Method for analyzing causes of bank slope instability, program product and electronic device

Abstract

This disclosure provides a method, program product, and electronic equipment for analyzing the causes of reservoir bank slope instability, relating to the field of engineering data processing technology. The method includes: acquiring monitoring data of the reservoir bank slope, including monitoring data corresponding to various instability inducing parameters, including static topographic and geological parameters and dynamic inducing parameters; constructing an initial graphical model using each instability inducing parameter and the instability state parameter of the reservoir bank slope as nodes; determining constraint information based on the physical correlation between different instability inducing parameters and the physical correlation between instability inducing parameters and instability state parameters; injecting the monitoring data and constraint information into a pre-constructed objective function, updating the initial graphical model through the objective function to obtain an interpretable causal graphical model; and determining the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope based on the interpretable causal graphical model. This disclosure improves the accuracy and interpretability of reservoir bank slope instability inducing factor analysis.

Description

Technical Field

[0001] This disclosure relates to the field of engineering data processing technology, and more specifically, to methods, program products, and electronic equipment for analyzing the causes of instability of reservoir bank slopes. Background Technology

[0002] Reservoir bank slopes are a crucial component of large reservoirs, pumped-storage power stations, hydropower stations, and other water conservancy and hydropower projects. Their stability directly impacts the safety of the project and the safety of people and property in the surrounding area. The hydrogeological environment of reservoir bank slopes is unique, subject to the coupled effects of various factors such as reservoir water level fluctuations, reservoir water infiltration, bank erosion, rainfall, and the wet-dry cycle of the soil and rock mass. Their instability is typically driven by a combination of factors, resulting in a complex mechanism.

[0003] To prevent and manage reservoir bank slope instability, relevant technologies typically rely on threshold comparisons of single monitoring indicators to trigger early warnings. This approach only indicates the presence of instability risk but struggles to identify the specific triggers, leading engineers to resort to general reinforcement and drainage measures, resulting in a lack of targeted control. Furthermore, single-indicator threshold warnings fail to adequately consider the synergistic effects of multiple triggers and the coupling relationship between these triggers and slope response, making false alarms or missed alarms more likely.

[0004] To further identify the causes of instability, related technologies have attempted to model multi-source monitoring data using machine learning models, and then interpret the model output using feature importance analysis methods. The "feature importance" given by this approach is essentially still a description of statistical correlation, making it difficult to distinguish between correlation and causation. Furthermore, its modeling process typically does not incorporate prior knowledge of geotechnical mechanics, making it difficult to correlate the causal analysis results with actual engineering mechanisms. In addition, some related technologies employ purely data-driven causal structure learning methods to construct correlation graphs between instability causal parameters and instability state variables. However, such schemes rely entirely on the statistical regularities exhibited in the data, easily leading to correlation paths that contradict geotechnical mechanics mechanisms. Moreover, they usually only provide one-time static analysis results, failing to reflect the temporal evolution characteristics of the causal influence as reservoir conditions change.

[0005] Therefore, how to accurately identify the various instability causes of reservoir bank slopes and ensure that the analysis results are consistent with the geotechnical mechanics mechanism and reflect the dynamic changes of the working conditions is a technical problem that urgently needs to be solved in related technologies. Summary of the Invention

[0006] This disclosure provides methods, programs, and electronic devices for analyzing the causes of instability on reservoir bank slopes, in order to at least partially address problems in related technologies.

[0007] According to a first aspect of this disclosure, a method for analyzing the causes of instability on a reservoir bank slope is provided. The method includes: acquiring monitoring data of the reservoir bank slope, the monitoring data including monitoring data corresponding to various instability inducing parameters, the instability inducing parameters including static topographic and geological parameters and dynamic inducing parameters; constructing an initial graphical model using each instability inducing parameter and the instability state parameter of the reservoir bank slope as nodes; determining constraint information based on the physical correlation between different instability inducing parameters and the physical correlation between instability inducing parameters and instability state parameters; injecting the monitoring data and the constraint information into a pre-constructed objective function, updating the initial graphical model through the objective function to obtain an interpretable causal graphical model; and determining the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope based on the interpretable causal graphical model.

[0008] According to a second aspect of this disclosure, a device for analyzing the causes of instability of a reservoir bank slope is provided. The device includes: a monitoring data acquisition module configured to acquire monitoring data of the reservoir bank slope, the monitoring data including monitoring data corresponding to various instability inducing parameters, the instability inducing parameters including static topographic and geological parameters and dynamic inducing parameters; a graph model construction module configured to construct an initial graph model using each instability inducing parameter and the instability state parameter of the reservoir bank slope as nodes; a constraint information determination module configured to determine constraint information based on the physical correlation between different instability inducing parameters and the physical correlation between instability inducing parameters and instability state parameters; a graph model update module configured to inject the monitoring data and the constraint information into a pre-constructed objective function, and update the initial graph model through the objective function to obtain an interpretable causal graph model; and a contribution determination module configured to determine the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope based on the interpretable causal graph model.

[0009] According to a third aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method of the first aspect described above and possible implementations thereof.

[0010] According to a fourth aspect of this disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the method of the first aspect and possible implementations thereof by executing the executable instructions.

[0011] The technical solution disclosed herein has the following beneficial effects: Firstly, the acquired monitoring data covers both static topographic and geological parameters and dynamic instability induction parameters. In terms of feature composition, it avoids the problem of incomplete coverage of instability induction factors caused by relying on only a single indicator or only focusing on static geological conditions. This lays a relatively complete data foundation for subsequent identification of the impact of various induction factors on the instability of the reservoir bank slope.

[0012] Secondly, by constructing an initial graphical model using instability inducing parameters and instability state parameters as nodes, and then updating this model with an objective function, the relationships between instability inducing parameters and between instability inducing parameters and instability state parameters can be characterized at the graph structure level. Compared to directly providing statistical correlations based on feature importance, the graph structure facilitates tracing back the inducing influence process along directed paths between nodes, making the analysis results readable and understandable by engineers.

[0013] Thirdly, the introduced constraint information originates from the physical correlations between different instability inducing parameters and between these parameters and the instability state parameters. This constraint information works in conjunction with the monitoring data in the objective function, ensuring that the final interpretable causal graph model is supported by the statistical regularities reflected in the monitoring data and constrained by the physical correlations in geotechnical mechanics. This approach, from a mechanistic perspective, suppresses the occurrence of correlation paths that contradict physical mechanisms, ensuring that the causal analysis results are no longer divorced from engineering practice.

[0014] Fourthly, the interpretable causal graph model determines the dynamic contribution of each instability inducing factor parameter to the instability of the reservoir bank slope, rather than a one-time static evaluation result. The dynamic contribution can continuously characterize the strength changes of each inducing factor in different time periods as the monitoring data is updated, which facilitates engineering technicians to take targeted prevention and control measures for the inducing factors that play a dominant role under the current working conditions. Attached Figure Description

[0015] Figure 1 A flowchart illustrating the method for analyzing the causes of reservoir bank slope instability in one embodiment of this disclosure is shown. Figure 2 This diagram illustrates a sub-flowchart of one embodiment of the present disclosure; Figure 3 This diagram illustrates a sub-flowchart of one embodiment of the present disclosure; Figure 4 This diagram illustrates a sub-flowchart of one embodiment of the present disclosure; Figure 5 This diagram illustrates a method architecture in one embodiment of the present disclosure. Figure 6 This diagram illustrates a graphical model of one embodiment of the present disclosure. Figure 7This diagram illustrates the contribution level in one embodiment of the present disclosure. Figure 8 This diagram illustrates a comparison between embodiments of the present disclosure and related technologies. Figure 9 A schematic diagram of a device for analyzing the causes of instability of a reservoir bank slope is shown in one embodiment of this disclosure; Figure 10 A schematic diagram of an electronic device according to one embodiment of the present disclosure is shown. Detailed Implementation

[0016] Exemplary embodiments of this disclosure will be described more fully below with reference to the accompanying drawings.

[0017] The accompanying drawings are schematic illustrations of this disclosure and are not necessarily drawn to scale. Some block diagrams shown in the drawings may be functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in hardware modules or integrated circuits, or in networks, processors, or microcontrollers. Implementations can be carried out in various forms and should not be construed as limited to the examples set forth herein. The features, structures, or characteristics described in this disclosure can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a full description of the technical solutions of this disclosure. However, those skilled in the art will recognize that one or more specific details may be omitted when implementing the technical solutions provided in this disclosure, or other methods, components, apparatuses, steps, etc., may be used to replace one or more specific details.

[0018] The hydrogeological environment of reservoir bank slopes is complex. Multiple factors, including reservoir water level fluctuations, rainfall infiltration, changes in the seepage field, and the wet-dry cycle of the soil and rock mass, all influence slope stability. These factors exhibit synergistic effects and varying degrees of time lag. One approach to related technologies relies on thresholds for single monitoring indicators for early warning. This approach can only determine the presence of instability risk but struggles to identify the truly dominant contributing factors. Consequently, engineers cannot develop targeted prevention and control measures. Furthermore, it is prone to false alarms when multiple contributing factors are synergistically approaching the instability threshold, but any single indicator has not yet exceeded its threshold. False alarms are also likely when an occasional single indicator fluctuation exceeds its threshold without corresponding changes in other indicators. Another approach utilizes machine learning models to model multi-source monitoring data and interprets the model output through feature importance analysis. The feature importance given is merely a measure of the statistical correlation between each feature and the model output, and cannot distinguish between correlation and causation. It may misjudge features that appear only occasionally as significant contributing factors, and it does not incorporate prior knowledge of geotechnical mechanics into the modeling process, resulting in a gap between the given explanation and engineering practice. Another approach is a data-driven causal structure learning method that constructs a correlation graph between instability contributing factors and instability state variables. While it introduces the concept of causal structure, it relies entirely on statistical regularities in the data, easily leading to paths that contradict physical mechanisms, such as slope response pointing to reservoir water level. Furthermore, it typically only provides a one-time analysis of data for a specific time period, lacking characterization of the evolution of contributing factors over time for dynamic conditions such as reservoir water level adjustments, rainfall variations, and soil-rock wet-dry cycles.

[0019] In view of one or more of the above problems, this disclosure provides a method for analyzing the causes of instability of reservoir bank slopes. Figure 1 An exemplary procedure for analyzing the causes of reservoir bank slope instability is shown, including the following steps: S110, acquire monitoring data of reservoir bank slopes, including monitoring data corresponding to various instability inducing parameters, including static topographic and geological parameters and dynamic inducing parameters; S120, using each instability induction parameter and the instability state parameter of the reservoir bank slope as nodes, constructs an initial graphical model; S130, determine constraint information based on the physical relationship between different instability inducing parameters and the physical relationship between instability inducing parameters and instability state parameters; S140, The monitoring data and constraint information are injected into the pre-built objective function, and the initial graph model is updated through the objective function to obtain an interpretable causal graph model; S150, based on an interpretable causal graph model, determines the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope.

[0020] Based on the above method, the acquired monitoring data simultaneously covers both static topographic and geological parameters and dynamic causal parameters, avoiding the incomplete coverage problem caused by relying on only a single indicator or focusing solely on static geological conditions. An initial graphical model is constructed using instability causal parameters and instability state parameters as nodes, and then updated using an objective function. This allows the relationships between parameters to be presented in a graphical structure, facilitating the tracing of causal influence processes along directed paths between nodes. The introduced constraint information originates from the physical relationships between instability causal parameters and between instability causal parameters and instability state parameters. These constraints work in conjunction with the monitoring data in the objective function, ensuring that the resulting interpretable causal graphical model is supported by the statistical regularities reflected in the monitoring data and constrained by the physical relationships in geotechnical mechanics. This mechanism suppresses the occurrence of correlation paths that contradict physical mechanisms. Based on this interpretable causal graph model, the dynamic contribution of each instability inducing factor parameter to the instability of the reservoir bank slope is determined. It can characterize the changes in the strength of each inducing factor at different time periods as the monitoring data is updated, which makes it easier for engineering technicians to take targeted prevention and control measures for the inducing factors that play a dominant role under the current working conditions.

[0021] The following describes, in conjunction with one or more embodiments and related accompanying drawings, Figure 1 Each step in the process will be explained in detail.

[0022] refer to Figure 1 In step S110, monitoring data of the reservoir bank slope is acquired. The monitoring data includes monitoring data corresponding to various instability inducing parameters, including static topographic and geological parameters and dynamic inducing parameters.

[0023] Monitoring data can be obtained from measured data collected by various monitoring devices deployed on the reservoir bank slopes and in the reservoir area, such as global navigation satellite system displacement monitoring points, seepage monitoring wells, inclinometers, meteorological monitoring stations, and water level monitoring stations. Instability inducing parameters refer to physical quantities that affect the instability process of the reservoir bank slope. Static topographic and geological parameters refer to parameters reflecting the inherent topography and geological conditions of the reservoir bank slope; these parameters do not change significantly over a short time scale and can be obtained through on-site investigation, laboratory tests, remote sensing interpretation, and ground-penetrating radar detection. Dynamic inducing parameters refer to parameters whose values ​​change significantly over time and reflect the dynamic processes of reservoir water, rainfall, seepage, and the slope's own response; these are mainly collected in real time by the reservoir area monitoring system.

[0024] In one implementation, instability parameters can be used to characterize the current stability state of the reservoir bank slope. For example, the slope safety factor value can be calculated using the limit equilibrium method based on the currently monitored soil and rock parameters and working conditions. When the safety factor value is less than 1, it indicates that the slope is in an unstable state; when the safety factor value is greater than or equal to 1, it indicates that the slope is in a stable state.

[0025] In one implementation, static topographic geological parameters may include one or more of the following parameters: slope α The angle between the slope surface of the reservoir bank and the horizontal plane, expressed in degrees (°), with a range of 0° < α ≤ 90°, reflects the steepness of the slope; Elevation H ele The absolute elevation of each monitoring point on the slope, in meters (m), reflects the elevation of the slope and the extent of reservoir inundation. Lithology coefficient K lith The dimensionless coefficient is assigned based on the slope rock and soil type, with a value range of (0,1]. The value approaches 1 for hard rock and approaches 0 for soft rock, reflecting the basic shear strength of the rock and soil. Fault distance L fault The horizontal distance from the slope monitoring point to the nearest fault is measured in meters (m). It reflects the degree of influence of the fault structure on the slope stability. The smaller the distance, the more significant the influence.

[0026] In one implementation, the dynamic inducing parameters include one or more of the following: reservoir water level dynamic parameters, seepage field dynamic parameters, rainfall dynamic parameters, and slope response parameters.

[0027] Reservoir water level dynamic parameters reflect the effect of the dynamic process of reservoir water level rise and fall on the slope. In one embodiment, the reservoir water level dynamic parameters may include one or more of the following parameters: Water level fluctuation Δ H w Δ represents the maximum change in reservoir water level within a set time interval, expressed in meters (m). The formula for this is: Δ H w =| H w ( t )- H w ( t -Δ t )|, where H w ( t )for t Reservoir water level at any time (unit: m). H w( t -Δ t )for t -Δ t Reservoir water level at any time (in meters), Δ t The time interval (in days) reflects the magnitude of changes in reservoir water level. Rate of change of water level v w The rate of change of reservoir water level over time, expressed in meters per day (m / d), is calculated using the following formula: v w =Δ H w / Δ t Positive values ​​indicate rising water levels, while negative values ​​indicate falling water levels, reflecting the drastic changes in reservoir water levels. Water level fluctuation frequency f w The number of times the reservoir water level completes its rise and fall cycle per unit of time, expressed in times per month (times per 30 days), is calculated using the following formula: f w =N w / 30, of which N w The number of cycles of rise and fall in reservoir water level within 30 days (in times) reflects the frequency of reservoir water level fluctuations. Duration of extreme high water levels t high The continuous duration during which the reservoir water level is higher than the set warning water level is measured in days (d). The warning water level is set according to the slope safety level and reflects the long-term effect of extreme water levels on the slope.

[0028] The dynamic parameters of the seepage field reflect the changes in the seepage state inside the slope after reservoir water infiltration. In one embodiment, the dynamic parameters of the seepage field include one or more of the following parameters: seepage velocity v seep Based on the estimation of changes in the phreatic line and the seepage path length, it reflects the direct driving intensity of seepage dynamics on the slope, with the unit being meters per day (m / d). Pore ​​water pressure ratio r u The ratio of pore water pressure to effective overburden pressure in slope soil and rock mass, dimensionless, and calculated using the following formula: r u = u / σ v ,in u The pore water pressure is (kPa). σ v The effective overburden pressure (unit: kPa) reflects the degree to which pore water pressure weakens the strength of the rock and soil mass. Hydraulic gradient i grad The ratio of head loss along the seepage direction to the seepage path length in the slope seepage field is dimensionless and calculated using the following formula: i grad =Δh / L seep , where Δh is the head difference (in meters) between the two ends of the seepage path. L seep The seepage path length (m) reflects the magnitude of the seepage force.

[0029] Rainfall dynamic parameters reflect the impact of rainfall infiltration on slope stability. In one implementation, rainfall dynamic parameters include one or more of the following parameters: cumulative rainfall R cum The total rainfall within a set time interval, expressed in millimeters (mm), is collected in real time by the reservoir area meteorological station to reflect the cumulative effect of rainfall. Rainfall intensity I rain Rainfall per unit time, expressed in millimeters per hour (mm / h), is calculated using the following formula: I rain= R cum / t rain ,in t rain The duration of rainfall (in hours) reflects the intensity of the rainfall. Rain type coefficient K rain It is a dimensionless coefficient that reflects the type of rainfall, such as frontal rainfall. K rain =0.8~1.0, continuous rainfall K rain =0.3~0.7, which can be determined by the characteristics of the rainfall time series curve, reflecting the differences in the impact of different rainfall types on slope infiltration.

[0030] Slope response parameters reflect the dynamic response of a slope under the influence of multiple induced factors. In one embodiment, slope response parameters include one or more of the following parameters: Displacement rate v disp The rate of change of displacement of the slope monitoring points over time, expressed in millimeters per day (mm / d), is collected in real time by a GNSS monitoring system. The calculation formula is as follows: v disp =|d( t )-d( t -Δ t )| / Δt , where d( t ) represents the displacement (mm) at time t, d( t -Δ t )for t -Δ t Displacement at any moment (mm); displacement acceleration a disp The rate of change of slope displacement, expressed in millimeters per day² (mm / d²), is calculated using the following formula: a disp =| v disp ( t )- v disp ( t -Δ t )| / Δ t This reflects the changing trend of slope displacement; a positive value indicates accelerated displacement and increased risk. Inclination change Δ θ The change in the tilt angle of the slope monitoring point over time is expressed in arcseconds (″). It is collected in real time by an inclinometer and reflects the degree of slope tilt deformation.

[0031] By further subdividing the instability inducing parameters as described above, various natural forces and the slope's own response can be specifically reflected in the monitoring data. Among them, static topographic and geological parameters characterize the inherent conditions of the slope, reservoir water level dynamic parameters, seepage field dynamic parameters, and rainfall dynamic parameters characterize the effects of the external hydrological environment on the slope, and slope response parameters characterize the kinematic characteristics of the slope, such as deformation and tilting, under various forces. This provides more comprehensive data support for subsequent graphical model construction and inducing factor analysis.

[0032] In one implementation, instability inducing parameters include slope response parameters and other inducing parameters; reference Figure 2 As shown, the method also includes the following steps: S210, after acquiring the monitoring data of the reservoir bank slope, the monitoring data of other inducing parameters at different times are used to form an inducing time series, and the monitoring data of the slope response parameters at different times are used to form a slope response time series. S220, substitute the induced time series and the slope response time series into the cross-correlation function, and determine the optimal time delay between other induced parameters and slope response parameters by calculating the maximum value of the cross-correlation function; S230, adjust the time information of the monitoring data corresponding to other inducing parameters according to the optimal time delay time, so as to align the other inducing parameters and slope response parameters in time.

[0033] During reservoir bank slope instability, the influence of other inducing parameters such as rainfall infiltration and reservoir water level fluctuations on slope displacement and tilt responses is usually not immediate. It often takes time for these parameters to propagate into the slope and ultimately cause changes in response parameters; this phenomenon is commonly referred to as the time lag effect in engineering. Other inducing parameters refer to various parameters other than slope response parameters among the instability inducing parameters, such as dynamic rainfall parameters, dynamic reservoir water level parameters, dynamic seepage field parameters, and quantifiable items in static topographic and geological parameters. An inducing time series refers to a sequence of monitoring values ​​corresponding to a certain inducing parameter arranged in chronological order; a slope response time series refers to a sequence of monitoring values ​​corresponding to a certain slope response parameter arranged in chronological order. The cross-correlation function is used to measure the similarity of two time series under different time lags. The optimal time lag is the time lag value that maximizes the cross-correlation function, reflecting the time misalignment condition under which the two time series are most correlated. Adjust the time information of the monitoring data corresponding to other inducing parameters according to the optimal time delay time. That is, shift the time series of other inducing parameters on the time axis so that they are aligned with the slope response time series in time. This ensures that the subsequent graphical model construction and inducing factor analysis based on these data are contemporaneous data with physical causal transmission relationships.

[0034] By introducing time delay correction, the data included in the subsequent graphical model construction no longer exhibits false correlation or uncorrelation due to time misalignment, thereby improving the accuracy of characterizing the correlation between various instability inducing parameters and instability state parameters.

[0035] In one implementation, the cross-correlation function is: (1) in, τ Represents the time delay; E[·] represents the mathematical expectation; X( t Y( represents the time series of the triggers; Y( t () represents the slope response time series; μ X The time series X represents the cause. t The mean of ) μ Y The slope response time series Y( t The mean of ) σ X The time series X represents the cause. t The standard deviation of ) σ Y The slope response time series Y( t The standard deviation of ρ( τ When the maximum value is reached, the corresponding τ Optimal time delay τopt .

[0036] The range of the time delay τ can be set according to the actual working conditions of the reservoir area, such as 0 to 30 days or 0 to 60 days. Within this range, τ is scanned, and the corresponding cross-correlation function values ​​are calculated. The τ corresponding to the maximum value is taken as the optimal time delay. τ opt Between the rate of change of reservoir water level and the rate of displacement, the optimal time lag is usually shorter; between the rate of change of cumulative rainfall and the rate of displacement, the optimal time lag is relatively longer. Furthermore, the causal time series X( t Translation τ opt The corrected cause time series X' is obtained. t )=X( t - τ opt This method achieves precise time axis alignment between the causal time series and the slope response time series, i.e., time alignment between other causal parameters and slope response parameters. Using this form of cross-correlation function allows for a normalized measurement of the correlation between two time series at different time lags, avoiding interference from mean drift and dimensional differences on the correlation measurement results. This makes the obtained optimal time lag more reliable and further ensures the accuracy of time lag correction.

[0037] Continue to refer to Figure 1 In step S120, an initial graphical model is constructed using each instability induction parameter and the instability state parameter of the reservoir bank slope as nodes.

[0038] The initial graphical model refers to the graph structure that provides the starting point for subsequent causal structure learning. It includes nodes corresponding to each instability trigger parameter and nodes corresponding to the instability state parameters of the reservoir bank slope. The initial graphical model can take various forms, such as an initial undirected graph or a causal graph skeleton obtained after preliminary conditional independence tests. Organizing the instability trigger parameters and instability state parameters in a graph structure allows the subsequent objective function processing to unfold around the edges and directions between nodes, facilitating the explicit expression of the relationships between parameters.

[0039] In one implementation, the initial graph model is a causal graph skeleton; see reference. Figure 3 As shown, the above-mentioned initial graphical model is constructed using each instability induction parameter and the instability state parameter of the reservoir bank slope as nodes, including the following steps: S310, using each instability induction parameter and the instability state parameter of the reservoir bank slope as nodes, construct an initial undirected graph; there is an undirected edge between any two nodes in the initial undirected graph; S320: Perform a conditional independence test on any two nodes in the initial undirected graph. If the two nodes are not conditionally independent, delete the undirected edge between the two nodes to obtain the intermediate undirected graph. S330. Perform a conditional independence test on any two nodes in the intermediate undirected graph. If the two nodes are conditionally independent under the set condition set, delete the undirected edge between the two nodes. Iterate and update the intermediate undirected graph by gradually increasing the size of the condition set until the size of the condition set reaches its maximum and the remaining undirected edges cannot be deleted, thus obtaining the causal graph skeleton.

[0040] For example, the set of parameters for the instability causes of the reservoir bank slope is denoted as . Where m is the total number of instability-inducing parameters. For example, the above 16 instability-inducing parameters can be selected, including slope. α Elevation H ele lithology coefficient K lith Fault distance L fault Water level fluctuation Δ H w Rate of water level change v w Water level fluctuation frequency f w Duration of extreme high water levels t high seepage velocity v seep pore water pressure ratio r u Hydraulic gradient i grad Cumulative rainfall R cum Rainfall intensity I rain Displacement rate v disp Displacement acceleration a disp Inclination change Δ θ That is, m=16, each x i The parameters for instability causes are denoted by Y after time delay correction; the parameters for instability of the reservoir bank slope are denoted by Y and are expressed using the slope safety factor (less than 1 indicates instability, and greater than or equal to 1 indicates stability), which can be calculated by the limit equilibrium method.

[0041] It should be noted that the rain pattern coefficient K rain To provide auxiliary descriptive features for rainfall types, in actual causal modeling, factors must be considered in conjunction with other rainfall features (such as...) Rcum , I rain Significant collinearity exists, and the physical causal chain is not direct, so it is removed during the graph model construction process.

[0042] This disclosure requires the construction of a directed acyclic graph (DAG), denoted as . G =(Vector,Edge), where the node set Vector = X∪{Y}, and each node corresponds to a type of instability trigger parameter or instability state parameter; and the directed edge set Edge. Vector Vector, directed edge x i → x j express x i yes x j The direct cause is the directed edge x i →Y represents x i It is the direct cause of slope instability; the constraint condition is a graphical model. G There are no loops in the middle, and all directed edges conform to the basic principles of reservoir bank slope geomechanics, avoiding reverse causal and physically meaningless causal paths.

[0043] A causal graph skeleton refers to a graph structure that retains only undirected edges that truly establish statistical dependencies between nodes. It can be understood as a systematic screening process to determine the existence of dependencies between nodes before determining edge directions. The initial undirected graph provides the starting point for this screening process, where every two nodes are connected by an undirected edge, essentially assuming that all nodes may have dependencies. The conditional independence test determines whether the dependency between two nodes still holds given a set of other nodes (i.e., the condition set). If it no longer holds, the association between the two nodes is actually transmitted through other nodes in the condition set, and their undirected edges can be deleted. The condition set size refers to the number of conditional nodes participating in a conditional independence test, gradually increasing from 0 to the maximum allowed value. This allows for different granularities of discrimination between node pairs, thus progressively eliminating redundant undirected edges.

[0044] For example, the PC algorithm (Peter-Clark, a constraint-based causal discovery algorithm) is used to automatically discover the skeleton of the causal graph. The PC algorithm is based on the causal Markov condition and the fidelity assumption. It constructs the initial causal graph skeleton by progressively performing conditional independence tests and eliminating edges with no statistical correlation. The specific steps are as follows: Initialization: Construct a complete undirected graph containing all node vectors. G 0, There is an undirected edge between any two nodes; Phase 1 (Unconditional Independence Test): For any two nodes v1, v2 ∈ Vector, test their unconditional independence. If v1 v2| (meaning v1 and v2 are unconditionally independent) then delete. G The undirected edge between v1 and v2 in 0 yields an undirected graph. G 1; The second stage (conditional independence test): Define the size k of the condition set S (k starts from 1 and gradually increases, with a maximum of m-2). For... G k Given any two adjacent nodes v1 and v2, select k nodes adjacent to v1 and v2 to form a condition set S, and test v1. If v2|S (meaning v1 and v2 are conditionally independent under a given condition set S), then delete the undirected edge between v1 and v2 to obtain an undirected graph. G k+1 ; Termination condition: When the size k of the condition set reaches its maximum and no new edges can be deleted, the test stops. The resulting undirected graph is the skeleton of the causal graph. G skeleton .

[0045] In one implementation, the edge directions of the causal graph skeleton can also be initially determined. For example, for a V-shaped structure v1-v3-v2 (v1 and v2 are not adjacent, and v3 is adjacent to both v1 and v2) in the causal graph skeleton, if there is no condition set S such that v1 and v2 are conditionally independent under this condition set and v3 does not belong to this condition set, then the edge direction can be determined as v1→v3←v2, completing the direction labeling of some edges.

[0046] By constructing the initial graph model as the aforementioned causal graph skeleton, redundant edges can be removed based on the conditional independence relationships shown by the monitoring data before the introduction of constraint information and objective function. This significantly reduces the size of the graph structure that the subsequent objective function needs to process, and allows for the concentration of subsequent direction discrimination and constraint application on the retained edges, thereby improving the targeting of the processing.

[0047] In one implementation, when performing a conditional independence test on two nodes, a Bayesian estimation is introduced to correct the test statistic. The correction formula is as follows: (2) in, This represents the original test statistic. This represents the corrected test statistic; This represents the variance of the monitoring data corresponding to the parameters that cause instability. This represents the variance of data noise. The original test statistic refers to the statistic calculated directly from the monitoring data in the conditional independence test, for example, it can be constructed based on the chi-square distribution. The corrected test statistic refers to the statistic obtained after adjusting the original test statistic, and is used in the actual discrimination process. (Data noise variance) The dispersion of random noise introduced into monitoring data by factors such as measurement errors and environmental interference can be obtained through calibration of monitoring equipment or statistical estimation of data during the static steady-state period. The correction formula introduces an exponential decay factor related to the ratio of noise variance to data variance to shrink the original test statistic. This reduces the value of the original test statistic when noise levels are high, avoiding spurious statistical significance caused by data noise and making the determination of conditional independence between nodes more robust. In actual engineering, the environment in which reservoir bank slope monitoring equipment is located is complex, and monitoring data generally contains varying degrees of noise. Using the above correction formula, the impact of noise on the judgment results can be effectively reduced at key stages of causal diagram framework discovery, thereby further improving the accuracy of the causal diagram framework.

[0048] Continue to refer to Figure 1 In step S130, constraint information is determined based on the physical correlation between different instability inducing parameters and the physical correlation between instability inducing parameters and instability state parameters.

[0049] The physical correlation between different instability inducing parameters refers to the interaction relationships that are clearly defined by the basic principles of geotechnical mechanics and hydrology, whether they exist or not. For example, rainfall infiltration affects the seepage field inside the slope, while slope deformation usually does not affect the rainfall process itself. The physical correlation between instability inducing parameters and instability state parameters refers to the basic mechanism by which the instability inducing parameters affect slope stability. For example, an increase in the pore water pressure ratio weakens the effective shear strength of the soil and rock mass. Constraint information refers to the limiting conditions formed based on the above physical correlations, which can act on the existence or direction of directed edges in the graphical model. Constraint information can include constraints that are strictly valid and cannot be violated, or constraints that are valid with a certain probability. By introducing constraint information, prior knowledge in geotechnical mechanics is transformed from human experience into formal conditions that can be handled by the objective function, thereby guiding the existence and direction of edges during the construction of the graphical model.

[0050] Continue to refer to Figure 1 In step S140, the monitoring data and constraint information are injected into the pre-built objective function, and the initial graph model is updated through the objective function to obtain an interpretable causal graph model.

[0051] The pre-constructed objective function refers to a function designed before the graph model update to measure the degree of conformity between the current graph model and the monitoring data and constraints. Monitoring data enters the objective function through a likelihood term, while constraint information enters through corresponding constraint terms, both influencing the value of the objective function. An interpretable causal graph model refers to a graph model obtained from the objective function update, where nodes correspond to instability inducing parameters or instability state parameters, and directed edges correspond to causal relationships between nodes. "Interpretable" means that its structure is supported by monitoring data and satisfies physical constraints in geotechnical mechanics; therefore, its structure and the relationships represented by directed edges can be understood and accepted by engineering technicians. In practical applications, the current graph model can be iteratively adjusted within the search space composed of all directed acyclic graphs, using the minimum objective function value as the criterion, ultimately obtaining the graph model corresponding to the minimum objective function value as the interpretable causal graph model.

[0052] In one implementation, the constraint information includes hard constraints and soft constraints; hard constraints include required edges and prohibited edges. Required edges may include: Required side 1, rainfall dynamic parameters (e.g.) R cum , I rain ) refers to the dynamic parameters of the seepage field (such as h phr , r u The directed edge of the slope is Rain→Per, which means that the infiltration of rainfall will directly lead to the change of the seepage field inside the slope. This is the basic mechanism of rainfall-induced slope instability in geotechnical mechanics. Therefore, it is a necessary edge, denoted as Rain→Per (Rain represents the dynamic parameter of rainfall, and Per represents the dynamic parameter of seepage field).

[0053] Side 2 is required; dynamic parameters of the seepage field (e.g.) h phr , i grad ) points to slope response parameters (such as v disp , a disp The directed edge of the slope, i.e., the change of the seepage field, will directly cause slope displacement, tilt and other responses, denoted as Per→Rsp (Rsp represents the slope response parameter).

[0054] Side 3 is required; dynamic parameters of reservoir water level (such as Δ) H w , v w ) refers to the dynamic parameters of the seepage field (such as v seep , r uThe directed edge of the reservoir water level, i.e., the rise and fall of the reservoir water level will directly change the dynamic parameters of the seepage field such as the seepage velocity and the pore water pressure ratio, is denoted as WL→Perc (WL represents the dynamic parameters of the reservoir water level).

[0055] Forbidden edges can include: Edge 1 is prohibited. A directed edge that points the slope response parameter (Rsp) to the reservoir water level dynamic parameter (WL) means that the slope displacement, tilt and other responses do not affect the reservoir water level change. The reverse causal path does not hold, and such a directed edge is prohibited.

[0056] Edge 2 is prohibited. A directed edge whose slope response parameter (Rsp) points to the rainfall dynamic parameter (Rain) is prohibited. This means that the slope response does not affect the rainfall process. Such a directed edge is prohibited.

[0057] Forbidden edge 3, static topographic and geological parameters (Geo, such as α , L fault Directed edges pointing to dynamic inducing parameters (such as WL, Rain, Per) are prohibited. This means that static geological conditions will not change dynamic processes such as reservoir water level, rainfall, and seepage, but will only affect the intensity of the effect of dynamic inducing factors on the slope.

[0058] Soft constraints can include: Soft constraint 1: The probability of a directed edge between the reservoir water level dynamic parameter (WL) and the slope response parameter (Rsp) is positively correlated with the lithology coefficient of the reservoir bank slope. Such edges are considered possible edges only if the hydraulic gradient... i grad Reaching the set threshold (based on lithology coefficient) K lith Sure, K lith The smaller the threshold (the lower the threshold), the greater the probability constraint P(WL→Rsp) = K lith That is, the softer the soil and rock mass, the higher the probability that the reservoir water level will directly trigger a slope response.

[0059] Soft constraint 2, static topographic and geological parameters (Geo, such as...) α , K lith The probability of a directed edge pointing to the instability state parameter (Y) is negatively correlated with the lithology coefficient of the reservoir bank slope. Such edges are considered possible edges, and the probability constraint is P(Geo→Y)=1- K lith That is, the softer the soil and rock mass and the steeper the slope, the greater the direct impact of static geological conditions on slope instability.

[0060] Soft constraint 3: The probability of the existence of a directed edge pointing from the slope response parameter (Rsp) to the instability state parameter (Y) is a preset value (which can be set according to experience or specific needs, such as 0.9). Such edges are high-probability edges, and the probability constraint is P(Rsp→Y)=0.9, that is, the slope displacement, tilt and other responses are direct characteristics of slope instability and have a very strong causal relationship.

[0061] By specifying the physical associations as the aforementioned hard and soft constraints, the constraint information processed by the objective function includes both explicit conditions that must be met or excluded, and probabilistic information that flexibly guides the direction of edges based on physical characteristics. This further refines the form of the constraint information based on the scheme described in claim 1, thereby further improving the physical interpretability of the final interpretable causal graph model.

[0062] In one implementation, the objective function is: (3) in, G This represents the current graph model, where DAG stands for Directed Acyclic Graph; D represents the monitoring data. e L represents a directed edge; G The value of ) represents the objective function; the smaller the value, the stronger the graphical model. G The higher the fit with actual monitoring data and the more the causal path conforms to the geotechnical mechanism, the better the graphical model; P(D| G ) represents the monitoring data D in the graphical model G The likelihood of a model under a given condition reflects the degree of fit between the model and the data; E forbid E represents the set of prohibited edges, dimensionless, containing all causal edges that contradict the geotechnical mechanisms of reservoir bank slopes and have no physical meaning (such as anti-causal edges); must E represents the set of required edges, dimensionless, containing all causal edges that conform to the basic mechanisms of reservoir bank slope geotechnical mechanics and must exist (e.g., rainfall → seepage field, seepage field → slope response); soft I represents the set of soft constraints, dimensionless, containing all causal edges that conform to geotechnical mechanics mechanisms and have a certain probability of existence (e.g., reservoir water level → slope response); e ∈ G ) is an indicator function. e ∈ G The value is 1 when the condition is met, and 0 otherwise; I( e G ) is an indicator function. e G The value is 1 when the condition is met, and 0 otherwise; P( e ) represents a soft-constrained edge eThe prior probability is dimensionless and is set based on the geotechnical properties of the reservoir bank slope (such as lithology coefficient), reflecting the edge... e The a priori possibility of existence; Represents soft-constrained edges e In graphical models G The posterior probability in the equation is dimensionless and reflects the outcome after incorporating monitoring data. e The actual possibility of its existence; 1. 2. 3 indicates the constraint weight, which is set according to the actual project requirements to meet the following conditions. 1= 2> 3. Ensure that hard constraints have higher priority than soft constraints.

[0063] The first term in the objective function is the fitting term obtained by taking the negative of the log-likelihood of the monitoring data under the current graphical model, reflecting the degree to which the current graphical model interprets the monitoring data. The second term, as an indicator function, penalizes the occurrence of prohibited edges in the current graphical model, accumulating the penalty by weight λ1 for each prohibited edge. The third term, also as an indicator function, penalizes the absence of required edges in the current graphical model, accumulating the penalty by weight λ2 for each missing required edge. The fourth term guides the direction of soft-constrained edges in the form of prior and posterior probabilities, ensuring that the presence of soft-constrained edges in the graphical model is as consistent as possible with their prior probabilities. The constraint weights λ1, λ2, and λ3 can be set according to the specific engineering application. Typically, λ1 and λ2 are set to the same large value, while λ3 is set to a relatively small value, so that hard constraints have higher priority than soft constraints.

[0064] By minimizing the above objective function under the premise that the search space is limited to a directed acyclic graph, a trade-off can be achieved between the statistical support provided by the monitoring data and the physical requirements reflected by the constraint information. This results in an interpretable causal graph model that can both fit the monitoring data and meet the physical correlation requirements, further ensuring the physical interpretability of the graph model from the perspective of objective function construction.

[0065] Continue to refer to Figure 1 In step S150, the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope is determined based on the interpretable causal graph model.

[0066] The dynamic contribution refers to the quantitative result of the influence of each instability inducing parameter on the instability of the reservoir bank slope at different time periods. Its value varies with the time period corresponding to the monitoring data, and can reflect the influence of the evolution of the working condition on the relative magnitude of the inducing factor's influence. Based on the interpretable causal graph model, the influence of changing the value of the instability inducing parameter on the instability state parameter can be quantified along the directed path from the instability inducing parameter node through the intermediate node to the instability state node. The quantification results of each instability inducing parameter are then normalized to obtain the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope.

[0067] In one implementation, the determination of the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope based on the interpretable causal graph model includes the following steps: Based on the interpretable causal graph model, the average causal effect of each instability inducing parameter on the instability state parameter is quantified by the backdoor criterion or the frontdoor criterion. The average causal effect was normalized to obtain the dynamic contribution of each instability inducing factor parameter to the instability of the reservoir bank slope.

[0068] The average causal effect refers to the change in the expected value of the unstable state parameter caused by a change in the value of the instability trigger parameter under a given intervention method. It is used to characterize the pure causal influence of the instability trigger parameter on the unstable state parameter. The backdoor criterion is a processing criterion that uses a set of confounding variables to block all backdoor paths between the instability trigger parameter and the target variable, thereby identifying the average causal effect from the observed data. It is applicable when a suitable set of confounding variables can be found in an interpretable causal graphical model. The frontdoor criterion is a processing criterion that decomposes the causal effect between the instability trigger parameter and the target variable into several parts identifiable by the observed data through mediating variables, thereby completing the identification of the causal effect. It is applicable when there are suitable mediating variables between the instability trigger parameter and the target variable, and there is no unobserved confounding between the instability trigger parameter and the mediating variables.

[0069] For example, the backdoor criterion is applied under the condition that: if there exists a set of confounding variables Z(Z... X, and Z does not contain X i In the offspring of Z, X can block i (All backdoor paths between Y), then the backdoor criterion is used to calculate X. i The average causal effect on Y. The front-door criterion is applied when X... i There is a mediating variable M (all) between Y and Y. All causal paths to Y pass through M and X. i There is no unobserved contamination from M to Y, and the contamination from M to Y can be detected by X. i If the blockage occurs, the front-door criterion is used to calculate X. iThe average causal effect on Y.

[0070] Normalizing the average causal effect of each instability trigger parameter obtained by quantification can make the dynamic contribution of each instability trigger parameter fall within a uniform range, which is convenient for horizontal comparison between different triggers and makes the sum of the contributions of all triggers equal to 1, thus more intuitively reflecting the relative importance of each trigger in the current period.

[0071] By specifying the determination of dynamic contribution as a process of first quantifying the average causal effect and then normalizing it, and by using the backdoor criterion or frontdoor criterion to ensure that the quantified result is a pure causal effect rather than a simple statistical correlation, the interpretability and quantification accuracy of the analysis results are further improved.

[0072] In one implementation, the formula for calculating the average causal effect using the backdoor criterion is as follows: (4) Among them, ACE ( x i →Y) represents the parameters of the instability induction factor. x i The larger the value of the average causal effect on the instability state parameter Y, the better. x i The stronger the driving force on slope instability; do( () indicates the interference parameter, representing the externally forced change of the instability trigger parameter. x i The value of is used to eliminate the interference of confounding variable Z and ensure that the calculation results are pure causal effects rather than statistical correlations; express x i The upper limit (corresponding to a high-risk state); express x i The lower limit value (corresponding to a low-risk state); Z represents the confounding variable or set of confounding variables; Given a confounding variable Z=z, When the instability state parameter Y is in the specified state, the conditional expectation is given. Given a confounding variable Z=z, When the instability state parameter Y is in the range of z, P(Z=z) represents the probability that the confounding variable Z takes the value z.

[0073] By using the confounding variable set Z and the weighted summation of each value of Z, the backdoor path can be blocked, and the instability trigger parameters can be calculated from the observation data. x iThe average causal effect on the instability state parameter Y. This form of calculation formula is directly geared towards the application scenario of the backdoor criterion, making the quantification process of the average causal effect mathematically clear and executable. Compared with simple feature importance analysis, it can give quantitative results with causal meaning, making the analysis conclusions easier to corroborate with the geotechnical mechanism, and further improving the physical credibility of the quantification.

[0074] In one implementation, the determination of the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope based on the interpretable causal graph model includes the following steps: Within each time window, the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope is determined based on an interpretable causal graph model, and the dynamic contribution evolution sequence of each instability inducing parameter is formed according to the dynamic contribution of each instability inducing parameter in different time windows.

[0075] A time window refers to a period of time extracted from the monitoring data time axis at a set length. Its length can be set in conjunction with the instability evolution pattern of the reservoir bank slope, as detailed below: Time window length T This refers to the duration of a fixed time window, measured in days (d), with a range of 20-40 days, preferably... T =30d, to ensure that there is enough monitoring data within the time window, while avoiding calculation delays caused by excessive data volume.

[0076] sliding step step This refers to the time interval for advancing the time window, measured in days (d), with a range of 1 to 7 days, preferably... step =1d, enabling daily real-time updates of the contribution of incentives.

[0077] Number of time windows N This refers to the total number of sliding time windows, calculated using the following formula: N = ( T total - T ) / step +1, where T total Total monitoring duration (d). T The time window length is (d). step The sliding step size is (d). This indicates rounding down to the nearest integer.

[0078] Within each time window, only the monitoring data corresponding to that window is used to quantify the dynamic contribution of each instability trigger parameter based on an interpretable causal graph model. As time progresses, the time window slides with a set step size, and the above process is repeated sequentially within new windows to obtain the dynamic contribution of each instability trigger parameter in different time windows. The dynamic contribution of each instability trigger parameter in different time windows is arranged in chronological order, thus forming the evolution sequence of the dynamic contribution of each instability trigger parameter. (The last sentence appears to be incomplete and possibly refers to a time window.) t The end time of the time window, the corresponding time window is [ t - T , t The process of determining the dynamic contribution evolution sequence is as follows: Extraction Time Window The monitoring data within the time window is used to construct a dataset. ,in x i ( t ')for t 'Time-based instability trigger parameters' x i The value of Y( t ')for t The values ​​of the instability state parameters of the reservoir bank slope at time '.

[0079] Based on dataset D t Interpretable causal graphical models that integrate physical information Perform parameter updates, recalculate the causal effect strength of each directed edge, and focus on updating the parameters of each instability inducing factor. x i The average causal effect of Y on ACE t ( x i ).

[0080] Set the time window [ t - T , t The average causal effect of each instability-inducing parameter within the ACE t ( x i After normalization, the dynamic contribution C of each instability inducing factor parameter is obtained. t ( x i The calculation formula is: (5) Among them, C t ( x i (time window) t (i.e., time window) t - T ,t ], abbreviated here as t Internal factors x i The dynamic contribution, ranging from 0 to 1, indicates that the larger the value, the greater the contribution within that time window. x i The stronger the dominant effect on slope instability, the greater the sum of the contributions of all instability-inducing parameters, which equals 1.

[0081] Output Time Window t The contribution vector of the internal factors C t =[C t ( x 1 ),C t ( x 2 ),…,C t ( x m [)] At the same time, the dynamic contribution curves of each instability inducing factor parameter over time are plotted.

[0082] Advance the end time of the time window to t + step Repeat the above steps to obtain the dynamic contribution evolution sequence C of the instability inducing parameters. t C t+step C t+2step ,…,C t+(N-1)step This enables real-time tracking of dynamic contributions.

[0083] By using time windows and dynamic contribution evolution sequences, the relative magnitude of the effects of each instability-causing parameter in different time periods can be continuously tracked. For example, the dominant instability-causing factors may be different during the flood season and dry season, or during periods of high and low reservoir water levels. The dynamic contribution evolution sequence can explicitly reflect these differences, enabling the dynamic contribution to further characterize the temporal evolution and facilitating engineers to focus on key prevention and control measures during the current period.

[0084] In one implementation, reference Figure 4 As shown, the method also includes the following steps: S410, determine the first statistical feature of the monitoring data in the current time window and the second statistical feature of the monitoring data in the previous time window, and calculate the statistical feature deviation between the first statistical feature and the second statistical feature; S420, if the statistical characteristic deviation meets the offset criterion, the initial graphical model is reconstructed and a new interpretable causal graphical model is obtained, so as to determine the dynamic contribution of each instability cause parameter in the current time window based on the new interpretable causal graphical model; S430, if the statistical characteristic deviation does not meet the offset criterion, then update the parameters of the interpretable causal graph model to determine the dynamic contribution of each instability cause parameter in the current time window based on the updated interpretable causal graph model.

[0085] Here, the first statistical characteristic refers to the statistical distribution characteristics of the monitoring data within the current time window, and the second statistical characteristic refers to the similar statistical distribution characteristics of the monitoring data within the previous time window. For example, the current time window is calculated separately. t and the previous time window t - step The statistical distribution characteristics of the monitoring data corresponding to all instability-causing parameters, including mean, variance, and median, are used to obtain the first and second statistical characteristics. The first statistical characteristic includes... μ t , σ t 2 , Med t , respectively representing the current window t The mean, variance, and median; the second statistical characteristic includes μ t-step , σ t-step 2 , Med t-step , respectively representing the previous window t - step The mean, variance, and median.

[0086] Calculate the statistical characteristic deviation between the first and second statistical characteristics, which may include the statistical characteristic deviations of the mean, variance, and median (referred to as the relative deviation of the mean, relative deviation of the variance, and relative deviation of the median, respectively), referring to the following formula: (6) in, μ This represents the relative deviation from the mean. σ This represents the relative deviation of the variance. Med This represents the relative deviation of the median.

[0087] Offset criteria are used to determine whether statistical characteristic deviations reflect significant changes in the distribution of monitored data; for example, a threshold for relative deviation can be set. th (This value can be set based on the stability of the reservoir monitoring data, typically 0.2~0.3). When the relative deviations of multiple components all exceed this threshold, the offset standard is considered met. μ > th , σ > th , Med > th If any two of the three conditions are met, the offset criterion is satisfied, and the data distribution is considered to have shifted significantly.

[0088] If the statistical characteristic deviation meets the offset criterion, it means that the working conditions of the slope may have changed significantly relative to the previous time window. The original interpretable causal graph model may no longer match the correlation reflected by the current data. In this case, the initial graph model is reconstructed to obtain a new interpretable causal graph model, which serves as the basis for determining the dynamic contribution. If the statistical characteristic deviation does not meet the offset criterion, it is considered that the working conditions have not changed significantly, and the original interpretable causal graph model remains structurally applicable. Only the parameters such as the causal effects of its directed edges need to be updated.

[0089] By introducing the aforementioned adaptive triggering mechanism, unnecessary graphical model reconstruction is avoided to save computational costs when the slope conditions do not change significantly. When the conditions do change significantly, the original model is replaced by a new interpretable causal graphical model in a timely manner, thereby ensuring that the identification of dynamic contribution is always consistent with the actual state of the current slope, further improving the adaptability and robustness of the method to dynamic conditions.

[0090] In one implementation, the method further includes the following steps: If the dynamic contribution of any instability-causing parameter increases within k consecutive time windows and the cumulative increase exceeds the preset range, the corresponding first warning information will be triggered. If the dynamic contribution of any instability trigger parameter within a single time window is greater than the high contribution threshold, and the average causal effect corresponding to the dynamic contribution exceeds the risk threshold, then the corresponding second warning information is triggered.

[0091] The term "increase within k consecutive time windows" refers to the increase in the dynamic contribution of a certain instability trigger parameter within each of the k adjacent time windows arranged chronologically compared to the previous window. The value of k can be set based on the time scale of the instability evolution of the reservoir bank slope, such as 7. The cumulative increase refers to the difference in the dynamic contribution of the instability trigger parameter between the first and last windows of the k consecutive time windows. The preset amplitude is a threshold used to determine whether this difference is large enough, and can be set based on engineering experience, such as 0.3. The first warning information is used to indicate that a certain instability trigger parameter is gradually becoming the dominant trigger, reflecting the upward trend in the dynamic contribution evolution sequence, making it easy to detect before the trigger reaches a dominant position.

[0092] The high contribution threshold is used to determine whether a certain instability trigger parameter has become the dominant trigger within a single time window; for example, it could be 0.4. The risk threshold is used to determine whether the corresponding average causal effect has reached a level requiring close attention. The second warning information is used to indicate that a certain instability trigger parameter is currently the dominant trigger and its causal effect is strong, reflecting the characteristics of high contribution and high causal effect within a single time window, which facilitates timely identification of key triggers.

[0093] For example, the dynamic contribution curve of each instability inducing parameter can be monitored. If a certain instability inducing parameter... x i The dynamic contribution curve will trigger the corresponding warning if it meets the following conditions: Within k consecutive windows, x i Its contribution has continued to rise, and the cumulative increase has exceeded γ C t+(k-1)step ( x i )-C t ( x i )>γ, where k is the number of consecutive windows (e.g., k=7). γ A preset threshold (e.g., γ=0.3) is used to measure the cumulative increase, triggering a first warning message (e.g., an intermediate warning message), indicating that the trigger is gradually becoming the dominant failure mode.

[0094] Within a single window, x i Contribution C t ( x i The ACE value is greater than β (β is the high contribution threshold, such as β=0.4), and the corresponding average causal effect ACE is... t ( x iIf the risk threshold is exceeded, a second warning message (such as an advanced warning) will be triggered, indicating that the trigger has become the dominant cause of slope instability and that prevention and control measures should be taken immediately.

[0095] With the help of the two types of early warning information mentioned above, a two-level early warning system for the risk of instability of reservoir bank slopes can be further formed on the basis of the dynamic contribution evolution sequence. This will enable engineering technicians to take control measures of different strengths for trend changes and current critical states, thereby further transforming the evolution results of dynamic contribution into actionable early warning outputs.

[0096] Figure 5 The schematic diagram illustrates the methodological architecture, which is divided into three progressive stages. The first stage involves the spatiotemporal refinement of instability inducing parameters and their adaptation to the reservoir bank scenario. This includes the collection of multi-source instability inducing parameters (including static geotechnical parameters, dynamic hydrological parameters, and time response parameters), the construction of a comprehensive dynamic inducing parameter set (including parameters related to reservoir water level, seepage, rainfall, and displacement), and time-delay effect correction (aligning the time axis based on cross-correlation functions). The second stage involves the construction of an interpretable causal graph model incorporating physical information, including PC algorithm causal graph skeleton discovery, injection of geotechnical mechanics prior constraints, generation of directed acyclic graphs, and quantification of average causal effects. The third stage is the real-time identification of dynamic contributions driven by a sliding time window, including sliding time window rolling updates, recalculation of inducing causal effect intensity, and adaptive triggering of model updates (distribution offset detection). These stages are executed sequentially and work synergistically to ensure the accuracy, interpretability, and real-time nature of inducing factor identification. Finally, the dynamic contribution evolution trajectory of the inducing factors is output for real-time risk warning and to provide dynamic and accurate decision support.

[0097] The following example, using the upper reservoir bank slope of a large hydropower station, further illustrates and verifies the method of this disclosure. The reservoir bank slope is approximately 800m long, with a maximum height of 65m and a slope of... α =35°~45°, elevation H ele =890~955m; the slope soil and rock mass is mainly composed of interlayered silty clay and weathered sandstone, with a lithology coefficient of 890~955m. K lith =0.45 (moderately weak rock mass); horizontal distance of slope from the nearest fault L fault =120m, less affected by fault structures; the reservoir area receives an average annual rainfall of 1200mm, and the reservoir water level is frequently adjusted, with a water level fluctuation Δ H wThe maximum water level can reach 8m, and the duration of extreme high water levels (950m) can be up to 15 days, exhibiting significant time lag and wet-dry cycle effects, consistent with typical hydrogeological characteristics of reservoir bank slopes. Three GNSS displacement monitoring points, two seepage monitoring wells, and one meteorological monitoring station were deployed on the slope to collect multi-source data in real time, including displacement, seepage, rainfall, and reservoir water level. The monitoring cycle is 180 days, with a sampling frequency of once per day to ensure data continuity and integrity.

[0098] First, static topographic and geological parameters are collected. Slope can be extracted from high-precision DEM (Digital Elevation Model) data. α The slopes of the three monitoring points were 38°, 42°, and 40°, with an average of 40°; the elevations were obtained through on-site leveling measurements. H ele The lithology coefficients were determined to be 910m, 930m, and 950m, respectively; based on indoor tests and field investigations, the lithology coefficients were determined. K lith =0.45; the fault distance was determined using ground-penetrating radar. L fault =120m.

[0099] Then, dynamic induced parameters are collected. Based on 180 days of monitoring data from the monitoring system, four types of dynamic induced parameters can be extracted. Some monitoring data (selected from 30 consecutive days of data) are shown in Table 1 below.

[0100] Table 1. 30-day continuous monitoring data of some dynamic trigger parameters

[0101] It should be noted that the cumulative rainfall in the table... R cum The rate of change of water level is the cumulative value for the current day and the previous 7 days. v w The displacement rate is calculated based on the difference between the water level on the current day and the water level on the previous day. v disp Calculations were made based on GNSS monitoring data, and all data underwent outlier removal to ensure accuracy.

[0102] Time lag correction is applied to the monitoring data. For example, cumulative rainfall is selected. R cum Reservoir water level change rate v w Two core dynamic inducing parameters, and displacement rate v disp (i.e., slope response parameters) are time-delayed corrected, and the working conditions of the test reservoir area are combined to set... τ=40d, the optimal time delay was calculated using the cross-correlation function. The cumulative rainfall was calculated. R cum With displacement rate v disp Optimal time delay τ =5d, rate of change of reservoir water level v w With displacement rate v disp Optimal time delay τ =3d, shift the time series of the two dynamic induced parameters to the corresponding optimal time delay time to complete the time delay correction, ensure that the time axis of the dynamic induced parameters and the slope response parameters are accurately aligned, and eliminate the attribution bias caused by time misalignment.

[0103] Determine the set of parameters X (m=16) for instability induction factors. Calculate the instability state parameter Y (safety factor value) of the reservoir bank slope. The daily safety factor value over 180 days can be calculated using the limit equilibrium method. The following parameters are selected for the test slope soil and rock mass: cohesion c=18kPa, internal friction angle... =22°, unit weight γ=19kN / m³, the calculation results show that the safety factor value within 180 days is between 1.12 and 1.35, which meets the requirement of being greater than or equal to 1, indicating that the slope is in a stable state during the test, but there are local displacement anomalies (the displacement rate exceeds 0.3mm / d in some time windows), which is suitable for carrying out the verification of the cause attribution.

[0104] An optimized PC algorithm is adopted, and a Bayesian estimation is introduced to correct the test statistic. Based on the characteristics of the experimental data, the data noise variance is set. σ n ² = 0.02, the variance of the monitoring data corresponding to the instability inducing parameter. σ data ²=0.15. The initial undirected graph contains 17 nodes (16 instability trigger parameters and 1 instability state parameter) and 50 undirected edges. After conditional independence testing, 12 edges with no statistical correlation are removed, resulting in a causal graph skeleton containing 38 undirected edges, as shown below. Figure 6 As shown, its skeletal structure conforms to the basic laws of the inducing effects of reservoir bank slopes.

[0105] Figure 6 The meanings of the physical quantities are as follows: Geo1 represents the lithology coefficient ( K lith Geo2 represents the slope gradient. α Geo3 represents the slope elevation ( H ele Geo4 represents the fault distance ( L fault WL1 represents the reservoir water level fluctuation (Δ).H w WL2 represents the rate of change of reservoir water level ( v w WL3 indicates the duration of extreme high water levels. t high WL4 indicates the frequency of water level fluctuations. f w Per1 represents the pore water pressure ratio ( r u Per2 represents the seepage velocity ( v seep Per3 represents the hydraulic gradient ( i grad Rain1 represents the cumulative rainfall ( ). R cum Rain2 represents rainfall intensity ( I rain Rsp1 represents the displacement rate ( v disp Rsp2 represents the change in tilt (Δ) θ Rsp3 represents displacement acceleration ( a disp Y represents the slope safety factor (i.e., the instability state parameter).

[0106] Based on hard and soft constraints, prior knowledge of geotechnical mechanics was incorporated. Three types of mandatory edges (rainfall dynamic parameters → seepage field dynamic parameters, seepage field dynamic parameters → slope response parameters, reservoir water level dynamic parameters → seepage field dynamic parameters) were retained, and none of the six prohibited edges appeared, meeting the requirements of hard constraints. Soft-constrained edges were guided by probability, with the probability of reservoir water level dynamic parameters → slope response parameters being 0.45 (corresponding to the lithology coefficient). K lith =0.45 (consistent), the probability of static topographic geological parameters → instability parameters is 0.55, and the probability of slope response parameters → instability parameters is 0.9. Set constraint weights λ1=λ2=10, λ3=5, and minimize the objective function L( G This leads to the final interpretable causal graphical model that integrates physical information. G It clearly presents the directional causal path between each instability inducing parameter and the instability state parameter, and there is no path with anti-causality or physical meaning.

[0107] The backdoor criterion is used to calculate the average causal effect (ACE) of each instability trigger parameter on Y (safety factor value). For example, Figure 2 The lithology coefficient is shown. K lith pore water pressure ratio ru Displacement rate v disp Cumulative rainfall R cum Reservoir water level fluctuation Δ H w The ACE values ​​of these five instability-inducing parameters.

[0108] Table 2. Average Causal Effect (ACE) Values ​​of Some Instability-Inducing Parameters

[0109] As shown in Table 2, the displacement rate v disp lithology coefficient K lith pore water pressure ratio r u It is the core factor affecting slope stability, and its absolute ACE value is greater than 0.75, which is highly consistent with the mechanism of geotechnical mechanics, verifying the rationality and accuracy of the cause-effect diagram model.

[0110] The time window length was set based on the experimental monitoring period (180 days). T =30d, sliding step size step =1d, calculate the total number of time windows. N = (180-30) / 1 +1 = 151. The dynamic contribution of each instability trigger parameter is updated daily in real-time according to a time window, balancing computational accuracy and efficiency.

[0111] Using the 30th, 60th, 90th, 120th, 150th, and 180th days as the end times of the time windows, the monitoring data within each time window are extracted, the causal effect strength is updated, and the dynamic contribution is calculated. Figure 7 The evolution of the dynamic contribution of some instability-causing parameters is shown, including displacement rate. v disp The contribution of pore water pressure was consistently the highest (0.23~0.27), making it the dominant cause of slope instability during the experiment; pore water pressure ratio r u and lithology coefficient K lith The contribution of cumulative rainfall was second only to that of other rainfall sources. R cum The change in reservoir water level Δ H w The contribution of each instability inducing factor parameter is relatively low, which is consistent with the ACE quantification results. Moreover, the contribution of each instability inducing factor parameter shows dynamic evolution over time, which is consistent with the dynamic working condition change law of the reservoir bank slope.

[0112] Set offset threshold δ th =0.25, a distribution offset test was performed on 151 time windows, specifically examining whether the statistical characteristic deviation between the first statistical characteristic of the monitoring data in each time window and the second statistical characteristic of the monitoring data in the previous time window met the offset criteria. The model was retrained on the 89th and 132nd time windows (satisfying that the relative deviations of the mean and variance statistical characteristics exceeded the threshold). After retraining, the causal structure of the graphical model did not change significantly, but after updating the causal effect parameters, the attribution accuracy improved by 8.3%, verifying the effectiveness of the adaptive update mechanism and ensuring that the graphical model always adapts to the dynamic response characteristics of the slope.

[0113] With k=7, γ=0.3, and β=0.4, the dynamic contribution of each instability inducing parameter within each time window was monitored. No advanced warning was triggered during the experiment. From day 145 to 151 (7 consecutive windows), the pore water pressure ratio... r u The dynamic contribution increased from 0.21 to 0.24, with a cumulative increase of 0.03. It did not reach γ=0.3 and did not trigger a medium-level warning, indicating that the reservoir bank slope was generally stable during the test, which was consistent with the actual monitoring results (safety factor value ≥1.12), and verified the rationality of the warning logic.

[0114] To further verify the accuracy and superiority of this implementation method, it is compared with the traditional SHAP (SHapley Additive exPlanations) attribution method and the static causal attribution method. The comparison indicators include attribution accuracy, interpretability score, and real-time performance (computation time per single time window). The specific comparison results are as follows: Figure 8 As shown. By Figure 8 It can be seen that the attribution accuracy of this implementation method is 14.2% higher than that of the traditional SHAP method and 11.4% higher than that of the static causal attribution method. The interpretability score is significantly higher than that of the two traditional methods. This is because this implementation method integrates the physical mechanism of geotechnical mechanics, the causal path is traceable and the effect is quantifiable, and it solves the black box defect of the traditional method. Although the calculation time of a single time window is slightly longer than that of the traditional method, the calculation time of 12.8s can meet the real-time prevention and control needs of engineering, and the overall performance is optimal.

[0115] Meanwhile, the core instability induction parameters output by this implementation method (such as displacement rate, pore water pressure ratio, and lithology coefficient) are highly consistent with the actual hydrogeological characteristics of the test slope and the results of rock and soil mechanics analysis, further verifying the scientific nature, feasibility, and engineering applicability of this implementation method.

[0116] In summary, the embodiments disclosed herein have the following technical advantages: By refining the spatiotemporal representation of multidimensional instability inducing parameters and correcting for time lag effects, the fundamental accuracy of inducing factor identification is improved, addressing the shortcomings of traditional parameters being one-sided and having misaligned time lags. A comprehensive parameter system is constructed, encompassing static topographic and geological parameters as well as dynamic inducing parameters. This system covers fundamental characteristics such as static slope and lithology coefficients, as well as core dynamic characteristics such as reservoir water level dynamics, seepage field dynamics, rainfall dynamics, and slope response. The optimal time lag is calculated using a cross-correlation function, and the inducing factor time series is shifted and corrected, achieving precise alignment of the instability inducing parameters and slope response parameters on their time axes. This overcomes the shortcomings of traditional inducing parameter systems being one-sided and static, eliminates the time misalignment between inducing factors and response caused by time lag effects, and provides high-quality, scenario-based parameter input for subsequent causal modeling, reducing attribution bias at its source. It is adapted to the unique hydrogeological scenarios of reservoir bank slopes (such as reservoir water level fluctuations, seepage field changes, etc.), with comprehensive parameter coverage and close alignment with actual engineering conditions. The time delay correction method is scientific and repeatable, eliminating the need for additional monitoring equipment and directly compatible with existing monitoring data, thus reducing data preprocessing costs.

[0117] By constructing an interpretable causal graphical model that integrates physical information, this approach breaks through the black-box nature of traditional machine learning algorithms, achieving a leap from statistical correlation to physical causality and improving the interpretability and scientific rigor of attribution. A three-step method is employed: data-driven causal graph skeleton discovery, injection of physical prior constraints, and quantification of causal effects. Based on an optimized PC algorithm, the causal graph skeleton is discovered, and Bayesian estimation is introduced to correct the test statistics and enhance noise resistance. Prior knowledge of reservoir bank slope geotechnical mechanics is transformed into hard constraints of mandatory and forbidden edges and soft constraints guided by probability, injected into the causal structure learning process. The average causal effect of each inducement is quantified using backdoor or frontdoor criteria. This approach addresses the shortcomings of traditional machine learning algorithms, such as pure data-driven approaches, lack of physical mechanism constraints, and black-box interpretability. It ensures that the final graphical model conforms to statistical laws and strictly adheres to the basic principles of geotechnical mechanics, clearly tracing the complete causal chain from inducement to mediating variables to slope instability, achieving traceability and interpretability of attribution results. The causal effect is quantified accurately with an error controlled within 5%. The attribution results are highly consistent with engineering practice and geotechnical mechanics mechanisms, avoiding attribution paths that are contrary to causality or have no physical meaning, and providing a reliable basis for engineering decision-making.

[0118] This system achieves real-time identification and adaptive model updates of the dynamic contribution of various instability inducing parameters driven by a sliding time window, enabling dynamic and real-time attribution of instability causes and adapting to dynamic slope conditions. By setting reasonable sliding time window parameters (such as time window length and sliding step size), monitoring data is extracted through the sliding time window, updating the causal effect intensity and calculating the dynamic contribution of each instability inducing parameter. An adaptive model update triggering mechanism is introduced, which automatically triggers model retraining when the statistical distribution of monitoring data shifts significantly, and only updates parameters when there is no shift. This overcomes the limitation of traditional static attribution schemes that cannot track the temporal evolution of inducing factor contribution, achieving real-time updates of inducing factor contribution and enabling timely capture of dynamic hydrogeological conditions of reservoir bank slopes (such as changes in reservoir water level scheduling and differences in rainfall distribution). The adaptive update mechanism ensures that the model always remains synchronized with the current dynamic response characteristics of the slope, improving the robustness and adaptability of dynamic identification. Balancing computational accuracy and efficiency, the single-window computation time is only 12.8 seconds, meeting the real-time control requirements of engineering projects. The model is flexible in updating, avoiding the efficiency loss caused by blind retraining and preventing attribution bias caused by sudden changes in data distribution. It can adapt to various dynamic working conditions of reservoir bank slopes.

[0119] By employing dynamic early warning triggering logic and a closed-loop technology across the entire chain, early warning of slope instability risks is achieved, enhancing the timeliness and targeted nature of safety control. Based on the dynamic contribution evolution trajectory of instability inducing parameters, multi-level early warning triggering conditions are set (e.g., a cumulative increase in dynamic contribution over multiple consecutive time windows triggers a medium-level early warning, while a high contribution within a single time window triggers a high-level early warning). Combined with causal effect quantification results, a complete technical closed loop of "data input → feature processing → model building → dynamic identification → early warning output" is formed. This allows for early identification of the evolution trend of dominant inducing factors, timely triggering of medium and high-level early warnings, reserving sufficient time for slope safety control, effectively reducing the probability of slope instability disasters, and avoiding losses such as engineering damage and casualties. The early warning logic aligns with engineering realities, resulting in low false alarm and missed alarm rates. The early warning threshold can be flexibly adjusted according to the actual working conditions of the reservoir area, adapting to different types of reservoir bank slopes. The early warning results are directly correlated with the contribution of inducing factors and causal effects, facilitating targeted prevention and control measures by engineering technicians, avoiding blind prevention and control, and improving prevention and control efficiency.

[0120] By ensuring the technology's universality through design and engineering implementation, its adaptability and feasibility are enhanced, reducing engineering application costs and expanding its application scope. Adopting a modular design with a fixed core technical architecture, it can be adapted to different types of reservoir bank slopes through feature adjustments and parameter optimization (such as window parameters, constraint weights, and early warning thresholds). A universal data interface is developed, compatible with existing monitoring systems, and implementation support measures such as equipment adaptation, personnel training, and operation and maintenance management are provided. It can be widely adapted to various reservoir bank slope scenarios such as pumped storage power stations, large reservoirs, and hydropower stations, without requiring major modifications to the core technical architecture. Existing monitoring equipment can be directly reused, reducing engineering application costs and implementation difficulty. The engineering process is clear and the steps are repeatable, facilitating technology promotion and application. It is highly versatile and scalable, allowing for rapid adaptation to the hydrogeological characteristics of different reservoir areas. Implementation costs are low, requiring no additional dedicated equipment, and operation and maintenance are simple, making it suitable for large-scale engineering promotion. It also lays the foundation for subsequent technology optimization and multi-source data fusion expansion.

[0121] This disclosure also provides a device for analyzing the causes of reservoir bank slope instability. (Reference) Figure 9 As shown, the reservoir bank slope instability cause analysis device 900 includes: The monitoring data acquisition module 910 is configured to acquire monitoring data of the reservoir bank slope. The monitoring data includes monitoring data corresponding to various instability inducing parameters, including static topographic and geological parameters and dynamic inducing parameters. The graph model construction module 920 is configured to construct an initial graph model using each instability inducing factor parameter and the instability state parameter of the reservoir bank slope as nodes. The constraint information determination module 930 is configured to determine constraint information based on the physical relationship between different instability inducing parameters and the physical relationship between instability inducing parameters and instability state parameters; The graph model update module 940 is configured to inject the monitoring data and the constraint information into a pre-constructed objective function, and update the initial graph model through the objective function to obtain an interpretable causal graph model. The contribution determination module 950 is configured to determine the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope based on the interpretable causal graph model.

[0122] In one embodiment, the instability inducing parameters include slope response parameters and other inducing parameters; the device is further configured to: After acquiring the monitoring data of the reservoir bank slope, the monitoring data at different times corresponding to the other inducing parameters are used to form an inducing time series, and the monitoring data at different times corresponding to the slope response parameters are used to form a slope response time series. Substitute the time series of the precipitating factors and the time series of the slope response into the cross-correlation function, and determine the optimal time delay between the other precipitating factor parameters and the slope response parameters by calculating the maximum value of the cross-correlation function; Adjust the time information of the monitoring data corresponding to the other inducing parameters according to the optimal time lag time, so as to align the other inducing parameters and the slope response parameters in time.

[0123] In one implementation, the cross-correlation function is: ; in, τ Represents the time delay; E[·] represents the mathematical expectation; X( t Y( represents the time series of the triggers; Y( t () represents the slope response time series; μ X The time series X represents the cause. t The mean of ) μ Y The slope response time series Y( t The mean of ) σ X The time series X represents the cause. t The standard deviation of ) σ Y The slope response time series Y( t The standard deviation of ρ( τ When the maximum value is reached, the corresponding τ The optimal time delay is given.

[0124] In one implementation, the initial graphical model is a causal graph skeleton; the construction of the initial graphical model using each instability inducing factor parameter and the instability state parameters of the reservoir bank slope as nodes includes: An initial undirected graph is constructed using each instability induction parameter and the instability state parameter of the reservoir bank slope as nodes; an undirected edge exists between any two nodes in the initial undirected graph; For any two nodes in the initial undirected graph, perform a conditional independence test. If the two nodes are not conditionally independent, delete the undirected edge between the two nodes to obtain the intermediate undirected graph. For any two nodes in the intermediate undirected graph, perform a conditional independence test. If the two nodes are conditionally independent under the set of conditions, delete the undirected edge between the two nodes. Iterate and update the intermediate undirected graph by gradually increasing the size of the condition set until the size of the condition set reaches its maximum and the remaining undirected edges cannot be deleted, thus obtaining the causal graph skeleton.

[0125] In one implementation, when performing a conditional independence test on two nodes, a Bayesian estimation is introduced to correct the test statistic. The correction formula is as follows: ; in, This represents the original test statistic. This represents the corrected test statistic; This represents the variance of the monitoring data corresponding to the parameters that cause instability. This represents the variance of data noise.

[0126] In one implementation, the constraint information includes hard constraints and soft constraints; the hard constraints include required edges and prohibited edges. The required edges include: directed edges from rainfall dynamic parameters to seepage field dynamic parameters, directed edges from seepage field dynamic parameters to slope response parameters, and directed edges from reservoir water level dynamic parameters to seepage field dynamic parameters. The prohibited edges include: directed edges where slope response parameters point to reservoir water level dynamic parameters, directed edges where slope response parameters point to rainfall dynamic parameters, and directed edges where static topographic and geological parameters point to dynamic inducing parameters. The soft constraints include: the probability of the existence of the directed edge of the reservoir water level dynamic parameter pointing to the slope response parameter is positively correlated with the lithology coefficient of the reservoir bank slope; the probability of the existence of the directed edge of the static topographic and geological parameter pointing to the instability state parameter is negatively correlated with the lithology coefficient of the reservoir bank slope; and the probability of the existence of the directed edge of the slope response parameter pointing to the instability state parameter is a preset value.

[0127] In one implementation, the objective function is: ; in, G This represents the current graph model, where DAG stands for Directed Acyclic Graph; D represents the monitoring data. e L represents a directed edge; G P(D|) represents the objective function value; G ) represents the monitoring data D in the graphical model G likelihood of E forbid E represents the set of forbidden edges; must E represents the set of required edges; soft I represents the set of soft constraints; e ∈ G ) is an indicator function. e ∈ G The value is 1 when the condition is met, and 0 otherwise; I( e G ) is an indicator function. e G The value is 1 when the condition is met, and 0 otherwise; P( e ) represents a soft-constrained edge e The prior probability; Represents soft-constrained edges e In graphical models G The posterior probability in; 1. 2. 3 indicates the constraint weight.

[0128] In one implementation, determining the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope based on the interpretable causal graph model includes: Based on the interpretable causal graph model, the average causal effect of each instability inducing parameter on the instability state parameter is quantified by the backdoor criterion or the frontdoor criterion. The average causal effect is normalized to obtain the dynamic contribution of each instability inducing factor parameter to the instability of the reservoir bank slope.

[0129] In one implementation, the formula for calculating the average causal effect using the backdoor criterion is as follows: ; Among them, ACE ( x i →Y) represents the parameters of the instability induction factor. x i The average causal effect on the instability state parameter Y; do( () indicates the interference parameter, representing the externally forced change of the instability trigger parameter. x i The value of ; express x i The upper limit; express x i The lower bound value; Z represents the set of mixed variables; Given a confounding variable Z=z, When the instability state parameter Y is in the specified state, the conditional expectation is given. Given a confounding variable Z=z, When the instability state parameter Y is in the range of z, P(Z=z) represents the probability that the confounding variable Z takes the value z.

[0130] In one implementation, determining the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope based on the interpretable causal graph model includes: Within each time window, the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope is determined based on the interpretable causal graph model, and an evolution sequence of the dynamic contribution of each instability inducing parameter is formed according to the dynamic contribution of each instability inducing parameter in different time windows.

[0131] In one embodiment, the device is further configured to: Determine the first statistical feature of the monitoring data within the current time window and the second statistical feature of the monitoring data within the previous time window, and calculate the statistical feature deviation between the first statistical feature and the second statistical feature; If the statistical characteristic deviation meets the offset criterion, the initial graphical model is reconstructed and a new interpretable causal graphical model is obtained, so as to determine the dynamic contribution of each instability inducing parameter in the current time window based on the new interpretable causal graphical model; If the statistical characteristic deviation does not meet the offset criterion, the parameters of the interpretable causal graph model are updated to determine the dynamic contribution of each instability instability cause parameter within the current time window based on the updated interpretable causal graph model.

[0132] In one embodiment, the device is further configured to: If the dynamic contribution of any instability-causing parameter increases within k consecutive time windows and the cumulative increase exceeds the preset range, the corresponding first warning information will be triggered. If the dynamic contribution of any instability trigger parameter within a single time window is greater than the high contribution threshold, and the average causal effect corresponding to the dynamic contribution exceeds the risk threshold, then the corresponding second warning information is triggered.

[0133] In one embodiment, the static topographic and geological parameters include slope, elevation, lithology coefficient, and fault distance; The dynamic inducing parameters include reservoir water level dynamic parameters, seepage field dynamic parameters, rainfall dynamic parameters, and slope response parameters; wherein, the reservoir water level dynamic parameters include water level amplitude, water level change rate, water level fluctuation frequency, and duration of extreme high water levels; the seepage field dynamic parameters include seepage velocity, pore water pressure ratio, and hydraulic gradient; the rainfall dynamic parameters include cumulative rainfall and rainfall intensity; and the slope response parameters include displacement rate, displacement acceleration, and slope change.

[0134] The specific details of each part of the above-mentioned device have been described in detail in the method section of the implementation plan. For any undisclosed details, please refer to the implementation plan of the method section, and therefore will not be repeated here.

[0135] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to exemplary embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0136] This disclosure also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the method steps of various exemplary embodiments of this disclosure.

[0137] In one implementation, the computer program product can be a tangible product, such as a computer-readable storage medium storing a computer program. The readable storage medium can be based on electrical, magnetic, optical, electromagnetic, infrared, or other signals, and includes, but is not limited to: Random Access Memory (RAM), Read-Only Memory (ROM), magnetic tape, floppy disk, flash memory, Hard Disk Drive (HDD), Solid State Disk (SSD), etc. For example, the computer program product can be a non-volatile storage medium storing a computer program, such as read-only memory, NAND flash memory, etc.

[0138] In one implementation, the computer program product can be an intangible product. For example, the computer program product can be a virtual digital product, such as an executable file or installation package containing a computer program.

[0139] Computer program code can be written in one or more programming languages. Examples of programming languages ​​include C, Java, and C++. Program code can execute entirely on the user's computing device, partially on the user's computing device, or as a standalone software package. It can also execute partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, such as a Local Area Network (LAN) or a Wide Area Network (WAN), or it can be connected to an external computing device (e.g., via an internet connection provided by a mobile network operator).

[0140] Computer programs can be carried or transmitted via signals such as electrical, magnetic, optical, electromagnetic, and infrared rays. Electronic devices can convert the signals carrying computer programs into digital signals, thereby running the computer programs. When a computer program runs on an electronic device, its code is used to cause the electronic device to execute (more specifically, to be executed by the processor of the electronic device) the method steps of various embodiments of this disclosure, such as... Figure 1 The method and steps.

[0141] Implementing the above method steps through a computer program achieves the following technical effects: the acquired monitoring data simultaneously covers both static topographic and geological parameters and dynamic causal parameters, avoiding the incomplete coverage problem caused by relying on a single indicator or focusing only on static geological conditions. An initial graphical model is constructed using instability causal parameters and instability state parameters as nodes, and then updated using an objective function. This allows the relationships between parameters to be presented in a graphical structure, facilitating the tracing of causal influence processes along directed paths between nodes. The introduced constraint information originates from the physical relationships between instability causal parameters and between instability causal parameters and instability state parameters. These constraints work in conjunction with the monitoring data in the objective function, ensuring that the resulting interpretable causal graphical model is supported by the statistical regularities reflected in the monitoring data and constrained by the physical relationships in geotechnical mechanics, thus mechanistically suppressing the occurrence of correlation paths that contradict physical mechanisms. Based on this interpretable causal graph model, the dynamic contribution of each instability inducing factor parameter to the instability of the reservoir bank slope is determined. It can characterize the changes in the strength of each inducing factor at different time periods as the monitoring data is updated, which makes it easier for engineering technicians to take targeted prevention and control measures for the inducing factors that play a dominant role under the current working conditions.

[0142] This disclosure also provides an electronic device. The electronic device includes a processor and a memory. The memory stores executable instructions for the processor, such as computer programs. The processor executes the executable instructions to perform the method steps of various exemplary embodiments of this disclosure.

[0143] The following is for reference. Figure 10 The electronic device is illustrated by way of a general-purpose computing device. It should be understood that... Figure 10 The electronic device 1000 shown is merely an example and should not be construed as limiting the functionality or scope of this disclosure.

[0144] like Figure 10 As shown, the electronic device 1000 may include: a processor 1010, a memory 1020, a bus 1030, an I / O (input / output) interface 1040, and a network adapter 1050.

[0145] The memory 1020 may include volatile memory, such as RAM 1021 and cache unit 1022, and may also include non-volatile memory, such as ROM 1023. The memory 1020 may also include one or more program modules 1024, such program modules 1024 including, but not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. For example, program module 1024 may include the modules described above.

[0146] The processor 1010 may include one or more processing units, such as an AP (Application Processor), a modem processor, a GPU (Graphics Processing Unit), an ISP (Image Signal Processor), a controller, an encoder, a decoder, a DSP (Digital Signal Processor), a baseband processor, and / or an NPU (Neural-Network Processing Unit).

[0147] The processor 1010 can be used to execute executable instructions stored in the memory 1020 to perform method steps of various embodiments of this disclosure, such as... Figure 1 The method and steps.

[0148] By executing the above method steps through processor 1010, the following technical effects are achieved: the acquired monitoring data simultaneously covers both static topographic and geological parameters and dynamic causal parameters, avoiding the incomplete coverage problem caused by relying on only a single indicator or focusing only on static geological conditions. An initial graphical model is constructed using instability causal parameters and instability state parameters as nodes, and then the graphical model is updated using an objective function. This allows the relationships between parameters to be presented in a graphical structure, facilitating the tracing of the causal influence process along directed paths between nodes. The introduced constraint information originates from the physical relationships between instability causal parameters and between instability causal parameters and instability state parameters. This information works in conjunction with the monitoring data in the objective function, ensuring that the resulting interpretable causal graphical model is supported by the statistical regularities reflected in the monitoring data and constrained by the physical relationships in geotechnical mechanics. This mechanism suppresses the occurrence of correlation paths that contradict physical mechanisms. Based on this interpretable causal graph model, the dynamic contribution of each instability inducing factor parameter to the instability of the reservoir bank slope is determined. It can characterize the changes in the strength of each inducing factor at different time periods as the monitoring data is updated, which makes it easier for engineering technicians to take targeted prevention and control measures for the inducing factors that play a dominant role under the current working conditions.

[0149] Bus 1030 is used to connect different components of electronic device 1000, and may include data bus, address bus and control bus.

[0150] Electronic device 1000 can communicate with one or more external devices 1100 (such as keyboard, mouse, external controller, etc.) through I / O interface 1040.

[0151] Electronic device 1000 can communicate with one or more networks via network adapter 1050. For example, network adapter 1050 can provide mobile communication solutions such as 3G / 4G / 5G, or wireless communication solutions such as wireless LAN, Bluetooth, and near-field communication. Network adapter 1050 can communicate with other modules of electronic device 1000 via bus 1030.

[0152] In one embodiment, the electronic device 1000 further includes a display for displaying a graphical user interface.

[0153] although Figure 10 As not shown in the diagram, other hardware and / or software modules may also be configured in the electronic device 1000, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (Redundant Arrays of Independent Disks) systems, tape drives, and data backup storage systems.

[0154] As can be seen from the above, the technical solutions disclosed herein can be implemented as methods, apparatus, systems, computer program products, storage media, electronic devices, etc. Those skilled in the art will understand that various aspects of this disclosure can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or an implementation combining hardware and software aspects. Exemplarily, these three forms can be referred to as "circuit," "module," and "system," respectively.

[0155] It should be understood that this disclosure is not limited to the specific methods, steps, or structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. Those skilled in the art will readily conceive of other embodiments based on the specific implementations provided in this disclosure. Therefore, the specific implementations provided in this disclosure are merely exemplary, and the scope and spirit of this disclosure are indicated by the claims, and should cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary technical means in the art not disclosed in this disclosure.

Claims

1. A method for analyzing the causes of instability on a reservoir bank slope, characterized in that, The method includes: Acquire monitoring data of the reservoir bank slope, the monitoring data including monitoring data corresponding to various instability inducing parameters, the instability inducing parameters including static topographic and geological parameters and dynamic inducing parameters; An initial graphical model is constructed using each instability induction parameter and the instability state parameter of the reservoir bank slope as nodes. Constraint information is determined based on the physical relationships between different instability inducing parameters and the physical relationships between instability inducing parameters and instability state parameters; The monitoring data and the constraint information are injected into a pre-constructed objective function, and the initial graph model is updated through the objective function to obtain an interpretable causal graph model. The dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope is determined based on the interpretable causal graph model. The constraint information includes hard constraints and soft constraints. Hard constraints include mandatory edges and prohibited edges. Mandatory edges include directed edges pointing from rainfall dynamic parameters to seepage field dynamic parameters, directed edges pointing from seepage field dynamic parameters to slope response parameters, and directed edges pointing from reservoir water level dynamic parameters to seepage field dynamic parameters. Prohibited edges include directed edges pointing from slope response parameters to reservoir water level dynamic parameters, directed edges pointing from slope response parameters to rainfall dynamic parameters, and directed edges pointing from static topographic and geological parameters to dynamic inducing parameters. Soft constraints include: the probability of a directed edge pointing from reservoir water level dynamic parameters to slope response parameters is positively correlated with the lithology coefficient of the reservoir bank slope; the probability of a directed edge pointing from static topographic and geological parameters to instability state parameters is negatively correlated with the lithology coefficient of the reservoir bank slope; and the probability of a directed edge pointing from slope response parameters to instability state parameters is a preset value. The objective function is: ; in, G This represents the current graph model, where DAG stands for Directed Acyclic Graph; D represents the monitoring data. e L represents a directed edge; G P(D|) represents the objective function value; G ) represents the monitoring data D in the graphical model G likelihood of E forbid E represents the set of forbidden edges; must E represents the set of required edges; soft I represents the set of soft constraints; e ∈ G ) is an indicator function. e ∈ G The value is 1 when the condition is met, and 0 otherwise; I( e G ) is an indicator function. e G The value is 1 when the condition is met, and 0 otherwise; P( e ) represents a soft-constrained edge e The prior probability; Represents soft-constrained edges e In graphical models G The posterior probability in; 1.

2. 3 indicates the constraint weight.

2. The method according to claim 1, characterized in that, The instability inducing parameters include slope response parameters and other inducing parameters; the method further includes: After acquiring the monitoring data of the reservoir bank slope, the monitoring data at different times corresponding to the other inducing parameters are used to form an inducing time series, and the monitoring data at different times corresponding to the slope response parameters are used to form a slope response time series. Substitute the time series of the precipitating factors and the time series of the slope response into the cross-correlation function, and determine the optimal time delay between the other precipitating factor parameters and the slope response parameters by calculating the maximum value of the cross-correlation function; Adjust the time information of the monitoring data corresponding to the other inducing parameters according to the optimal time lag time, so as to align the other inducing parameters and the slope response parameters in time.

3. The method according to claim 2, characterized in that, The cross-correlation function is: ； in, τ Represents the time delay; E[·] represents the mathematical expectation; X( t Y( represents the time series of the triggers; Y( t () represents the slope response time series; μ X The time series X represents the cause. t The mean of ) μ Y The slope response time series Y( t The mean of ) σ X The time series X represents the cause. t The standard deviation of ) σ Y Represents the slope response time series Y( t The standard deviation of ρ( τ When the maximum value is reached, the corresponding τ The optimal time delay is given.

4. The method according to claim 1, characterized in that, The initial graph model is a causal graph skeleton; the initial graph model is constructed by using each instability inducing factor parameter and the instability state parameters of the reservoir bank slope as nodes, including: An initial undirected graph is constructed using each instability induction parameter and the instability state parameter of the reservoir bank slope as nodes; an undirected edge exists between any two nodes in the initial undirected graph; For any two nodes in the initial undirected graph, perform a conditional independence test. If the two nodes are not conditionally independent, delete the undirected edge between the two nodes to obtain the intermediate undirected graph. For any two nodes in the intermediate undirected graph, perform a conditional independence test. If the two nodes are conditionally independent under the set of conditions, delete the undirected edge between the two nodes. Iterate and update the intermediate undirected graph by gradually increasing the size of the condition set until the size of the condition set reaches its maximum and the remaining undirected edges cannot be deleted, thus obtaining the causal graph skeleton.

5. The method according to claim 4, characterized in that, When performing a conditional independence test on two nodes, Bayesian estimation is introduced to correct the test statistic. The correction formula is as follows: ； in, This represents the original test statistic. This represents the corrected test statistic; This represents the variance of the monitoring data corresponding to the parameters that cause instability. This represents the variance of data noise.

6. The method according to claim 1, characterized in that, The determination of the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope based on the interpretable causal graph model includes: Based on the interpretable causal graph model, the average causal effect of each instability inducing parameter on the instability state parameter is quantified by the backdoor criterion or the frontdoor criterion. The average causal effect is normalized to obtain the dynamic contribution of each instability inducing factor parameter to the instability of the reservoir bank slope.

7. The method according to claim 6, characterized in that, The formula for calculating the average causal effect using the backdoor criterion is: ； Among them, ACE ( x i →Y) represents the parameters of the instability induction factor. x i The average causal effect on the instability state parameter Y; do( () indicates the interference parameter, representing the externally forced change of the instability trigger parameter. x i The value of ; express x i The upper limit; express x i The lower bound value; Z represents the set of mixed variables; Given a confounding variable Z=z, When the instability state parameter Y is in the specified state, the conditional expectation is given. Given a confounding variable Z=z, When the instability state parameter Y is in the range of z, P(Z=z) represents the probability that the confounding variable Z takes the value z.

8. The method according to claim 1, characterized in that, The determination of the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope based on the interpretable causal graph model includes: Within each time window, the dynamic contribution of each instability inducing parameter to the instability of the reservoir bank slope is determined based on the interpretable causal graph model, and an evolution sequence of the dynamic contribution of each instability inducing parameter is formed according to the dynamic contribution of each instability inducing parameter in different time windows.

9. The method according to claim 8, characterized in that, The method further includes: Determine the first statistical feature of the monitoring data within the current time window and the second statistical feature of the monitoring data within the previous time window, and calculate the statistical feature deviation between the first statistical feature and the second statistical feature; If the statistical characteristic deviation meets the offset criterion, the initial graphical model is reconstructed and a new interpretable causal graphical model is obtained, so as to determine the dynamic contribution of each instability inducing parameter in the current time window based on the new interpretable causal graphical model; If the statistical characteristic deviation does not meet the offset criterion, the parameters of the interpretable causal graph model are updated to determine the dynamic contribution of each instability instability cause parameter within the current time window based on the updated interpretable causal graph model.

10. The method according to claim 8, characterized in that, The method further includes: If the dynamic contribution of any instability-causing parameter increases within k consecutive time windows and the cumulative increase exceeds the preset range, the corresponding first warning information will be triggered. If the dynamic contribution of any instability trigger parameter within a single time window is greater than the high contribution threshold, and the average causal effect corresponding to the dynamic contribution exceeds the risk threshold, then the corresponding second warning information is triggered.

11. The method according to any one of claims 1 to 10, characterized in that, The static topographic and geological parameters include slope, elevation, lithology coefficient, and fault distance; The dynamic inducing parameters include reservoir water level dynamic parameters, seepage field dynamic parameters, rainfall dynamic parameters, and slope response parameters; wherein, the reservoir water level dynamic parameters include water level amplitude, water level change rate, water level fluctuation frequency, and duration of extreme high water levels; the seepage field dynamic parameters include seepage velocity, pore water pressure ratio, and hydraulic gradient; the rainfall dynamic parameters include cumulative rainfall and rainfall intensity; and the slope response parameters include displacement rate, displacement acceleration, and slope change.

12. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 11.

13. An electronic device, characterized in that, include: Processor and memory; The memory is used to store executable instructions of the processor; the processor is configured to implement the method of any one of claims 1 to 11 by executing the executable instructions.

Images