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107330224A (en) * 2017-07-24 2017-11-07 中国地质大学（武汉） A kind of Analysis of Slope Stability slices method of the non-hypothesis in slitting intermolecular forces inclination angle
CN109840541A (en) * 2018-12-05 2019-06-04 国网辽宁省电力有限公司信息通信分公司 A kind of network transformer Fault Classification based on XGBoost
CN111784070A (en) * 2020-07-09 2020-10-16 中国地质环境监测院 Intelligent landslide short-term early warning method based on XGboost algorithm
CN112329349A (en) * 2020-11-16 2021-02-05 中南大学 Slope reliability assessment method

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107330224A (en) * 2017-07-24 2017-11-07 中国地质大学（武汉） A kind of Analysis of Slope Stability slices method of the non-hypothesis in slitting intermolecular forces inclination angle
CN109840541A (en) * 2018-12-05 2019-06-04 国网辽宁省电力有限公司信息通信分公司 A kind of network transformer Fault Classification based on XGBoost
CN111784070A (en) * 2020-07-09 2020-10-16 中国地质环境监测院 Intelligent landslide short-term early warning method based on XGboost algorithm
CN112329349A (en) * 2020-11-16 2021-02-05 中南大学 Slope reliability assessment method

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
Comparative Analysis of Gradient Boosting Algorithms for Landslide Susceptibility Mapping;Emrehan Kutlug Sahin;《Geocarto International》;20201005;1-32 *
基于RF-ELM模型的边坡稳定性预测研究;邵良杉等;《中国安全生产科学技术》;20150330(第03期);95-100 *
基于多源信息融合的车门匹配精度预测控制方法;肖灵 等;《 农业装备与车辆工程》;20191217;29-34 *
基于网格搜索支持向量机的边坡稳定性系数预测;王健伟等;《铁道建筑》;20190520(第05期);99-102 *

Also Published As

Publication number Publication date
CN112966425A (en) 2021-06-15

Similar Documents

Publication Publication Date Title
CN112966425B (en) Slope stability prediction and evaluation method
US20230214557A1 (en) Method for dynamically assessing slope safety
CN109611087B (en) Volcanic oil reservoir parameter intelligent prediction method and system
CN102269972B (en) Method and device for compensating pipeline pressure missing data based on genetic neural network
CN112257140A (en) Safety coefficient calculation method for stability of seabed slope
CN113033108B (en) Side slope reliability judging method based on AdaBoost algorithm
CN115310361B (en) Underground coal mine dust concentration prediction method and system based on WGAN-CNN
CN105493100A (en) Static earth model calibration methods and systems
CN112016212B (en) Reservoir longitudinal heterogeneity evaluation method based on seepage control equation
CN114154427A (en) Volume fracturing fracture expansion prediction method and system based on deep learning
CN114002129B (en) High-water-pressure-crack rock mass seepage test platform
CN116931085A (en) Method and device for predicting natural gas hydrate of sandy reservoir
CN114036831A (en) Real-time detection method for geotechnical parameters of side slope of engineering field to be detected
CN112507438B (en) Slope rock mass deformation control method, computer program product and readable storage medium
CN103898890A (en) Soil layer quantization layering method based on double-bridge static sounding data of BP neural network
CN111779477B (en) Fractal theory-based dynamic evaluation method for complexity of hydraulic fracture
CN113723706B (en) Shale gas well repeated fracturing productivity prediction method, device, terminal and storage medium
CN114036829B (en) Geological profile generation method, system, equipment and storage medium
CN110119522A (en) A kind of stability ranking method excavated rock side slope and destroy risk analysis
Fu et al. Slope stability analysis based on big data and convolutional neural network
CN112182888A (en) Method and device for identifying mechanical parameters of main control structural plane of small-sized sliding dangerous rock mass
CN117933103B (en) Carbon sequestration model uncertainty analysis method based on Bayesian deep learning
Gorucu et al. A neurosimulation tool for predicting performance in enhanced coalbed methane and CO2 sequestration projects
CN116467943A (en) Method for determining landslide rainfall threshold curve based on machine learning
CN116499532B (en) Complex marine environment deep water pile group construction monitoring system constructed based on hydrologic model

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant