Bayesian network-based reservoir bank soil slope stability prediction analysis method

Through Bayesian network analysis of the influencing factors of the bank shore slope, a stability prediction model was constructed, which solved the problem of fuzzy prediction results of the bank shore slope stability in the existing technology, and achieved the trend of changing slope stability and the most dangerous The accurate identification of working conditions and the introduction of preventive measures have improved the safety guarantee of the bank shore slope.

CN119066960APending Publication Date: 2024-12-03WUHAN SURVEYING GEOTECHN RES INST OF MCC
View PDF 0 Cites 7 Cited by

Patent Information

Application Number
CN202411095082.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-12
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The existing technology has influencing factors in the prediction of the stability of the reservoir shore slope. The buffer area is too large, which cannot deeply reveal the slope deformation mechanism, resulting in vague prediction results, unable to effectively avoid accidents and propose targeted response measures.

Method used

Using the Bayesian network-based prediction analysis method for soil slope stability in the bank, the influencing factors were screened through statistical analysis and finite element numerical simulation, the machine learning sample library was established, K-means clustering analysis and correlation analysis were carried out, and the Bais was constructed. Yes network model, performs forward and reverse reasoning, and obtains the trend of slope stability change and the most dangerous working conditions.

Benefits of technology

Accurate prediction of the change trend of the reservoir shore slope stability is achieved, the most dangerous working conditions are found, and targeted preventive measures are made in combination with mechanism analysis to provide technical support and safety guarantees for the long-term safe operation of reservoirs and tailings ponds.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066960A_ABST
    Figure CN119066960A_ABST
Patent Text Reader

Abstract

The invention relates to a reservoir bank soil slope stability prediction analysis method based on a Bayesian network. The method comprises the steps of performing statistical analysis on influence factors of reservoir bank slope stability, performing single-factor sensitivity analysis on the statistical influence factors, perfecting disaster-inducing factors and grade division of reservoir bank soil slope stability evaluation indexes, and establishing a prediction evaluation index system; an improved interpretation structure model is applied, the sequence and interlayer directivity connection of influence factors are determined, a construction method of a Bayesian network structure is improved, and a reservoir bank slope stability prediction model based on the Bayesian network is established; and based on causal reasoning and diagnostic reasoning functions of a Bayesian network model, reservoir bank soil slope stability prediction and instability mechanism analysis are carried out. According to the method, the stability of the reservoir bank slope can be predicted, hidden dangers can be found in time, corresponding measures are taken immediately for treatment, and technical support and safety guarantee are provided for long-term safe operation of reservoirs and tailings ponds.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of slope engineering disaster prediction, and specifically is a reservoir bank soil slope stability prediction and analysis method system based on Bayesian network. Background Art

[0002] Landslide is the process in which a slope composed of rocks, soil or debris deposits slides downward as a whole along a weak surface (sliding surface) under the action of gravity and the influence of surface water, groundwater or earthquakes. General landslide disasters can destroy villages and towns, disrupt transportation, cause property losses and casualties, while secondary disasters caused by reservoir bank landslides can also block rivers, cause floods, and even induce surges, impact mountains to form mudslides, causing more serious losses.

[0003] Landslides on the soil slopes along the reservoir bank are a type of geological disaster caused by changes in the seepage field, stress field, and physical and mechanical conditions of the slopes due to fluctuations in the reservoir water level and rainfall conditions. Changes in the reservoir water level may increase the sliding force of the reservoir bank slopes and reduce the shear strength of the soil, which in turn reduces the stability of the reservoir bank slopes and even destroys them, threatening the safety of downstream reservoir facilities. Therefore, accurate analysis of the stability of the reservoir bank slopes and prediction of landslide risks are of great scientific significance and engineering value.

[0004] At present, the research on reservoir bank slope stability analysis and slope failure mechanism by domestic and foreign scholars has been relatively mature, but the research results on reservoir bank slope stability prediction only focus on macro-regional landslide susceptibility evaluation. This method has the problem of setting a buffer area that is too large for influencing factors and cannot deeply reveal the slope deformation mechanism, resulting in vague prediction results. It may be impossible to avoid accidents and it is impossible to propose targeted countermeasures.

[0005] Machine learning algorithms can realize comprehensive analysis of multiple factors and are an effective method for predicting the stability of reservoir bank slopes under the influence of multiple factors. At present, the commonly used machine learning methods for reservoir bank slope stability analysis include logistic regression algorithms, various neural network algorithms, support vector machines, and random forest models. When the above-mentioned quantitative analysis machine learning methods are applied to slope stability analysis, when the samples are insufficient or different types of slopes consider different characteristic variables, overfitting problems often occur. Therefore, based on existing research results, this paper will statistically screen influencing factors, collect engineering data, and apply Bayesian network methods to establish a reservoir bank slope stability prediction and analysis model. Summary of the invention

[0006] In view of the above shortcomings, the present invention proposes a reservoir bank soil slope stability prediction and analysis method based on Bayesian network, introduces a probability calculation method, and establishes a reservoir bank slope stability prediction and analysis system based on machine learning method for reservoir bank slope stability prediction and destruction mechanism analysis; it can analyze and predict the stability change trend of the reservoir bank slope, infer the instability mechanism and the most dangerous working condition of the reservoir bank slope, and find out the most dangerous working condition before a certain reservoir bank slope becomes unstable, and make targeted preventive measures in combination with mechanism analysis, so as to provide technical support and safety guarantee for the long-term safe operation of reservoirs and tailings ponds.

[0007] In order to achieve the above technical objectives, the present invention provides a method for predicting and analyzing the stability of reservoir bank soil slope based on Bayesian network, which is characterized by comprising the following steps:

[0008] S1. Statistical analysis of the factors affecting slope stability in the literature on reservoir bank slope stability analysis in recent years;

[0009] S2. Use finite element numerical simulation software to model and calculate each influencing factor counted in step S1 for single factor analysis, screen all influencing factors, and eliminate adverse influencing factors; collect reservoir bank slope stability analysis engineering data based on the remaining effective influencing factors, and establish a machine learning sample library including effective influencing factors of slope stability and safety factors;

[0010] S3. The effective influencing factors and safety factors of slope stability in the machine learning sample library established in step S2 are classified based on K-menas cluster analysis, and the selected influencing factors and safety factor level interval division are revised in combination with engineering specifications to establish a prediction analysis library bank slope stability evaluation index system;

[0011] S4. Conduct a bivariate correlation analysis on the safety factor and the effective influencing factors of slope stability in step S2, objectively obtain the hierarchical arrangement of the influencing factors based on the results of the correlation analysis, determine the directional connection between the layers, and establish an explanatory structural model of the influencing factors of slope stability;

[0012] S5. Introduce the causal graph correction principle, correct the explanatory structural model of slope stability influencing factors established in step S4, add cross-level causal relationships, optimize the learning method of the Bayesian network structure, and construct a Bayesian network structural model of slope stability;

[0013] S6: Based on the parameter learning method, the conditional probability table of the Bayesian network model of slope stability is constructed, and the reservoir bank slope stability prediction model based on the Bayesian network is established;

[0014] S7: Based on the Bayesian network reasoning function, input the measured values ​​of the influencing factors of the reservoir bank slope, obtain the probability curve of the safety factor level of the reservoir bank slope by forward reasoning, and obtain the probability curve of the level of the slope stability influencing factors by reverse diagnostic reasoning. Combined with the prediction and analysis reservoir bank slope stability evaluation index system established in step S2 and the reservoir bank slope stability prediction model based on the Bayesian network established in step S6, a reservoir bank slope stability prediction and analysis method system is established;

[0015] S8: Select the actual influencing factor data of the reservoir bank slope and predict and analyze the stability of the slope through the reservoir bank slope stability prediction and analysis method system established in step S7; specifically, the actual influencing factors of the selected reservoir bank slope are graded through the prediction and analysis of the reservoir bank slope stability evaluation index system and then inserted into the reservoir bank slope stability prediction model based on the Bayesian network, and the safety factor level probability curve and the influencing factor level probability curve are obtained by controlling the level probability of the influencing factor node and the safety factor node, and the safety factor change trend and the most sensitive influencing factor when the safety factor changes and the node combination with the maximum probability of the edge of the influencing factor level when the safety factor node is at the instability level are obtained from the curve, and the stability of the slope is comprehensively predicted and analyzed from three aspects: the stability change trend, the most sensitive influencing factor and the most dangerous instability condition.

[0016] The preferred technical solution of the present invention is as follows: the influencing factors statistically analyzed in the S1 step include 13 influencing factors, namely, soil type, natural bulk density, cohesion, internal friction angle, permeability coefficient, elastic modulus of sliding soil, Poisson's ratio of sliding soil, pore water pressure, slope height, slope angle, calculation method, water level change rate and water level change amount; in the S2 step, numerical simulation calculation is performed with the slope safety factor and the evolution process of the stress field and displacement field of the slope under different influencing factors as evaluation criteria, and the 13 influencing factors are analyzed on a single factor basis to determine the rationality of the selected factors and eliminate the undesirable factors; finally, the elastic modulus of sliding soil and the Poisson's ratio of sliding soil are eliminated from the ranks of influencing factors, and 11 factors that have a significant impact on the stability of the reservoir bank slope are retained as effective influencing factors, and the calculation results of the 11 effective influencing factors and the corresponding safety factors are collected as a reservoir bank slope stability analysis engineering data sample library, and 80% of the samples of the collected engineering data sample library are used for training, and 20% of the samples are used for testing.

[0017] A further technical solution of the present invention: In the step S3, the data in the machine learning sample library is objectively graded based on K-menas cluster analysis to classify the influencing factors, and the selected influencing factors and the safety factor grade interval division are corrected in combination with engineering specifications. The specific steps of establishing a prediction and analysis library bank slope stability evaluation index system are as follows:

[0018] (1) Select k samples from the N evaluation index data sample library and initialize these k samples as cluster centers (C1, C2, ..., C k-1 , C k ) The cluster center is set by itself or randomly selected by the system;

[0019] (2) Calculate the Euclidean distance from each sample in the library to each cluster center, compare the distance from each sample to each cluster center in turn, and assign the sample to the cluster with the closest cluster center to form k clusters (S1, S2, ..., S k-1 , S k );

[0020] (3) Update the cluster according to the following formula:

[0021]

[0022] Where: C i For cluster S i The central sample of; x is the sample point in the sample library;

[0023] k is the number of clusters; d is the minimized square error of cluster division;

[0024] Recalculate the cluster center of the new k clusters. The calculation formula is as follows:

[0025]

[0026] Where: n i For cluster S i The number of samples included; x is S i The sample points in ;

[0027] Select the initial C i Examine all data samples and divide the initial cluster S i , from the initial C i With S i Calculate the size of d. The smaller d is, the higher the similarity of samples in the cluster.

[0028] (4) Repeat steps (2) to (3) until the conditions are met; modify the node levels divided by the cluster analysis results in combination with existing engineering specifications to establish a prediction and analysis bank slope stability evaluation index system.

[0029] A further technical solution of the present invention is as follows: in step S4, a correlation analysis of two variables is performed on the safety factor and the effective influencing factors of the remaining slope stability, and the Sig value, i.e., the significance P value, is used to represent whether the correlation is significant; based on the result of the correlation analysis, the hierarchical arrangement of the influencing factors is objectively obtained, and according to the functional relationship of the variables in the same or adjacent levels, and in accordance with the rule that the high-level elements point to the low-level elements, the high-level variables are on the top, and the low-level ones are on the bottom, lines are sequentially connected to draw a multi-level directed topological diagram, and the relationship between the cross-level variables is ignored, and finally an explanatory structural model of the influencing factors of slope stability is established;

[0030] The S4 step includes a safety factor and an influencing factor, a total of 12 variables, for a bivariate correlation analysis, and the Sig value (two-tailed value) or the significant P value is used to represent whether the correlation is significant. After the analysis, the correlation results of all variables in pairs are obtained. When the significant P<0.05, it represents a significant correlation, and when P<0.01, it represents an extremely significant correlation. The construction of the explanatory structural model specifically includes the following steps:

[0031] (1) Based on the results of correlation analysis, construct the interaction matrix of variables;

[0032] (2) Based on theoretical knowledge and expert experience, determine two elements X in the set i With X j The influence relationship between ij . Convert the influence relationship between factors into an adjacency matrix If a ij =1, indicating that X i X j If a ij =0, indicating that X i X i No impact;

[0033] (3) The adjacency matrix is ​​further calculated to obtain the reachability matrix. The reachability matrix is ​​a matrix that describes the degree to which nodes in a directed graph can be reached after a path of a certain length passes through the nodes. The Warshall algorithm is often used to calculate the reachability matrix. The calculation formula is as follows:

[0034] M=(A+I) n+1 =(A+I) n ≠(A+I) n-1 ≠…≠(A+I)

[0035] Where n is the path length of M, and I is the unit matrix; when the equality sign holds, it means that the reachable matrix is ​​obtained;

[0036] (4) When dividing the system into levels, the judgment basis is R(X i )=R(X i )∩Q(Xi ), find the elements or variables corresponding to the columns with the value of 1 in the i-th row of the reachable matrix M to form a set called the reachable set R(X i ); Find the elements or variables corresponding to the row with column j as 1 in the reachable matrix M to form a set called the antecedent set Q(X i ). The elements of the solution variable set L1 must conform to R(X i )=R(X i )∩Q(X i ) This judgment condition; after obtaining the set L1, the rows and columns corresponding to the elements in the original M are removed, and a corresponding new reachable matrix is ​​formed. Similarly, for the matrix Need to be based on R(X i )=R(X i )∩Q(X i ) This judgment condition is used to find a new L2; then this step is repeated until all system elements of the system are divided into different solution sets to form a skeleton matrix structure;

[0037] (5) On the basis of the skeleton matrix structure, according to the interaction relationship between variables at the same or adjacent levels, and in accordance with the rule that high-level elements point to low-level elements, and high-level variables are above and low-level ones are below, lines are connected in sequence to draw a multi-level directed topological diagram, ignoring the relationship between cross-level variables, and finally an explanatory structural model is established.

[0038] A further technical solution of the present invention is as follows: the causal graph correction method introduced in step S5 is to correct the explanatory structure model of the factors affecting slope stability, so that there is a feedback loop between the levels of the explanatory structure model, and the main relationship between the parameters is not easily missed, and finally a subjective Bayesian network model of slope stability is established. The specific correction process is as follows:

[0039] (1) Check the conditional independence relationship between parameters: Based on the results of correlation analysis, check whether the influencing factors are conditionally independent, that is, whether they are correlated. If they are not conditionally independent, add cross-level dotted arrows in the figure to indicate the connection;

[0040] (2) Distinguish direct or indirect relationships between parameters: Based on the results of correlation analysis, check whether the influencing factors are strongly or weakly correlated. For strong correlation, add cross-level dotted lines to represent the influence relationship;

[0041] (3) Check the inversion of causal relationships between parameters; if there is inversion of causal relationships, the direction of the arrows of the connecting lines needs to be adjusted;

[0042] (4) Check for unreasonable directional connections; if there are obviously unreasonable directional connections, delete the connection;

[0043] (5) Delete the loop relationship.

[0044] A further technical solution of the present invention: the learning method for optimizing the Bayesian network structure in the step S5 is specifically to optimize the K2 algorithm based on the explanatory structural model of the factors affecting slope stability, and to obtain the Bayesian network model structure based on the optimized K2 algorithm, providing the K2 algorithm with the order of nodes and the possible maximum number of parent nodes: 7; the optimization of the K2 algorithm is specifically based on the sample data of the training set, and according to the directional connection between some Bayesian network nodes specified in advance, the search strategy is determined to be a greedy search algorithm; the search puts the parent node variable with the largest scoring function into the set; when the scoring function cannot be increased, the search is stopped, and a hybrid Bayesian network model structure is established; the specific calculation process is:

[0045] (1) Determine the scoring function and use P(G, D) as the scoring function. The formula is:

[0046]

[0047] The meanings of the variables in the scoring function are as follows: D-dataset; G-network model to be selected; n-number of variables; P(G)-prior probability distribution of network structure G; r i -Change node x i The number of values; i-node number; j-node x i The value number of the parent node; π(x i )-node x i The parent node of - π(x i ) value number, N ijk -Node x i The parent node π(x i ) takes the jth value, x i The number of samples when taking the kth value; N ij -Node x i Parent node set π(x i )’s number of samples corresponding to the j-th state combination;

[0048] (2) According to the node sorting of the Bayesian network model structure learning method and the maximum number of parent nodes allowed for each node is 7, determine the value of the node number i (1, 2, 3...11) and q i The value of

[0049] (3) Determine the partial π(x) based on the directional connections between the nodes of the Bayesian network specified in advance i ) value;

[0050] (4) Determine the search strategy as a greedy search algorithm; search and put the parent node variable with the largest score function into the set; when the score function cannot be increased, stop searching and find the optimal model;

[0051] (5) Based on the sample data of the training set, a hybrid Bayesian network model structure is established through the optimized K2 algorithm mentioned above.

[0052] A further technical solution of the present invention is as follows: In step S6, the expectation maximization method is used to learn the parameters of the Bayesian network model, and the conditional probability table is constructed using the inference calculation of the Bayesian network; the specific calculation of the parameter learning of the Bayesian network model using the expectation maximization algorithm is as follows:

[0053] If the parameter Θ is known, the value of the optimal latent variable Z can be inferred from the training data, that is, the E step; conversely, if the value of Z is known, the maximum likelihood estimation of the parameter Θ can be conveniently performed, that is, the M step;

[0054] Let X denote the set of observed variables, Z denote the set of latent variables, Θ denote the model parameters, and the initial value Θ 0 As a starting point, perform the following steps until convergence:

[0055] ①. Based on Θ t Infer the expectation of the latent variable Z, denoted as Z t ;

[0056] ②. Based on observed variables X and Z t Make a maximum likelihood estimate of the parameter Θ, denoted by Θ t+1 ; This is the prototype of the EM algorithm;

[0057] ③. Based on Θ t Calculate the probability distribution P(Z|X, Θ) of the latent variable Z t ), then the two steps of the EM algorithm are:

[0058] Step E (Expectation): With the current parameter Θ t Infer the latent variable distribution P(Z|X, Θ t ), and calculate the log-likelihood LL(Θ|X, Z) with respect to the expectation of Z

[0059] M step (Maximization): Find the parameters that maximize the expected likelihood Θ t+1 = argmax Q(Θ|Θ t );

[0060] The EM algorithm is calculated alternately in two steps: the first step is expectation, i.e., E step, which uses the current estimated parameter value to calculate the expected value of the log-likelihood; the second step is maximization step, i.e., M step, which searches for the parameter value that maximizes the expected likelihood generated by the E step, and then the newly obtained parameter value is reused in the E step, ..., until the increase in the likelihood function is less than the set threshold δ, i.e., |Θ t+1 -Θ t |≤δ, it is considered to have converged to the local optimal solution.

[0061] In step S6, the conditional probability table is constructed using the inference calculation of the Bayesian network. The specific calculation process is as follows: Assume that the safety factor of the reservoir bank slope calculated by the Bayesian network model is represented by the variable D, and the influencing factors are simplified to A, B, and C; A is the root node, D is the leaf node, A is the parent node of B and C, and B and C are the parent nodes of D respectively. The values ​​of the influencing factors A, B, and C are known. The calculation method for calculating the safety factor of the slope is:

[0062] P(D / A,B,C)=P(D)*P(A,B,C / D) / P(A,B,C),

[0063] At this time, A, B, and C are not independent of each other. Because the relationship between the three nodes A, B, and C is established in the Bayesian network model, the above relationship can be effectively analyzed in depth. At this time, another Markov chain rule formula needs to be introduced:

[0064] P(x1,...,x n )=P(x1)P(x2|x1)...P(x n |x1,x2,...,x n-1 )

[0065] where x1, ..., x n Represents the node variables of the Bayesian network,

[0066] The chain rule above indicates that, based on the sorting of variables, the joint probability distribution of each variable is equal to the product of the conditional probability of each variable. Therefore, for the above formula P(D|A, B, C)=P(D)*P(A, B, C|D) / P(A, B, C), further analysis can be done as follows:

[0067]

[0068] P(A,B,C|D)=P(A,B,C,D) / P(D)=P(A).P(B|A)P(C|A,B)P(D / A,B,C) / P(D)

[0069] In the above three formulas, the prior probability can be directly obtained from the sample library, while the conditional probability needs further analysis. At this time, according to the dependency relationship between the variables in the Bayesian network, the conditional independence formula shown below is introduced:

[0070]

[0071] Among them, π(x i ) is the variable x i The set of parent nodes; at this time, the conditional terms of each conditional probability in the above formula only retain its parent node terms and eliminate other terms. At this time, each conditional probability can be obtained from the node parameter CPT of the Bayesian network, thereby realizing the calculation of the formula.

[0072]

[0073] Thus we can get:

[0074]

[0075] A further technical solution of the present invention is as follows: The specific process of obtaining the probability curve of the safety factor level of the reservoir bank slope by applying the forward reasoning of the Bayesian network in step S7 is as follows:

[0076] (1) Extracting the reservoir bank slope evaluation index based on the prediction and analysis reservoir bank slope stability evaluation index system established in step S3; discretizing the evaluation index corresponding data with reference to the reservoir bank slope stability evaluation index classification standard in step S3;

[0077] (2) Assuming a single slope, the discretized evaluation index is input into the slope stability Bayesian network model in step S5 to obtain the initial grade probability table of each node;

[0078] (3) Limit the status of some nodes, adjust the node level of a single variable in turn, record the corresponding safety factor node level posterior probability release, and obtain the safety factor level probability curve of the reservoir bank slope with a single variable as the independent variable and the safety factor level probability as the dependent variable; at the same time, record the pore water pressure level probability change curve.

[0079] A further technical solution of the present invention is as follows: The specific process of obtaining the grade probability curve of the factors affecting slope stability by applying the reverse diagnostic reasoning of the Bayesian network in step S7 is as follows:

[0080] (1) Extracting the reservoir bank slope evaluation index based on the prediction and analysis reservoir bank slope stability evaluation index system established in step S3; discretizing the evaluation index corresponding data with reference to the reservoir bank slope stability evaluation index classification standard in step S3;

[0081] (2) Assuming a single slope, the discretized evaluation index is input into the slope stability Bayesian network model in step S5 to obtain the initial grade probability table of each node;

[0082] (3) The node levels of the slope safety factor are adjusted in sequence, and the probability of each level of the corresponding slope stability control factor node is recorded to obtain a level probability curve with the safety factor level as the independent variable and the slope stability influencing factor level probability as the dependent variable.

[0083] A further technical solution of the present invention is as follows: the specific process of selecting the actual influencing factor data of the reservoir bank slope in the step S8 to predict and analyze the stability of the slope is as follows: the actual influencing factor number of the selected reservoir bank slope is obtained through predictive analysis of the reservoir bank slope stability evaluation index system to obtain the slope influencing factor parameter grade distribution table; the parameters in the table are imported into the reservoir bank slope stability prediction model based on the Bayesian network to obtain the slope safety factor grade probability change curve, and the stability change trend of the slope under different working conditions is predicted by comparing the evaluation index system grading standard to obtain the slope influencing factor grade probability change curve; through the slope influencing factor grade probability change curve, the most sensitive influencing factor of the slope is analyzed by comparing the evaluation index system grading standard; through the slope influencing factor grade probability change curve, the evaluation index system grading standard is compared to obtain the influencing factor grade probability maximum node combination when the safety factor grade is in an unstable state; the stability change trend under different working conditions, the most sensitive influencing factor of the slope and the instability state influencing factor node combination are comprehensively evaluated to perform stability state prediction, analysis and evaluation, and targeted preventive measures are proposed.

[0084] The present invention is based on statistical reservoir bank slope stability analysis cases, collects and screens influencing factors to construct a slope stability evaluation index system; applies the method of improving the explanatory structural model, determines the hierarchical relationship and variable causal relationship, improves the Bayesian network model structure learning method, and combines the parameter learning method to construct a Bayesian network model for predicting the stability of the reservoir bank soil slope; applies the forward reasoning and reverse reasoning functions of the trained reservoir bank slope stability Bayesian network model to predict the stability of the reservoir bank slope under different working conditions and analyze the control factors of the instability and destruction of the reservoir bank slope, analyzes and predicts the stability change trend of the reservoir bank slope, and if hidden dangers are found, immediately takes corresponding measures to control them, provides technical support and safety guarantee for the long-term safe operation of reservoirs and tailings ponds, and is of great significance for local governments and decision makers to truly realize disaster prevention and mitigation and implement disaster prevention measures.

[0085] The present invention has the following outstanding technical advantages:

[0086] (1) The present invention combines the forward causal reasoning and reverse diagnostic reasoning capabilities of the Bayesian network model to conduct a comprehensive predictive analysis of the stability of the reservoir bank slope. Compared with the general model that only predicts the stability of the reservoir bank slope, the method system established by the present invention can find the most dangerous working condition before a reservoir bank slope becomes unstable, and make targeted preventive measures in combination with mechanism analysis.

[0087] (2) The present invention uses the slope safety factor Fs and the changes in the stress field and displacement field of the slope as evaluation indicators, and performs single-factor sensitivity analysis on the statistical influencing factors, thereby solving the problem of difficulty in screening relatively important and easy-to-obtain influencing factors for effectively discriminating the stability of the soil slope on the reservoir bank, and improving the accuracy and feasibility of the prediction model.

[0088] (3) The present invention applies K-means cluster analysis combined with engineering specifications to objectively establish a reservoir bank slope stability evaluation index system, and classifies the safety factor and 11 influencing factors into levels; compared with the general Bayesian network model node level division, the present invention uses the amount of collected data as the division standard, so that the marginal probability obtained by the conditional probability table reasoning after parameter learning is more objective and scientific.

[0089] (4) The present invention establishes an explanatory structural model by obtaining a mutual influence relationship matrix based on the results of correlation analysis between the influencing factors, and introduces a causal diagram method to increase the connection relationship between influencing factors across levels; thus, the shortcoming that the explanatory structural model cannot cross levels is solved, making the traditional explanatory structural model more complete and objective. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 It is a flow chart of the establishment of a prediction and analysis method system for reservoir bank soil slope stability based on Bayesian network of the present invention;

[0091] Figure 2 It is a hierarchical structure relationship diagram of various influencing factors in the embodiment of the present invention;

[0092] Figure 3 It is a schematic diagram of the subjective Bayesian network structure constructed by the present invention based on the causal graph method to improve the explanatory structure model;

[0093] Figure 4 It is a schematic diagram of a hybrid Bayesian network structure obtained by improving the K2 scoring search algorithm based on the subjective Bayesian network structure of the present invention;

[0094] Figure 5 It is a schematic representation of the conditional probability of each node after the Bayesian network parameters are learned in an embodiment of the present invention;

[0095] Figure 6-1 It is a schematic diagram of the initial condition probability of the safety factor level node under the water level rising condition in the embodiment of the present invention;

[0096] Figure 6-2 is a probability distribution curve of the safety factor node level under the water level rising condition in the embodiment of the present invention;

[0097] Figure 7-1 It is a schematic diagram of the initial condition probability of the safety factor level node under the water level drop condition in the embodiment of the present invention;

[0098] Figure 7-2 is a probability distribution curve of the safety factor node level under the water level drop condition in the embodiment of the present invention;

[0099] Figure 8 It is a schematic representation of the conditional probability of the node level of the influencing factor under different safety factors in the embodiment of the present invention;

[0100] Figure 9-1 is a probability distribution curve of the water level variation node level obtained by changing the safety factor node level in the embodiment of the present invention;

[0101] Figure 9-2 It is a probability distribution curve of the water level change rate node level obtained by changing the safety factor node level in the embodiment of the present invention. DETAILED DESCRIPTION

[0102] The present invention is described in detail below in conjunction with the accompanying drawings and specific implementation methods. It should be understood that the preferred embodiments described below are only used to explain the present invention and will not limit the protection scope of the present invention.

[0103] The specific process of obtaining the probability curve of the safety factor level of the reservoir bank slope by applying the forward reasoning of the Bayesian network in the embodiment of the present invention is as follows:

[0104] (1) Extracting the reservoir bank slope evaluation index based on the prediction and analysis reservoir bank slope stability evaluation index system established in step S3; discretizing the evaluation index corresponding data with reference to the reservoir bank slope stability evaluation index classification standard in step S3;

[0105] (2) Assuming a single slope, the discretized evaluation index is input into the slope stability Bayesian network model in step S5 to obtain the initial grade probability table of each node;

[0106] (3) Limit the status of some nodes, adjust the node level of a single variable in turn, record the corresponding safety factor node level posterior probability release, and obtain the safety factor level probability curve of the reservoir bank slope with a single variable as the independent variable and the safety factor level probability as the dependent variable; at the same time, record the pore water pressure level probability change curve.

[0107] The specific calculation principle of the probability curve of the safety factor level of the reservoir bank slope obtained by forward reasoning is as follows:

[0108] Assume that the safety factor of the reservoir bank slope calculated by the Bayesian network model is represented by the variable Y, the levels of the nodes are divided into Y = 1 and Y = 2, the probability of taking the value of 1 is P(Y = 1), and the probability of taking the value of 2 is P(Y = 2), then P(Y = 1) + P(Y = 2) = 1; the influencing factors are simplified to be represented by X1 and X2, and the node levels are divided into X1 = 1, X1 = 2, X2 = 1, and X2 = 2; the probability distribution calculation process of each level of the safety factor Y based on the conditional probability formula is as follows:

[0109] P(Y)=P(Y / X1=1, X2=1)P(X1=1 / X2=1)P(X2=1)+P(Y / X1=1, 2)+P(Y / X1=2, X2=1)P(X1=2 / X2=1)P(X2=1)+P(Y / X1=2,

[0110] But when the safety factor Y has m levels, the influencing factors are expanded to X1, X2, ..., X n And each node has n levels, X1, X2, ..., X n The joint probability table controlling the Y node level is as follows:

[0111]

[0112] The joint probability formula will become more complicated. At this time, another Markov chain rule formula can be introduced based on the conditional independence assumption:

[0113] P(x1,...,x n )=P(x1)P(x2|x1)...P(x n |x1,x2,...,x n-1 )

[0114] where x1, ..., x n Represents the node variables of the Bayesian network,

[0115] The above chain rule indicates that, based on the sorting of variables, the joint probability distribution of each variable is equal to the product of the conditional probability of each variable. Combining the above formula, the joint probability can be simplified, and the marginal probability of the safety factor node level can be determined by marginalization. The marginal probability of the safety factor node level can be predicted by the following formula:

[0116]

[0117] Where Y is the safety factor node variable, X1, X2, …, X nis the node variable of each influencing factor, P(Y) is the node level probability (prior probability) of the safety factor, P(Y|X i ) is the safety factor node level probability (posterior probability) under a certain state of a certain influencing factor.

[0118] The specific process of obtaining the grade probability curve of slope stability influencing factors by applying the reverse diagnostic reasoning of the Bayesian network in the embodiment of the present invention is as follows:

[0119] (1) Extracting the reservoir bank slope evaluation index based on the prediction and analysis reservoir bank slope stability evaluation index system established in step S3; discretizing the evaluation index corresponding data with reference to the reservoir bank slope stability evaluation index classification standard in step S3;

[0120] (2) Assuming a single slope, the discretized evaluation index is input into the slope stability Bayesian network model in step S5 to obtain the initial grade probability table of each node;

[0121] (3) The node levels of the slope safety factor are adjusted in sequence, and the probability of each level of the corresponding slope stability control factor node is recorded to obtain a level probability curve with the safety factor level as the independent variable and the slope stability influencing factor level probability as the dependent variable.

[0122] The specific calculation principle of the node level probability of the disaster factor calculated by the reverse diagnosis reasoning is as follows: Assume that the safety factor of the reservoir bank slope calculated by the Bayesian network model is represented by the variable Y, the node level is divided into Y=1 and Y=2, the probability of taking the value as 1 is P(Y=1), and the probability of taking the value as 2 is P(Y=2), then P(Y=1)+P(Y=2)=1; the influencing factors are simplified to be represented by X1 and X2, and the node levels are divided into X1=1, X1=2, X2=1, and X2=2;

[0123] According to Bayesian theorem, the probability formula for calculating each level of influencing factor X1 is as follows:

[0124]

[0125] Similarly, the probability of each level of the influencing factor X2 can also be calculated;

[0126] But when the safety factor Y has m levels, the influencing factors are expanded to X1, X2, ..., X n And each node has n states, X1, X2, ..., X n The joint probability of controlling the level of node Y is shown in the conditional probability table in right 8. According to the Bayesian theorem, when the probability of the safety factor Y = m is known, the influencing factor X can be diagnosed and analyzed. i The level probability of , the calculation process is as follows:

[0127]

[0128] Based on the total probability formula:

[0129]

[0130] Substituting in, we get:

[0131]

[0132] Where Y is the safety factor node variable, X1, X2, …, X n are the node variables of each influencing factor, x1, x2, …, x n is the node level of each influencing factor, P(Y=m|X n =x n ) is the probability that the safety factor node level is m when the level of influencing factors is determined.

[0133] The embodiment provides a Bayesian network-based reservoir bank soil slope stability prediction and analysis method, and its construction process is as follows: Figure 1 As shown, the specific steps include:

[0134] S1. Statistical analysis of the influencing factors of slope stability in the literature on reservoir bank slope stability analysis in recent years; the statistical influencing factors specifically include soil type, natural bulk density, cohesion, internal friction angle, permeability coefficient, elastic modulus of sliding soil, Poisson's ratio of sliding soil, pore water pressure, slope height, slope angle, calculation method, water level change rate and water level change, a total of 13 influencing factors.

[0135] S2. Use finite element numerical simulation software to model and calculate each influencing factor counted in step S1 for single factor analysis, screen the 13 influencing factors in step S1, determine the rationality of the selected factors, and eliminate the undesirable factors; finally, eliminate the elastic modulus of sliding soil and the Poisson's ratio of sliding soil from the list of influencing factors, retain 11 factors that have a significant impact on the stability of the reservoir bank slope as effective influencing factors, and collect the calculation results of the 11 effective influencing factors and the corresponding safety factors as a reservoir bank slope stability analysis engineering data sample library, and use 80% of the samples in the collected engineering data sample library for training and 20% of the samples for testing.

[0136] S3. The effective influencing factors and safety factors of slope stability in the machine learning sample library established in step S2 are classified based on K-menas cluster analysis, and the selected influencing factors and safety factor level interval division are revised in combination with engineering specifications to establish a slope stability evaluation index system for the prediction analysis library. The specific steps are as follows:

[0137] (1) Select k samples from the N evaluation index data sample library and initialize the k cluster centers (C1, C2, ..., Ck-1 , C k ) The cluster center is set by itself or randomly selected by the system;

[0138] (2) Calculate the Euclidean distance from each sample to each cluster center, compare the distance from each sample to each cluster center in turn, and assign the sample to the cluster with the closest cluster center to form k clusters (S1, S2, ..., S k-1 , S k );

[0139] (3) Update the cluster according to the following formula:

[0140]

[0141] Where: C i For cluster S i The central sample of; k is the number of clusters; d is the minimized square error of cluster division;

[0142] Recalculate the cluster center of the new k clusters. The calculation formula is as follows:

[0143]

[0144] Where: n i For cluster S i The number of samples included;

[0145] Select the initial C i Examine all data samples and divide the initial cluster S i , from the initial C i With S i Calculate the size of d. The smaller d is, the higher the similarity of samples in the cluster.

[0146] (4) Repeat steps (2) to (3) until the conditions are met; modify the node levels divided by the cluster analysis results in combination with existing engineering specifications to establish a prediction and analysis bank slope stability evaluation index system.

[0147] The influencing factors are classified according to the existing engineering specifications, as shown in Table 1 and Figure 2 As shown in Table 1, the classification table of factors affecting the stability of the soil slope of the reservoir bank is shown in Table 2. Similarly, the interval classification table of the safety factor is shown in Table 2:

[0148] Table 1 is a table of the classification of 11 factors affecting the stability of the reservoir bank slope

[0149]

[0150] Table 2 Safety factor classification table

[0151]

[0152] Based on the level division and interval division of the above-mentioned influencing factors and evaluation indicators, the case library is processed to form a sample library that meets the requirements of the Bayesian network model.

[0153] S4. Carry out a bivariate correlation analysis on the above safety factor and the effective influencing factors of slope stability, and use the Sig value, that is, the significant P value, to represent whether the correlation is significant. Based on the results of the correlation analysis, objectively obtain the hierarchical arrangement of the influencing factors. According to the functional relationship of variables in the same or adjacent levels, and in accordance with the rule that high-level elements point to low-level elements, high-level variables are above, and low-level variables are below, connect the lines in sequence to draw a multi-level directed topological diagram, ignoring the relationship between cross-level variables, and finally establish an explanatory structural model of slope stability influencing factors; the specific steps are as follows:

[0154] (1) Select the relevant variables that constitute the key issues affecting the system: In step S2, 11 influencing factors and safety factors suitable for predicting the stability of the reservoir bank slope have been determined and numbered to establish an adjacency matrix: X1 soil type, X2 natural bulk density, X3 slope height, X4 slope angle, X5 cohesion, X6 internal friction angle, X7 saturated permeability coefficient, X8 calculation method, X9 water level change, X10 water level change rate, X11 pore water pressure, and X12 safety factor Fs.

[0155] (2) The results of the correlation analysis between the 12 influencing factors are used to obtain the mutual influence relationship matrix, as shown in Table 3. In the table, A represents that the horizontal influencing factor has an influence on the vertical influencing factor, B represents that the vertical influencing factor has an influence on the horizontal influencing factor, and 0 represents that there is no influence between them:

[0156] Table 3 shows the results of the correlation analysis between the 12 influencing factors and the mutual influence relationship matrix.

[0157] <![CDATA[X1]]> <![CDATA[X2]]> <![CDATA[X3]]> <![CDATA[X4]]> <![CDATA[X5]]> <![CDATA[X6]]> <![CDATA[X7]]> <![CDATA[X8]]> <![CDATA[X9]]> <![CDATA[X 10 ]]> <![CDATA[X 11 ]]> <![CDATA[X 12 ]]> A O O A A A O O O O A <![CDATA[X1]]> B O A A O O O O O A <![CDATA[X2]]> O O O O O O O O A <![CDATA[X3]]> O O O O O O O A <![CDATA[X4]]> O O O B O O A <![CDATA[X5]]> O O B O O A <![CDATA[X6]]> O O O A A <![CDATA[X7]]> O O O A <![CDATA[X8]]> O A A <![CDATA[X9]]> A A <![CDATA[X 10 ]]> A <![CDATA[X 11 ]]> <![CDATA[X 12 ]]>

[0158] (3) Convert the causal relationship graph between variables into an adjacency matrix A:

[0159] A=(a ij )N×N, where N=13, The adjacency matrix obtained after transformation is:

[0160]

[0161] (4) Calculate the reachability matrix R of the system through matrix operations:

[0162] According to the Boolean matrix operation rules, let I be the unit matrix, and perform the power operation of matrix A+I. When A+I≠(A+I) 2 ≠......≠(A+I)n-1 =(A+I) n When , we get the reachable matrix R = (A + I) n ; The reachable matrix is ​​calculated as:

[0163]

[0164] (5) Perform regional decomposition and inter-level decomposition on the reachable matrix R: calculate the influencing factors X respectively i The reachable set R(X i ) and the antecedent set Q(X i ), when R(X i )∩Q(X i )=R(X i ), Si is the highest level element that meets the conditions at present. First, find the highest level elements that meet the conditions, remove them from the reachable matrix to get a new matrix, and then find the highest level elements that meet the conditions in the new matrix. This method is used to obtain the final hierarchical division of influencing factors layer by layer, as shown in Table 4. The hierarchical division of influencing factors according to the steps in the ISM principle is shown in Table 4. The final division result: the first layer L1 = {X 12}, the second layer L2 = {X3,X4,X5,X6,X8,X 11}, the third layer L3 = {X2,X7,X9,X 10}, the fourth layer L4 = {X1}. Finally, the hierarchical distribution of influencing factors is obtained.

[0165] Table 4 Classification of influencing factors

[0166]

[0167]

[0168] (6) Establish an explanatory structural model: Based on the reachability matrix hierarchical division table and the causal relationship between influencing factors, draw an explanatory structural model, as shown in the attached figure. Figure 2 As shown;

[0169] The selected influencing factors are divided into 4 levels, which affect each level from bottom to top. The top level represents the evaluation index of the reservoir bank slope stability, which is used to evaluate whether the slope instability occurs: the bottom level factor is the soil type, which means it is the most comprehensive and fundamental influencing factor. The change of soil type will directly lead to the change of the physical and mechanical parameters of the slope soil, and indirectly change the safety factor Fs of the slope through multiple influencing factors; the elements of the second level directly act on the evaluation index of the top level, which is the most direct influencing factor affecting the stability of the reservoir bank slope: the water level change, permeability coefficient and water level change rate of the third level point to the pore water pressure of the second level, which shows that the pore water pressure is affected by multiple influencing factors at the same time. The first layer is the indirect influencing factor, which affects the stability of the reservoir bank slope through the elements of the second layer. The slope height, slope angle and calculation method of the second layer have no directional relationship with the other three layers, which means that they are independent and directly affect the safety factor Fs, and can be directly determined in the calculation and analysis. The bottom layer is the most fundamental comprehensive influencing factor, which can directly or indirectly affect multiple influencing factors. It is usually the most initial discriminant factor and only affects the upper layer.

[0170] S5. The explanatory structure model of the factors affecting slope stability is modified so that there is a feedback loop between the levels of the explanatory structure model, making the hierarchical structure of the model more consistent with professional knowledge. The specific modification process is as follows:

[0171] (1) Check the conditional independence relationship between parameters; for example, soil type will affect the permeability coefficient, and the internal friction angle and cohesion will affect the safety factor; the water level change will also affect the cohesion and internal friction angle. It is necessary to add dotted arrows in the figure to indicate the connection.

[0172] (2) Distinguish between direct and indirect relationships between parameters; for example, the natural bulk density of soil can indirectly affect the safety factor, but it can also directly affect the safety factor; the soil type can indirectly affect the cohesion and internal friction angle, but it can also directly affect the cohesion and internal friction angle; the soil type can affect the matrix suction of unsaturated soil. The absolute value of the matrix suction of unsaturated soil is approximately equal to the absolute value of the pore water pressure, so the soil type can also directly affect the pore water pressure, etc. Therefore, new lines need to be added between them to indicate the influence relationship.

[0173] (3) Check the inversion of causal relationships between parameters; if there is a causal inversion, the direction of the arrows of the connecting lines needs to be adjusted. There is no causal inversion in this interpretation of the structural model.

[0174] (4) Check for unreasonable directional connections. If there are obviously unreasonable directional connections, delete the connections.

[0175] (5) Delete loop relationships. There are no loop relationships in this interpretation structure model.

[0176] The final directed acyclic graph is obtained by modifying the interpretation structure model using the causal graph method. In fact, it is also a more complete manual modeling of the subjective Bayesian network model of the factors affecting the stability of the reservoir bank slope. Figure 3 The hierarchical relationship in the subjective Bayesian network is combined with the K2 algorithm to provide the K2 structure learning algorithm with the order of nodes and the maximum possible number of parent nodes, and to limit the algorithm accordingly, reduce the search space, and improve the computational efficiency.

[0177] S6: According to the subjective Bayesian network model, the node order is determined, as shown in Table 4, which is a node order table; according to the subjective Bayesian network diagram, the maximum number of parent nodes allowed for each node is determined to be 7. In order to reduce the search space and improve the computational efficiency, the node structure is defined according to the optimized interpretation structure model, and a directed connection is made between the influencing factors, as shown in Table 5, which is a directed connection relationship table;

[0178] Table 4 is the node sequence table of 12 influencing factors

[0179]

[0180]

[0181] Table 5 Directivity connection relationship table

[0182] Directional connection relationship dag=1 dag=0 dag(1,[2,7]) √ dag(2,[5,6,12]) √ dag(7,11) √ dag(9,[5,6,11]) √ dag(10,[7,11]) √ dag(5,12) √ dag(6,[3,12]) √ dag(11,12) √ dag(3,12) √ dag(4,12) √ dag(8,12) √ dag(10,8) √ dag(9,8) √

[0183] In Table 5, dag represents a directed acyclic graph; dag = 1 means that there is a directional connection between nodes; dag = 0 means that there is no directional connection between nodes. For example, dag (1, [2, 7]) = 1 means that there is a connection between the sliding soil type (node ​​number 1) and the natural gravity (node ​​number 2) and the permeability coefficient (node ​​number 7), and dag [10, 8] = 0 means that there is no directional connection between the water level change rate (node ​​number 10) and the calculation method (node ​​number 8). Other node relationships that are not specified are determined by learning through database samples.

[0184] S7: Based on the optimized K2 algorithm, the Bayesian network model structure is obtained. The specific calculation process is as follows:

[0185] (1) Determine the scoring function and use P(G,D) as the scoring function. The formula is:

[0186]

[0187]

[0188] The meanings of the variables in the above scoring function are shown in Table 6:

[0189] Table 6 Explanation of the meaning of each variable in the scoring function

[0190] Variable Symbols Variable meaning D Dataset G Proposed network model n Number of variables P(G) Prior probability distribution of network structure G <![CDATA[r i ]]> <![CDATA[The number of values taken by variable node x i > i Node number j <![CDATA[Value number of the parent node of node x i > <![CDATA[π(x i )]]> <![CDATA[The parent node of node x i > <![CDATA[q i ]]> <![CDATA[π(x i )]]> <![CDATA[N ijk ]]> <![CDATA[The parent node π(x i ) of node x i takes the j-th value, and the sample quantity when x i takes the k-th value]]> <![CDATA[N ij ]]> <![CDATA[Node x i The number of samples corresponding to the j-th state combination of the set of parent nodes π(x i )

[0191] (2) According to step S6, the values ​​of the node numbers i (1, 2, 3...12) are determined as shown in Table 5, and the maximum value of qi is 7. i ) is shown in Table 6. Determine the search strategy as the greedy search algorithm; search and put the parent node variable with the largest score function into the set; when the score function cannot be increased, stop searching, and the optimal model can be found;

[0192] Based on the sample data of the training set, the hybrid Bayesian network model structure is established through the above optimized K2 algorithm, as shown in the attached Figure 4 shown.

[0193] In step S7, the expectation maximization method is used to learn the parameters of the Bayesian network model. The specific calculation method for learning the parameters of the Bayesian network model using the expectation maximization algorithm is:

[0194] The specific calculation method for parameter learning of the Bayesian network model using the expectation maximization algorithm is as follows: if the parameter Θ is known, the value of the optimal latent variable Z can be inferred from the training data (E step); conversely, if the value of Z is known, the maximum likelihood estimation of the parameter Θ can be conveniently performed (M step). Let X represent the set of observed variables, Z represent the set of latent variables, Θ represent the model parameters, and take the initial value Θ as 0 As a starting point, perform the following steps until convergence:

[0195] ①. Based on Θ t Infer the expectation of the latent variable Z, denoted as Z t ;

[0196] ②. Based on observed variables X and Z t Make a maximum likelihood estimate of the parameter Θ, denoted by Θ t+1 ; This is the prototype of the EM algorithm;

[0197] ③. Based on Θ t Calculate the probability distribution P(Z|X,Θ) of the latent variable Z t ), then the two steps of the EM algorithm are:

[0198] Step E (Expectation): With the current parameter Θ t Infer the latent variable distribution P(Z|X,Θt ), and calculate the log-likelihood LL(Θ|X,Z) with respect to the expectation of Z

[0199]

[0200] M step (Maximization): Find the parameters that maximize the expected likelihood

[0201] Θ t+1 = argmaxQ(ΘΘ t )

[0202] In simple terms, the EM algorithm is calculated alternately in two steps: the first step is the expectation (E) step, which uses the current estimated parameter value to calculate the expected value of the log-likelihood; the second step is the maximization step (M) step, which looks for the parameter value that maximizes the expected likelihood generated by the E step, and then the newly obtained parameter value is reused in the E step, ..., until the increase in the likelihood function is less than the set threshold δ, that is, |Θ t+1 -Θ t |≤δ, it is considered to have converged to the local optimal solution.

[0203] Through the above method, the parameter learning of the Bayesian network model is completed. Finally, after the database sample learning, the conditional probability table of the Bayesian network model is established to complete the reservoir bank slope stability prediction model based on the Bayesian network. Since the conditional probability table is too large, only the edge probability of each node is shown in the attached figure. Figure 5 shown.

[0204] S8. Substitute the test set sample data into the Bayesian network model, train and test the model, introduce the gradient descent method for parameter learning to establish a control Bayesian network model, and introduce logistic regression and random forest methods to establish a reservoir bank slope stability prediction model to verify the accuracy of the Bayesian network prediction model; the present invention is not limited to the comparison of the above-mentioned introduced parameter learning method and machine learning method, and other control models can also be introduced. The accuracy test control results are shown in Table 7. It can be seen that the prediction ability of the Bayesian network model is better than that of the logistic regression model and the random forest model, and in the Bayesian network model, the prediction ability of the maximum likelihood estimation method is slightly higher than the gradient descent method.

[0205] Table 7

[0206]

[0207] S9: Substitute the influencing factors obtained from the actual engineering survey into the Bayesian network model. In this example, a Three Gorges reservoir bank slope is taken, and the water level change is used as a variable to calculate the probability change of the safety factor level of the reservoir bank slope. Using the evaluation index grading standard in the established evaluation index system, the geometric factor level of the slope, the physical and mechanical parameter level of the landslide body, and the specific classification results of the formulated variable water level change level can be obtained as shown in Tables 8 and 9, where Table 8 is a table of the node parameter values ​​of each influencing factor and the variable level classification under the condition of rising water level, and Table 9 is a table of the node parameter values ​​of each influencing factor and the variable level classification under the condition of falling water level;

[0208] Table 8 Node parameter values ​​of various influencing factors and classification of influence variables under water level rise conditions

[0209]

[0210]

[0211] Table 9 shows the node parameter values ​​and variable level division of each influencing factor under the water level drop condition

[0212]

[0213] The node levels of the influencing factors in the table are imported into the reservoir bank slope stability prediction model based on the Bayesian network after training. The initial condition probability of the safety factor level node under the water level rising condition is expressed as follows: Figure 6-1 As shown in the figure, the initial condition probability of the safety factor level node under the water level drop condition is expressed as follows: Figure 7-1 As shown in the figure; the safety factor level probability change curve obtained by changing the water level change node level in sequence is shown in the figure. Figure 6-2 and Figure 7-2 The influencing factors and safety factor classification are shown in the previous table and will not be mentioned here. Figure 6-1 , Figure 6-2 , Figure 7-1 and Figure 7-2 The prediction results can predict the changing trend of the soil slope stability of the reservoir bank under this working condition.

[0214] S10: Substitute the values ​​of the influencing factors obtained from the actual engineering survey into the Bayesian network model. In this example, a slope on the bank of the Three Gorges Reservoir is taken, and the safety factor is used as a variable. The reverse diagnostic reasoning is used to calculate the probability changes of the levels of each influencing factor. By using the evaluation index grading standard in the established evaluation index system, the geometric factor level of the slope, the physical and mechanical parameter level of the landslide body, and the specific classification results of the safety factor level can be obtained as shown in Table 10.

[0215] Table 10

[0216]

[0217]

[0218] The node levels of the influencing factors in the table are imported into the reservoir bank slope stability prediction model based on the Bayesian network after training. The conditional probability table of the influencing factor nodes under different safety factors is as follows: Figure 8 As shown in the figure, the probability change curve of each influencing factor level is obtained by changing the safety factor node level in sequence. In this example, the water level change and water level change rate are taken as examples, as shown in the attached figure. Figure 9-1 and Figure 9-2 As shown in the figure, the most dangerous water level change condition can be obtained from the drawing according to the maximum principle, revealing the slope instability mechanism when the safety factor changes from stable to unstable step by step, as well as the node level combination of the most dangerous working condition influencing factors.

[0219] S11: Combine the most dangerous working condition prediction and mechanism analysis results of the reverse reasoning of step S10 with the forward reasoning results of step S9, the Bayesian network prediction model established in step S7, and the reservoir bank slope stability evaluation index system established in step S3, and finally establish a reservoir bank slope stability prediction analysis method system based on the Bayesian network model. For the reservoir bank slope of a certain Three Gorges reservoir area selected in the example, the reservoir bank slope stability prediction analysis method system based on the Bayesian network model is applied. It can be seen that under the selected working conditions, the safety factor level of the reservoir bank slope is always in a relatively stable level range, and the change of working conditions makes the safety factor tend to decrease; when the safety factor decreases from the stable level to the unstable level, the water level change rate is the most sensitive influencing factor, and the most dangerous working condition combination during instability is that the reservoir water level decreases at a rate of 1-2m / d, and the reservoir water level drops by more than 30m. The comprehensive evaluation shows that the slope is always relatively stable under common working conditions. Under extreme working conditions, when the water level drops suddenly and the reservoir water level drops by more than 30m, the slope of the reservoir bank will become unstable. Therefore, it is necessary to control the rate of change of the water level when the reservoir water drops to avoid a sudden drop in the water level.

[0220] In view of the complex and diverse factors affecting the stability of soil slopes on reservoir banks, unclear classification of evaluation indicators, large amount of calculation and low efficiency in calculating slope reliability analysis, and inability to prevent and control slope instability in a targeted manner, the present invention establishes a method system for predicting and analyzing the stability of soil slopes on reservoir banks based on a Bayesian network. If the data is sufficient, the method system can calculate the stability development trend of a single slope and the most dangerous working condition leading to instability before the slope becomes unstable, and has strong engineering applicability.

[0221] The preferred implementation modes of the present application are described in detail above in conjunction with the accompanying drawings. Typical known structures and common knowledge technologies in the preferred implementation modes are not described in detail here. Ordinary technicians in the relevant field can improve and implement the technical solutions of the present invention based on their own abilities under the inspiration given by the present implementation modes. Some typical known structures, known methods or common knowledge technologies should not become obstacles for ordinary technicians in the relevant field to implement the present application.

[0222] The scope of protection required by this application shall be based on the content of its claims, and the content recorded in the content of the invention, specific implementation methods and drawings of the specification shall be used to interpret the claims. Within the scope of the technical concept of this application, several modifications may be made to the specific implementation methods of this application, and these modified specific implementation methods shall also be deemed to be within the scope of protection of this application.

Claims

1. A prediction and analysis method for reservoir bank soil slope stability based on Bayesian network. It is characterized in that The following steps are involved: S1. Statistical analysis of the factors affecting slope stability in the literature on reservoir bank slope stability analysis in recent years; S2. Use finite element numerical simulation software to model and calculate each influencing factor counted in step S1 for single factor analysis, screen all influencing factors, and eliminate adverse influencing factors; collect reservoir bank slope stability analysis engineering data based on the remaining effective influencing factors, and establish a machine learning sample library including effective influencing factors of slope stability and safety factors; S3. The effective influencing factors and safety factors of slope stability in the machine learning sample library established in step S2 are classified based on K-menas cluster analysis, and the selected influencing factors and safety factor level interval division are revised in combination with engineering specifications to establish a prediction analysis library bank slope stability evaluation index system; S4. Conduct a bivariate correlation analysis on the safety factor and the effective influencing factors of slope stability in step S2, objectively obtain the hierarchical arrangement of the influencing factors based on the results of the correlation analysis, determine the directional connection between the layers, and establish an explanatory structural model of the influencing factors of slope stability; S5. Introduce the causal graph correction principle, correct the explanatory structural model of slope stability influencing factors established in step S4, add cross-level causal relationships, optimize the learning method of the Bayesian network structure, and construct a Bayesian network structural model of slope stability; S6: Based on the parameter learning method, the conditional probability table of the Bayesian network model of slope stability is constructed, and the reservoir bank slope stability prediction model based on the Bayesian network is established; S7: Based on the Bayesian network reasoning function, input the measured values ​​of the influencing factors of the reservoir bank slope, obtain the probability curve of the safety factor level of the reservoir bank slope by forward reasoning, and obtain the probability curve of the level of the slope stability influencing factors by reverse diagnostic reasoning. Combined with the prediction and analysis reservoir bank slope stability evaluation index system established in step S2 and the reservoir bank slope stability prediction model based on the Bayesian network established in step S6, a reservoir bank slope stability prediction and analysis method system is established; S8: Select the actual influencing factor data of the reservoir bank slope and predict and analyze the stability of the slope through the reservoir bank slope stability prediction and analysis method system established in step S7; Specifically, the actual number of influencing factors of the selected reservoir bank slope is graded through the prediction and analysis of the reservoir bank slope stability evaluation index system and then introduced into the reservoir bank slope stability prediction model based on the Bayesian network. The safety factor grade probability curve and the influencing factor grade probability curve are obtained by controlling the grade probability of the influencing factor node and the safety factor node. The safety factor change trend and the most sensitive influencing factor when the safety factor changes, as well as the node combination with the maximum probability of the edge of the influencing factor grade when the safety factor node is at the instability level are obtained from the curve. The stability of the slope is comprehensively predicted and analyzed from three aspects: the stability change trend, the most sensitive influencing factor and the most dangerous instability condition.

2. According to claim 1, a method for predicting and analyzing the stability of a reservoir bank slope based on a Bayesian network, Features: The influencing factors counted in step S1 include 13 influencing factors, namely, soil type, natural bulk density, cohesion, internal friction angle, permeability coefficient, elastic modulus of sliding soil, Poisson's ratio of sliding soil, pore water pressure, slope height, slope angle, calculation method, water level change rate and water level change amount; in step S2, numerical simulation calculation is performed with the slope safety factor and the evolution process of the stress field and displacement field of the slope under different influencing factors as evaluation criteria, and the 13 influencing factors are analyzed on a single factor basis to determine the rationality of the selected factors and eliminate the undesirable factors; finally, the elastic modulus of sliding soil and the Poisson's ratio of sliding soil are eliminated from the list of influencing factors, and 11 factors that have a significant impact on the stability of the reservoir bank slope are retained as effective influencing factors, and the calculation results of the 11 effective influencing factors and the corresponding safety factors are collected as a reservoir bank slope stability analysis engineering data sample library, and 80% of the samples of the collected engineering data sample library are used for training, and 20% of the samples are used for testing.

3. A method for predicting and analyzing the stability of a reservoir bank slope based on a Bayesian network according to claim 1 or 2, It is characterized in that In the step S3, the data in the machine learning sample library is objectively graded based on K-menas cluster analysis to classify the influencing factors, and the selected influencing factors and safety factor grade intervals are corrected in combination with engineering specifications. The specific steps for establishing the bank slope stability evaluation index system of the prediction analysis library are as follows: (1) Select k samples from the N evaluation index data sample library and initialize these k samples as cluster centers (C 1 , C 2 , …, C k-1 , C k ) The cluster center is set by itself or randomly selected by the system; (2) Calculate the Euclidean distance from each sample in the library to each cluster center, compare the distance from each sample to each cluster center in turn, and assign the sample to the cluster closest to the cluster center to form k clusters (S 1 , S 2 , …, S k-1 , S k ); (3) Update the cluster according to the following formula: Where: C i For cluster S i The central sample of; x is the sample point in the sample library; k is the number of clusters; d is the minimized square error of cluster division; Recalculate the cluster center of the new k clusters. The calculation formula is as follows: Where: n i For cluster S i The number of samples included; x is S i The sample points in ; Select the initial C i Examine all data samples and divide the initial cluster S i , from the initial C i With S i Calculate the size of d. The smaller d is, the higher the similarity of samples in the cluster. (4) Repeat steps (2) to (3) until the conditions are met; modify the node levels divided by the cluster analysis results in combination with existing engineering specifications to establish a prediction and analysis bank slope stability evaluation index system.

4. A method for predicting and analyzing reservoir bank slope stability based on Bayesian network according to claim 1 or 2, Its characteristics are: In step S4, a bivariate correlation analysis is performed on the safety factor and the effective influencing factors of slope stability, and the Sig value, i.e., the significance P value, is used to represent whether the correlation is significant. Based on the results of the correlation analysis, the hierarchical arrangement of the influencing factors is objectively obtained. According to the functional relationship of variables in the same or adjacent levels, and in accordance with the rule that high-level elements point to low-level elements, high-level variables are above, and low-level variables are below, lines are connected in sequence to draw a multi-level directed topological diagram, ignoring the relationship between cross-level variables, and finally an explanatory structural model of slope stability influencing factors is established.

5. A method for predicting and analyzing reservoir bank slope stability based on Bayesian network according to claim 1 or 2, Its characteristics are: The causal graph correction method introduced in step S5 is to correct the explanatory structure model of the factors affecting slope stability, so that there is a feedback loop between the levels of the explanatory structure model, and the main relationship between the parameters is not easily missed, and finally a subjective Bayesian network model of slope stability is established. The specific correction process is as follows: (1) Check the conditional independence relationship between parameters: Based on the results of correlation analysis, check whether the influencing factors are conditionally independent, that is, whether they are correlated. If they are not conditionally independent, add cross-level dotted arrows in the figure to indicate the connection; (2) Distinguish direct or indirect relationships between parameters: Based on the results of correlation analysis, check whether the influencing factors are strongly or weakly correlated. For strong correlation, add cross-level dotted lines to represent the influence relationship; (3) Check for inversion of causal relationships between parameters; if there is inversion of causal relationships, the arrow direction of the connecting line needs to be adjusted; (4) Check for unreasonable directional connections; if there are obviously unreasonable directional connections, delete the connection; (5) Delete the loop relationship.

6. A method for predicting and analyzing reservoir bank slope stability based on Bayesian network according to claim 1 or 2, Its characteristics are: The learning method for optimizing the Bayesian network structure in the S5 step is specifically to optimize the K2 algorithm based on the explanatory structural model of the factors affecting slope stability, and based on the optimized K2 algorithm, obtain the Bayesian network model structure of slope stability, and provide the K2 algorithm with the order of nodes and the possible maximum number of parent nodes: 7; the optimization of the K2 algorithm is specifically based on the sample data of the training set, and according to the directional connection between some Bayesian network nodes specified in advance, the search strategy is determined to be a greedy search algorithm; the search puts the parent node variable with the largest scoring function into the set; when the scoring function cannot be increased, stop the search and establish a hybrid Bayesian network model structure.

7. A method for predicting and analyzing the stability of a reservoir bank slope based on a Bayesian network according to claim 1 or 2, Features: In step S6, the expectation maximization method is used to learn the parameters of the Bayesian network model, and the conditional probability table is constructed using the inference calculation of the Bayesian network. The specific calculation of the parameter learning of the Bayesian network model using the expectation maximization algorithm is as follows: If the parameter Θ is known, the value of the optimal latent variable Z can be inferred from the training data, that is, the E step; conversely, if the value of Z is known, the maximum likelihood estimation of the parameter Θ can be conveniently performed, that is, the M step; Let X denote the set of observed variables, Z denote the set of latent variables, Θ denote the model parameters, and the initial value Θ 0 As a starting point, perform the following steps until convergence: ①. Based on Θ t Infer the expectation of the latent variable Z, denoted as Z t ; ②. Based on observed variables X and Z t Make a maximum likelihood estimate of the parameter Θ, denoted by Θ t+1 ; This is the prototype of the EM algorithm; ③. Based on Θ t Calculate the probability distribution P(Z|X, Θ) of the latent variable Z t ), then the two steps of the EM algorithm are: Step E (Expectation): With the current parameter Θ t Infer the latent variable distribution P(Z|X, Θ t ), and calculate the log-likelihood LL(Θ|X, Z) with respect to the expectation of Z M step (Maximization): Find the parameters that maximize the expected likelihood Θ t+1 = arg max Q(Θ|Θ t ); The EM algorithm is calculated alternately in two steps: the first step is expectation, i.e., E step, which uses the current estimated parameter value to calculate the expected value of the log-likelihood; the second step is maximization step, i.e., M step, which searches for the parameter value that maximizes the expected likelihood generated by the E step, and then the newly obtained parameter value is reused in the E step, ..., until the increase in the likelihood function is less than the set threshold δ, i.e., |Θ t+1 -Θ t |≤δ, it is considered to have converged to the local optimal solution.

8. A method for predicting and analyzing reservoir bank slope stability based on Bayesian network according to claim 1 or 2, Its characteristics are: The specific process of obtaining the probability curve of the safety factor grade of the reservoir bank slope by applying the forward reasoning of the Bayesian network in step S7 is as follows: (1) Extracting the reservoir bank slope evaluation index based on the prediction and analysis reservoir bank slope stability evaluation index system established in step S3; discretizing the evaluation index corresponding data with reference to the reservoir bank slope stability evaluation index classification standard in step S3; (2) Assuming a single slope, the discretized evaluation index is input into the slope stability Bayesian network model in step S5 to obtain the initial grade probability table of each node; (3) Limit the status of some nodes, adjust the node level of a single variable in turn, record the corresponding safety factor node level posterior probability release, and obtain the safety factor level probability curve of the reservoir bank slope with a single variable as the independent variable and the safety factor level probability as the dependent variable; at the same time, record the pore water pressure level probability change curve.

9. A method for predicting and analyzing reservoir bank slope stability based on Bayesian network according to claim 1 or 2, Its characteristics are: The specific process of obtaining the grade probability curve of slope stability influencing factors by applying the reverse diagnostic reasoning of the Bayesian network in step S7 is as follows: (1) Extracting the reservoir bank slope evaluation index based on the prediction and analysis reservoir bank slope stability evaluation index system established in step S3; discretizing the evaluation index corresponding data with reference to the reservoir bank slope stability evaluation index classification standard in step S3; (2) Assuming a single slope, the discretized evaluation index is input into the slope stability Bayesian network model in step S5 to obtain the initial grade probability table of each node; (3) The node levels of the slope safety factor are adjusted in sequence, and the probability of each level of the corresponding slope stability control factor node is recorded to obtain a level probability curve with the safety factor level as the independent variable and the slope stability influencing factor level probability as the dependent variable.

10. A method for predicting and analyzing reservoir bank slope stability based on Bayesian network according to claim 3, Its characteristics are: The specific process of selecting the actual influencing factor data of the reservoir bank slope in the step S8 to predict and analyze the stability of the slope is as follows: the actual influencing factor number of the selected reservoir bank slope is obtained through the predictive analysis of the reservoir bank slope stability evaluation index system to obtain the parameter grade distribution table of the slope influencing factor; the parameters in the table are imported into the reservoir bank slope stability prediction model based on the Bayesian network to obtain the probability change curve of the slope safety factor grade, and the stability change trend of the slope under different working conditions is predicted by comparing with the evaluation index system grading standard to obtain the slope influencing factor grade probability change curve; through the slope influencing factor grade probability change curve, the most sensitive influencing factor of the slope is analyzed by comparing with the evaluation index system grading standard; through the slope influencing factor grade probability change curve, the evaluation index system grading standard is compared to obtain the node combination with the maximum probability of the influencing factor grade when the safety factor grade is in an unstable state; the stability change trend under different working conditions, the most sensitive influencing factor of the slope and the node combination of instability state influencing factors are comprehensively considered to perform stability state prediction, analysis and evaluation of the slope, and targeted preventive measures are proposed.

Citation Information

Cited By

  • Deep and long tunnel water burst prediction method based on Bayesian network

    CN115049124A

  • Power transmission line slope stability early warning method and system for short-time heavy rainfall

    CN119830050A

  • Metal chip detection system and method for narrow space, wide drainage basin and high signal-to-noise ratio

    CN120539026A

  • Basic BN model of temporary sliding displacement, prediction model construction method and prediction method

    CN120782061A

  • Method and device for analyzing instability risk of bedding slope containing weak intercalated layer and medium

    CN121189040A