A geological disaster multi-parameter dynamic threshold adjustment early warning method

Abstract

The present application relates to the technical field of geological disaster monitoring and early warning, and discloses a geological disaster multi-parameter dynamic threshold adjustment early warning method. The method synchronously collects original monitoring data streams of at least two different types of geological sensors and performs fusion; signals reflecting different physical processes in the fused data stream are separated, potential multi-field coupling events are identified according to the overlapping degree and energy distribution characteristics of the signals in the time scale; a temporary monitoring grid is established based on the space-time range covered by the event, and an initial monitoring threshold is assigned to the nodes in the grid; the initial threshold is dynamically adjusted according to the data evolution path of each node in the continuous observation window to generate a gridded dynamic threshold profile; the morphological change of the profile is continuously tracked, and when the change meets the preset instability criterion, an early warning signal for the temporary monitoring grid is activated. The present application realizes accurate identification and dynamic early warning of multi-field coupling geological disaster precursors.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring and early warning technology, specifically a method for adjusting dynamic thresholds for multiple parameters in geological disasters. Background Technology

[0002] Existing geological disaster early warning technologies typically rely on multiple types of sensors for monitoring, but generally employ a simple overlay or weighted fusion approach to multi-source data. These methods treat monitoring signals from different physical processes as a single, mixed whole, making it difficult to distinguish whether data anomalies stem from changes in a single physical factor or from the interaction and reinforcement of multiple physical fields. This results in the system being insensitive to the multi-field coupled evolution processes commonly seen in geological disaster precursors, leading to insufficient targeting and clarity of the physical mechanisms in early warning systems.

[0003] Regarding the setting of early warning thresholds, current technologies mostly adopt fixed thresholds based on historical statistics or set uniform thresholds for the entire monitoring area. This static and global threshold management model cannot adapt to the dynamic characteristics of localized and migratory development during the formation of geological disasters. When anomalies begin locally in space and gradually evolve and spread, static thresholds either fail to detect in the early stages due to insufficient sensitivity, or continuously trigger alarms throughout the entire area without being able to locate the core risk area, resulting in a high false alarm rate and poor spatial accuracy in early warnings.

[0004] Effectively extracting and identifying precursors of coupled events, which are the result of the synergistic effects of multiple physical processes, from fused data has become crucial for improving the accuracy of early warning mechanisms. Simultaneously, constructing a localized, gridded early warning threshold system that can adaptively adjust in sync with the spatiotemporal development of such dynamically evolving coupled events is another core challenge for achieving precise early warning. Summary of the Invention

[0005] The purpose of this invention is to provide a method for early warning of geological disasters by adjusting dynamic thresholds for multiple parameters, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a method for early warning of geological disasters by adjusting dynamic thresholds for multiple parameters, the method comprising:

[0007] Select at least two different types of geological sensors within the monitoring area, simultaneously collect raw monitoring data streams, and fuse the raw monitoring data streams to obtain a fused data stream;

[0008] The signals reflecting different physical processes in the fused data stream are separated, and potential multi-field coupling events are identified based on the degree of overlap and energy distribution characteristics of the separated signals on the time scale.

[0009] Based on the spatiotemporal range covered by the potential multi-field coupling events, a temporary monitoring grid corresponding to the spatiotemporal range is established, and an initial monitoring threshold is assigned to each node in the temporary monitoring grid.

[0010] Based on the data evolution path of each node in the temporary monitoring grid within the continuous observation window, the initial monitoring threshold allocated to each node is dynamically adjusted to generate a gridded dynamic threshold profile.

[0011] The morphological changes of the gridded dynamic threshold profile are continuously tracked. When the morphological changes meet the preset instability criteria, an early warning signal is activated for the temporary monitoring grid.

[0012] Preferably, the process of fusing the original monitoring data stream to obtain the fused data stream specifically includes the following steps:

[0013] It receives raw monitoring data streams from geological sensors of different physical types, maps each raw monitoring data stream to the same time reference and spatial coordinate system, and generates a standardized data stream with spatiotemporal alignment.

[0014] The periodic interference patterns implied in the standardized data stream are analyzed, a filter bank matching the periodic interference patterns is constructed, and the standardized data stream is subjected to parallel filtering using the filter bank to obtain multiple clean data streams after denoising.

[0015] Extract the feature vectors of the multiple clean data streams within the same time slice, calculate the correlation matrix between the feature vectors, and determine the fusion weight of each data stream based on the spectral structure of the correlation matrix;

[0016] The multiple clean data streams are weighted and superimposed according to the fusion weights, and the phase delay introduced by the differences in sensor spatial layout is compensated during the superposition process, so as to finally generate a fused data stream containing multi-source information features.

[0017] Preferably, the operation of identifying potential multi-field coupling events specifically includes the following steps:

[0018] Empirical mode decomposition is performed on the fused data stream to obtain a series of intrinsic mode components arranged from high frequency to low frequency;

[0019] From the intrinsic modal components, at least three target components with an energy percentage exceeding a set ratio are selected, and the instantaneous frequency and instantaneous amplitude of the target components in the time-frequency domain are calculated respectively.

[0020] The instantaneous frequency and instantaneous amplitude of the target component are projected onto the same time axis. The clustered and dispersed regions of the projections of different target components on the time axis are observed, and the dense time intervals where the projection overlap exceeds the threshold are recorded.

[0021] Analyze the energy transfer direction and intensity of each target component within the dense time interval. If there is a pattern in which energy is directionally transferred from high-frequency components to low-frequency components and the intensity continues to increase, then the dense time interval and its corresponding energy transfer pattern are jointly marked as potential multi-field coupling events.

[0022] Preferably, the step of establishing a temporary monitoring grid corresponding to the spatiotemporal range and assigning initial monitoring thresholds to each node within the temporary monitoring grid specifically includes the following steps:

[0023] Based on the projection boundary of the potential multi-field coupling events in three-dimensional space, a minimum convex hull space region containing the projection boundary is defined;

[0024] The minimum convex hull space region is divided into regular or irregular spatial units. The vertex or center point of each spatial unit is defined as a monitoring node. All monitoring nodes and their spatial connection relationships constitute a temporary monitoring grid.

[0025] By retrospectively analyzing the historical sensor readings corresponding to the locations of each monitoring node during a period prior to the occurrence of the potential multi-field coupling event, the mean and variance of the historical sensor readings are statistically analyzed.

[0026] By combining the mean and variance, and introducing a scaling factor related to the energy level of the potential multi-field coupling event, a unique initial monitoring threshold is calculated and assigned to each node in the temporary monitoring grid.

[0027] Preferably, the generation of the meshed dynamic threshold profile specifically includes the following steps:

[0028] Obtain continuous monitoring data of each node within the temporary monitoring grid within the latest complete observation window, and fit the continuous monitoring data into a data evolution path that changes over time;

[0029] Compare the current form of the data evolution path with the path form of the node in the previous observation window, and calculate three quantitative indicators: path curvature change rate, path direction turning angle, and path point distribution entropy.

[0030] The three quantification metrics are input into a pre-trained nonlinear mapping network, which outputs a threshold adjustment factor for the node.

[0031] The threshold adjustment factor is used to scale the initial monitoring threshold previously assigned to the node, and this adjustment process is performed synchronously on all nodes in the temporary monitoring grid to generate a gridded dynamic threshold profile that reflects the threshold status of each node at the current moment.

[0032] Preferably, the method further includes the following operations:

[0033] After generating the gridded dynamic threshold profile, new sensor data from the adjacent area outside the temporary monitoring grid is monitored in real time.

[0034] If the newly added sensor data indicates that a new abnormal signal has appeared at the boundary of the temporary monitoring grid, the range of the temporary monitoring grid is dynamically expanded to encompass the new abnormal signal, and a temporary monitoring threshold obtained by interpolation based on the threshold of neighboring nodes is quickly assigned to the newly added nodes.

[0035] Preferably, the step of quickly assigning a temporary monitoring threshold to the newly added node based on the threshold interpolation of neighboring nodes specifically includes the following steps:

[0036] Identify the spatial proximity relationship between the new abnormal signal and the original boundary nodes of the temporary monitoring grid, and select at least three of the closest original boundary nodes as reference nodes;

[0037] Read the dynamic threshold currently held by the reference node in the gridded dynamic threshold profile, and obtain the spatial distance between the reference node and the newly added node;

[0038] An inverse distance weighted interpolation algorithm is used to calculate the dynamic threshold of the reference node based on the spatial distance, and the calculation result is used as the temporary monitoring threshold of the newly added node.

[0039] The newly added nodes and their temporary monitoring thresholds are incorporated into the management system of the temporary monitoring grid and the gridded dynamic threshold profile.

[0040] Preferably, activating the early warning signal for the temporary monitoring grid specifically includes the following steps:

[0041] The geometric morphological differences of the gridded dynamic threshold profile in two adjacent update cycles are continuously calculated. The geometric morphological differences are characterized by the average curvature change of the profile surface and the degree of topological change of the profile contour lines.

[0042] When the average curvature change continuously exceeds a set threshold and the topological change degree shows that the contour lines are broken or the closed loops suddenly shrink, the morphological change is determined to meet the preset instability criterion.

[0043] Based on the location of the region satisfying the instability criterion within the temporary monitoring grid, the core unstable node group is located, and the warning level is determined according to the number and distribution density of the core unstable node group;

[0044] Centered on the spatial coordinates of the core unstable node group, a spatially directional early warning signal is generated. The content of the early warning signal includes the identifier of the temporary monitoring grid, the early warning level, and the coordinate set of the core unstable node group.

[0045] Preferably, the three quantitative indicators used to calculate the path curvature change rate, path direction turning angle, and path point distribution entropy include:

[0046] Extract the discrete data point sequence arranged in chronological order within the current observation window of the data evolution path;

[0047] Cubic spline interpolation is performed on the discrete data point sequence to generate a smooth and continuous data evolution path curve;

[0048] Calculate the curvature of the data evolution path curve at each data point, and obtain the average rate of change of the curvature of all data points in the current observation window relative to the curvature of the corresponding data points in the previous observation window, as the path curvature change rate.

[0049] Connect adjacent data points to form path segments, calculate the direction angle of each path segment within the current observation window, and take the sum of the absolute values ​​of the differences in direction angles between adjacent path segments as the path direction turning angle;

[0050] The distribution probability of data points within the current observation window in the numerical distribution interval is statistically analyzed, and the Shannon entropy value of the data point distribution is calculated based on the distribution probability, which is used as the path point distribution entropy.

[0051] Preferably, the continuous calculation of the geometrical differences of the meshed dynamic threshold profile in two adjacent update cycles includes:

[0052] The gridded dynamic threshold profile is represented as a three-dimensional spatial surface, where the planar coordinates correspond to the monitoring node positions and the height coordinates correspond to the node dynamic thresholds.

[0053] The three-dimensional spatial surface of the current update cycle is triangulated, the average curvature of each triangular facet is calculated, and the average value of the average curvature of all triangular facets of the entire surface is obtained as the average curvature of the profile surface in the current cycle.

[0054] Calculate the absolute difference between the average curvature of the profile surface in the current cycle and the average curvature of the profile surface in the previous update cycle, and use it as the change in the average curvature of the profile surface.

[0055] Extract the set of contour lines from the current update cycle's gridded dynamic threshold profile, analyze the connectivity and closure of each contour line, and identify the number of newly generated contour line breakpoints and the area reduction ratio of existing closed contour line loops.

[0056] The topological change of the profile contour lines is obtained by summing the total number of break points of all contour lines and the number of loops whose closed loop area shrinkage exceeds a threshold.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] By separating signals from different physical processes in multi-sensor fusion data streams and analyzing the temporal overlap and energy distribution characteristics of the separated signals, the coupling relationships between changes in individual physical fields such as displacement, seepage, and stress can be effectively identified. This analysis, based on signal separation and spatiotemporal-energy correlation, can distinguish between simple random interference and genuine multi-field coordinated degradation processes. This elevates early warning judgment from being based on isolated parameter exceedances to identifying coupled events based on mutual corroboration of multiple physical mechanisms, enhancing the physical reliability of early warning signals and the accuracy of event identification.

[0059] For identified potential coupled events, a temporary monitoring grid is established within the spatiotemporal range of their impact. The node thresholds are dynamically adjusted based on the data evolution path of each node within a continuous observation window, generating a gridded dynamic threshold profile whose shape changes with the event's development. Tracking the morphological changes of this profile serves as an instability criterion, allowing early warnings to no longer rely on breakthroughs of fixed values, but rather on "morphological instability" reflecting the overall instability trend of the system. This method endows the early warning threshold with spatial local adaptability and temporal dynamic evolution capabilities, enabling earlier detection of locally initiated anomalous evolutions and more accurately characterizing the expansion process of risk areas, thereby reducing missed alarms and false alarms. Attached Figure Description

[0060] Figure 1 This is a schematic diagram illustrating the working principle of the multi-parameter dynamic threshold adjustment and early warning method for geological disasters described in this invention.

[0061] Figure 2 A flowchart for the fusion of raw monitoring data streams;

[0062] Figure 3 A flowchart for establishing a temporary monitoring grid and assigning initial thresholds;

[0063] Figure 4 Grouped bar chart showing the fusion effect of multi-source sensor data for geological disaster monitoring;

[0064] Figure 5 This is a 3D surface map showing the spatial distribution of dynamic thresholds in a geological disaster monitoring area. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] Please see Figure 1 This invention provides a method for early warning of geological disasters using multi-parameter dynamic threshold adjustment. The method includes: selecting at least two different types of geological sensors within a monitoring area, such as displacement sensors, pressure sensors, and acoustic emission sensors, which simultaneously acquire raw monitoring data streams reflecting different physical characteristics of the geological body. The acquired raw monitoring data streams are fused to obtain a fused data stream. Signals reflecting different physical processes in the fused data stream are separated, and potential multi-field coupling events are identified based on the degree of overlap and energy distribution characteristics of the separated signals over time. Based on the spatiotemporal range covered by the identified potential multi-field coupling events, a temporary monitoring grid corresponding to this spatiotemporal range is established, and initial monitoring thresholds are assigned to each node within the temporary monitoring grid. The initial monitoring thresholds assigned to each node are dynamically adjusted according to the data evolution path of each node within a continuous observation window, thereby generating a gridded dynamic threshold profile. The morphological changes of the gridded dynamic threshold profile are continuously tracked, and when the morphological change meets a preset instability criterion, an early warning signal is activated for the temporary monitoring grid.

[0067] In one embodiment of the present invention, see [reference] Figure 2 In specific implementation, taking the monitoring of a rock slope as an example, three different types of geological sensors were selected: displacement sensor, pressure sensor and acoustic emission sensor. The displacement sensor outputs displacement change data stream, the pressure sensor outputs pore water pressure data stream, and the acoustic emission sensor outputs acoustic emission event rate data stream. These sensors synchronously collect data once per second, forming three raw monitoring data streams.

[0068] During data fusion, raw monitoring data streams from displacement sensors, pore water pressure sensors, and acoustic emission sensors are received. The timestamps of each raw monitoring data stream are aligned to a unified BeiDou satellite navigation system time reference, and the spatial coordinates of all sensors are transformed to an independent three-dimensional coordinate system with the slope toe as the origin, generating a standardized data stream with spatiotemporal alignment. The daily variation interference patterns with a 24-hour period and the mechanical vibration interference patterns with a period of several minutes are analyzed within the standardized data stream. A filter bank consisting of band-stop filters and low-pass filters with corresponding cutoff frequencies is constructed. This filter bank is used to perform parallel filtering on the standardized data stream, removing daily variation signals and high-frequency vibration noise, resulting in three clean data streams for displacement, pore water pressure, and acoustic emission event rate after denoising. Within a five-minute time slice, the feature vectors of the displacement pure data stream (standard deviation, kurtosis), the pore water pressure pure data stream (mean, slope), and the acoustic emission event rate pure data stream (event count, energy sum) are extracted. A correlation matrix is ​​calculated among these three feature vectors. The fusion weights of each data stream are determined based on the eigenvector component corresponding to the largest eigenvalue of the correlation matrix. The three pure data streams are weighted and superimposed with a fusion weight of 0.4 for the displacement data stream, 0.35 for the pore water pressure data stream, and 0.25 for the acoustic emission data stream. During the superposition process, phase delays introduced by differences in sensor spatial deployment are compensated based on sensor location and wave velocity models. Finally, a fused data stream containing multi-source information features is generated.

[0069] In identifying potential multi-field coupling events, empirical mode decomposition (EMD) is performed on the fused data stream to obtain seven intrinsic mode components (IMF1 to IMF7). From these seven IMFs, three target components—IMF2, IMF3, and IMF5—with a cumulative energy percentage exceeding 80% are selected. The instantaneous frequencies and amplitudes of these target components in the time-frequency domain are calculated. The instantaneous frequencies and amplitudes of the target components IMF2, IMF3, and IMF5 are projected onto the same time axis, and the clustering and dispersion regions of the projections on the time axis are observed. The dense time intervals T1 to T3, where the projection overlap exceeds 60%, are recorded. Analyzing the energy transfer direction and intensity of each target component within the dense time intervals T1 to T3, within the dense time interval T2, there exists a pattern where energy is directionally transferred from target component IMF2 to target component IMF5, and the transfer intensity continuously increases from 0.5 units per second to 2.1 units per second. Therefore, the dense time interval T2 and its corresponding energy transfer pattern are jointly labeled as potential multi-field coupling events.

[0070] In some embodiments, the process of calculating the spectral structure of the correlation matrix between eigenvectors to determine the fusion weights is performed based on the following relationship:

[0071] ;

[0072] in: Indicates the first The fusion weight of pure data streams. This represents the eigenvector corresponding to the largest eigenvalue of the calculated correlation matrix. One portion, This represents the total number of clean data streams participating in the fusion. This indicates the absolute value operation, where the denominator is summed over the absolute values ​​of all components of the eigenvector to achieve normalization.

[0073] It is understood that the design of filter banks is not limited to band-stop and low-pass types. In specific implementations, for environments with significant periodic electromagnetic interference, filter banks may additionally include notch filters. Optionally, the elements constructing the feature vector may include time-domain statistical features, frequency-domain features, or time-frequency-domain features. In some embodiments, the projection overlap is calculated by statistically analyzing the proportion of sample points whose instantaneous frequency curves and instantaneous amplitude curves of different target components simultaneously fall within their respective preset value intervals within a sliding time window, relative to the total window length. Optionally, the determination of the energy transfer mode must simultaneously satisfy three conditions: directional consistency, intensity monotonicity, and a lower limit of duration.

[0074] In one embodiment of the present invention, see [reference] Figure 3In practical implementation, taking the monitoring of a potential landslide as an example, after identifying a potential multi-field coupling event lasting five minutes, the projection boundary of the potential multi-field coupling event in three-dimensional space is determined by the maximum envelope of the displacement monitoring data projected onto the horizontal plane and the projection depth range of the acoustic emission event on the vertical profile. Using the spatial coordinates of these boundary points (100, 50, -20), (150, 50, -20), (150, 80, -5), and (100, 80, -5), etc., as vertices, the minimum convex hull spatial region containing all these vertices is calculated. This minimum convex hull spatial region is an irregular polyhedron. The minimum convex hull spatial region is divided into regular square grids at 10-meter intervals on the horizontal plane, and into three layers vertically according to geological strata. The intersection of each grid point and the stratum is defined as a monitoring node. All monitoring nodes and their spatial proximity connections constitute a temporary monitoring grid covering the potential risk area. The temporary monitoring grid contains a total of 156 monitoring nodes. By retrospectively analyzing the historical sensor readings of displacement and pore water pressure sensors at each monitoring node's location over a 30-day period prior to potential multi-field coupling events, the mean and variance of the historical displacement and pore water pressure sensor readings were calculated for each monitoring node. Combining these historical readings, and introducing a scaling factor related to the energy level of the potential multi-field coupling event—obtained from the maximum energy transfer intensity recorded at the time of event marking through a linear mapping function—a unique initial monitoring threshold was calculated and assigned to each node in the temporary monitoring grid.

[0075] In some embodiments, the formula for calculating the initial monitoring threshold is expressed as follows:

[0076] ;

[0077] in: This represents the calculated initial monitoring threshold for a certain monitoring parameter at a specific monitoring node. This represents the average of the historical sensor readings corresponding to this monitoring node. This represents the variance of the historical sensor readings corresponding to that monitoring node. This represents a scaling factor associated with the energy level of a potential multi-field coupled event. The numerical range is between 1.5 and 3.0, determined by the energy transfer intensity and a preset mapping table. It is understandable that the length of the historical period is adjusted based on the historical activity cycle of the specific geological hazard type; for creeping landslides, the historical period may be extended to several months. Optionally, the spatial units can be divided irregularly based on geological structural lines or rock strata interfaces instead of a regular grid, allowing monitoring nodes to be more densely deployed on key geological units.

[0078] In some embodiments, the calculation of the minimum convex hull spatial region is implemented using a three-dimensional fast convex hull algorithm to ensure that the temporary monitoring grid can tightly enclose the spatial projection of the abnormal signal. Optionally, when assigning initial monitoring thresholds to nodes in the temporary monitoring grid, for newly added virtual nodes lacking direct historical sensor readings, their initial monitoring thresholds can be obtained by spatial interpolation, using the mean and variance interpolated from neighboring actual monitoring nodes with historical sensor readings. It is understood that the scaling factor... The introduction of this feature enables the initial monitoring threshold to dynamically respond to the initial energy level of potential multi-field coupled events; the higher the event energy level, the larger the scaling factor. The larger the value, the higher the calculated initial monitoring threshold. relative to historical average The larger the upward adjustment, the more sensitive the triggering conditions can be set in the early stages of the warning.

[0079] In one embodiment of the present invention, taking a landslide area with a temporary monitoring grid containing 156 monitoring nodes as an example, a gridded dynamic threshold profile is generated. Continuous monitoring data from the fused data stream is obtained for each node within the temporary monitoring grid within the latest complete observation window, for example, a ten-minute time window. This continuous monitoring data is a one-dimensional sequence changing over time. The continuous monitoring data is fitted into a data evolution path that changes over time using the least squares method. The current form of the data evolution path of node A is compared with the path form of node A in the previous observation window, and three quantitative indicators are calculated: the path curvature change rate, the path direction turning angle, and the path point distribution entropy. Sixty discrete data point sequences of node A's data evolution path are extracted in chronological order within the current observation window. Cubic spline interpolation is performed on the discrete data point sequences to generate a smooth and continuous data evolution path curve. The curvature of the data evolution path curve at each discrete data point is calculated, and the average rate of change of curvature of all sixty data points in the current observation window relative to the curvature of the corresponding data point in the previous observation window is obtained. This average is taken as the path curvature change rate. For example, the calculated path curvature change rate for node A is 0.15 radians per second. Fifty-nine path segments are formed by connecting adjacent discrete data points. The direction angle of each path segment in the current observation window is calculated, and the sum of the absolute values ​​of the differences between the direction angles of adjacent path segments is obtained. This sum is taken as the path direction turning angle. For example, the path direction turning angle for node A is 4.7 radians. The distribution probability of the sixty discrete data points in the current observation window within the numerical distribution interval is statistically analyzed. The numerical range is divided into ten equally spaced intervals, and the number of data points falling into each interval is counted and the probability is calculated. Based on the distribution probability, the Shannon entropy value of the data point distribution is calculated. This entropy value is taken as the path point distribution entropy.

[0080] In some embodiments, path point distribution entropy The calculation is based on the following formula:

[0081] ;

[0082] in: Represents the path point distribution entropy. This represents the total number of intervals in the numerical distribution. This indicates that the data point in the current observation window falls into the first... The probability of a numerical distribution interval. This represents a logarithmic operation with base 2. It can be understood as the total number of intervals in the numerical distribution. The settings can be adjusted according to the data range and accuracy requirements, for example, when the data changes gradually. The value can be appropriately reduced to smooth the probability distribution. Optionally, the rate of change of path curvature can be calculated using the central difference method, and the direction angle of the path segment can be calculated based on the coordinate difference between the two endpoints of the segment.

[0083] In practice, the calculated path curvature change rate, path direction turning angle, and path point distribution entropy are input into a pre-trained nonlinear mapping network. This nonlinear mapping network is a fully connected neural network with one hidden layer. It receives three inputs and outputs a scalar value, which serves as the threshold adjustment factor for node A. For example, the threshold adjustment factor output by the nonlinear mapping network is 1.32. The initial monitoring threshold previously assigned to node A is scaled using the threshold adjustment factor of 1.32. Assuming the initial monitoring threshold for node A is 50 mm, the adjusted dynamic threshold is 66 mm. This data evolution path fitting, calculation of the three quantification indicators, nonlinear mapping network processing, and threshold scaling process are simultaneously performed on all 156 nodes within the temporary monitoring grid. This generates a gridded dynamic threshold profile reflecting the current threshold state of each node. The gridded dynamic threshold profile is a three-dimensional data structure, where the planar coordinates correspond to the spatial position of the nodes, and the height coordinates correspond to the dynamic thresholds of the nodes.

[0084] In some embodiments, the nonlinear mapping network is trained using historical monitoring data and expert-annotated threshold-adjusted sample pairs. Optionally, in addition to the least squares method, moving average or exponential smoothing methods can also be used to fit the data evolution path. It is understood that the length of the observation window needs to be set according to the evolution rate of the geological hazard; for rapid landslides, the observation window may be shortened to the minute level, while for slow creep, the observation window may be extended to the hour level.

[0085] In one embodiment of the present invention, taking a landslide area under monitoring as an example, after generating a gridded dynamic threshold profile reflecting the threshold status of each node at the current moment, the system monitors data from newly added sensors deployed in the adjacent area outside the temporary monitoring grid in real time. The newly added sensors include a tilt sensor located approximately 15 meters outside the original eastern boundary of the temporary monitoring grid. Over three consecutive sampling periods, the new sensor data transmitted back by the tilt sensor shows that its X-axis tilt angle value consistently exceeds twice the standard deviation of the historical baseline value. This new sensor data indicates a new anomalous signal at the eastern boundary of the temporary monitoring grid. The system then dynamically expands the range of the temporary monitoring grid to encompass the new anomalous signal. The expansion logic extends the grid coverage eastward to include the area surrounding the location of the new anomalous signal and quickly assigns a temporary monitoring threshold, obtained by interpolating the thresholds of neighboring nodes, to the newly added nodes.

[0086] The spatial proximity of the new anomalous signal to the original boundary nodes of the temporary monitoring grid is identified. The geographic coordinates of the new anomalous signal are (165, 75, -10). The three closest original boundary nodes are selected as reference nodes, with coordinates of Node_E1 (155, 70, -8), Node_E2 (150, 80, -12), and Node_E3 (160, 85, -5). The dynamic thresholds held by reference nodes Node_E1, Node_E2, and Node_E3 in the gridded dynamic threshold profile are read, and the spatial Euclidean distance between each reference node and the new node is obtained. An inverse distance weighted interpolation algorithm is used to calculate the dynamic thresholds of the reference nodes based on the spatial distance. The calculation results are used as the temporary monitoring thresholds for the new nodes (see Table 1).

[0087] Table 1: Spatial distance and dynamic threshold data between reference nodes and newly added nodes

[0088]

[0089] In practice, inverse distance weighted interpolation is used to calculate the temporary monitoring threshold for newly added nodes. The process is performed according to the following formula:

[0090] ;

[0091] in: This represents the calculated temporary monitoring threshold for newly added nodes. This represents the total number of reference nodes, as used in this specific implementation. , Indicates the first The dynamic threshold of each reference node, Indicates the first The weights of each reference node, Indicates the first Spatial distance between reference nodes and newly added nodes This represents the power-law parameter of distance decay, typically with a value of 2. It can be understood that the power-law parameter of distance decay... The value of affects the decay rate of the weights of neighboring nodes. The larger the value, the smaller the contribution of distant reference nodes to the interpolation result. Optionally, the number of reference nodes is not limited to three; the system can automatically select the N nearest neighbors to participate in the calculation based on the distribution density of nodes around the newly added node.

[0092] In some embodiments, the dynamic expansion of the temporary monitoring grid occurs not only when new anomalous signals are detected, but also, in specific implementations, when the system predicts that the scope of disaster impact may expand, based on preset expansion rules, the grid will be expanded in advance. New nodes and their calculated temporary monitoring thresholds are incorporated into the management system of the temporary monitoring grid and the gridded dynamic threshold profile. New nodes participate in the dynamic threshold adjustment calculations for all subsequent nodes and also serve as potential reference nodes for future grid expansion. Optionally, the assignment of temporary monitoring thresholds is not the final state. After a new node is incorporated into the system, its temporary monitoring threshold will serve as the node's initial monitoring threshold, participating in the data evolution path analysis and dynamic threshold adjustment process of the next observation window. It can be understood that the inverse distance weighted interpolation algorithm ensures that the threshold assignment of a new node is closely related to the current state of its spatially neighboring nodes, thus maintaining spatial continuity in the expanded gridded dynamic threshold profile.

[0093] See Figure 4 This is a grouped bar chart showing the fusion effect of multi-source sensor data for geological disaster monitoring. It primarily displays the weight of different types of sensors in data fusion and the performance improvement after denoising. The "signal-to-noise ratio improvement rate" of all sensors is significantly higher than the "data fusion weight," indicating that denoising processing has a significant effect on optimizing the quality of data from various sensors. The comparison of these two indicators reflects the priority and preprocessing effect of different sensors in data fusion. This type of chart is used in the multi-source data fusion stage to help determine the fusion weight allocation strategy for sensors and to verify the effectiveness of denoising preprocessing. The chart reflects the "fusion value difference" of sensors: sensors whose data are directly related to geological deformation, such as displacement and tilt angle, have a higher proportion in the fusion process, and their data shows a more significant quality improvement after denoising, providing high-quality basic data for subsequent identification of multi-field coupled events.

[0094] In one embodiment of the present invention, taking a temporary monitoring grid under continuous monitoring as an example, the temporary monitoring grid contains 162 nodes, and the gridded dynamic threshold profile is updated every five minutes. The geometric difference of the gridded dynamic threshold profile in two adjacent update cycles is continuously calculated. The gridded dynamic threshold profile of the current update cycle is represented as a three-dimensional spatial surface, where the planar coordinates (X, Y) correspond to the projection position of the monitoring node on the topographic map, and the height coordinate (Z) corresponds to the node's dynamic threshold. The unit of the height coordinate is millimeters. The three-dimensional spatial surface of the current update cycle is triangulated. Based on the planar coordinates of the nodes, a surface model seamlessly stitched together by triangular facets is generated. The average curvature of each triangular facet is calculated, and the arithmetic mean of the average curvature of all triangular facets on the entire surface is obtained. This arithmetic mean is used as the average curvature of the profile surface in the current cycle. The absolute difference between the average curvature of the profile surface in the current cycle and that in the previous update cycle is calculated. This absolute difference is taken as the change in the average curvature of the profile surface. For example, between cycle t and cycle t-1, the change in average curvature is calculated to be 0.023 mm to the power of -1. The set of contour lines in the meshed dynamic threshold profile of the current update cycle is extracted. Contour lines are spatial curves formed by connecting nodes with the same dynamic threshold. The connectivity and closure of each contour line are analyzed, identifying the number of newly generated contour line breakpoints and the area reduction ratio of existing closed contour line loops. A breakpoint refers to a discontinuous position on a contour line, and the area reduction ratio of a closed loop is calculated by comparing the area of ​​the closed loop in the current cycle with the corresponding closed loop area in the previous cycle. The total number of breakpoints and the number of loops with an area reduction ratio exceeding 10% are counted. The total number of breakpoints is weighted at 0.6, and the number of loops with significantly reduced area is weighted at 0.4. The weighted sum is then used to obtain the topological change degree of the profile contour lines. For example, the calculated topological change degree is 15.7.

[0095] In some embodiments, the degree of topological variation of profile contour lines The calculation formula is:

[0096] ;

[0097] in: Indicates the degree of topological variation of contour lines in a cross-section. This represents the total number of newly identified breakpoints across all contour lines. This indicates that the reduction in the area of ​​the closed loop exceeds a preset threshold. The number of closed contour loops. and Let represent the weighting coefficients for the total number of fracture points and the number of contraction rings, respectively, and satisfy . This is understandable; a preset threshold... with weighting coefficients , The numerical values ​​need to be calibrated based on the specific type of geological hazard and the monitoring accuracy. Optionally, the Delaunay triangulation algorithm can be used for triangular meshing to ensure the quality of the generated triangular patches.

[0098] In practical implementation, when the average curvature change of the profile surface exceeds the set threshold of 0.02 mm -1 for three consecutive update cycles, and simultaneously, the topological change of the contour lines shows that the contour lines are broken or the closed loops suddenly shrink, specifically manifested as a topological change value greater than the preset criterion threshold of 10, the system determines that the morphological change of the gridded dynamic threshold profile meets the preset instability criterion. Based on the location of the area that meets the instability criterion within the temporary monitoring grid, the core instability node group is located. The core instability node group is the set of nodes located in the region of abrupt change in average curvature and surrounded by broken contour lines or abruptly shrinking closed loops. For example, a core instability node group containing 23 nodes was located. The warning level is determined based on the number and distribution density of the core instability node group. When the number of core instability node groups is between 10 and 30 and the distribution density is greater than 0.1 nodes per square meter, the warning level is determined to be orange level two.

[0099] See Figure 5 This is a 3D surface map showing the spatial distribution of dynamic thresholds in a geological hazard monitoring area, primarily displaying the differences in dynamic thresholds at different spatial locations within the monitoring grid. The spatial heterogeneity of dynamic thresholds is visually presented through a 3D surface and color gradient, with the morphology of high-threshold areas directly correlated with the distribution of geological risks. This type of chart is used in the generation stage of gridded dynamic threshold profiles to help visualize the risk distribution in the monitoring area, providing spatial location data for subsequent instability assessments. This map reflects the "spatial differentiation characteristics" of geological risks: high dynamic threshold areas are geologically deformation-sensitive zones requiring focused monitoring resource allocation; low-threshold areas can have their monitoring frequency appropriately reduced, thereby optimizing the resource allocation efficiency of the early warning system.

[0100] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0101] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for early warning of geological disasters with multi-parameter dynamic threshold adjustment, characterized in that, The method includes: Select at least two different types of geological sensors within the monitoring area, simultaneously collect raw monitoring data streams, and fuse the raw monitoring data streams to obtain a fused data stream; The signals reflecting different physical processes in the fused data stream are separated, and potential multi-field coupling events are identified based on the degree of overlap and energy distribution characteristics of the separated signals on the time scale. Based on the spatiotemporal range covered by the potential multi-field coupling events, a temporary monitoring grid corresponding to the spatiotemporal range is established, and an initial monitoring threshold is assigned to each node in the temporary monitoring grid. Based on the data evolution path of each node in the temporary monitoring grid within the continuous observation window, the initial monitoring threshold allocated to each node is dynamically adjusted to generate a gridded dynamic threshold profile. The morphological changes of the gridded dynamic threshold profile are continuously tracked. When the morphological changes meet the preset instability criteria, an early warning signal is activated for the temporary monitoring grid.

2. The method for early warning of geological disasters with multi-parameter dynamic threshold adjustment according to claim 1, characterized in that, The process of fusing the original monitoring data stream to obtain the fused data stream specifically includes the following steps: It receives raw monitoring data streams from geological sensors of different physical types, maps each raw monitoring data stream to the same time reference and spatial coordinate system, and generates a standardized data stream with spatiotemporal alignment. The periodic interference patterns implied in the standardized data stream are analyzed, a filter bank matching the periodic interference patterns is constructed, and the standardized data stream is subjected to parallel filtering using the filter bank to obtain multiple clean data streams after denoising. Extract the feature vectors of the multiple clean data streams within the same time slice, calculate the correlation matrix between the feature vectors, and determine the fusion weight of each data stream based on the spectral structure of the correlation matrix; The multiple clean data streams are weighted and superimposed according to the fusion weights, and the phase delay introduced by the differences in sensor spatial layout is compensated during the superposition process, so as to finally generate a fused data stream containing multi-source information features.

3. The method for early warning of geological disasters with multi-parameter dynamic threshold adjustment according to claim 2, characterized in that, The operation of identifying potential multi-field coupling events specifically includes the following steps: Empirical mode decomposition is performed on the fused data stream to obtain a series of intrinsic mode components arranged from high frequency to low frequency; From the intrinsic modal components, at least three target components with an energy percentage exceeding a set ratio are selected, and the instantaneous frequency and instantaneous amplitude of the target components in the time-frequency domain are calculated respectively. The instantaneous frequency and instantaneous amplitude of the target component are projected onto the same time axis. The clustered and dispersed regions of the projections of different target components on the time axis are observed, and the dense time intervals where the projection overlap exceeds the threshold are recorded. Analyze the energy transfer direction and intensity of each target component within the dense time interval. If there is a pattern in which energy is directionally transferred from high-frequency components to low-frequency components and the intensity continues to increase, then the dense time interval and its corresponding energy transfer pattern are jointly marked as potential multi-field coupling events.

4. The method for early warning of geological disasters with multi-parameter dynamic threshold adjustment according to claim 3, characterized in that, The establishment of a temporary monitoring grid corresponding to the spatiotemporal range, and the assignment of initial monitoring thresholds to each node within the temporary monitoring grid, specifically includes the following steps: Based on the projection boundary of the potential multi-field coupling events in three-dimensional space, a minimum convex hull space region containing the projection boundary is defined; The minimum convex hull space region is divided into regular or irregular spatial units. The vertex or center point of each spatial unit is defined as a monitoring node. All monitoring nodes and their spatial connection relationships constitute a temporary monitoring grid. By retrospectively analyzing the historical sensor readings corresponding to the locations of each monitoring node during a period prior to the occurrence of the potential multi-field coupling event, the mean and variance of the historical sensor readings are statistically analyzed. By combining the mean and variance, and introducing a scaling factor related to the energy level of the potential multi-field coupling event, a unique initial monitoring threshold is calculated and assigned to each node in the temporary monitoring grid.

5. The method for early warning of geological disasters with multi-parameter dynamic threshold adjustment according to claim 4, characterized in that, The generation of the gridded dynamic threshold profile specifically includes the following steps: Obtain continuous monitoring data of each node within the temporary monitoring grid within the latest complete observation window, and fit the continuous monitoring data into a data evolution path that changes over time; Compare the current form of the data evolution path with the path form of the node in the previous observation window, and calculate three quantitative indicators: path curvature change rate, path direction turning angle, and path point distribution entropy. The three quantification metrics are input into a pre-trained nonlinear mapping network, which outputs a threshold adjustment factor for the node. The threshold adjustment factor is used to scale the initial monitoring threshold previously assigned to the node, and this adjustment process is performed synchronously on all nodes in the temporary monitoring grid to generate a gridded dynamic threshold profile that reflects the threshold status of each node at the current moment.

6. The method for early warning of geological disasters with multi-parameter dynamic threshold adjustment according to claim 1, characterized in that, The method also includes the following operations: After generating the gridded dynamic threshold profile, new sensor data from the adjacent area outside the temporary monitoring grid is monitored in real time. If the newly added sensor data indicates that a new abnormal signal has appeared at the boundary of the temporary monitoring grid, the range of the temporary monitoring grid is dynamically expanded to encompass the new abnormal signal, and a temporary monitoring threshold obtained by interpolation based on the threshold of neighboring nodes is quickly assigned to the newly added nodes.

7. The method for early warning of geological disasters with multi-parameter dynamic threshold adjustment according to claim 6, characterized in that, The process of quickly assigning a temporary monitoring threshold to newly added nodes, based on interpolation of thresholds from neighboring nodes, specifically includes the following steps: Identify the spatial proximity relationship between the new abnormal signal and the original boundary nodes of the temporary monitoring grid, and select at least three of the closest original boundary nodes as reference nodes; Read the dynamic threshold currently held by the reference node in the gridded dynamic threshold profile, and obtain the spatial distance between the reference node and the newly added node; An inverse distance weighted interpolation algorithm is used to calculate the dynamic threshold of the reference node based on the spatial distance, and the calculation result is used as the temporary monitoring threshold of the newly added node. The newly added nodes and their temporary monitoring thresholds are incorporated into the management system of the temporary monitoring grid and the gridded dynamic threshold profile.

8. The method for early warning of geological disasters with multi-parameter dynamic threshold adjustment according to claim 5, characterized in that, Activating the early warning signal for the temporary monitoring grid specifically includes the following steps: The geometric morphological differences of the gridded dynamic threshold profile in two adjacent update cycles are continuously calculated. The geometric morphological differences are characterized by the average curvature change of the profile surface and the degree of topological change of the profile contour lines. When the average curvature change continuously exceeds a set threshold and the topological change degree shows that the contour lines are broken or the closed loops suddenly shrink, the morphological change is determined to meet the preset instability criterion. Based on the location of the region satisfying the instability criterion within the temporary monitoring grid, the core unstable node group is located, and the warning level is determined according to the number and distribution density of the core unstable node group; Centered on the spatial coordinates of the core unstable node group, a spatially directional early warning signal is generated. The content of the early warning signal includes the identifier of the temporary monitoring grid, the early warning level, and the coordinate set of the core unstable node group.

9. The method for early warning of geological disasters with multi-parameter dynamic threshold adjustment according to claim 5, characterized in that, The three quantitative indicators used in the calculation include: the rate of change of path curvature, the path direction turning angle, and the path point distribution entropy. Extract the discrete data point sequence arranged in chronological order within the current observation window of the data evolution path; Cubic spline interpolation is performed on the discrete data point sequence to generate a smooth and continuous data evolution path curve; Calculate the curvature of the data evolution path curve at each data point, and obtain the average rate of change of the curvature of all data points in the current observation window relative to the curvature of the corresponding data points in the previous observation window, as the path curvature change rate. Connect adjacent data points to form path segments, calculate the direction angle of each path segment within the current observation window, and take the sum of the absolute values ​​of the differences in direction angles between adjacent path segments as the path direction turning angle; The distribution probability of data points within the current observation window in the numerical distribution interval is statistically analyzed, and the Shannon entropy value of the data point distribution is calculated based on the distribution probability, which is used as the path point distribution entropy.

10. The method for early warning of geological disasters with multi-parameter dynamic threshold adjustment according to claim 8, characterized in that, The continuous calculation of the geometrical differences of the gridded dynamic threshold profile between two adjacent update cycles includes: The gridded dynamic threshold profile is represented as a three-dimensional spatial surface, where the planar coordinates correspond to the monitoring node positions and the height coordinates correspond to the node dynamic thresholds. The three-dimensional spatial surface of the current update cycle is triangulated, the average curvature of each triangular facet is calculated, and the average value of the average curvature of all triangular facets of the entire surface is obtained as the average curvature of the profile surface in the current cycle. Calculate the absolute difference between the average curvature of the profile surface in the current cycle and the average curvature of the profile surface in the previous update cycle, and use it as the change in the average curvature of the profile surface. Extract the set of contour lines from the current update cycle's gridded dynamic threshold profile, analyze the connectivity and closure of each contour line, and identify the number of newly generated contour line breakpoints and the area reduction ratio of existing closed contour line loops. The topological change of the profile contour lines is obtained by summing the total number of break points of all contour lines and the number of loops whose closed loop area shrinkage exceeds a threshold.

Images