A method, device and medium for analyzing the instability risk of a bedding slope containing a weak interlayer

Abstract

The present application relates to the technical field of engineering slope instability risk analysis, and particularly relates to a soft interlayer containing bedding slope instability risk analysis method, equipment and medium. The technical scheme of the present application converts the complex and uncertain slope engineering problem into a computable and inferable probability graph model. The whole life cycle disaster control factors from natural conditions, geological conditions, engineering design, construction and operation maintenance are systematically sorted out, the omission of key factors is avoided, and the comprehensiveness of risk analysis is ensured. The traditional qualitative causal analysis is upgraded to quantitative probability dependence relationship, which provides a mathematical model basis for subsequent risk diagnosis and prediction. The risk analysis has realized the leap from "static evaluation" to "dynamic reasoning". The model can not only give a static risk value, but also dynamically update the probability of all related variables after obtaining new evidence (such as monitoring data), and realize intelligent diagnosis and prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of engineering slope instability risk analysis, and particularly relates to a soft interlayer-containing bedding slope instability risk analysis method, device and medium. BACKGROUND

[0002] A bedding slope is formed when engineering construction excavates a stratified sediment structure slope, and the rock layer interface and the slope direction are close, which is prone to sliding instability. Rainfall is one of the typical inducements for geological disasters of the bedding slope. Existing researches believe that the influence of heavy rainfall on the stability of the bedding slope is mainly the effect of surface water infiltration on increasing the slope body gravity, deteriorating the rock-soil strength and the permeability, especially for the soft interlayer-containing bedding slope, the water-induced deterioration of the soft interlayer is an important inducement for instability and sliding. The soft interlayer refers to the soft and weak structural plane or weak zone in the rock mass with a certain thickness. The causes include primary deposition, volcanic debris, sedimentary metamorphism, interlayer dislocation and fracture, secondary filling and groundwater cementation, etc. Obviously, the soft interlayer-containing bedding slope needs to focus on the influence of rainfall and the soft interlayer on the stability.

[0003] However, the vegetation coverage of the slope in a rainy mountainous area is relatively good. After the surface heavy rainfall, the surface runoff will have a significant scouring and dragging effect on the vegetation, especially when the vegetation coverage of grass and shrubs is relatively high, the surface runoff will have a strong downslope load on the slope surface, thereby increasing the sliding force of the bedding slope. This is less concerned in the existing researches.

[0004] The Bayesian network can process complex dependency relationships based on the dependency relationships and probability analysis among variables, and can support the risk analysis of the soft interlayer-containing bedding slope, so as to provide technical support for the decision-making of the engineering construction party and the operation party in the evaluation, diagnosis and prediction of the slope risk.

[0005] Therefore, for the soft interlayer-containing bedding slope, considering the two important natural conditions of rainfall intensity and vegetation coverage, and combining the geological conditions, design protection, construction time sequence and operation maintenance information of the soft interlayer-containing bedding slope which is prone to instability and sliding, a soft interlayer-containing bedding slope instability risk analysis model based on the Bayesian network is established, which has important significance for the instability cause diagnosis and instability risk prediction of the slope engineering construction and operation in a rainy mountainous area. SUMMARY

[0006] The purpose of the present application is to solve the problem that the existing stability analysis of the soft interlayer-containing bedding slope does not consider the surface runoff effect, and to establish a soft interlayer-containing bedding slope instability risk analysis model based on the Bayesian network. The model can diagnose the causes of slope instability, and can also predict the slope instability probability under given conditions, which can provide technical support for the decision-making of the engineering construction party and the operation party in the diagnosis and evaluation of the slope risk.

[0007] In a first aspect, the present application provides a method for analyzing the risk of instability of a bedding slope containing a weak interlayer, comprising the following steps:

[0008] Step 1, identifying a plurality of disaster control factors related to the risk of instability of a bedding slope containing a weak interlayer, and dividing the plurality of disaster control factors into root nodes, intermediate nodes, and taking the slope instability as a target node;

[0009] Step 2, establishing the causal relationship between all nodes, expressing through directed edges, and constructing a directed graph structure reflecting the dependency relationship between variables; based on the directed graph structure, a Bayesian network-based risk analysis model for a bedding slope containing a weak interlayer is constructed;

[0010] Step 3, defining the prior probability of each root node based on historical data, national standards or engineer experience;

[0011] Step 4, estimating the conditional probability of the intermediate nodes and the target node using Bayesian neural network, and generating a corresponding conditional probability table based on each intermediate node;

[0012] Step 5, based on the risk analysis model for the bedding slope containing a weak interlayer, the prior probability and the conditional probability table, the cause diagnosis or prediction analysis of slope instability is carried out.

[0013] The technical scheme of the present application converts the complex and uncertain slope engineering problem into a calculable and inferable probability graph model. The whole life cycle disaster-causing factors from natural conditions, geological conditions, engineering design, construction and operation maintenance are systematically sorted out, avoiding the omission of key factors and ensuring the comprehensiveness of risk analysis. The traditional qualitative causal analysis is upgraded to quantitative probability dependency relationship, providing a mathematical model basis for subsequent risk diagnosis and prediction. It realizes the leap from "static evaluation" to "dynamic reasoning" in risk analysis. The model not only gives a static risk value, but also dynamically updates the probability of all related variables after obtaining new evidence (such as monitoring data), realizing intelligent diagnosis and prediction.

[0014] Preferably, in step 1, the root nodes include five categories of natural conditions, geological conditions, design protection, construction sequence and operation maintenance; wherein the natural conditions include rainfall intensity and vegetation coverage; the geological conditions include slope height, slope length, slope and weak interlayer thickness; the design protection includes safety factor and protection structure type; the construction sequence includes construction season and protection sequence; the operation maintenance includes monitoring system effectiveness, patrol frequency and drainage smoothness.

[0015] In the scheme, the root node is explicitly defined as 13 specific variables in five categories, so that the risk analysis of different engineering projects can be based on unified standards, ensuring the comparability and repeatability of the analysis results. The operation and maintenance nodes are included, which extends the view of risk analysis from the traditional survey and design stage to the entire operation cycle of the slope, captures the key influence of later management and maintenance on slope stability, and realizes the whole life cycle risk management.

[0016] Preferably, the intermediate nodes include surface runoff, underground seepage, soft interlayer state, and free face stability, totaling four. The target node is slope instability.

[0017] In the scheme, the intermediate nodes and target nodes are explicitly defined. The technical effect lies in accurately depicting the typical failure path and core mechanical process of the soft interlayer containing bedding slope instability. The intermediate nodes such as "surface runoff erosion", "underground seepage", "soft interlayer state" are set, and their physical mechanisms are clear, which can simulate the two main effects of rainfall leading to instability: the increasing effect of surface runoff on the sliding force, and the deterioration effect of underground water on the shear strength of soft interlayer. This directly solves the problem of insufficient attention to the effect of surface runoff mentioned in the background technology. Taking slope instability as the only target node makes the model output focused and clear, which is convenient for engineering decision-making.

[0018] Preferably, the directed edge table between the root node and the intermediate node expresses:

[0019] The rainfall intensity points to the surface runoff erosion, the underground seepage and the free face stability;

[0020] The vegetation coverage points to the surface runoff erosion and the underground seepage;

[0021] The slope height, slope and protection timing point to the free face stability;

[0022] The slope length points to the surface runoff erosion;

[0023] The soft interlayer thickness points to the soft interlayer state;

[0024] The construction season points to the underground seepage and the surface runoff erosion;

[0025] The drainage unobstructedness points to surface runoff and underground seepage, indirectly affecting the soft interlayer state through underground seepage;

[0026] The protection structure type points to the soft interlayer state and the free face stability.

[0027] In this scheme, the abstract causal relationship is further concretized as the structure of the model by limiting the connection relationship of the directed edge between the root node and the intermediate node, ensuring the logical correctness of the network topology. Among them, by clearly pointing out that the rainfall intensity is directed to both surface runoff and underground seepage, the two different migration paths of water in the slope and their coexistence phenomenon under different rainfall intensities are accurately simulated. By clearly pointing out that the type of protective structure is directed to the state of soft interlayer and the stability of the free surface, the dual reinforcement mechanism of the engineering measures on the slope stability (i.e. direct reinforcement of rock mass and inhibition of soft interlayer deterioration) is quantified, so that the model can scientifically evaluate the effectiveness of different engineering schemes.

[0028] Preferably, the step of estimating the conditional probability of the intermediate node using the Bayesian neural network comprises:

[0029] Step 41, extracting a sample data set of the parent node and the child node from historical data;

[0030] Step 42, defining prior constraints on the state relationship of the node in combination with expert experience;

[0031] Step 43, training a Bayesian neural network model with the parent node as the feature and the child node as the label;

[0032] Step 44, predicting the state probability of the child node for each parent node state combination, and generating a conditional probability table.

[0033] This scheme further solves the problem of determining CPT in traditional methods. Make full use of existing data and engineering cases, and let the data-driven model learn. At the same time, the valuable experience of the field experts is integrated into the model in the form of computable probability constraints, solving the "black box" and unreasonable problems caused by data scarcity in pure data-driven methods in geotechnical engineering, and enhancing the credibility of the model. The Bayesian network automatically captures the complex and nonlinear interaction between the parent node and the child node, which cannot be achieved by manually compiling CPT or using simple regression methods. The final generated conditional probability table is a high-quality probability query table that meets both data rules and physical common sense, laying a solid foundation for accurate reasoning of the entire Bayesian network.

[0034] Preferably, the definition of the prior constraint is according to the following steps:

[0035] S1, identify the dominant parent node; S2, define extreme or typical working conditions; S3, probability limit assignment; S4, machine learning iteration feedback correction to obtain the final prior constraint.

[0036] For example: the prior constraint of the soft interlayer state is:

[0037] When the drainage is unobstructed and the rainfall intensity I is less than the surface infiltration rate Ir, the probability of soft interlayer softening should be less than 5%.

[0038] When the rainfall intensity I is greater than the surface infiltration rate Ir, the probability that the soft interlayer is dry should be less than 10%.

[0039] When the rainfall intensity I is greater than the surface infiltration rate Ir, the probability that the soft interlayer is dry should be less than 10%.

[0040] Preferably, the conditional probability table includes a conditional probability table of surface runoff scouring, a conditional probability table of underground seepage, a conditional probability table of the state of the soft interlayer, and a conditional probability table of the stability of the free face.

[0041] Preferably, in step 5, the diagnosis or prediction analysis of the slope instability risk includes: for the slope that has occurred instability, the diagnosis of the control disaster factor is performed; for the slope that has not occurred instability, the prediction of the instability probability under given conditions is performed.

[0042] When the slope instability occurs, the most possible key control disaster factor can be quickly located, a scientific basis is provided for rescue treatment and responsibility definition, and intelligent post-event diagnosis is realized. In the slope design or operation stage, the instability probability under different scenarios (such as different rainfall and different support schemes) can be quantitatively evaluated, and pre-event risk prediction is realized.

[0043] In a second aspect, the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the above-mentioned slope instability risk analysis method for the bedding slope containing a soft interlayer when executing the program.

[0044] In a third aspect, the present application provides a computer-readable storage medium, which stores a computer program, wherein the program is executed by a processor to implement the above-mentioned slope instability risk analysis method for the bedding slope containing a soft interlayer.

[0045] In the present solution, the complex and multi-step risk analysis method is solidified into a computer program, and data analysis is completed in a relatively short time. The problems of low efficiency and long cycle of manual analysis are solved, so that immediate response can be made in emergency rescue, scheme comparison and other scenes, and the timeliness of engineering decision-making is greatly improved.

[0046] The inherent defects of strong subjectivity and poor consistency in the traditional evaluation depending on expert experience are overcome. This makes the risk evaluation result no longer vary from person to person, but a standardized product that can be repeated and verified, and improves the public credibility and comparability of the result.

[0047] Compared with the prior art, the present application has the following beneficial effects:

[0048] The technical scheme of the present application converts the complex and uncertain slope engineering problem into a calculable and inferable probability graph model. The whole life cycle disaster control factors from natural conditions, engineering design to construction operation are systematically combed, avoiding the omission of key factors and ensuring the comprehensiveness of risk analysis. The traditional qualitative causal analysis is upgraded to quantitative probability dependence relationship, providing a mathematical model basis for subsequent risk diagnosis and prediction. The risk analysis has realized the leap from "static evaluation" to "dynamic reasoning". The model can not only give a static risk value, but also dynamically update the probability of all related variables after obtaining new evidence (such as monitoring data), realizing intelligent diagnosis and prediction. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 The risk analysis method flowchart of the present application.

[0050] Figure 2 The directed edge table expression graph of the root node and the intermediate node taking rainfall intensity as an example.

[0051] Figure 3 The risk analysis model graph of the present application based on the Bayesian network for the bedding slope with soft interlayer.

[0052] Figure 4 The conditional probability estimation method flowchart of the present application based on the Bayesian network. DETAILED DESCRIPTION

[0053] The present application will be further described in detail below in combination with specific embodiments. However, it should not be understood that the above-mentioned subject matter of the present application is limited to the following embodiments only, and any technology realized based on the content of the present application falls within the scope of the present application.

[0054] Example 1

[0055] The present embodiment discloses a risk analysis method for the bedding slope with soft interlayer based on the Bayesian network, which combines the risk analysis method for the bedding slope with soft interlayer and the Bayesian network. Figure 1 As shown in the figure, the method specifically comprises the following steps:

[0056] Step 1, identifying a plurality of disaster control factors related to the risk of the bedding slope with soft interlayer, and dividing the plurality of disaster control factors into root nodes and intermediate nodes, and taking the slope instability as a target node;

[0057] Among them, the root node represents the node without parent node in the network, that is, without any arrow pointing to it. It is the starting point or source of the causal chain. The probability of the root node does not depend on any other variable in the network, so its probability table is a prior probability.

[0058] In this embodiment, five types of key disaster control factors are selected as root nodes based on engineering experience from natural conditions, geological conditions, design protection, construction sequence, and operation and maintenance. Among them, the natural conditions include two root nodes of rainfall intensity and vegetation coverage; the geological conditions include four root nodes of slope height, slope length, slope gradient, and soft interlayer thickness; the design protection includes two root nodes of safety factor and protection structure type; the construction sequence includes two root nodes of construction season and protection sequence; the operation and maintenance includes three root nodes of monitoring system effectiveness, patrol frequency, and drainage smoothness; and the total number of root nodes is 13.

[0059] An intermediate node is a node that has both parent nodes (pointed to it by arrows) and child nodes (it points to other nodes). It connects the upstream and downstream in the causal chain. The probability of an intermediate node depends entirely on the state of its parent nodes, described by its conditional probability table, i.e., P(node | parents). Intermediate nodes are the hubs of information transmission and probability calculation. They transmit the influence of root nodes down and eventually affect the target nodes. They are the core of the network for complex reasoning.

[0060] In this embodiment, the intermediate nodes include four types of surface runoff, underground seepage, soft interlayer state, and free face stability.

[0061] A target node is the node that the invention ultimately cares about and wants to predict or infer. It usually has no child nodes (i.e., it is a leaf node), but not absolutely. Any node can become a "target node" in a specific problem. In this embodiment, the target node is slope instability.

[0062] The above root nodes, intermediate nodes, and target nodes are connected by directed edges, forming a chain of causal paths. The influence strength is quantified by the conditional probability table.

[0063] Step 2, establish the causal relationship between all nodes, express it through directed edges, and build a directed graph structure reflecting the dependency relationship between variables; based on the directed graph structure, build a soft interlayer bedding slope risk analysis model based on Bayesian network;

[0064] Among them, the directed edge represents the direct causal relationship or probability dependency relationship between nodes, and the directed edge is represented by an arrow. The arrow direction indicates that the "cause" points to the "effect". In this embodiment, taking rainfall intensity as an example, the surface infiltration rate is introduced as a reference. When the rainfall intensity ≥ the surface infiltration rate, surface runoff erosion and surface water infiltration occur at the same time; when the rainfall intensity < the surface infiltration rate, only underground seepage occurs.

[0065] Therefore, the directed edge expression of the rainfall intensity (root node) as a parent node is:

[0066] The directed edges of the intermediate node as the parent node are expressed as: rainfall intensity→surface runoff, rainfall intensity→underground seepage, rainfall intensity→open surface stability, and the directed edges of the rainfall intensity as the parent node are expressed as: rainfall intensity→surface runoff, rainfall intensity→underground seepage, rainfall intensity→open surface stability. Figure 2 .

[0067] The directed edges of the root node as the parent node are expressed as: vegetation coverage→surface runoff erosion, vegetation coverage→underground seepage.

[0068] The directed edges of the intermediate node as the parent node are expressed as: vegetation coverage→surface runoff erosion, vegetation coverage→underground seepage.

[0069] The directed edges of the root node as the parent node are expressed as: slope height, slope length, slope gradient, and weak interlayer thickness.

[0070] The directed edges of the intermediate node as the parent node are expressed as: slope height→open surface stability, slope length→surface runoff erosion, slope gradient→open surface stability, and weak interlayer thickness→weak interlayer state.

[0071] The directed edges of the root node as the parent node are expressed as: safety factor and protection structure type.

[0072] The directed edges of the intermediate node as the parent node are expressed as: safety factor→slope instability, protection structure type→weak interlayer state, and protection structure type→open surface stability.

[0073] The directed edges of the root node as the parent node are expressed as: construction season and protection timing.

[0074] The directed edges of the intermediate node as the parent node are expressed as: construction season→underground seepage, construction season→surface runoff erosion, and protection timing→open surface stability.

[0075] The root node is: monitoring system effectiveness, patrol frequency, and drainage unobstructedness.

[0076] The directed edges of the intermediate node as the parent node are expressed as: monitoring system effectiveness→surface runoff, monitoring system effectiveness→open surface stability.

[0077] The directed edges of the intermediate node as the parent node are expressed as: patrol frequency→surface runoff, patrol frequency→open surface stability.

[0078] The directed edges of the intermediate node as the parent node are expressed as: drainage unobstructedness→surface runoff, drainage unobstructedness→underground seepage.

[0079] The directed edges of the intermediate node as the parent node are expressed as:

[0080] The directed edges of the intermediate node as the parent node are expressed as: underground seepage→weak interlayer state, and surface runoff→open surface stability.

[0081] The directed edges of the intermediate node as the parent node are expressed as: rainfall intensity→surface runoff erosion, rainfall intensity→underground seepage, and rainfall intensity→open surface stability.

[0082] Wherein, the underground seepage will automatically find the seepage channel in the slope body, and the cracks in the slope body will become the water passage, and the soft interlayer will have the water isolation effect because it is mainly composed of fine particles, so that the water in the slope body will finally gather at the top of the soft interlayer.

[0083] When the water content of the soft interlayer continuously increases to the near-saturation-saturation state, the soft interlayer will be mudified, and the strength will be exponentially reduced, which greatly weakens the resistance to the sliding of the upper slope to the slope toe, and when the anti-sliding force is weakened to the critical state, the bedding slope will deform and then slide.

[0084] The surface runoff erosion will change the geometric state of the slope surface, on the one hand, the water drag on the plant roots will increase the sliding load, and on the other hand, the gully formed by the runoff will destroy the superficial stability, especially near the free face, after the surface runoff erosion, the superficial sliding often occurs, which is one of the reasons why the slope is easy to be unstable after heavy rain. In fact, the instability and damage of the slope after heavy rain is mainly small-scale superficial damage.

[0085] The vegetation coverage is directed to the surface runoff erosion and the underground seepage;

[0086] The slope height, slope and protection timing are directed to the free face stability;

[0087] The slope length is directed to the surface runoff erosion;

[0088] The soft interlayer thickness is directed to the soft interlayer state;

[0089] The construction season is directed to the underground seepage and the surface runoff erosion;

[0090] The drainage unobstructedness is directed to the surface runoff and the underground seepage, and indirectly affects the soft interlayer state through the underground seepage;

[0091] The protection structure type is directed to the soft interlayer state and the free face stability.

[0092] As shown in Figure 3 It is a schematic diagram of the Bayesian network.

[0093] Step 3, based on historical data, national specifications and / or engineer experience, define the prior probability of each root node to form a prior probability table.

[0094] Assign a probability value to each different state of each root node, and the sum of the probabilities of all states of the same root node is 1.

[0095] The state and prior probability of the root node of the natural condition class are defined as:

[0096] Rainfall intensity node: the prior probability of state "I ≥ Ir" is 0.4, and the prior probability of state "I < Ir" is 0.6;

[0097] wherein I is the rainfall intensity, and Ir is the surface infiltration rate;

[0098] Vegetation coverage node: the prior probability of state "FVC ≥ 50%" is 0.5, and the prior probability of state "FVC < 50%" is 0.5.

[0099] wherein the state and prior probability of the geological condition class root node are defined as:

[0100] Slope height node: the prior probability of state "H ≥ 20m" is 0.5, and the prior probability of state "H < 20m" is 0.5;

[0101] Slope length node: the prior probability of state "L ≥ 100m" is 0.4, and the prior probability of state "L < 100m" is 0.6;

[0102] Slope gradient node: the prior probability of state "α ≤ 30°" is 0.4, the prior probability of state "30° < α ≤ 60°" is 0.4, and the prior probability of state "α > 60°" is 0.2.

[0103] Weak interlayer thickness node: the prior probability of state "d < 0.1m" is 0.3, the prior probability of state "0.1m ≤ d ≤ 0.5m" is 0.4, and the prior probability of state "d > 0.5m" is 0.3.

[0104] wherein the state and prior probability of the design protection class root node are defined as:

[0105] Safety factor node: the prior probability of state "1.2" is 0.3, the prior probability of state "1.25" is 0.4, and the prior probability of state "1.3" is 0.3.

[0106] Protection structure type node: the prior probability of state "anchoring" is 0.4, the prior probability of state "anti-slide pile" is 0.4, and the prior probability of state "pile-anchor combined structure" is 0.3.

[0107] wherein the state and prior probability of the construction timing class root node are defined as:

[0108] Construction season node: the prior probability of state "rainy season" is 0.4, and the prior probability of state "non-rainy season" is 0.6.

[0109] Protection timing node: the prior probability of state "protection after excavation" is 0.3, and the prior probability of state "protection while excavating" is 0.7.

[0110] The state of the operation and maintenance class root node is defined as a prior probability:

[0111] The prior probability of the monitoring system effectiveness node in the state "effective" is 0.5, and the prior probability of the state "ineffective" is 0.5.

[0112] The prior probability of the patrol frequency node in the state "no less than once a year" is 0.6, and the prior probability of the state "ineffective" is 0.4.

[0113] The prior probability of the drainage unobstructedness node in the state "unobstructed" is 0.4, the prior probability of the state "obstructed" is 0.3, and the prior probability of the state "ineffective" is 0.3.

[0114] The prior probability summary table is shown in Table 1 as follows:

[0115] Table 1 is a prior probability table between root nodes and intermediate nodes

[0116]

[0117] The Bayesian neural network can output a probability distribution, which perfectly matches the conditional probability determination requirements of the intermediate nodes and target nodes in the application. The experience of engineers as expert knowledge fusion can be integrated into the model as Bayesian prior, which enables the Bayesian neural network to still make reasonable estimates when data samples are scarce.

[0118] Most importantly, the Bayesian neural network can automatically capture complex nonlinear relationships between parent nodes. For example, when studying the state of the soft interlayer, there is a certain correlation between the parent nodes such as rainfall intensity, vegetation coverage, and soft interlayer thickness. Traditional simple estimates cannot express this complex relationship.

[0119] Step 4, estimating the conditional probability of the intermediate nodes and the target nodes by using the Bayesian neural network, and generating a corresponding conditional probability table for each intermediate node; the specific steps include:

[0120] Step 41, extracting sample data sets of parent nodes and child nodes from historical data;

[0121] Step 42, combining expert experience to define prior constraints on node state relationships;

[0122] Step 43, training a Bayesian neural network model with parent nodes as features and child nodes as labels;

[0123] Step 44, for each parent node state combination, predicting the state probability of the child node, and generating a conditional probability table.

[0124] The step of estimating the conditional probability of the intermediate node using the Bayesian neural network is as shown in Figure 4

[0125] Suppose that the parent nodes of a certain intermediate node Y are X1, X2, …, X n ;

[0126] ① Extract a data set D containing X1, X2, …, X n and Y from historical data;

[0127] ② Define expert constraints as prior constraints in combination with engineering experience; wherein the prior constraints are set according to the following manner:

[0128] S1, identify the dominant parent node:

[0129] S2, define extreme or typical working conditions:

[0130] S3, give probability limits:

[0131] S4, cross-validation:

[0132] In this way, 2-4 core prior constraints are established for each intermediate node, which can effectively guide the model to learn the rules consistent with engineering common sense during the training of the Bayesian neural network, and especially in the case of scarce slope engineering data, greatly improve the reliability and generalization ability of the model.

[0133] The present application defines three prior constraints for the "soft interlayer state":

[0134] When the drainage is unobstructed and the rainfall intensity I is less than the surface infiltration rate Ir, the probability of soft interlayer mudification should be less than 5%;

[0135] When the drainage is obstructed or fails, and the rainfall intensity I is greater than the surface infiltration rate Ir, the probability of soft interlayer drying should be less than 10%;

[0136] When the soft interlayer thickness d is greater than 0.5m, and the rainfall intensity I is greater than the surface infiltration rate Ir, the probability of soft interlayer mudification should be greater than 30%.

[0137] Where different people have different constraint conditions, considering the corresponding probabilities as 10%, 20% and 40%, whether the prior constraints are accurate can be judged by whether the results of the Bayesian neural network learning conform to the actual situation, if there is a large deviation from the true situation, then the prior constraints based on expert experience should be adjusted, and the iteration is continued until the results of machine learning meet the expectations, that is, the iterative feedback correction.

[0138] ③ Based on the data set D, taking X1, X2, …, X n as features and Y as labels, train the Bayesian neural network model; ​

[0139] (4) For each possible combination of states of the parent nodes (x1, x2, …, xn), predict the probability of each state y of the node Y using the trained Bayesian network neural model, i.e., P(Y=y | X1=x1, X2=x2, …, XN=xn);

[0140] (5) Calculate the resulting probability and fill in the corresponding row of the conditional probability table of the node Y.

[0141] The conditional probability table CPT of the soft interlayer state obtained by training the Bayesian neural network is as shown in Table 2.

[0142] Table 2: Conditional probability table of soft interlayer state

[0143]

[0144] Based on the above conditional probability table, the most favorable condition combination of the intermediate node: soft interlayer state can be obtained:

[0145] Thin interlayer (d < 0.1 m), weak rainfall (I < Ir), unobstructed drainage, and pile-anchor combination protection, at this time the probability of the soft interlayer state being “dry” is as high as 92%, the probability of being “softened” is only 7%, and the probability of being “mudified” is only 1%;

[0146] The most unfavorable condition combination is:

[0147] Thick interlayer (d > 0.5 m), strong rainfall (I ≥ Ir), ineffective drainage, and no protection, at this time the probability of the soft interlayer state being “mudified” is as high as 70%, the probability of being “softened” is 25%, and the probability of being “dry” is only 5%.

[0148] Verify whether the conditional probability table meets the constraints by comparing the prior constraints:

[0149] Constraint ①: When drainage is unobstructed and I < Ir, the probability of mudification is ≤5%, which meets the constraint;

[0150] Constraint ②: When drainage is obstructed and I ≥ Ir, the probability of dryness is ≤10%, which meets the constraint;

[0151] Constraint ③: When the thickness is > 0.5 m and I ≥ Ir, the probability of mudification is ≥30%, which meets the constraint.

[0152] By the conditional probability table above, the geomechanics problem of weak interlayer state, which is difficult to calculate accurately, is transformed into a probability problem determined by four key observable or designed variables (weak interlayer thickness, rainfall intensity, drainage efficiency, and protection structure type). This allows us to predict the state of the weakest link of the slope by monitoring or designing these variables.

[0153] By the conditional probability table above, the key control factors and their mutual influence can be clearly defined:

[0154] For example, regardless of other conditions, the dry probability will always significantly increase and the mud probability will always decrease when the protection structure type is upgraded from "no protection" to "pile-anchor combination".

[0155] The most unfavorable combination (thick interlayer + heavy rainfall + ineffective drainage + no protection) results in a mud probability of up to 70%, which is a clear high-risk red line. Our goal is to avoid this combination at all costs.

[0156] The most favorable combination (thin interlayer + weak rainfall + unobstructed drainage + pile-anchor combination) results in a dry probability of up to 92%, which is the ideal state we strive for in slope design and reinforcement.

[0157] The three prior constraints (such as "unobstructed drainage + weak rainfall, mud probability <5%") ensure the reasonableness of the CPT in physical meaning and engineering experience. When applying this model to new areas and new projects, domain experts establish the above constraints to ensure the reliability of the model output.

[0158] Step 5: Risk analysis of weak interlayer bedding slope based on Bayesian network model.

[0159] For weak interlayer bedding slopes that have already failed, according to the Bayesian network model established in step 3, combined with the state changes of the disaster control factors before and after the failure, the disaster causes are diagnosed by cause-to-effect.

[0160] For weak interlayer bedding slopes where the control factors will change, according to the Bayesian network model established in step 3, the failure probability is predicted through the prior probability and conditional probability table.

[0161] First aspect: For existing or under-construction slopes (diagnosis and risk assessment) including the following steps:

[0162] 1. Data collection and state assessment:

[0163] Field investigation and exploration: accurately obtain the weak interlayer thickness (d) of the slope, and other geological condition root node data such as slope height / length / gradient.

[0164] Monitoring system construction / verification:

[0165] Set up rain gauges to obtain rainfall intensity (I) data and compare with the surface infiltration rate (Ir) obtained from tests.

[0166] Check the status of the drainage system and assess its patency.

[0167] Record the type of protection structure implemented.

[0168] Inference of weak interlayer state:

[0169] With the collected data (d, I vs Ir, drainage patency, protection type) as input, query this CPT (or run the trained Bayesian network model) to obtain the probability distribution of the weak interlayer being dry, softened, or mudified under the current conditions. This achieves a probabilistic and quantitative judgment of the hidden critical state.

[0170] 2. Risk diagnosis and cause analysis:

[0171] If the model calculates a high probability of mudification and there are already signs of deformation in the field, we can use the diagnostic reasoning function of the Bayesian network.

[0172] Set evidence: set "slope instability" or "weak interlayer state = mudification" as known results.

[0173] Diagnose causes: the model will calculate the posterior probability of each node in reverse. For example, it may find that the probability of "drainage patency = ineffective" has significantly increased from the prior 30% to more than 80%, which accurately locates the main cause of instability and provides a clear direction for treatment.

[0174] Second aspect: for slopes to be built or designed (prediction and design optimization) including the following steps:

[0175] 1. Scenario simulation: use CPT for a large number of "What-if" analyses. For example, under given geological conditions (d > 0.5m) and climate conditions (I > Ir probability of 0.4), compare the effects of different protection schemes:

[0176] Scheme A (no protection): mudification probability = 70%

[0177] Scheme B (anchoring): mudification probability = 55%

[0178] Scheme C (pile-anchor combination): mudification probability = 35%

[0179] Cost-benefit analysis: the above quantitative results provide clear evidence for decision-makers. The optimal scheme that reduces the mudification probability to an acceptable level within the budget can be selected.

[0180] 2. Establish dynamic early warning thresholds:

[0181] The output of CPT (mud probability) can be used as an early warning indicator.

[0182] For example, define mud probability > 30% as yellow warning, > 50% as red warning.

[0183] Linkage monitoring: Develop a system to receive rainfall intensity and drainage unobstructed data in real time (can be judged by video monitoring or simple sensor), automatically query CPT and calculate real-time risk probability, once exceeding the threshold, immediately issue a warning, remind to strengthen patrol or take emergency measures.

[0184] 3. Guide construction organization design:

[0185] The construction season and protection time sequence in the root node remind us to avoid large-scale excavation in the rainy season as much as possible, and follow the principle of "excavation and protection at the same time" to reduce the risk during construction. We can simulate the risk probability under different time sequences to quantify and demonstrate the pros and cons of these construction organization schemes.

[0186] The above only describes the preferred embodiments of the present application and does not limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for analyzing the instability risk of a bedding slope containing weak interlayers, characterized in that, Includes the following steps: Step 1: Identify multiple disaster control factors related to the instability risk of bedding slopes with weak interlayers, and divide the multiple disaster control factors into root nodes and intermediate nodes, with slope instability as the target node; The root nodes include five major categories: natural conditions, geological conditions, design protection, construction sequence, and operation and maintenance. Natural conditions include rainfall intensity and vegetation cover. Geological conditions include slope height, slope length, slope gradient, and thickness of weak interlayers. Design protection includes safety factors and types of protective structures. Construction sequence includes construction season and protection sequence. Operation and maintenance includes the effectiveness of the monitoring system, inspection frequency, and drainage patency. The intermediate nodes include surface runoff erosion, underground seepage, weak interlayer conditions, and free surface stability. Step 2: Establish causal relationships between all nodes and construct a directed graph structure that reflects the dependencies between variables by expressing them through directed edges; based on the directed graph structure, construct a risk analysis model for a bedding slope with weak interlayers based on a Bayesian network. The directed edge representation between the root node and the intermediate node includes: The rainfall intensity refers to the surface runoff erosion, the underground seepage, and the stability of the free surface; The vegetation coverage rate refers to the surface runoff erosion and the underground seepage. The slope height, slope gradient, and protection sequence are related to the stability of the free face. The slope length points in the direction of surface runoff scouring; The thickness of the weak interlayer indicates the state of the weak interlayer; The construction season refers to the underground seepage and the surface runoff scouring; The drainage flow rate refers to the state of the weak interlayer. The type of protective structure refers to the state of the weak interlayer and the stability of the free surface; Step 3: Define the prior probability of each root node based on historical data, national standards, or engineer experience; Step 4: Use a Bayesian neural network to estimate the conditional probabilities of the intermediate nodes and the target node, and generate a corresponding conditional probability table for each intermediate node. Step 5: Based on the risk analysis model of the slope with weak interlayers, the prior probability and the conditional probability table, conduct etiological diagnosis or predictive analysis of slope instability.

2. The method for analyzing the instability risk of bedding slopes containing weak interlayers according to claim 1, characterized in that, The steps for estimating the conditional probability of intermediate nodes using a Bayesian neural network include: Step 41: Extract sample datasets of parent and child nodes from historical data; Step 42: Define prior constraints on node state relationships based on expert experience; Step 43: Train a Bayesian neural network model using parent nodes as features and child nodes as labels; Step 44: For each combination of parent node states, predict the state probability of the child nodes and summarize them to generate a conditional probability table.

3. The method for analyzing the instability risk of bedding slopes containing weak interlayers according to claim 2, characterized in that, The prior constraints are defined according to the following steps: S1. Identify the dominant parent node; S2. Define extreme or typical operating conditions; S3. Assign probability limits; S4. Obtain the final prior constraints through iterative feedback and machine learning correction.

4. The method for analyzing the instability risk of bedding slopes containing weak interlayers according to claim 2, characterized in that, The conditional probability tables include conditional probability tables for surface runoff scour, conditional probability tables for underground seepage, conditional probability tables for weak interlayer conditions, and conditional probability tables for free surface stability.

5. The method for analyzing the instability risk of bedding slopes containing weak interlayers according to claim 1, characterized in that, In step 5, the diagnosis or prediction analysis of slope instability risk includes: for slopes that have already become unstable, diagnosing the disaster control factors; for slopes that have not yet become unstable, being able to predict the probability of instability under given conditions.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the risk analysis method as described in any one of claims 1-5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the risk analysis method as described in any one of claims 1-5.

Images