Slope quality monitoring and early warning method and system based on seismic data

Abstract

The application provides a slope quality monitoring and early warning method and system based on seismic data, which first acquires a set of seismic monitoring data of a target slope area, covering seismic waveform data and environment-related parameters of a plurality of monitoring nodes within a preset time period, then dynamically preprocesses the set of seismic monitoring data, including data alignment, noise filtering and dimension unification, to obtain a standardized data set, then extracts a dynamic correlation feature set containing seismic energy distribution, slope response frequency and environment coupling fluctuation characteristics based on the standardized data, then calls a slope quality analysis model to predict the state of the feature set, generates a slope quality evaluation result according to the difference between the stable state characteristics and the abnormal fluctuation characteristics, and finally determines an early warning strategy according to the slope quality evaluation result and triggers an early warning signal output, thereby achieving effective monitoring and early warning of the slope quality.

Description

Technical Field

[0001] The present invention relates to the field of geological disaster early warning technology, and in particular to a slope quality monitoring and early warning method and system based on seismic data. Background Art

[0002] Slope quality monitoring and early warning are crucial for ensuring infrastructure safety and protecting people's lives and property. Existing technologies primarily rely on traditional monitoring methods. Some methods rely solely on displacement sensors to monitor slope displacement changes. This approach only provides limited information on slope displacement and fails to fully reflect the combined impact of internal and surrounding environmental factors on slope stability.

[0003] In addition, when evaluating slope quality, existing technologies mostly use simple threshold judgment methods, lack state prediction processing based on scientific models, cannot accurately distinguish between the stable state and abnormal fluctuations of the slope, and it is difficult to generate comprehensive and accurate slope quality assessment results.

[0004] In the early warning link, the existing early warning strategies are often relatively fixed and single, and cannot be flexibly adjusted according to different slope quality assessment results, resulting in insufficient timeliness and effectiveness of early warnings, and unable to provide scientific and reasonable response measures and suggestions for relevant personnel. Summary of the Invention

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a slope quality monitoring and early warning method based on seismic data, the method comprising:

[0006] Acquire a set of earthquake monitoring data for a target slope area, the set of earthquake monitoring data including earthquake waveform data and corresponding environmental correlation parameters collected by multiple monitoring nodes within a preset time period;

[0007] Performing a dynamic preprocessing operation on the earthquake motion monitoring data set to obtain a standardized earthquake motion monitoring data set, wherein the dynamic preprocessing operation includes data alignment, noise filtering, and dimension unification processing;

[0008] Extracting features based on the standardized seismic monitoring data set to generate a dynamic correlation feature set for the target slope area, wherein the dynamic correlation feature set includes seismic energy distribution features, slope response frequency features, and environmental coupling fluctuation features;

[0009] Calling a slope quality analysis model to perform state prediction processing on the dynamic correlation feature set to obtain stable state features and abnormal fluctuation features of the target slope area, and generating a slope quality assessment result based on the difference between the stable state features and the abnormal fluctuation features;

[0010] determine a pre-warning response strategy based on the slope quality evaluation result, and trigger a pre-warning signal output operation of the slope monitoring system according to the pre-warning response strategy.

[0011] In still another aspect, the embodiments of the present application also provide a slope quality monitoring and pre-warning system based on ground motion data, which comprises a processor and a machine readable storage medium, the machine readable storage medium is connected with the processor, the machine readable storage medium is used for storing programs, instructions or codes, and the processor is used for executing the programs, instructions or codes in the machine readable storage medium to realize the above method.

[0012] Based on the above aspects, the embodiments of the present application can effectively improve the quality and consistency of data by acquiring the ground motion monitoring data set of the target slope region including the ground motion waveform data of multiple monitoring nodes in the target slope region and the corresponding environmental associated parameters in a preset time period, performing the dynamic preprocessing operation including data alignment, noise filtering and dimension unification processing on the ground motion monitoring data set, avoiding the adverse effects caused by data deviation, noise interference and dimension inconsistency on subsequent analysis, extracting the dynamic associated feature set including the ground motion energy distribution feature, the slope response frequency feature and the environmental coupling fluctuation feature based on the standardized ground motion monitoring data set, depicting the dynamic characteristics of the target slope region from multiple key aspects, and being able to more accurately capture the potential changes and abnormal conditions of the slope. The slope quality analysis model is called to perform state prediction processing on the dynamic associated feature set, to obtain the stable state feature and the abnormal fluctuation feature of the target slope region, and to generate the slope quality evaluation result according to the difference between the two, to realize the scientific and accurate evaluation of the slope quality, and finally to determine the pre-warning response strategy based on the slope quality evaluation result and to trigger the pre-warning signal output operation, to timely and effectively pre-warn the potential dangerous conditions of the slope, to help the relevant personnel to take measures in advance, to reduce the loss caused by the slope disaster, and to improve the overall efficiency of slope monitoring and pre-warning. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 is the execution flow schematic diagram of the slope quality monitoring and pre-warning method based on ground motion data provided by the embodiments of the present application.

[0014] Figure 2 is the schematic diagram of the exemplary hardware and software components of the slope quality monitoring and pre-warning system based on ground motion data provided by the embodiments of the present application. DETAILED DESCRIPTION

[0015] The present application will be described in detail below with reference to the accompanying drawings, Figure 1FIG1 is a flow chart of a slope quality monitoring and early warning method based on seismic motion data provided by an embodiment of the present invention. The slope quality monitoring and early warning method based on seismic motion data is introduced in detail below.

[0016] Step S110: Acquire a set of earthquake monitoring data of a target slope area, wherein the set of earthquake monitoring data includes earthquake waveform data and corresponding environment-related parameters collected by multiple monitoring nodes within a preset time period.

[0017] In this embodiment, in order to achieve effective monitoring and early warning of the quality of the target slope area, the first step is to obtain a set of seismic monitoring data for the target slope area. The target slope area usually has complex geological structures and environmental conditions, and its stability is affected by the combined effects of seismic motion and various environmental factors. In order to comprehensively and accurately collect relevant data, multiple monitoring nodes can be reasonably distributed in the target slope area. The distribution of these monitoring nodes is determined based on factors such as the topography, geological structure, and potential risk areas of the target slope area to ensure that the entire target slope area can be covered and key seismic motion and environmental information can be captured.

[0018] Each monitoring node is equipped with seismic monitoring equipment for collecting seismic waveform data and environmental parameter monitoring sensors for collecting environmental parameters. The seismic monitoring equipment can record the vibrations generated by seismic motion in the target slope area in real time, generating seismic waveform data. This seismic waveform data is presented in the form of a time series and contains a wealth of information, such as the amplitude, frequency, and duration of the seismic motion. Environmental parameter monitoring sensors are used to measure environmental factors closely related to the stability of the target slope area. Common environmental parameters include temperature, humidity, and air pressure. These environmental factors can affect the rock and soil properties and mechanical state of the slope, thereby affecting the stability of the slope.

[0019] The preset time period is set to ensure representative and timely data. The selection of this time period requires comprehensive consideration of factors such as the geological activity patterns of the target slope area, the frequency of seismic activity, and the actual monitoring needs. During this time period, each monitoring node continuously collects data at the set sampling frequency. The sampling frequency is determined based on the characteristics of seismic motion and environmental parameter changes to ensure accurate capture of data changes.

[0020] Assume that there are n monitoring nodes distributed within the target slope area, and let Ni (i = 1, 2, ..., n) represent the i-th monitoring node. Within a preset time period T, the seismic waveform data collected by each monitoring node Ni can be represented as a time series Di = {di1, di2, ..., dim}, where m represents the number of sampling points within time period T. The corresponding environmental parameters can be represented as a vector Ei = {ei1, ei2, ..., eik}, where k represents the number of types of environmental parameters. For example, ei1 can represent temperature, ei2 can represent humidity, ei3 can represent air pressure, etc.

[0021] By summarizing the seismic waveform data and environmental correlation parameters collected by all monitoring nodes within a preset time period, the seismic monitoring data sets D = {D1, D2, ..., Dn} and E = {E1, E2, ..., En} of the target slope area are obtained. Combining these two seismic monitoring data sets together constitutes a complete seismic monitoring data set.

[0022] Step S120: performing a dynamic preprocessing operation on the earthquake monitoring data set to obtain a standardized earthquake monitoring data set, wherein the dynamic preprocessing operation includes data alignment, noise filtering, and dimension unification processing.

[0023] After acquiring the seismic monitoring data set for the target slope area, dynamic preprocessing is required to obtain a standardized seismic monitoring data set, as this data may contain noise interference, missing data, and inconsistent dimensions. To improve the accuracy and reliability of subsequent analysis and processing, dynamic preprocessing is performed on the data to obtain a standardized seismic monitoring data set. This dynamic preprocessing operation mainly involves data alignment, noise filtering, and dimensional unification.

[0024] Step S121: performing noise component filtering and time-frequency conversion on the seismic waveform data to generate seismic spectrum data of each monitoring node, and performing frequency band energy normalization on the seismic spectrum data to obtain frequency band energy standardized data.

[0025] For seismic waveform data, noise filtering must first be performed to eliminate it. During the acquisition process, seismic waveform data may be subject to interference from various external factors, such as electromagnetic interference and mechanical vibration. These interferences can introduce noise, affecting data quality and the accuracy of subsequent analysis. To eliminate these noise components, filtering algorithms, such as low-pass, high-pass, or band-pass filtering, can be used. The specific filtering algorithm to be used depends on the frequency characteristics of the noise and the frequency range of the seismic signal.

[0026] Assuming that the earthquake motion waveform data is Di={di1, di2, ..., dim}, after being processed by the filtering algorithm, the filtered earthquake motion waveform data Di'={di1', di2', ..., dim'} is obtained.

[0027] Next, the filtered seismic waveform data undergoes time-to-frequency conversion. This process converts seismic waveform data from the time domain into the frequency domain, enabling better analysis of the frequency characteristics of seismic signals. Common time-to-frequency conversion methods include Fourier transform and wavelet transform. Through this process, the seismic waveform data from each monitoring node is converted into seismic spectrum data Si = {si1, si2, …, sinf}, where nf represents the number of points in the spectrum data.

[0028] Then, the seismic spectrum data is subjected to frequency band energy normalization. Since there may be large differences in the energy of different frequency bands, the frequency band energy needs to be normalized to facilitate subsequent analysis and comparison. The specific steps of frequency band energy normalization are as follows: First, the seismic spectrum data is divided into multiple frequency bands according to the frequency range. Assume that it is divided into nb frequency bands. The energy of each frequency band can be obtained by summing the spectrum data within the frequency band. Let the energy of the j-th frequency band be Ej, then Ej = ∑si (si belongs to the j-th frequency band). Then, calculate the total energy of all frequency bands Et = ∑Ej (j = 1, 2, ..., nb). Finally, the energy of each frequency band is normalized to obtain the frequency band energy standardized data Sj' = Ej / Et (j = 1, 2, ..., nb).

[0029] Step S122: performing data integrity verification on the environment-related parameters, performing interpolation filling operations on the missing parameters based on the verification results, and performing numerical range normalization on the filled environment-related parameters to obtain standardized environment parameter data.

[0030] In this embodiment, data integrity verification is first performed on the environment-related parameters. Because environmental parameter monitoring sensors may experience malfunctions, communication interruptions, and other issues, the collected environment-related parameters may contain missing values. Data integrity verification involves checking whether there are missing values ​​in the environment-related parameter vector Ei = {ei1, ei2, …, eik} of each monitoring node. This can be accomplished by setting a judgment condition, such as determining that a parameter value is missing when it is an invalid value (e.g., NaN).

[0031] Based on the verification results, interpolation is performed on the missing parameters. The purpose of interpolation is to estimate the missing parameter values ​​based on the existing valid data to ensure data integrity. Common interpolation methods include linear interpolation and spline interpolation. Assuming that the j-th environmental correlation parameter eij of the i-th monitoring node is missing, interpolation can be performed based on the other valid environmental correlation parameters of the monitoring node and the relevant parameters of the adjacent monitoring nodes. Let the filled environmental correlation parameter vector be Ei' = {ei1', ei2', ..., eik'}.

[0032] Considering that different environmental parameters have different physical dimensions—for example, temperature may be expressed in degrees Celsius, humidity in percentage, and air pressure in kilopascals—directly normalizing these parameters by numerical range can lead to erroneous results. Because the numerical ranges of parameters with different dimensions vary significantly, failing to eliminate the dimensionality effect can lead to over-compression or over-expansion of some parameters during the normalization process, thus affecting the accuracy of subsequent analysis. Therefore, before performing numerical range normalization, the populated environmental parameters must first be dimensionlessly normalized.

[0033] In detail, the Z-score standardization method is used here for dimensionless processing. First, for each environment-related parameter, its mean and standard deviation in the historical data are calculated. For each parameter eij' in the filled environment-related parameter vector Ei', the historical mean of the parameter is subtracted from eij', and then the difference is divided by the historical standard deviation of the parameter to obtain the standardized parameter eij". In this way, each environment-related parameter is converted into a dimensionless value, eliminating the influence of the dimension on the data, so that different types of environment-related parameters can be subsequently analyzed and processed under a unified scale, and finally the environmental parameter standardized data Ei" = {ei1", ei2", ..., eik"} is obtained.

[0034] Step S123: further perform global standardization on the frequency band energy standardization data and the environmental parameter standardization data and then perform timestamp alignment to generate standardized data units corresponding to the monitoring nodes one by one, and merge the various standardized data units to generate the standardized seismic monitoring data set.

[0035] After obtaining the frequency band energy normalized data and the environmental parameter normalized data, they need to be further globally normalized. The purpose of global normalization is to ensure that different types of data have the same scale and distribution across the entire data set, facilitating subsequent comprehensive analysis. A similar normalization method as in step S122 can be used to uniformly normalize the frequency band energy normalized data and the environmental parameter normalized data.

[0036] Assume that the frequency band energy standardization data is S = {S1', S2', ..., Sn'}, and the environmental parameter standardization data is E" = {E1", E2", ..., En"}. After global standardization, we get S' = {S1", S2", ..., Sn"} and E"' = {E1"', E2"', ..., En"'} respectively.

[0037] Next, timestamp alignment is performed. Because the acquisition of seismic waveform data and environmental parameters may differ in time, their timestamps need to be aligned to ensure that data from the same time point can be matched. For example, by finding the common parts of the timestamps, different data can be arranged and matched in chronological order.

[0038] After timestamp alignment, a standardized data unit Ui={Si”, Ei”’} (i=1, 2, …, n) corresponding to each monitoring node is generated. Finally, the standardized data units are merged, that is, all Ui are combined together to obtain the standardized seismic monitoring data set U={U1, U2, …, Un}.

[0039] Step S130: extracting features based on the standardized seismic monitoring data set to generate a dynamic correlation feature set of the target slope area, wherein the dynamic correlation feature set includes seismic energy distribution features, slope response frequency features, and environmental coupling fluctuation features.

[0040] After obtaining a standardized seismic monitoring data set, in order to deeply analyze the stability of the target slope area, it is necessary to extract relevant features from this data and generate a dynamic correlation feature set for the target slope area. The dynamic correlation feature set mainly includes seismic energy distribution characteristics, slope response frequency characteristics, and environmental coupling fluctuation characteristics.

[0041] Step S131: performing energy integration processing on the frequency band energy normalization data of each monitoring node to obtain the seismic energy distribution sequence of each monitoring node, and calculating the energy difference between adjacent monitoring nodes based on the seismic energy distribution sequence to generate energy distribution difference features.

[0042] First, energy integration is performed on the frequency band energy standardization data of each monitoring node. Frequency band energy standardization data reflects the energy distribution of different frequency bands. Energy integration can be used to determine how the total energy of each monitoring node changes over time across the entire frequency band, i.e., the seismic energy distribution sequence. Assuming that the frequency band energy standardization data of the i-th monitoring node is Si” = {si1”, si2”, …, sinb”}, energy integration is performed on this data to obtain the seismic energy distribution sequence Ei” = ∑sij” (j = 1, 2, …, nb).

[0043] Then, the energy difference between adjacent monitoring nodes is calculated based on the seismic energy distribution sequence. The energy difference between adjacent monitoring nodes can reflect the spatial uneven distribution of seismic energy, which is of great significance for analyzing slope stability. Suppose two adjacent monitoring nodes are Ni and Ni+1, and their seismic energy distribution sequences are Ei” and Ei+1” respectively. The energy difference can be obtained by calculating the absolute value of the difference between the two, that is, ΔEi = |Ei” - Ei+1”|.

[0044] By combining the energy differences between all adjacent monitoring nodes, an energy distribution difference feature ΔE={ΔE1, ΔE2, ..., ΔEn-1} is generated.

[0045] Step S132: performing a main frequency component extraction process on the frequency band energy normalized data to obtain a dominant frequency set of each monitoring node, and calculating the resonance response probability of the slope structure based on the dominant frequency set to generate a slope response frequency feature.

[0046] The primary frequency components of the frequency band energy-normalized data are extracted to identify the frequency components where the energy in the seismic signal of each monitoring node is primarily concentrated, i.e., the dominant frequencies. By sorting the frequency band energy-normalized data, the frequencies corresponding to the frequency bands with the largest energy values ​​are identified and designated as the dominant frequencies. Assuming that the frequency band energy-normalized data for the i-th monitoring node is Si" = {si1", si2", ..., sinb"}, the data is sorted from highest to lowest energy value. The frequencies corresponding to the first m frequency bands with the largest energy values ​​are selected to obtain the dominant frequency set Fi = {fi1, fi2, ..., fim} for this monitoring node.

[0047] Then, the probability of the slope structure's resonant response is calculated based on the dominant frequency set. The probability of the slope structure's resonant response is closely related to the slope's natural frequency and the dominant frequency of the seismic signal. When the dominant frequency of the seismic signal is close to the slope's natural frequency, the slope is more likely to resonate, increasing the risk of slope instability. The probability of the resonant response can be calculated by establishing a probabilistic model. Assuming the slope's natural frequency set is F0 = {f01, f02, …, f0p}, for each monitoring node's dominant frequency set Fi, the degree of frequency matching between it and the slope's natural frequency set F0 is calculated. The degree of frequency matching can be measured by calculating the absolute value of the frequency difference, i.e., |fi - f0j| (i = 1, 2, …, m; j = 1, 2, …, p). When the frequency difference is less than a certain threshold, a frequency match is considered to exist. The number of frequency matches is counted, and the probability of the resonant response Pi is then calculated using conventional probability calculation formulas in related art.

[0048] By combining the resonance response probabilities of all monitoring nodes, the slope response frequency characteristics P = {P1, P2, ..., Pn} are generated.

[0049] Step S133: performing dynamic correlation analysis on the standardized environmental parameter data to determine the coupling correlation between the environmental parameter changes and the seismic energy distribution, and generating an environmental coupling fluctuation feature based on the coupling correlation.

[0050] Dynamic correlation analysis is performed on standardized environmental parameter data to determine the degree of coupling between environmental parameter changes and seismic energy distribution. Changes in environmental factors can affect the properties and mechanical state of the slope's rock and soil, thereby influencing the distribution of seismic energy. Therefore, analyzing the degree of coupling between environmental parameter changes and seismic energy distribution is crucial for understanding slope stability.

[0051] Step S1331: Perform sliding window slicing processing on the standardized environmental parameter data according to a preset time granularity to generate a set of environmental parameter time series slices corresponding to each monitoring node, where each environmental parameter time series slice contains a standardized numerical sequence of temperature, humidity, and air pressure parameters in continuous time steps.

[0052] First, the environmental parameter standardized data is subjected to sliding window slicing processing according to the preset time granularity. The selection of the preset time granularity should be determined according to the characteristics of the environmental parameter changes and the needs of the analysis. Assume that the preset time granularity is Δt and the size of the sliding window is W. For the environmental parameter standardized data Ei''={ei1'', ei2'', ..., eik''} of the i-th monitoring node, sliding window slicing processing is performed in chronological order. Each sliding window contains a standardized numerical sequence of environmental parameters of continuous time steps, and generates a set of environmental parameter time series slices Ci={ci1, ci2, ..., cinw} corresponding to each monitoring node, where nw represents the number of slices, and each environmental parameter time series slice cij contains a standardized numerical sequence of parameters such as temperature, humidity, and air pressure at continuous time steps.

[0053] Step S1332: Slice the seismic energy distribution sequence at the same time granularity, and allow the environmental parameter time series slices to be slidably aligned within a preset lag time range to generate a set of seismic energy time series slices that take into account the environmental parameter lag effect.

[0054] Slice the seismic energy distribution sequence according to the same preset time granularity Δt. Assume that the seismic energy distribution sequence of the i-th monitoring node is Ei''. Slice it in chronological order to obtain the seismic energy time series slice set Di = {di1, di2, ..., dinw}.

[0055] Considering the potential hysteresis effect of changes in environmental parameters on seismic energy distribution, the environmental parameter time series slices are allowed to slide and align within a preset hysteresis time range. This preset hysteresis time range can be set based on actual conditions, for example, to ±Δt'. By adjusting the relative positions of the environmental parameter time series slices and the seismic energy time series slices within the preset hysteresis time range, the optimal alignment is found, generating a set of seismic energy time series slices Di' = {di1', di2', ..., dinw'} that takes into account the environmental parameter hysteresis effect.

[0056] Step S1333: Perform dynamic correlation analysis on the environmental parameter time series slices and seismic energy time series slices of the same monitoring node, combine the linear Pearson correlation coefficient and the nonlinear mutual information index, and generate a comprehensive correlation strength matrix of each parameter dimension and energy distribution dimension, wherein the element value in the comprehensive correlation strength matrix is ​​the weighted average of the linear and nonlinear correlation strengths.

[0057] When performing dynamic correlation analysis on the environmental parameter time series slices and the seismic energy time series slices of the same monitoring node, it is necessary to combine the linear Pearson correlation coefficient and the nonlinear mutual information index. Considering that the linear Pearson correlation coefficient and the nonlinear mutual information index have different dimensions and value ranges, the linear Pearson correlation coefficient has a value range from -1 to 1, which mainly measures the linear correlation between two variables; while the value of the nonlinear mutual information index is greater than or equal to 0, which is used to capture the nonlinear relationship between variables. Therefore, before calculating the comprehensive correlation strength, this embodiment needs to normalize the linear Pearson correlation coefficient and the nonlinear mutual information index separately. For the linear Pearson correlation coefficient, a mapping method is used to normalize it to the range of 0 to 1. Specifically, the linear Pearson correlation coefficient is added by 1 and then divided by 2. In this way, the linear Pearson correlation coefficient originally between -1 and 1 is mapped to the interval of 0 to 1, which is recorded as the normalized linear Pearson correlation coefficient ri'jl. For the nonlinear mutual information index, each index value is divided by the maximum value of the index in all data, and it is normalized to the range of 0 to 1, which is recorded as the normalized nonlinear mutual information index ri'jn.

[0058] After normalization, the normalized linear Pearson correlation coefficient ri'jl and the normalized nonlinear mutual information index ri'jn are weighted averaged. A weighting coefficient α is set between 0 and 1 to adjust the proportion of linear and nonlinear association strengths in the overall association strength. Multiply the normalized linear Pearson correlation coefficient ri'jl by α, then subtract 1 from α and multiply the normalized nonlinear mutual information index ri'jn. The two results are added to obtain the overall association strength rij.

[0059] By combining the comprehensive correlation strengths between the environmental parameter dimensions and the energy distribution dimensions of the same monitoring node, we can generate a comprehensive correlation strength matrix Ri = [rij] (i = 1, 2, ..., k; j = 1, 2, ..., nw) of each parameter dimension and the energy distribution dimension, where each element value rij in the matrix is ​​the weighted average of the linear and nonlinear correlation strengths obtained after normalization and weighted averaging.

[0060] Step S1334: performing a maximum pooling operation across parameter dimensions on the real-time correlation strength matrix, extracting the maximum coupling correlation strength between the environmental parameters and the seismic energy distribution at each time step, and generating a preliminary coupling correlation degree sequence.

[0061] Perform a maximum pooling operation across the parameter dimensions of the comprehensive correlation strength matrix Ri. The purpose of the maximum pooling operation is to extract the maximum coupling correlation strength between the environmental parameters and the ground motion energy distribution at each time step. For each column of the comprehensive correlation strength matrix Ri, find the maximum value, i.e., rij_max = max(rij) (i = 1, 2, ..., k).

[0062] The maximum coupling correlation strengths at all time steps are combined together to generate a preliminary coupling correlation degree sequence Ri'={ri1_max, ri2_max, ..., rinw_max}.

[0063] Step S1335: Based on the spatial gradient change of the preliminary coupling correlation degree sequence between adjacent monitoring nodes, the regional propagation consistency index of the coupling correlation strength is calculated, and the preliminary coupling correlation degree sequence is spatially smoothed and corrected to generate a corrected dynamic coupling correlation degree set.

[0064] In this embodiment, the coupling correlation strength between adjacent monitoring nodes should have a certain degree of continuity and consistency. If there is a large difference, it may be caused by local sensor anomalies or other interference factors. Suppose two adjacent monitoring nodes are Ni and Ni+1, and their preliminary coupling correlation degree sequences are Ri' and Ri+1' respectively. The spatial gradient between them can be calculated by calculating the difference in coupling correlation at the corresponding time step. For each time step t, the gradient value Δrit = |rit'-ri+1t'| is calculated, where rit' and ri+1t' are the coupling correlation values ​​of Ri' and Ri+1' at time step t, respectively. The gradient values ​​at all time steps are combined to obtain the spatial gradient sequence ΔRi = {Δri1, Δri2, ..., Δrinw}.

[0065] Next, the regional propagation consistency index of the coupling correlation strength is calculated based on the spatial gradient sequence. This can be achieved by performing statistical analysis on the spatial gradient sequence, for example, by calculating its mean μΔRi and standard deviation σΔRi. The regional propagation consistency index γi can be defined as a function related to the mean and standard deviation, for example, γi = f(μΔRi, σΔRi). The function f here can be selected based on the actual situation. A common approach is γi = 1 / (1 + μΔRi + σΔRi). The closer this index value is to 1, the more consistent the coupling correlation strength between adjacent monitoring nodes and the better the continuity of regional propagation.

[0066] In actual slope monitoring scenarios, for example, slope nodes in mountainous areas may be distributed linearly. If Euclidean distance is used to calculate the topology, the actual physical path may vary greatly due to the complex terrain. This will cause the selection of K nearest neighbor nodes to include irrelevant nodes, making the mean replacement operation based on K nearest neighbor nodes invalid.

[0067] To solve this problem, this embodiment uses geographically weighted distance instead of Euclidean distance. Geographically weighted distance takes into account actual factors such as terrain undulations and can more accurately reflect the actual physical distance between monitoring nodes. Geographically weighted distance can be calculated by obtaining the geographical location information (such as longitude and latitude) of the monitoring nodes and the elevation data of the terrain, combined with geographic information system (GIS) technology. For example, using the Dijkstra algorithm or the A* algorithm, taking into account factors such as the slope of the terrain and obstacles, the actual path length between the monitoring nodes is calculated as the geographically weighted distance.

[0068] At the same time, the K value is dynamically adjusted to accommodate changes in monitoring node density across different regions. In areas with high node density, the K value is appropriately increased to capture more neighboring node information and improve the accuracy of mean replacement. In areas with low node density, the K value is reduced to avoid introducing irrelevant nodes. A node density threshold can be pre-set based on the distribution of monitoring nodes. When the node density in a certain area exceeds this threshold, the K value is increased; when the node density is below this threshold, the K value is decreased.

[0069] Based on the adjusted topological distance and K value, the preliminary coupling correlation sequence is spatially smoothed and corrected. First, the first-order difference calculation in the time dimension is performed on the preliminary coupling correlation sequence to generate a coupling correlation change trend sequence for each monitoring node. For the preliminary coupling correlation sequence Ri' = {ri1_max, ri2_max, ..., rinw_max} of the i-th monitoring node, its first-order difference sequence ΔRi' = {Δri1', Δri2', ..., Δrinw-1'}, where Δrit' = rit+1_max - rit_max (t = 1, 2, ..., nw-1).

[0070] Then, the regional propagation consistency index is calculated based on the cosine similarity between the coupling correlation trend sequences of adjacent monitoring nodes. Assume that the coupling correlation trend sequences of adjacent monitoring nodes Ni and Ni+1 are ΔRi' and ΔRi+1', respectively. The cosine similarity cosθi between them can be calculated by calculating the vector dot product and the vector modulus: cosθi = (ΔRi'·ΔRi+1') / (||ΔRi'||×||ΔRi+1'||). When the cosine similarity falls below the preset threshold θ0, the corresponding node pair is marked as a region of spatial abnormal fluctuation.

[0071] For the preliminary coupling correlation of the spatial anomaly fluctuation region, a mean replacement operation based on the K nearest neighbor nodes is performed to eliminate the coupling correlation jump noise caused by local sensor anomalies. For the node Ni marked as the spatial anomaly fluctuation region, its K nearest neighbor nodes are found, and the preliminary coupling correlation sequence of these nearest neighbor nodes is set to Rj' (j∈Nk, Nk represents the set of Ni's K nearest neighbor nodes). For each value rit' in Ni's preliminary coupling correlation sequence Ri', it is replaced with the mean of the coupling correlation of its K nearest neighbor nodes at the same time step, that is, rit" = (1 / K) × ∑rjt'(j∈Nk).

[0072] Next, the modified preliminary coupling correlation is smoothed using a Gaussian kernel function, where the kernel bandwidth is adaptively adjusted based on the topological distance of the monitoring nodes. Assume that the modified preliminary coupling correlation sequence is Ri" = {ri1", ri2", ..., rinw"}. For each coupling correlation value rit" at time step t, the spatial neighborhood weighted smoothing result rit"' can be calculated as follows: First, determine the set of spatial neighborhood nodes Nj of Ni (including Ni itself). For each neighboring node Nj, calculate its geographically weighted distance dij to Ni. Then, based on the geographically weighted distance dij, calculate the Gaussian kernel value wij = exp(-dij2 / (2×h2)), where h is the kernel bandwidth. h can be adaptively adjusted based on the topological distance distribution of the monitoring nodes. For example, h can be taken as the mean of the geographically weighted distances of the neighboring nodes. Finally, rit"' = (∑wij × rjt") / (∑wij)(j∈Nj).

[0073] Finally, the smoothed dynamic coupling correlation is filtered using an exponentially weighted moving average filter in the time dimension to suppress high-frequency fluctuation noise and retain the trend component. Let the smoothed dynamic coupling correlation sequence be Ri'' = {ri1'', ri2'', ..., rinw''}, and the result after exponentially weighted moving average filtering be Ri'' = {ri1'', ri2'', ..., rinw''}, where ri1'' = ri1''. For t > 1, rit'' = β × rit'' + (1-β) × rit-1'', where β is the smoothing coefficient, ranging from (0 to 1) and can be adjusted according to actual conditions. Combining the corrected dynamic coupling correlation sequences of all monitoring nodes generates the corrected dynamic coupling correlation set R'' = {R1'', R2'', ..., Rn''}.

[0074] Step S1336: Input the dynamic coupling correlation degree set into a preset coupling strength grading model, map the coupling correlation degree of each monitoring node to a unified feature dimension through a fully connected layer, and generate an environmental coupling fluctuation feature vector that matches the characteristic dimension of the seismic energy distribution.

[0075] The corrected dynamic coupling correlation set R"" is input into the preset coupling strength grading model. The coupling strength grading model is a trained AI model whose main function is to process the coupling correlation of each monitoring node so that it can better reflect the coupling relationship between environmental parameters and seismic energy distribution.

[0076] The input of the coupling strength grading model is the modified dynamic coupling correlation set R””, where the dynamic coupling correlation sequence Ri”” of each monitoring node is a multi-dimensional vector. Through the fully connected layer in the coupling strength grading model, the coupling correlation of each monitoring node is mapped to a unified feature dimension. The role of the fully connected layer is to perform linear transformation and nonlinear activation on the input coupling correlation, so that the coupling correlation of different monitoring nodes can be represented in the same feature space. Suppose the weight matrix of the fully connected layer is W, the bias vector is b, and for the dynamic coupling correlation sequence Ri”” of the i-th monitoring node, the result after processing by the fully connected layer is Zi = f(W×Ri””+b), where f is a nonlinear activation function, such as a ReLU function.

[0077] The results of all monitoring nodes processed by the fully connected layer are combined to generate a new feature vector Z = {Z1, Z2, …, Zn}. The feature vector Z is then dimensionalized to match the characteristic dimension of the ground motion energy distribution. This can be achieved through operations such as interpolation and dimensionality reduction, ultimately generating an environmental coupling fluctuation feature vector Rf that matches the characteristic dimension of the ground motion energy distribution.

[0078] Step S134: Perform spatial topology correlation processing on the energy distribution difference feature, the slope response frequency feature, and the environmental coupling fluctuation feature to generate a dynamic correlation feature set of the target slope region.

[0079] The energy distribution difference feature ΔE, the slope response frequency feature P, and the environmental coupling fluctuation feature vector Rf are subjected to spatial topology correlation processing. The purpose of spatial topology correlation processing is to consider the spatial position relationship between monitoring nodes, and to integrate different types of features to more comprehensively reflect the dynamic characteristics of the target slope region.

[0080] First, according to the actual physical positions of the monitoring nodes, a spatial topology relationship graph G=(V, E) between the monitoring nodes is constructed, where V represents the set of monitoring nodes, and E represents the connection relationship between the monitoring nodes. The connection relationship can be determined according to the distance between the monitoring nodes, signal correlation, and other factors.

[0081] For the energy distribution difference feature ΔE, the slope response frequency feature P, and the environmental coupling fluctuation feature vector Rf, they are respectively correlated with the spatial topology relationship graph G. By corresponding each feature to the nodes in the spatial topology relationship graph according to the node numbers of the monitoring nodes, the feature values corresponding to each node can be obtained. For example, for the energy distribution difference feature ΔE, ΔEi is corresponded to node Ni in the spatial topology relationship graph.

[0082] Then, the graph convolution network (GCN) is used to process the correlated features. The graph convolution network can effectively capture the spatial topology information between the monitoring nodes, and propagate and aggregate the features. Let the input feature matrix be X=[ΔE; P; Rf], the adjacency matrix of the graph convolution network be A, and the weight matrix be W. After one layer of processing by the graph convolution network, the output feature matrix X' can be calculated as follows: X'=f(A×X×W), where f is a nonlinear activation function.

[0083] After multiple graph convolution network processing, different types of features are fully fused and propagated, and finally a dynamic correlation feature set F={F1, F2, …, Fn} of the target slope region is generated, where Fi represents the dynamic correlation feature vector corresponding to the i-th monitoring node.

[0084] Step S140: Call the slope quality analysis model to perform state prediction processing on the dynamic correlation feature set to obtain stable state features and abnormal fluctuation features of the target slope region, and generate a slope quality evaluation result according to the difference between the stable state features and the abnormal fluctuation features.

[0085] After obtaining the dynamic correlation feature set of the target slope area, the slope quality analysis model is called to perform state prediction processing. The slope quality analysis model is a trained AI model whose purpose is to predict the stability and abnormal fluctuation of the target slope area based on the input dynamic correlation feature set.

[0086] Step S141: Input the dynamic correlation feature set into the spatiotemporal coding network of the slope quality analysis model, perform global dependency modeling on the spatial topological correlation data in the dynamic correlation feature set through a multi-head self-attention mechanism, and generate a context coding vector containing cross-node spatiotemporal interaction information.

[0087] The dynamic correlation feature set F is input into the spatiotemporal coding network of the slope quality analysis model. The main function of the spatiotemporal coding network is to process the spatial topological correlation data in the dynamic correlation feature set and capture the global dependency and spatiotemporal interaction information between monitoring nodes.

[0088] First, the dynamic correlation feature set is sliced ​​in the time dimension to generate a set of spatiotemporal feature slices divided by the preset time window. Assuming the size of the preset time window is T and the length of the dynamic correlation feature set F in the time dimension is L, it can be divided into N = L / T spatiotemporal feature slice sets Ft = {F1t, F2t, ..., Fnt} (t = 1, 2, ..., N), where Fit represents the dynamic correlation feature vector of the i-th monitoring node in the t-th time window.

[0089] Based on the actual physical location distances of the monitoring nodes and the historical seismic signal correlations within each spatiotemporal feature slice, weighted connectivity relationships between the monitoring nodes are constructed, generating a node adjacency matrix and a characteristic node vector that reflect spatial proximity and signal correlation strength. For each spatiotemporal feature slice Ft, the distance matrix D = [dij] between nodes is calculated based on the actual physical location distances of the monitoring nodes, where dij represents the physical distance between nodes Ni and Nj. Simultaneously, the correlation matrix C = [cij] between nodes is calculated based on the historical seismic signal correlations, where cij represents the seismic signal correlation between nodes Ni and Nj. The distance matrix and the correlation matrix are then fused to obtain a weighted connectivity matrix W = [wij]. wij can be calculated using a fusion function f(dij, cij), for example, wij = α × (1 / dij) + (1-α) × cij, where α is the weighting coefficient.

[0090] Based on the weighted connection relationship matrix W, the node adjacency matrix A = [aij] is generated, where aij = 1 indicates that there is a connection relationship between nodes Ni and Nj, and aij = 0 indicates that there is no connection relationship. At the same time, the dynamic association feature vector Fit of each monitoring node is used as the feature node vector.

[0091] The node adjacency matrix A and the feature node vector are input into a multi-head self-attention mechanism. The multi-head self-attention mechanism can compute multiple attention heads in parallel, each of which can capture different aspects of dependencies. For each attention head, the feature node vector is first linearly transformed to obtain a query vector Q, a key vector K, and a value vector V. Let the feature node vector be X, the query vector Q = X × WQ, the key vector K = X × WK, and the value vector V = X × WV, where WQ, WK, and WV are learnable weight matrices.

[0092] Then, the attention score is calculated. For each node pair (i, j), the attention score sij can be obtained by taking the dot product of the query vector and the key vector, i.e., sij = Qi·Kj. To avoid the dot product being too large, the attention score is scaled, i.e., sij' = sij / √d, where d is the dimension of the query vector and the key vector.

[0093] Next, the attention scores are normalized by the softmax function to obtain the attention weight aij = softmax(sij'). Finally, the value vectors are weighted and aggregated according to the attention weights to obtain the output vector Oi = ∑aij × Vj for each node.

[0094] The output vectors of multiple attention heads are concatenated to obtain the output of the multi-head self-attention mechanism. Then, the output of the multi-head self-attention mechanism is linearly transformed and residually connected to obtain the final node-level attention weight matrix.

[0095] The feature node vectors are weighted and aggregated based on the node-level attention weight matrix to generate a spatially aggregated feature vector within each time window. For each time window t, the node-level attention weight matrix is ​​matrix multiplied with the feature node vector to obtain the spatially aggregated feature vector St = [S1t, S2t, …, Snt], where Sit represents the spatially aggregated feature vector of the i-th monitoring node in the t-th time window.

[0096] The spatially aggregated feature vectors of multiple time windows are positionally encoded and convolutionally fused in the time dimension. The purpose of positional encoding is to introduce information from the time dimension, enabling the model to distinguish features from different time steps. Sine-cosine positional encoding can be used to add positional encoding information to the spatially aggregated feature vectors of each time window. The positionally encoded spatially aggregated feature vectors are then convolutionally fused using a convolutional layer in the time dimension to capture feature changes in the time dimension. This ultimately generates a context encoding vector C = {C1, C2, …, Cn} that reflects the temporal evolution pattern, where Ci represents the context encoding vector corresponding to the i-th monitoring node.

[0097] Step S142: Input the context coding vector into the prediction branch of the slope quality analysis model, extract the seismic motion evolution patterns of different time scales through the dilated temporal convolutional network, output the energy distribution prediction sequence of multiple time steps in the future, and calculate the smoothness index characterizing the stability of the slope structure based on the fluctuation consistency of each time step in the energy distribution prediction sequence to generate the stable state characteristics.

[0098] The context encoding vector C is input into the prediction branch of the slope quality analysis model. The main function of the prediction branch is to predict the distribution of ground motion energy in multiple future time steps based on the context encoding vector and evaluate the stability of the slope.

[0099] First, a dilated temporal convolutional network (TCN) is used to extract seismic motion evolution patterns at different time scales. The TCN can capture feature information at different time scales using different dilation factors. Assuming the input context encoding vector C, the convolution kernel of the TCN is K, and the dilation factor is d, the output Y after a layer of TCN processing can be calculated as follows: Y = f(∑K[i]×C[ti×d]), where f is a nonlinear activation function, i represents the index of the convolution kernel, and t represents the time step.

[0100] Through multi-layer dilated temporal convolutional networks, seismic motion evolution patterns at different time scales are extracted. The extracted features are then fed into a fully connected layer, which outputs a sequence of energy distribution predictions for multiple future time steps, E' = {E1', E2', ..., Em'}, where m represents the number of future predicted time steps and Et' represents the energy distribution prediction vector for the tth time step.

[0101] A smoothness index characterizing slope stability is calculated based on the consistency of fluctuations at each time step in the energy distribution prediction sequence. Fluctuation consistency can be measured by calculating the difference between the energy distribution vectors at adjacent time steps in the energy distribution prediction sequence. Let Et' and Et+1' be the energy distribution prediction vectors at the tth and t+1th time steps, respectively. The difference between them can be calculated using the Euclidean distance, i.e., ΔEt = ||Et' - Et+1'||.

[0102] Perform a statistical analysis of the differences between all adjacent time steps, such as calculating the mean μΔE and standard deviation σΔE. The smoothness index S can be defined as a function related to the mean and standard deviation, for example, S = 1 / (1 + μΔE + σΔE). The closer this index is to 1, the smoother the fluctuations in the energy distribution prediction series and the more stable the slope structure. The smoothness index S is used as a characteristic of the stable state.

[0103] Step S143: Input the context coding vector into the anomaly detection branch of the slope quality analysis model, perform latent space reconstruction on the context coding vector through a variational autoencoder, generate a reconstructed feature vector that matches the historical normal state distribution, and use a wavelet packet decomposition algorithm to calculate the residual gradient of the energy distribution of the original coding vector and the reconstructed feature vector in each frequency band, extract the frequency band components and corresponding spatial positions where the residual gradient exceeds the historical fluctuation threshold, and generate the abnormal fluctuation feature.

[0104] The context encoding vector C is input into the anomaly detection branch of the slope quality analysis model. The main function of the anomaly detection branch is to detect whether there are abnormal fluctuations in the target slope area.

[0105] First, the context encoding vector is reconstructed into the latent space using a variational autoencoder (VAE). A variational autoencoder consists of an encoder network and a decoder network. The context encoding vector C is input into the encoder network of the variational autoencoder. The encoder network uses a multilayer perceptron to map the input features to a mean vector μ and a variance vector σ2 in the latent space. Assuming the encoder network's weight matrix W1 and the bias vector b1, the mean vector μ = f1(W1×C+b1) and the variance vector σ2 = f2(W1×C+b1), where f1 and f2 are nonlinear activation functions.

[0106] Random sampling is performed based on the mean vector μ and the variance vector σ2 to generate a latent space vector z that follows a Gaussian distribution. The sampling process can be achieved through the reparameterization technique, that is, z = μ + ε × √σ2, where ε is a random vector sampled from a standard normal distribution.

[0107] The latent space vector z is input into the decoder network of the variational autoencoder. The decoder network gradually restores the feature dimensions through deconvolutional layers and fully connected layers to generate a reconstructed feature vector C'. Assuming the weight matrix of the decoder network is W2 and the bias vector is b2, the reconstructed feature vector C' = f3(W2×z+b2), where f3 is a nonlinear activation function.

[0108] The original context coding vector C and the reconstructed eigenvector C' are expanded into time series signals according to the time step, and multi-scale wavelet packet decomposition is performed on the time series signals. Wavelet packet decomposition can decompose time series signals into components of different frequency bands, facilitating analysis of the signal's energy distribution in these bands. Assume that the scale of the wavelet packet decomposition is S, with multiple frequency bands at each scale. S-layer wavelet packet decomposition is performed on the time series signal Cc after the original context coding vector C is expanded, and the time series signal Cr after the reconstructed eigenvector C' is expanded.

[0109] In the s-th layer (s = 1, 2, ..., S) wavelet packet decomposition, the time series signal is decomposed into multiple frequency band signals. Let Ccsk be the signal of the k-th frequency band of the s-th layer, and let Cr be the signal of the k-th frequency band of the s-th layer. Calculate the energy of each frequency band signal. Energy calculation can be achieved by summing the squares of the frequency band signals. Let Ecsk be the energy of Ccsk, and Ecrsk be the energy of Crsk. That is, Ecsk is equal to the sum of the squares of each element in Ccsk, and Ecrsk is equal to the sum of the squares of each element in Crsk.

[0110] Next, the residual gradient of the energy distribution of the original code vector and the reconstructed feature vector in each frequency band is calculated. The residual gradient measures the difference in energy distribution between the original and reconstructed signals in each frequency band. For the kth frequency band in the sth layer, the residual gradient Gsk can be obtained by calculating the absolute value of the energy difference, that is, Gsk = |Ecsk - Ecrsk|.

[0111] In order to determine the frequency band components of abnormal fluctuations, it is necessary to set a historical fluctuation threshold. Based on the distribution boundary of the sliding window statistical residual gradient in the historical normal data set, the dynamic fluctuation threshold is determined. In the historical normal data set, a sliding window of a set length is selected in chronological order, and the statistical characteristics of the residual gradient of each frequency band in each window, such as the mean and standard deviation, are calculated. Assume that the mean value of the residual gradient of the kth frequency band in the sth layer of the historical normal data set in the sliding window is μsk, and the standard deviation is σsk. The dynamic fluctuation threshold Tsk can be determined based on the mean and standard deviation, for example, Tsk = μsk + α × σsk, where α is an adjustable coefficient used to control the strictness of the threshold.

[0112] The frequency band components and corresponding spatial positions that exceed the dynamic fluctuation threshold are marked as abnormal fluctuation features. For each frequency band, if its residual gradient Gsk is greater than the corresponding dynamic fluctuation threshold Tsk, it is considered that the frequency band has abnormal fluctuations. The numbers (s, k) of these abnormal fluctuation bands and the corresponding monitoring node number i are recorded. Since the context coding vector C corresponds one-to-one to the monitoring node, the spatial location where the abnormal fluctuation occurs can be clearly identified. The numbers of all abnormal fluctuation bands and the corresponding monitoring node numbers are combined together to generate an abnormal fluctuation feature set A = {(i1, s1, k1), (i2, s2, k2), ...}, where (ij, sj, kj) indicates that the ijth monitoring node has abnormal fluctuations in the kjth frequency band of the sjth layer.

[0113] After obtaining the stable state characteristics S and the abnormal fluctuation characteristics set A of the target slope area, the slope quality assessment result is generated based on the difference between the stable state characteristics and the abnormal fluctuation characteristics. The difference calculation can comprehensively consider multiple aspects of the stable state characteristics and the abnormal fluctuation characteristics.

[0114] The stability state feature S is an index reflecting the stability of the slope structure, and the closer the value is to 1, the more stable it is. The abnormal fluctuation feature set A contains the frequency band and spatial location information of the abnormal fluctuation. A difference degree function D can be defined to measure the difference between the two. First, for each abnormal fluctuation point (ij, sj, kj) in the abnormal fluctuation feature set A, a weight wij can be assigned to it, and the size of the weight can be determined according to the severity of the abnormal fluctuation, for example, it is related to the size of the residual gradient Gskj, wij can be a monotonically increasing function of Gskj.

[0115] Then, the weighted sum WA of all abnormal fluctuation points is calculated, WA = ∑ wij((ij, sj, kj) ∈ A). The difference degree function D can be defined as D = f(S, WA), where f is a function that comprehensively considers the stability state feature and the weighted sum of abnormal fluctuations. For example, D can be in the form of a linear combination, D = β × (1-S) + (1-β) × WA, where β is a weighting coefficient, the value range is between 0 and 1, used to adjust the importance of the stability state feature and the weighted sum of abnormal fluctuations in the difference degree calculation.

[0116] According to the value of the difference degree D, the slope quality assessment result is divided into different levels. For example, when D is greater than a certain higher threshold D1, it is considered that the slope quality is poor, and there is a higher risk of instability; when D is in an intermediate threshold range (between D2 and D1), it is considered that the slope quality is general and needs to be closely monitored; when D is less than a lower threshold D2, it is considered that the slope quality is good and is in a relatively stable state.

[0117] Step S150: determining a warning response strategy based on the slope quality assessment result, and triggering the warning signal output operation of the slope monitoring system according to the warning response strategy.

[0118] After obtaining the slope quality assessment result, a corresponding warning response strategy needs to be determined based on the result, and the warning signal output operation of the slope monitoring system is triggered to ensure the safety of the target slope area.

[0119] Step S151: dividing the warning level according to the size of the difference degree, wherein when the difference degree exceeds the first threshold, a first-level warning is triggered, when the difference degree is within the second threshold range, a second-level warning is triggered, and when the difference degree is lower than the third threshold, the monitoring state is maintained, wherein the first threshold > the upper limit of the second threshold range > the lower limit of the second threshold range > the third threshold.

[0120] In this embodiment, the warning level is divided according to the size of the difference degree D. Three key thresholds are set, which are the first threshold T1, the upper limit of the second threshold range T2u and the lower limit of the second threshold range T2l, and the third threshold T3, and satisfy T1>T2u>T2l>T3.

[0121] The extreme value theory (EVT) is used to dynamically set the threshold.

[0122] Extreme value theory is a statistical method specifically designed to handle extreme events. It can estimate the probability and threshold of future extreme events based on extreme values ​​in historical data. First, historical variance data for the target slope area is collected. This data captures the slope's stability and abnormal fluctuations over different time periods.

[0123] Perform extreme value analysis on historical variance data and select an appropriate extreme value model, such as the generalized extreme value distribution (GEV) model or the generalized Pareto distribution (GPD) model. For example, the generalized extreme value distribution model can describe the extreme value distribution characteristics of the data. Its distribution function includes location parameters, scale parameters, and shape parameters.

[0124] Parameters of the historical disparity data are estimated using methods such as maximum likelihood estimation to determine the specific parameter values ​​of the generalized extreme value distribution model. Based on the estimated parameters, different quantiles of the historical disparity data are calculated. For example, to determine the first threshold T1, the 99th percentile of the historical disparity data can be selected as the threshold. This means that only 1% of the disparity values ​​in the historical data will exceed this threshold. When the currently calculated disparity exceeds this threshold, it indicates that the slope anomaly has reached a critical level, requiring a Level 1 warning.

[0125] For the second threshold range (T2l to T2u), the 80th and 90th percentiles of the historical difference data can be selected as the lower limit T2l and upper limit T2u, respectively. When the difference falls within this range, it indicates that the slope is experiencing some abnormal fluctuations, but not yet reaching the severity of a first-level warning. In this case, a second-level warning is triggered, reminding relevant personnel to closely monitor the slope's condition.

[0126] The third threshold value T3 can be selected from the 20% quantile of the historical difference data. When the difference is lower than the threshold value, it indicates that the slope is in a relatively stable state and the normal monitoring state can be maintained.

[0127] According to the thresholds determined based on extreme value theory, when the difference D exceeds the first threshold T1, it indicates that the slope quality is extremely poor and there is a very high risk of instability, triggering a Level 1 warning. A Level 1 warning means that the most urgent and stringent measures must be taken to address the potential for slope instability.

[0128] When the difference D falls within the second threshold range (between T2l and T2u), the slope quality is fair, with a certain risk of instability requiring close attention. This triggers a Level 2 warning, which requires appropriate measures to strengthen slope monitoring and protection.

[0129] When the difference D is lower than the third threshold T3, it indicates that the slope quality is good and is in a relatively stable state. At this time, the monitoring state is maintained and the target slope area continues to be monitored according to the normal monitoring frequency and process.

[0130] Step S152: For the first-level warning, adjust the data collection frequency of the monitoring node to the first preset frequency, and synchronously update the sliding window time granularity and timestamp alignment rules in the preprocessing.

[0131] When a Level 1 warning is triggered, the data collection frequency of the monitoring nodes is adjusted to a first preset frequency to more accurately and promptly monitor changes in the target slope area. This frequency is typically higher than the normal monitoring frequency, allowing for more real-time data to be collected for a more detailed analysis of the slope's dynamics.

[0132] At the same time, the sliding window time granularity and timestamp alignment rules used in preprocessing need to be updated simultaneously. Due to the increased frequency of data collection, the original sliding window time granularity and timestamp alignment rules may no longer be applicable. The sliding window time granularity needs to be reduced accordingly to accommodate the higher frequency of data collection and ensure that more subtle changes can be captured. The timestamp alignment rules also need to be updated to ensure that data from different monitoring nodes accurately corresponds to the same point in time, facilitating subsequent analysis and processing.

[0133] For example, during normal monitoring, data collection is performed every minute, the sliding window granularity is 10 minutes, and the timestamp alignment rule is to align to the whole minute. When a level 1 alert is triggered, the data collection frequency is adjusted to every 30 seconds. Accordingly, the sliding window granularity is adjusted to 5 minutes, and the timestamp alignment rule is updated to align every 30 seconds.

[0134] Step S153: For the second-level warning, generate structural reinforcement recommendation parameters for the local area, and send the structural reinforcement recommendation parameters to the monitoring terminal for visual display.

[0135] When a Level 2 warning is triggered, it indicates that the slope is at risk of instability and requires structural reinforcement in a certain area. The local area requiring structural reinforcement is determined based on the spatial location information of the abnormal fluctuations recorded in the abnormal fluctuation feature set A.

[0136] By analyzing the severity and frequency of abnormal fluctuations, combined with factors such as the geological conditions and geotechnical properties of the target slope area, recommended structural reinforcement parameters for the local area are generated. These parameters may include reinforcement methods (such as anchor reinforcement and retaining wall reinforcement), material selection, specific location and scope of reinforcement, and strength requirements.

[0137] The generated structural reinforcement recommendation parameters are sent to a monitoring terminal for visual display. The monitoring terminal can be a computer or mobile device installed in a monitoring center. The recommended structural reinforcement parameters can be displayed on the monitoring terminal in the form of graphics, charts, and text descriptions, allowing relevant personnel to intuitively understand the areas requiring reinforcement and the specific reinforcement requirements.

[0138] Step S154: generating an early warning signal trigger instruction according to the early warning level, and outputting an audible and visual alarm signal and risk location identification information through the slope monitoring system.

[0139] For example, for a Level 1 warning, the triggering command requires the output of stronger and more obvious audible and visual alarm signals to draw the attention of relevant personnel. These signals can include high-decibel sirens and flashing warning lights. Simultaneously, the slope monitoring system's display screen uses eye-catching colors and symbols to highlight risk locations, such as red marking areas at risk of instability.

[0140] For Level 2 warnings, the triggering command requires the output of relatively weak audible and visual alarm signals, such as a moderately loud siren and slowly flashing warning lights. Risk location information is also displayed on the display, but with less conspicuous colors and markings, such as yellow to highlight areas requiring attention.

[0141] When the monitoring state is maintained, no sound or light alarm signal is triggered, but the slope monitoring system will continue to operate normally, displaying the monitoring data and status information of the target slope area in real time.

[0142] Through the above steps, the quality of the target slope area is monitored and warned based on the seismic data. Abnormal fluctuations in the slope can be detected in time, and corresponding early warning response measures can be taken to ensure the safety of the target slope area.

[0143] Figure 2 A schematic diagram illustrates exemplary hardware and software components of a seismic data-based slope quality monitoring and early warning system 100, which can implement the concepts of the present application, according to some embodiments of the present application. For example, a processor 120 can be used in the seismic data-based slope quality monitoring and early warning system 100 to perform the functions described in the present application.

[0144] The slope quality monitoring and early warning system 100 based on earthquake motion data can be a general-purpose server or a special-purpose server, both of which can be used to implement the slope quality monitoring and early warning method based on earthquake motion data of the present application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0145] For example, the slope quality monitoring and early warning system based on ground motion data 100 can include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as a disk, a ROM, or a RAM, or any combination thereof. Illustratively, the slope quality monitoring and early warning system based on ground motion data 100 can also include program instructions stored in a ROM, a RAM, or other types of non-transitory storage media, or any combination thereof. The methods of the present application can be implemented according to these program instructions. The slope quality monitoring and early warning system based on ground motion data 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0146] For ease of illustration, only one processor is described in the slope quality monitoring and early warning system based on ground motion data 100. However, it should be noted that the slope quality monitoring and early warning system based on ground motion data 100 in the present application can also include multiple processors, and thus the steps performed by one processor described in the present application can also be jointly performed by multiple processors or individually performed. For example, if the processor of the slope quality monitoring and early warning system based on ground motion data 100 performs steps A and B, it should be understood that steps A and B can also be jointly performed by two different processors or individually performed in one processor. For example, a first processor performs step A, a second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0147] In addition, the present application also provides a readable storage medium, in which computer executable instructions are preset, and when a processor executes the computer executable instructions, the slope quality monitoring and early warning method based on ground motion data is implemented.

[0148] It should be noted that, in order to simplify the expression of the present application and to help the understanding of one or more embodiments of the present application, in the foregoing description of the embodiments of the present application, various features are sometimes combined into one embodiment, figure, or description thereof.

Claims

1. A slope quality monitoring and early warning method based on seismic data, characterized in that: The method comprises: Acquire a set of earthquake monitoring data for a target slope area, the set of earthquake monitoring data including earthquake waveform data and corresponding environmental correlation parameters collected by multiple monitoring nodes within a preset time period; Performing a dynamic preprocessing operation on the earthquake motion monitoring data set to obtain a standardized earthquake motion monitoring data set, wherein the dynamic preprocessing operation includes data alignment, noise filtering, and dimension unification processing; Based on the standardized seismic monitoring data set, feature extraction is performed to generate a dynamic correlation feature set of the target slope area, wherein the dynamic correlation feature set includes seismic energy distribution features, slope response frequency features, and environmental coupling fluctuation features, wherein data integrity verification processing is performed on the environmental correlation parameters, an interpolation filling operation is performed on the missing parameters based on the verification results, and the filled environmental correlation parameters are normalized to obtain environmental parameter standardized data; dynamic correlation analysis processing is performed on the environmental parameter standardized data to determine the coupling correlation degree between environmental parameter changes and seismic energy distribution, and the environmental coupling fluctuation feature is generated based on the coupling correlation degree; Calling a slope quality analysis model to perform state prediction processing on the dynamic correlation feature set to obtain stable state features and abnormal fluctuation features of the target slope area, and generating a slope quality assessment result based on the difference between the stable state features and the abnormal fluctuation features; An early warning response strategy is determined based on the slope quality assessment result, and an early warning signal output operation of the slope monitoring system is triggered according to the early warning response strategy.

2. The slope quality monitoring and early warning method based on earthquake motion data according to claim 1 is characterized in that: The performing of a dynamic preprocessing operation on the earthquake motion monitoring data set to obtain a standardized earthquake motion monitoring data set includes: Performing noise component filtering and time-frequency conversion on the seismic motion waveform data to generate seismic motion spectrum data for each monitoring node, and performing frequency band energy normalization on the seismic motion spectrum data to obtain frequency band energy standardized data; The frequency band energy standardized data and the environmental parameter standardized data are further globally standardized and then timestamp aligned to generate standardized data units corresponding to the monitoring nodes one by one, and the standardized data units are merged to generate the standardized seismic monitoring data set.

3. The slope quality monitoring and early warning method based on earthquake motion data according to claim 2 is characterized in that: The extracting features based on the standardized seismic monitoring data set to generate a dynamic correlation feature set for the target slope area includes: Performing energy integration processing on the frequency band energy normalization data of each monitoring node to obtain the seismic energy distribution sequence of each monitoring node, and calculating the energy difference between adjacent monitoring nodes based on the seismic energy distribution sequence to generate an energy distribution difference feature; Performing a main frequency component extraction process on the frequency band energy normalized data to obtain a dominant frequency set of each monitoring node, and calculating the resonance response probability of the slope structure based on the dominant frequency set to generate a slope response frequency feature; The energy distribution difference characteristics, slope response frequency characteristics and environmental coupling fluctuation characteristics are subjected to spatial topological correlation processing to generate a dynamic correlation feature set of the target slope area.

4. The slope quality monitoring and early warning method based on earthquake motion data according to claim 3 is characterized in that: The performing of dynamic correlation analysis on the standardized environmental parameter data to determine the coupling correlation between the environmental parameter change and the seismic energy distribution, and generating an environmental coupling fluctuation feature based on the coupling correlation, includes: Performing sliding window slicing processing on the environmental parameter standardized data at a preset time granularity to generate a set of environmental parameter time series slices corresponding to each monitoring node, wherein each environmental parameter time series slice contains a standardized numerical sequence of temperature, humidity, and air pressure parameters at consecutive time steps; Slicing the seismic energy distribution sequence at the same time granularity, and allowing the environmental parameter time series slices to slide and align within a preset lag time range, to generate a set of seismic energy time series slices that take into account the lag effect of the environmental parameters; Dynamic correlation analysis is performed on the environmental parameter time series slices and the seismic energy time series slices of the same monitoring node. The linear Pearson correlation coefficient and the nonlinear mutual information index are combined to generate a comprehensive correlation strength matrix of each parameter dimension and the energy distribution dimension. The element value in the comprehensive correlation strength matrix is ​​the weighted average of the linear and nonlinear correlation strengths. Performing a maximum pooling operation across parameter dimensions on the comprehensive correlation strength matrix to extract the maximum coupling correlation strength between the environmental parameters and the seismic energy distribution at each time step, and generating a preliminary coupling correlation degree sequence; Based on the spatial gradient change of the preliminary coupling correlation degree sequence between adjacent monitoring nodes, a regional propagation consistency index of the coupling correlation strength is calculated, and the preliminary coupling correlation degree sequence is spatially smoothed and corrected to generate a corrected dynamic coupling correlation degree set; The dynamic coupling correlation degree set is input into a preset coupling strength grading model, and the coupling correlation degree of each monitoring node is mapped to a unified feature dimension through a fully connected layer to generate an environmental coupling fluctuation feature vector that matches the feature dimension of the seismic energy distribution.

5. The slope quality monitoring and early warning method based on earthquake motion data according to claim 4 is characterized in that: The performing spatial smoothing correction processing on the preliminary coupling correlation degree sequence to generate a corrected dynamic coupling correlation degree set includes: Performing first-order difference calculation in the time dimension on the preliminary coupling correlation degree sequence to generate a coupling correlation degree change trend sequence for each monitoring node; Based on the cosine similarity between the coupling correlation degree change trend sequences of adjacent monitoring nodes, the regional propagation consistency index is calculated, wherein the node pairs with the cosine similarity lower than the preset threshold are marked as spatial abnormal fluctuation areas; Performing a mean replacement operation based on K nearest neighbor nodes on the preliminary coupling correlation degree of the spatial abnormal fluctuation area to eliminate coupling correlation degree jump noise caused by local sensor abnormality; The Gaussian kernel function is used to perform spatial neighborhood weighted smoothing on the modified preliminary coupling correlation, where the kernel function bandwidth is adaptively adjusted according to the topological distance of the monitoring nodes. The smoothed dynamic coupling correlation degree is subjected to exponentially weighted moving average filtering in the time dimension to suppress high-frequency fluctuation noise and retain trend components, thereby generating the modified dynamic coupling correlation degree set.

6. The slope quality monitoring and early warning method based on earthquake motion data according to claim 1 is characterized in that: The calling of the slope quality analysis model to perform state prediction processing on the dynamic correlation feature set to obtain the stable state characteristics and abnormal fluctuation characteristics of the target slope area includes: Inputting the dynamic correlation feature set into the spatiotemporal coding network of the slope quality analysis model, performing global dependency modeling on the spatial topological correlation data in the dynamic correlation feature set through a multi-head self-attention mechanism, and generating a context coding vector containing cross-node spatiotemporal interaction information; The context encoding vector is input into the prediction branch of the slope quality analysis model, and the seismic motion evolution patterns at different time scales are extracted through a dilated temporal convolutional network. An energy distribution prediction sequence for multiple future time steps is output, and a smoothness index characterizing the stability of the slope structure is calculated based on the consistency of fluctuations at each time step in the energy distribution prediction sequence to generate the stable state feature. Furthermore, the context coding vector is input into the anomaly detection branch of the slope quality analysis model, and the context coding vector is reconstructed in latent space through a variational autoencoder to generate a reconstructed feature vector that matches the historical normal state distribution. The wavelet packet decomposition algorithm is used to calculate the residual gradient of the energy distribution of the original coding vector and the reconstructed feature vector in each frequency band, and the frequency band components and corresponding spatial positions where the residual gradient exceeds the historical fluctuation threshold are extracted to generate the abnormal fluctuation feature.

7. The slope quality monitoring and early warning method based on earthquake motion data according to claim 6 is characterized in that: The method of performing global dependency modeling on the spatial topological correlation data in the dynamic correlation feature set by using a multi-head self-attention mechanism to generate a context encoding vector includes: Performing time dimension slicing processing on the dynamic correlation feature set to generate a spatiotemporal feature slice set divided according to a preset time window; Based on the actual physical location distance of the monitoring nodes in each spatiotemporal feature slice and the correlation of historical seismic signals, a weighted connection relationship between the monitoring nodes is constructed, generating a node adjacency matrix and a characteristic node vector that reflect spatial proximity and signal correlation strength. Input the node adjacency matrix and the feature node vector into the multi-head self-attention mechanism, calculate the dynamic correlation strength between each node through the learnable attention weight, and generate a node-level attention weight matrix; Performing weighted aggregation on the feature node vectors based on the node-level attention weight matrix to generate a spatially aggregated feature vector within each time window; The spatially aggregated feature vectors of multiple time windows are positionally encoded and convolutionally fused in the time dimension to generate the context encoding vector containing the temporal evolution law.

8. The slope quality monitoring and early warning method based on earthquake motion data according to claim 6 is characterized in that: The step of performing latent space reconstruction on the context encoding vector by using a variational autoencoder to generate a reconstructed feature vector that matches the historical normal state distribution includes: Inputting the context encoding vector into the encoder network of the variational autoencoder, and mapping the input features to the mean vector and variance vector of the latent space through a multi-layer perceptron; Performing random sampling based on the mean vector and the variance vector to generate a latent space vector that obeys a Gaussian distribution; Inputting the latent space vector into the decoder network of the variational autoencoder, gradually restoring the feature dimension through the deconvolution layer and the fully connected layer to generate the reconstructed feature vector; Expand the original context coding vector and the reconstructed feature vector into a time series signal according to the time step, perform multi-scale wavelet packet decomposition on the time series signal, and calculate the relative error of the frequency band energy distribution at each scale; Determine a dynamic fluctuation threshold based on a distribution boundary of the relative error in a historical normal data set based on sliding window statistics; The spatial position and time point corresponding to the frequency band energy error exceeding the dynamic fluctuation threshold are marked as the abnormal fluctuation feature.

9. The slope quality monitoring and early warning method based on earthquake motion data according to claim 1 is characterized in that: The step of determining a warning response strategy based on the slope quality assessment result and triggering a warning signal output operation of the slope monitoring system according to the warning response strategy includes: The warning level is divided according to the size of the difference, wherein a first-level warning is triggered when the difference exceeds a first threshold, a second-level warning is triggered when the difference is within a second threshold range, and the monitoring state is maintained when the difference is lower than a third threshold, wherein the first threshold>the upper limit of the second threshold range>the lower limit of the second threshold range>the third threshold; For the first-level warning, adjust the data collection frequency of the monitoring node to the first preset frequency, and simultaneously update the sliding window time granularity and timestamp alignment rules in the preprocessing; For the second-level warning, generate structural reinforcement recommendation parameters for the local area and send the structural reinforcement recommendation parameters to the monitoring terminal for visual display; An early warning signal trigger instruction is generated according to the early warning level, and an audible and visual alarm signal and risk location identification information are output through the slope monitoring system.

10. A slope quality monitoring and early warning system based on seismic data, characterized in that: It includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the slope quality monitoring and early warning method based on seismic data as described in any one of claims 1 to 9.

Images