Network monitoring and early warning methods for geological disasters

Abstract

The present invention discloses a method for networked monitoring and early warning of geological disasters, which relates to the technical field of geological disaster monitoring and early warning, and includes the following steps: S1, collecting raw data of multiple types of monitoring equipment in the monitoring area, extracting high-amplitude mutation areas and frequency drift factors based on their time and frequency distribution, constructing a high-frequency disturbance perception matrix, and generating a disturbance feature index map for representing the spatial distribution of non-natural disturbance sources. The present invention realizes active identification and modeling of non-natural interference by constructing a high-frequency disturbance perception matrix and a disturbance index map, combines disturbance propagation analysis with residual difference to enhance precursor signal characteristics, and accurately judges the risk level through trend identification and causal analysis, and finally relies on the response control factor to dynamically optimize the warning model parameters, forming a closed-loop mechanism of interference identification, signal purification, trend extraction, risk judgment and strategy adjustment, significantly improving the warning stability, accuracy and practicality of the system in a high-interference environment.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster monitoring and early warning, and in particular to a method for networked geological disaster monitoring and early warning. Background Art

[0002] Networked monitoring and early warning of geological disasters refers to the use of modern information technology means such as the Internet of Things, big data and intelligent analysis to deploy a variety of sensing equipment (such as GNSS, inclinometers, inclinometers, water level gauges, rain gauges, etc.) in areas prone to geological disasters, collect key data such as surface deformation, rainfall, water level in real time, and transmit them to the data center through wired or wireless networks. Combined with geological models and intelligent algorithms, the data is comprehensively analyzed and risk assessed to dynamically identify potential disaster risks such as landslides, mudslides, and ground subsidence. When the monitoring indicators approach or exceed the preset threshold, the system will automatically issue an early warning message, prompting relevant departments and personnel to take timely response measures, thereby achieving early perception, rapid response and effective prevention and control of geological disasters.

[0003] The existing technology has the following deficiencies:

[0004] During the operation of a networked geological disaster monitoring and early warning system, if the monitored area is exposed to prolonged high-interference environments (such as vibration from large-scale infrastructure construction, impact from mining blasts, and frequent heavy train traffic), monitoring equipment is susceptible to continuously receiving a large number of high-frequency disturbance signals. These signals often exhibit complex characteristics such as nonlinear evolution, aperiodic triggering, and spatially diffuse transmission, significantly disrupting the spatiotemporal integrity of the original data stream. In this context, weak precursor signals generated in the early stages of geological disaster evolution (such as slow propagation of deep cracks, shear slip in rock and soil, and abnormal release of groundwater pressure) are often masked by these strong interference signals in terms of energy intensity and frequency. This makes it difficult for traditional early warning models to effectively separate disaster signs from external disturbances, resulting in the system misinterpreting disaster precursors as invalid noise or short-term anomalies, and failing to trigger the early warning response mechanism. If the actual disaster accelerates and the early warning system fails to respond in a timely manner, it is very likely that sudden geological disasters such as landslides, collapses, and debris flows will occur without warning, causing serious casualties and property losses, and seriously undermining the credibility and practicality of the early warning system.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present invention is to provide a networked monitoring and early warning method for geological disasters. By constructing a high-frequency disturbance perception matrix and a disturbance index map, the active identification and modeling of non-natural interference are realized, the disturbance propagation analysis and residual difference are combined to enhance the precursor signal characteristics, and the risk level is accurately judged through trend identification and causal analysis. Finally, the early warning model parameters are dynamically optimized based on the response control factor, forming a closed-loop mechanism of interference identification, signal purification, trend extraction, risk judgment and strategy adjustment, which significantly improves the warning stability, accuracy and practicality of the system in a high-interference environment, so as to solve the problems in the above-mentioned background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a method for networked monitoring and early warning of geological disasters, comprising the following steps:

[0008] S1 collects raw data from multiple types of monitoring equipment within the monitoring area, extracts high-amplitude mutation areas and frequency drift factors based on their time and frequency distribution, constructs a high-frequency disturbance perception matrix, and generates a disturbance feature index map to represent the spatial distribution of unnatural disturbance sources;

[0009] S2, based on the disturbance feature index map and the spatial location of the monitoring equipment, analyzes the propagation trajectory of the disturbance signal, derives its spatial propagation boundary, constructs the dynamic disturbance impact area, and generates a disturbance coverage map;

[0010] S3, based on the disturbance coverage map, establishes a disturbance propagation model, differentiates its output from the original monitoring signal, and constructs a signal residual map to enhance the weak abnormal signals of disaster precursors;

[0011] S4, based on the signal residual map, identifies asymmetric abnormal trends in time series data, extracts continuous trend drift areas, local fluctuation heterogeneity areas, and time series asynchronous response points, and constructs the initial trend contour line of potential disaster evolution;

[0012] S5, establish a causal consistency analysis mechanism to match the initial trend contour line with the historical disaster evolution path in time and space, determine whether it conforms to the reversible deformation or irreversible disaster evolution pattern, and generate high-confidence risk identification results;

[0013] S6, based on the confidence risk identification results, constructs the early warning response control factor, and dynamically adjusts the response threshold and parameter sensitivity of the early warning model in combination with the monitoring status to achieve stable identification of disaster precursor signals and robust early warning output.

[0014] Preferably, step S1 includes:

[0015] Deploy multiple types of monitoring equipment within the monitoring area, collect raw monitoring data, perform time alignment and frequency normalization processing, and construct the time-frequency response matrix of each monitoring node;

[0016] Based on the time-frequency response matrix, the local maximum amplitude point, frequency change rate and energy concentration interval are extracted to form a feature vector set. The high-disturbance feature area is extracted through the clustering algorithm to construct a high-frequency disturbance perception matrix.

[0017] A mapping relationship is established between the high-frequency disturbance perception matrix and the spatial coordinates of the monitoring equipment, and a disturbance feature index map is generated using interpolation modeling and weighting function to represent the spatial distribution of unnatural disturbance sources.

[0018] Preferably, step S2 includes:

[0019] Obtain the disturbance feature index map and the spatial position of the monitoring equipment, and establish a mapping relationship between the disturbance center point and the disturbance weight;

[0020] A spatial diffusion model is constructed based on the disturbance center point to deduce the propagation trajectory and spatial diffusion path of the disturbance signal;

[0021] Gradient edge detection and probability contour extraction algorithms are used to identify disturbance propagation boundaries and construct disturbance impact range layers;

[0022] The disturbance impact range layer is output as a disturbance coverage map, which is used as a spatial restriction template for subsequent abnormal signal screening.

[0023] Preferably, step S3 includes:

[0024] Extract monitoring node data within the disturbance coverage map and construct a disturbance propagation model that integrates physical diffusion modeling and time series prediction;

[0025] The disturbance propagation model is used to generate a disturbance response simulation sequence as the theoretical response of each node under the influence of the disturbance;

[0026] Perform differential operation on the original monitoring signal of each monitoring node and the corresponding disturbance response simulation sequence to obtain the signal residual sequence;

[0027] A signal residual graph is constructed based on the residual sequences of all monitoring nodes to enhance the identifiability and spatial aggregation characteristics of disaster precursor signals.

[0028] Preferably, constructing the disturbance propagation model includes:

[0029] Based on the disturbance intensity, frequency drift pattern and propagation direction information of the disturbance center point in the disturbance coverage map, a physical model including the spatial diffusion behavior of the disturbance is established;

[0030] Build a time series prediction model on the monitoring node to dynamically fit the disturbance response trend;

[0031] The output of the physical model is used as the prior input of the time series model to achieve joint modeling of spatial attenuation and temporal evolution;

[0032] The fusion modeling results generate a disturbance response simulation sequence of each monitoring node in the coverage area, which is used for differential analysis with the original monitoring signal.

[0033] Preferably, step S4 includes:

[0034] Extract the residual signal time series of each monitoring node in the signal residual graph, identify the continuous trend drift area and mark the trend paragraph;

[0035] Perform frequency decomposition and amplitude normalization on the local fluctuations in the residual signal to identify the fluctuation areas with heterogeneous frequency components and obvious amplitude mutations;

[0036] Compare the response phase difference and mutation response time of each monitoring node in the same time period to extract the time series asynchronous response points;

[0037] Combining trend drift areas, fluctuation heterogeneity areas and asynchronous response points, the initial trend contour line of the potential disaster evolution process is constructed.

[0038] Preferably, step S5 includes:

[0039] Extract the structural characteristic parameters of the initial trend contour line and call the standard evolution path template classified by disaster type in the historical geological disaster database;

[0040] The initial trend contour line and the historical evolution path are nonlinearly aligned in the time dimension using the dynamic time warping algorithm, and the response distribution similarity and path direction consistency are calculated in the spatial dimension;

[0041] Establish a causal consistency judgment mechanism to determine whether the current trend is a reversible deformation mode or an irreversible disaster evolution mode based on trend matching, trend stage structure and response characteristics;

[0042] Based on the discrimination results, a high-confidence risk identification result of geological hazards is generated, and the risk level, disaster type, impact range and response recommendation level are marked.

[0043] Preferably, step S6 specifically includes:

[0044] Extract risk level labels, trend type characteristics, matching confidence scores, abnormal response structure types and spatial impact ranges from high-confidence risk identification results to construct a risk status parameter set;

[0045] Based on the risk state parameter set, a set of early warning response control factors consisting of dynamic response threshold adjustment factor, model sensitivity enhancement factor, disturbance immunity weight factor and time series response window correction factor is constructed;

[0046] Combined with the current monitoring status parameters, the applicability of the control factors is judged and ranked through the rule engine or fuzzy control logic to generate a real-time response strategy model;

[0047] The response strategy model is applied to the early warning model in real time, and the parameter configuration and logical judgment path of the early warning model are dynamically adjusted to achieve closed-loop adaptive control of the early warning model.

[0048] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0049] The present invention starts from the signal source, constructs a high-frequency disturbance perception matrix and generates a disturbance feature index map, realizes active identification and spatial modeling of non-natural interference sources, and avoids the disturbance signal from masking the precursor information; through disturbance propagation modeling and residual differential analysis, effectively removes the interference components and strengthens the characteristics of weak abnormal signals; further combines asymmetric trend identification and dynamic causal consistency analysis to accurately depict the evolution path of disasters and judge their risk levels; finally, by constructing a response control factor, the model parameters are dynamically adjusted according to the monitoring status to achieve adaptive optimization of the early warning model. The overall solution forms a closed-loop mechanism from "interference identification" to "signal purification", from "trend extraction" to "risk reasoning", from "result judgment" to "strategy adjustment", so that the early warning system can not only operate stably under the interference background, but also has a high sensitivity and robustness to disaster precursors, significantly improving the accuracy, reliability and engineering adaptability of the early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction to the drawings required for use in the embodiments will be given below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0051] Figure 1 This is a flow chart of the method for networked monitoring and early warning of geological disasters of the present invention. DETAILED DESCRIPTION

[0052] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0053] The present invention provides Figure 1 The geological disaster network monitoring and early warning method shown includes the following steps:

[0054] S1 collects raw data from multiple types of monitoring equipment deployed in the monitoring area. Based on the time and frequency distribution of the raw data, it extracts high-amplitude mutation areas and frequency drift factors, constructs a high-frequency disturbance perception matrix, and generates a disturbance feature index map based on the high-frequency disturbance perception matrix to represent the basic spatial distribution information of unnatural disturbance sources.

[0055] To address the problem of disaster precursor signals being masked when the monitoring area is exposed to strong interference for a long time, this embodiment proposes a high-frequency disturbance feature extraction method to extract the spatial distribution characteristics of unnatural disturbances from the original monitoring data to assist in the identification and regulation of subsequent early warning models. The method specifically includes the following steps:

[0056] Raw monitoring data from various types of monitoring equipment deployed within the target monitoring area (including but not limited to surface displacement sensors, tilt sensors, pore water pressure gauges, strain gauges, rain gauges, and more) is collected through multi-source convergence. Data from all devices is aligned using a standardized sampling frequency and a unified timestamp synchronization mechanism to ensure spatiotemporal consistency across sensor types. During the collection process, a sliding time window strategy is used to segment each data type, extracting key fluctuation characteristics from each data segment for subsequent analysis.

[0057] Based on the time-synchronized, multi-type raw data, time-frequency distribution maps are constructed for each sensor data type. Time-frequency distribution analysis utilizes multidimensional decomposition methods such as wavelet transform, multiscale short-time Fourier transform, or empirical mode decomposition to extract energy density, frequency drift trajectories, and instantaneous amplitude changes from the data. A corresponding time-frequency response matrix is ​​constructed for each sensor node, and peak identification and gradient calculations are performed to determine whether the data contains preliminary signs of strong interference signals.

[0058] To realize the time-frequency distribution analysis based on monitoring data, each type of original monitoring signal is first uniformly preprocessed, including denoising, normalization and time alignment, to ensure data quality and time consistency; for each time series signal, a suitable time-frequency analysis method is selected for decomposition. For example, the wavelet transform can be used to realize local feature analysis of different frequency bands, which is suitable for multi-scale decomposition of non-stationary signals; if it is necessary to obtain short-time frequency components and energy concentration trends at the same time, the multi-scale short-time Fourier transform (MS-STFT) can be introduced to divide the signal into multiple short-time windows on the time axis, and extract the dominant frequency and local power spectral density in each window; if the signal has complex nonlinear perturbation characteristics, the empirical mode decomposition (EMD) can also be used to decompose it into multiple intrinsic mode functions (IMFs), and analyze their frequency drift trajectories and energy evolution processes respectively.

[0059] Through the above method, a high-resolution energy density map and instantaneous frequency curve can be constructed in the time-frequency two-dimensional space, and frequency drift points, high-amplitude energy mutation areas and multimodal interference structures can be further extracted to form a time-frequency response matrix that describes the disturbance characteristics, providing a stable and precise data basis for constructing a high-frequency disturbance perception matrix.

[0060] Based on the time-frequency response matrix of global sensors, statistical rules and clustering algorithms (such as K-means or density clustering) are used to extract areas of high-amplitude mutations and significant frequency drift, forming a high-frequency disturbance perception matrix. This matrix describes the intensity characteristics and frequency evolution trends of possible disturbance signals observed at each monitoring point within a certain time period, and normalizes the spatial distribution in a standardized form, making the disturbance responses of different sensor types comparable.

[0061] To extract regions of high-amplitude mutations and significant frequency drift, the time-frequency response matrix of each monitoring node is first characterized. Key indicators such as the local maximum amplitude point, frequency change rate, and energy concentration interval within a specific time window are extracted to form a set of feature vectors. Statistical rules are then used for preliminary screening to eliminate normal signals with low-amplitude changes and stable frequencies, retaining only samples with significant mutations or frequency drift as candidate disturbance points. Next, using K-means clustering or density-based spatial clustering algorithms (such as DBSCAN), these candidate points are divided into several disturbance type clusters based on feature similarity, thereby identifying a group of high-disturbance signals that are clustered in space and time and exhibit similar characteristics. The centroid of each disturbance cluster serves as a typical disturbance source, and the spatial distribution outline of the disturbance cluster represents the extraction result of the regions of high-amplitude mutation and frequency drift.

[0062] In this way, it is possible to adaptively identify multi-type, irregular disturbance features without relying on a fixed threshold, providing highly reliable input data for constructing a high-frequency disturbance perception matrix.

[0063] Based on the high-frequency disturbance perception matrix, it is integrated with the spatial coordinate system of the monitoring equipment, and a disturbance feature index map is constructed through interpolation modeling and spatial weighted mapping methods. This disturbance feature index map presents the distribution basis of unnatural disturbance sources in the region in a two-dimensional or three-dimensional geographic information system coordinate system, specifically reflecting the disturbance intensity gradient, main disturbance direction, and boundary change trends in each region. This map not only provides an intuitive representation of the spatial distribution of interference sources, but also serves as an important basis for subsequent disturbance propagation modeling and anomaly identification area limitation, enhancing the front-end anti-interference capability and data quality self-diagnosis capabilities in strong interference environments.

[0064] The process of generating a disturbance feature index map based on the fusion of the high-frequency disturbance perception matrix and the monitoring equipment spatial coordinate system is as follows:

[0065] A one-to-one mapping relationship is established between the disturbance perception indicators of each monitoring node (such as frequency drift amplitude, mutation energy value, disturbance duration, etc.) and its corresponding geographic coordinates to form a disturbance dataset with spatial attributes.

[0066] A unified spatial grid model is constructed within the entire monitoring area. Through spatial modeling algorithms such as inverse distance weighted interpolation (IDW) or Kriging interpolation, the disturbance indicators of discrete monitoring points are smoothly extended to areas where no equipment is deployed, thereby achieving continuous expression of disturbance characteristics within the area.

[0067] Combining spatial features such as terrain elevation, monitoring density, and disturbance boundaries, a spatial weighting function is set to enhance the disturbance expression weight in important areas or abnormal clustered areas, making the results more geologically and engineering-related.

[0068] The interpolation results are converted into two-dimensional or three-dimensional disturbance feature index maps and superimposed on the GIS platform to achieve continuous visualization of unnatural disturbance sources in space, providing spatial basis constraints for subsequent disturbance propagation analysis and abnormal signal screening.

[0069] The core role of this step is to build a front-end disturbance perception and spatial modeling mechanism for the geological disaster monitoring system, identify the intensity characteristics and spatial distribution patterns of non-natural disturbance signals from the source, and thus realize the active identification of abnormal signals and data hierarchical management in high-interference environments. Specifically, by collecting the original data of various types of monitoring equipment deployed in the monitoring area, combined with the time distribution and frequency distribution characteristics, it is possible to accurately extract signals with high-amplitude mutations or frequency drift behaviors. Such signals are often caused by non-geological factors, such as engineering vibrations, mechanical shocks or traffic loads. After extracting and classifying these disturbance signals from the original data, a high-frequency disturbance perception matrix is ​​further constructed, which can realize the quantitative expression of the intensity of disturbances suffered by different monitoring points at different time periods. On this basis, the spatial positioning information of the monitoring equipment is integrated, and a disturbance feature index map is generated through spatial interpolation and weighted modeling algorithms to realize the continuous mapping of non-natural disturbance sources in geographic space. This index map not only reveals the concentrated areas and boundary ranges of interference sources, but also provides regional limitation conditions and interference weight basis for subsequent abnormal signal screening, disturbance propagation modeling and disaster precursor identification. It significantly enhances the system's adaptability to complex background signals and the robustness of data processing, and is a basic link in pre-interference identification and data purification in the entire early warning process.

[0070] S2, based on the disturbance feature index map and the spatial location information of the monitoring equipment, analyzes the propagation trajectory of the disturbance signal within the spatial range, derives the spatial propagation boundary of the disturbance signal, constructs the dynamic boundary area of ​​the disturbance impact range, and generates a disturbance coverage map as the data limit range for subsequent abnormal signal screening;

[0071] In order to achieve accurate regional limitation of disturbance signals in a strong interference environment and improve the spatial accuracy of abnormal signal screening, a disturbance influence range modeling method based on the disturbance feature index map and the spatial location information of the monitoring equipment is provided to construct the disturbance propagation trajectory, spatial boundary and its dynamic evolution area. The specific steps include:

[0072] Obtain the constructed disturbance feature index map and extract key fields such as the disturbance intensity weight, disturbance time window information, and disturbance type label. Simultaneously, read the spatial coordinate information of all monitoring devices to form a spatial positioning matrix. Map the disturbance features to the geographic locations of the monitoring devices one by one to obtain the starting point set and high-response node set of the disturbance signal in the spatial domain. By setting a disturbance threshold, select nodes that may serve as disturbance centers or propagation sources, which serve as the starting point set for subsequent propagation path derivation.

[0073] Based on the disturbance center and its weight, spatial diffusion modeling is used to deduce the disturbance propagation trajectory. Weighted Voronoi diagrams, Delaunay triangulations, or path probability modeling methods based on dynamic Bayesian networks are preferred to simulate the propagation of disturbance signals in complex terrain and with unevenly distributed equipment. The propagation model comprehensively considers multiple factors, such as the disturbance energy attenuation rate, terrain slope, and geological structure continuity, dynamically adjusting the propagation direction and speed to more realistically reflect the disturbance's expansion trend.

[0074] The specific steps for implementing the spatial diffusion modeling method based on disturbance center points and their disturbance weights are as follows: the disturbance center points, i.e., the monitoring nodes where the disturbance intensity reaches a significant level, are extracted from the disturbance feature index map, and a disturbance weight is assigned to each center point. This weight takes into account factors such as the disturbance duration, amplitude change rate, and frequency drift intensity.

[0075] Based on these disturbance centers, a two-dimensional or three-dimensional geospatial grid model is constructed, and a spatial diffusion function, such as a Gaussian diffusion kernel or an exponential decay model, is introduced to simulate the propagation of disturbance signals to neighboring areas. In this process, spatial anisotropy weights can be set according to the terrain slope, geotechnical structure, or sensor density to reflect the non-uniform propagation characteristics of disturbances in different directions.

[0076] Use weighted Voronoi diagram or Delaunay triangulation to partition all monitoring nodes into spatial partitions, and calculate the propagation trajectory path and energy attenuation curve within the influence area of ​​each disturbance center;

[0077] By aggregating the diffusion results of each disturbance center, a disturbance propagation map is formed, which includes the diffusion direction, propagation speed, and boundary intensity gradient of the disturbance. This provides a precise spatial evolution basis for subsequent boundary identification and anomaly screening. This method not only has dynamic response capabilities but also simulates the actual propagation behavior of disturbances in complex geological environments, effectively supporting the spatial sensitivity modeling of early warning systems in high-disturbance scenarios.

[0078] Based on the disturbance propagation trajectory map, a disturbance influencing factor field (such as disturbance duration and frequency mutation density) is introduced to construct a disturbance propagation boundary identification mechanism. Specifically, gradient edge detection combined with a probabilistic contour extraction algorithm is used to enhance the edge of the disturbance influence field, forming the spatial propagation boundary of the disturbance signal. The boundary region includes not only areas of significant disturbance intensity but also weak disturbance radiation belts, ensuring redundant coverage of potential disturbance areas. Furthermore, a time sliding window mechanism is used to update the disturbance boundary in real time, enabling modeling of the dynamic evolution of the disturbance range.

[0079] The specific steps for implementing edge enhancement processing on the disturbance influence field and forming the spatial propagation boundary of the disturbance signal are as follows:

[0080] The disturbance perception matrix is ​​spatially interpolated to generate a continuous disturbance influence field, that is, the disturbance intensity around each monitoring node is distributed in the region in numerical form;

[0081] Use gradient edge detection methods (such as Sobel, Laplacian or Canny operators) to calculate the intensity change gradient of each position in the disturbance field, thereby identifying the location of sudden changes in disturbance intensity and preliminarily extracting candidate areas of disturbance boundaries;

[0082] A probability contour extraction algorithm is introduced. Based on the disturbance probability density function in the disturbance intensity distribution map, multiple disturbance confidence intervals (such as 70%, 85%, 95%, etc.) are set, and corresponding contour lines are extracted in the disturbance intensity field. These contour lines reflect the influence range boundaries of the disturbance signal in space.

[0083] By superimposing and analyzing the gradient detection results with the multi-layer contour line results, the credibility and closure of the disturbance boundary are comprehensively judged, isolated pseudo-edges are eliminated and the boundary is smoothed to generate a continuous, closed and credible disturbance propagation boundary layer, providing spatial boundary support for the subsequent construction of the dynamic disturbance coverage map.

[0084] This method combines mutation detection and probabilistic modeling, can accurately identify the diffusion boundary of disturbances and has real-time update capabilities. It is particularly suitable for geological environments where multiple sources of disturbances coexist and the propagation paths are complex.

[0085] The disturbance propagation boundary area is aligned with the original geographic coordinate system to construct a dynamic boundary layer of the disturbance's impact range, which is then output as a disturbance coverage map. This map serves as a spatially restricted template for abnormal signal screening and provides a basis for interference shielding in subsequent signal residual analysis and trend identification. Target signal feature enhancement and residual analysis are performed only within the disturbance coverage area, effectively avoiding the false interference effect of high-frequency interference in non-correlated areas, significantly improving the accuracy and robustness of the system's recognition of disaster precursor signals.

[0086] The core function of this step is to model the propagation behavior of unnatural disturbance signals within a spatial range, identify their boundaries, and dynamically constrain them. This provides a precise spatial constraint framework for the geological disaster monitoring system, improving the accuracy and reliability of subsequent abnormal signal screening and precursor identification. In a strong interference environment, disturbance signals often do not occur in isolation, but instead exhibit distinct spatial diffusion characteristics and dynamic evolution paths. For example, shock waves caused by construction blasting, heavy machinery vibration, or train operation exhibit a certain regular propagation trajectory and attenuation range in the geological medium. By fusing the disturbance feature index map with the spatial location information of the monitoring equipment, the starting position, high-response nodes, and propagation direction of the disturbance signal within the monitoring area can be identified. The spatial diffusion model is then used to deduce its propagation trajectory, and the effective influence boundary of the disturbance signal is derived by combining energy attenuation laws and topographic geological factors. The disturbance influence range constructed based on the above process can dynamically reflect the changing trend of the disturbance coverage area in different time periods, and ultimately generate a disturbance coverage map, which serves as a spatial constraint template for eliminating abnormal noise and focusing on reliable precursors in the subsequent signal processing process. This not only effectively shields misleading information from interference sources, but also avoids global misjudgment and waste of computing resources, significantly enhances the sensitivity and anti-interference ability to disaster precursor signals, and is a key intermediate link in achieving robust early warning.

[0087] S3, based on the disturbance coverage map, establish a disturbance propagation model, perform differential calculations on the output of the disturbance propagation model and the original monitoring signal in the disturbance coverage map, and construct a signal residual map to enhance the characteristics of weak abnormal signals corresponding to disaster precursors;

[0088] To achieve highly sensitive identification of geological disaster precursor signals under strong interference background, a weak signal enhancement method based on disturbance propagation modeling and residual difference analysis is proposed. It is used to remove non-natural interference signals from the original monitoring signals within the disturbance coverage map and extract potential disaster precursor features. The specific steps include:

[0089] Based on the generated disturbance coverage map, data from monitoring nodes within the disturbance space boundary is extracted, and regions with significant disturbance characteristic responses are selected as modeling input regions. A disturbance propagation model is constructed based on the disturbance intensity weights, frequency drift patterns, and propagation direction information contained in the disturbance coverage map. This model can be developed by combining physical diffusion equations with data-driven mechanisms. The physical diffusion component simulates the attenuation trend of disturbance energy in space, while the data-driven component characterizes the local dynamic behavior of the disturbance signal through time series modeling (such as ARIMA models and LSTM neural networks), thereby achieving joint modeling and prediction of the disturbance signal in both spatial and temporal dimensions.

[0090] A "significant disturbance characteristic response" refers to the presence of distinct dynamic response characteristics to disturbance events at certain monitoring nodes within the spatial range defined by the disturbance coverage map within a specific time period. The significance of these characteristics can be determined through a comprehensive evaluation of multiple quantitative indicators. Specifically, this significance includes the following three aspects: First, a high intensity of signal amplitude mutation, meaning that the monitoring signal experiences a large change in a short period of time, with its first-order difference or local gradient exceeding a set threshold; Second, drift behavior in the frequency component, meaning that time-frequency analysis reveals a significant shift in the main frequency or a shift in energy distribution across frequency bands, indicating that the system is sensitive to disturbances; Third, a long duration of response, meaning that the disturbance signal at this node is not instantaneous but rather has a significant time span, possibly accompanied by a decaying tail or periodic echo. Based on these three characteristics, a "disturbance response score" can be assigned to each monitoring node using a weighted scoring model or anomaly integration method. Nodes with the highest scores are identified as regions with significant disturbance characteristic responses. These regions are prioritized as input points for disturbance propagation modeling because they contain more effective disturbance information and more accurately represent the characteristics of the disturbance source and its propagation trend.

[0091] To implement a disturbance propagation model that combines the physical diffusion equation with a data-driven mechanism, it can be constructed as follows:

[0092] Establish a basic mathematical model of disturbance propagation at the physical level. Usually, a two-dimensional or three-dimensional diffusion partial differential equation (such as Fick's diffusion law or heat conduction equation) is used to simulate the spatial attenuation process of disturbance energy in geological media. Parameters such as the disturbance source intensity, medium diffusion coefficient, and terrain gradient are introduced into the equation to achieve a preliminary characterization of the propagation rate and range of disturbance in different directions and regions.

[0093] A time series model is established at each monitoring node to fit the temporal variation trend of the disturbance response. The ARIMA model can be used to capture linear trends and periodic disturbances, or the LSTM (long short-term memory) neural network can be introduced to identify nonlinear and long-term dependent disturbance behaviors.

[0094] Using the disturbance estimates output by the physical model as prior inputs or structural constraints of the data-driven model makes time series predictions more consistent with physical laws.

[0095] By integrating the prediction results, a joint disturbance propagation model with spatial attenuation characteristics and temporal prediction capabilities is formed. This model can be used to simulate the response changes of any monitoring node under the influence of disturbances, achieving comprehensive modeling and dynamic prediction of disturbance behavior. This model combines highly explanatory physical mechanisms with highly predictive learning algorithms, resulting in enhanced robustness and adaptability, making it particularly suitable for complex geological settings and irregular disturbance scenarios.

[0096] A disturbance propagation model is used to generate a sequence of simulated disturbance values ​​for each monitoring node within the coverage area. This represents the theoretical response behavior each node should exhibit under the influence of a disturbance. This simulated value not only considers the distance between the disturbance source and the monitoring point but also incorporates weighting factors such as local geological conditions, equipment type, and sampling frequency for normalization. This ensures that the simulated data has consistent temporal resolution and physical units with the actual monitoring data, providing a precise foundation for subsequent differential analysis.

[0097] For each monitoring node within the disturbance coverage area, a point-by-point difference operation is performed between the original signal sequence collected in real time and the simulated disturbance response sequence generated by the disturbance propagation model, resulting in the corresponding signal residual sequence. This residual reflects the portion of the actual observation data that is not explained by the disturbance model, namely, the weak non-disturbance response that may contain disaster precursor information. To further enhance the identifiability of the precursor signal, the residual sequence can be smoothed, denoised, and outlier enhancement processed. For example, a sliding mean filter and Z-score detection method can be used to remove background noise while retaining significant mutation structures.

[0098] The point-by-point difference operation between the original signal sequence of the monitoring node in the disturbance coverage area and the simulated disturbance response sequence is performed as follows:

[0099] For each monitoring node within the disturbance coverage map, extract the original monitoring signal sequence collected within a specific time period. Ensure that the sequence has been pre-processed, including time synchronization, denoising, and normalization, to ensure data quality and comparison consistency.

[0100] Obtain the simulated disturbance response sequence generated by the disturbance propagation model. This sequence reflects the theoretical response trajectory that the node should present under the influence of the disturbance factor, including dynamic characteristics such as disturbance trend, amplitude change, and response rhythm;

[0101] The original signal sequence and the simulated response sequence are compared and analyzed point by point, and the difference between them is calculated item by item to form a residual sequence. This residual sequence represents the abnormal part of the original monitoring data that cannot be explained by the disturbance model and is very likely to contain weak geological disaster precursor signals.

[0102] Further processing of the residual sequence, including removing background noise, highlighting mutation structures, and marking unusual inflection points, enhances the significance and stability of abnormal signals, making potential disaster signs easier to identify and track in subsequent analysis. This differential analysis process effectively strips away systematic disturbances from a strong interference background, purifying potential abnormal evolution signals and providing a solid data foundation for the high-precision response of subsequent early warning models.

[0103] The residual results of all monitoring nodes are spatially reconstructed to form a complete signal residual map. Based on spatial coordinates, this residual map maps the residual signal strength of each node in the time dimension into a regional distribution map, which can intuitively display the spatial aggregation trends, abnormal movement boundaries, and evolutionary contours of potential disaster precursor signals. The signal residual map not only provides structured input for subsequent asymmetric anomaly trend identification but also provides a basis for adjusting the credibility weight of the early warning model. It significantly improves the ability to distinguish precursor signals and the accuracy of early warning responses in the presence of strong interference.

[0104] This step aims to accurately identify and enhance weak precursor signals during the evolution of geological hazards in complex, high-interference environments, significantly improving the early warning system's sensitivity and ability to identify true anomalies. In geological hazard monitoring, strong interference signals (such as blasting vibrations, construction impacts, and train traffic) often flood the monitoring area with high energy and short periods, drowning out small, slowly varying precursor signals within the overall data. Traditional models struggle to distinguish disturbance responses from true precursors, leading to underreporting or misjudgment. To address this issue, this step first constructs a disturbance propagation model to simulate the response trends of the disturbance signal at each monitoring node within the disturbance coverage area, generating a set of theoretical expected data. This simulated response is then compared point by point with the original monitoring signal to extract the residual signal between them. This residual signal represents the portion of the original data not explained by the disturbance model and is likely to contain low-energy, slowly varying, and non-periodic characteristics of geological hazard precursors. This differential mechanism effectively removes the influence of interference and allows weak anomaly signals to emerge from the background data. The signal residual map constructed further not only preserves spatial distribution information but also reflects the intensity and morphological changes of abnormal signals at each node. This serves as an important input for subsequent trend extraction, contour identification, and causal determination, directly determining the accuracy and response speed of the early warning model. Therefore, this step plays a key role in the system's "data purification, precursor purification, and enhanced identification," and is one of the core technical links for achieving robust early warning.

[0105] S4, performs asymmetric abnormal trend identification analysis on the time series data in the signal residual graph, extracts the areas with continuous trend drift, local fluctuation heterogeneity and time series asynchronous response points, and constructs the initial trend contour line of the potential disaster evolution process based on the areas with continuous trend drift, local fluctuation heterogeneity and time series asynchronous response points;

[0106] In order to further analyze the residual signal in a high-interference environment and deeply explore the evolution trend of geological disaster precursors, an asymmetric abnormal trend recognition method based on the signal residual map is proposed. The dynamic characteristic area of ​​the potential evolution path of the disaster is extracted from the residual time series, and the initial trend contour line of the potential disaster process is constructed based on it. The method includes the following steps:

[0107] The residual signal time series for each monitoring node is extracted from the signal residual graph and analyzed for continuous trend changes. By setting a sliding time window and combining multiscale regression fitting with trend offset calculation, asymmetric trend segments, where the residual values ​​continuously increase monotonically, decrease slowly, or fluctuate, are identified. These trend segments often reflect the evolutionary paths of physical behaviors that are precursors to disasters, such as stress accumulation in the geological body, propagation of microfractures, or release of pore water pressure, and are highly indicative.

[0108] To identify asymmetric trend segments in residual signals, a sliding time window must first be set for the residual time series of each monitoring node. Within each window, a multi-scale regression method (such as linear, quadratic, or locally weighted regression) is used to fit the sequence to obtain the changing trend of the signal at different time scales. Subsequently, the trend offset within each time window is calculated, that is, the mean difference or slope difference between the fitting curve of the current window and the fitting curve of the previous window, to quantify the change in trend direction and intensity. When the trend offset of multiple consecutive windows shows a unilateral change, showing an asymmetric structure such as continuous rise, continuous decline, or slow oscillation rise, it can be marked as a potential trend segment. At the same time, these segments are screened for intensity based on the duration of the trend and the amplitude of the change. Only trend segments with significant characteristics in amplitude, time, or stability are retained to construct the initial outline of the disaster evolution. This process not only enhances the ability to identify weak trends in the signal, but also suppresses the false trend misjudgment caused by random fluctuations, significantly improving the accuracy and spatial continuity of precursor trend extraction in strong interference environments.

[0109] For each residual time series, local fluctuations outside the trend segment are discretized and analyzed, focusing on extracting those fluctuation segments whose frequency structure is clearly inconsistent with the previous disturbance pattern. Using frequency decomposition and amplitude normalization methods, we identify regions of fluctuation with heterogeneous characteristics: signal segments exhibiting the coexistence of different frequency components, frequent amplitude fluctuations, and a lack of periodicity. These regions often indicate uneven stress release or micro-regional deformation behavior in geotechnical structures and provide important supplementary information for constructing hazard profiles.

[0110] To identify the fluctuation regions with heterogeneous characteristics in the signal residual sequence, the following steps can be followed: perform frequency decomposition on the residual signal sequence, preferably using a multi-component decomposition method such as wavelet transform or empirical mode decomposition, to decompose the original non-stationary signal into several sub-signal components with different frequency characteristics, each component representing a fluctuation behavior at a frequency scale;

[0111] Perform amplitude normalization on each frequency component sequence, that is, convert its amplitude into a unified dimension through maximum and minimum value scaling or Z-score normalization, so that the amplitude characteristics between different frequency bands are comparable;

[0112] Sudden changes in the local amplitude of each frequency component are detected. The amplitude jump area can be identified by the standard deviation, coefficient of variation or volatility changes within the sliding window, and its mutation location and duration are recorded.

[0113] Comprehensively analyzing the mutation characteristics of all frequency components, if multiple frequency components experience mutations simultaneously within the same time period, with significant differences in the magnitude and duration of the mutations across different components and inconsistent rhythms of change, this time period can be identified as a fluctuation region with heterogeneous characteristics. Such regions typically exhibit multimodal superposition of signal perturbation behavior, reflecting potentially complex geological processes or local anomalies, and warrant special attention in subsequent hazard trend profile modeling.

[0114] Time series comparisons of residual signals between different monitoring nodes are performed to identify asynchronous points with different responses. By analyzing characteristics such as phase differences, amplitude reversal times, and sudden response lags in the signal responses of multiple nodes over the same time period, points of asynchronous response are identified. These points often occur at the intersection of asymmetric propagation paths of disaster expansion or at the interface of local geological tectonic changes, and serve as key anchor points in trend profile modeling.

[0115] Based on the continuous trend drift regions, local fluctuation heterogeneity regions, and time-series asynchronous response points extracted above, an initial trend contour line of the potential disaster evolution process is constructed in accordance with the chronological order and spatial layout relationship. This contour line, with the dynamic evolution structure as the skeleton, depicts the possible starting location, expansion direction, and impact range of the precursor signal, providing structural support for subsequent causal path identification and early warning response. The overall method combines multi-dimensional technical paths such as asymmetric trend mining, frequency structure identification, and cross-node time series analysis. It has significant creativity and non-obviousness, and is particularly suitable for monitoring scenarios in complex backgrounds with weak disaster precursors and obvious nonlinear characteristics of spatiotemporal evolution.

[0116] This step aims to accurately extract key spatiotemporal features of potential geological hazard precursor signals through in-depth analysis of the time series data in the signal residual map, thereby constructing an initial structure of the hazard's evolutionary trend and laying the foundation for subsequent causal identification and early warning response. In complex interference environments, early signs of geological hazards often manifest as small, nonlinear, and irregular anomalies, making them difficult to identify using traditional thresholding or single-point anomaly detection methods. The signal residual map has already been stripped of major disturbance components in the previous step, leaving a signal closer to the true precursor information. This step further structures these residual signals into evolutionary patterns with physical indicative significance. Specifically, regions of continuous trend drift reflect stable evolutionary signals generated by stress accumulation or slow deformation in the geological body, representing the main path of hazard development. Regions of localized fluctuation heterogeneity reveal multi-frequency, non-uniform response behavior in spatially or structurally inhomogeneous media, potentially corresponding to multi-point instability or crack propagation. Temporally asynchronous response points reflect inconsistent responses from different nodes in the monitoring network, suggesting directional propagation or staged release in hazard evolution. By comprehensively modeling these three key features, the resulting initial trend profile not only depicts the anomaly's starting point, expansion, and acceleration phases in time, but also provides spatial evidence for the propagation path of precursor signals, helping to improve the accuracy of the early warning system's discriminant logic and triggering mechanisms. Therefore, this step is a critical bridge in the transition from "anomaly" to "disaster mode," a crucial analytical link that connects the entire intelligent early warning system and offers strong interpretability, structured functionality, and high credibility.

[0117] S5. Establish a dynamic causal consistency analysis mechanism to match the initial trend contour line with the historical geological disaster evolution path for spatiotemporal characteristics, determine whether the current trend conforms to the reversible deformation characteristic pattern or the irreversible disaster evolution triggering pattern, and generate a high-confidence geological disaster risk identification result based on this;

[0118] In order to achieve high-confidence identification of disaster precursor trends and effectively distinguish whether their evolution process belongs to a reversible deformation mode or an irreversible disaster triggering process, a dynamic causal consistency analysis mechanism based on trend contours and historical disaster evolution paths is proposed. Through structural matching of spatiotemporal characteristics, a logical causal relationship discrimination model is established to achieve high-confidence identification of geological hazard risks. The specific implementation of this mechanism includes the following steps:

[0119] Structural characteristic parameters are extracted from the constructed initial trend contours, including the time span of the trend segment, the amplitude of change, the directionality, the distribution density of the response nodes, the strength of the signal consistency, and the phased characteristics of the trend change (such as the three-stage structure of slow change, sudden change, and stability). At the same time, the system uses typical evolution path templates from the historical geological disaster database. Based on a large number of actual disaster cases, this template is classified according to the type of disaster (such as landslide, debris flow, and ground subsidence). The template records the temporal evolution process, spatial expansion trajectory, precursor signal change characteristics, topographic conditions, and external excitation factors in a standardized form, forming a multi-dimensional historical evolution model library.

[0120] A temporal and spatial feature matching comparison is performed between the current initial trend profile and the historical path template. In the temporal dimension, a dynamic time warping (DTW) algorithm is used to nonlinearly align the trend change rhythm, ensuring that precursor processes can still be compared equivalently despite differences in time scales. In the spatial dimension, node response distribution similarity measurement, path direction consistency analysis, and slope deformation synergy calculation are used to assess the similarity between the spatial propagation characteristics of the current trend path and the historical disaster path. In addition, a multi-feature fusion weight model is introduced to normalize multiple matching indicators such as time, space, frequency structure, and fluctuation pattern, and then fuse them into a score to obtain a matching score between the trend profile and each historical path.

[0121] The key to achieving nonlinear alignment of trend contours with historical disaster evolution paths in the temporal dimension is to use the Dynamic Time Warping (DTW) algorithm to address the difficulty in comparing trends due to the tempo differences between them. The specific process is as follows:

[0122] The time series of the current initial trend contour line and the time series in the historical evolution path are extracted into a set of multidimensional data points arranged in chronological order. Each data point can contain characteristic dimensions such as residual signal strength, change slope, and response node density.

[0123] The DTW algorithm is used to gradually calculate the shortest "alignment path" between two time series at different time points. This path allows one series to be "stretched" or "compressed" in certain time periods to better fit the evolution rhythm of the other series, thereby eliminating the superficial differences caused by different time scales.

[0124] By calculating the difference between the aligned sequences, such as cumulative distance, local similarity and structural consistency, the dynamic similarity between the trend profile and the historical path is quantified;

[0125] Ultimately, the alignment results are used as one of the inputs for causal consistency determination. This process effectively addresses the problem of comparison distortion caused by the varying speeds of disaster evolution and signal change rates, enabling equivalent comparison of trend patterns across different time scales and improving the accuracy and robustness of the overall determination.

[0126] Based on the matching analysis results, a causal consistency discrimination mechanism is established. This mechanism takes as input the matching threshold between the trend contour and the historical path, the structural correspondence, and the trend evolution stage marker, and performs logical reasoning and judgment: If the current trend change exhibits characteristics that are highly consistent with known reversible deformation patterns, such as rapid recovery after a short-term anomaly, limited change amplitude, localized response range, and the absence of synchronous external triggering factors, the trend is judged to be reversible. Conversely, if the trend evolution process exhibits highly irreversible characteristics, such as continued expansion of the trend amplitude, rapid expansion of the spatial response range, high synchronization of node response times, a high degree of consistency between the signal anomaly stage and historical disasters, and the presence of a typical acceleration growth stage, then it is judged to be an irreversible disaster evolution pattern. To improve the robustness of the discrimination, Bayesian reasoning or fuzzy logic frameworks can be introduced to perform fuzzy classification processing on boundary trends to avoid the risk of misjudgment caused by binary decision-making.

[0127] Introducing Bayesian reasoning or fuzzy logic frameworks aims to improve the robustness and error tolerance of trend identification, particularly in complex situations where boundary trend features are unclear, data volatility is high, or trend matching is critical. This effectively reduces the risk of misclassification. Traditional binary decision-making mechanisms often directly classify trends as "reversible" or "irreversible." This rigid classification is susceptible to noise, missing data, or local fluctuations in practice, leading to misclassification of borderline cases. By introducing Bayesian reasoning, the posterior probability of trend attribution can be dynamically updated based on prior probabilities and observed data, quantifying the uncertainty of trend classification as a probability distribution. Fuzzy logic also allows the system to set multiple fuzzy membership functions, progressively assessing the reversibility, uncertainty, and trigger probability of a trend. This allows regions with fuzzy boundaries to be identified as intermediate states, such as "partially reversible" or "high potential risk." This flexible classification mechanism maintains warning sensitivity while avoiding false negatives and underreporting caused by extreme judgments, enabling a more robust, intelligent, and adaptive risk identification strategy.

[0128] Based on the results of causal consistency analysis, a high-confidence risk identification result for geological hazards is generated. This result not only includes the current trend's risk level (e.g., high, medium, or low), but also identifies the possible hazard type, potential impact range, temporal evolution stage, and recommended response level. This result is then fed into the early warning and response module as input, triggering subsequent threshold adjustment, coordinated response, and information release processes. This step allows for the identification of potential hazard pathways at an early stage of a trend, allowing for early detection of whether hazard evolution signals are physically evolvable. This completes a closed-loop logical chain from "signal perception" to "trend inference" to "risk assessment." This analysis mechanism combines structural modeling, path identification, and logical judgment, integrating a physical hazard evolution model with a data-driven algorithm. It is highly creative and non-obvious, making it particularly suitable for geological monitoring applications in multi-source, heterogeneous data environments, where identifying hazard precursors is difficult, trend evolution pathways are complex, and extremely low false alarm rates are required.

[0129] This step aims to establish a trend identification mechanism with interpretable and logical reasoning capabilities. This allows the geological hazard early warning system to proactively identify risk levels and determine whether the evolution of current monitored trends, before a disaster manifests itself, is converging with historical disaster evolution pathways. This dynamic causal consistency analysis mechanism not only identifies surface anomalies in monitoring data but also deduces the underlying physical logic of the trend. By structurally matching temporal evolutionary features (such as trend onset, rate of change, and amplitude gradient) and spatial extension features (such as response node distribution, propagation direction, and local response synchronization) extracted from the current trend profile with typical historical disaster processes, it effectively determines whether the trend is highly consistent with known disaster processes, thereby identifying a high-risk state for disaster evolution. In particular, when trends are in a "fuzzy evolution" phase, potentially representing both normal fluctuations and disaster precursors, this mechanism uses Bayesian reasoning or fuzzy logic to flexibly determine the reversibility or irreversibility of the trend, effectively avoiding both false positives and false negatives. The resulting high-confidence risk identification results not only identify the evolving state of the trend but also deduce its likely direction, impact scope, and triggering conditions, providing a reliable basis for subsequent warning threshold adjustment, response level setting, and emergency response strategies. Therefore, this step is the core link in elevating "data anomalies" to "disaster patterns," serving as a key technological bridge from intelligent identification to decision-making support. It plays a dual role in ensuring both accuracy and response efficiency within the overall monitoring and early warning system.

[0130] S6, based on the high-confidence risk identification results, constructs early warning response control factors, and dynamically adjusts the response threshold and parameter sensitivity weights of the early warning model in combination with the current monitoring status to achieve stable identification of disaster precursor signals and robust early warning output in high-interference scenarios;

[0131] To achieve robust identification and high-confidence response to disaster precursor signals in strong interference environments, a mechanism is proposed to construct a warning response control factor based on high-confidence risk identification results. This mechanism introduces dynamic control parameters to adaptively adjust the response threshold and sensitivity weights in the warning model, ensuring that the model maintains high recognition accuracy and response stability under different risk levels and disturbance backgrounds. The specific implementation of this mechanism includes the following steps:

[0132] Based on the high-confidence risk identification results generated in the previous step, key indicators such as risk level labels, trend type characteristics (such as reversible deformation or irreversible disaster evolution), matching confidence scores, abnormal response structure types (such as sudden, gradual, and mixed), and spatial impact range are extracted to construct a risk state parameter set. This set not only reflects the risk level of the current trend evolution but also characterizes the complexity of the disturbance and the uncertainty of the response. It serves as the core input for the subsequent construction of regulatory factors.

[0133] Based on the risk state parameter set, a multi-dimensional early warning response control factor set is constructed. The control factors include but are not limited to: dynamic response threshold adjustment factor, model sensitivity enhancement factor, disturbance immunity weight factor and time series response window correction factor. Among them, the dynamic response threshold adjustment factor is used to adjust the trigger threshold of the early warning model up and down, and elastically adjust it according to the risk level of the current trend; the model sensitivity enhancement factor strengthens the model's response ability to weak abnormal signals by increasing the weight of specific features (such as trend slope, node synchronization, etc.); the disturbance immunity weight factor is used to reduce the probability of the model over-responding to known interference types and improve the ability to suppress false alarms; the time series response window correction factor is used to adjust the length and overlap of the sliding analysis window in the early warning model to adapt to trend changes at different development rates.

[0134] Combined with the current monitoring status parameters, such as the activation status of the monitoring network nodes, the health status of the equipment, the data integrity level, the current disturbance environment interference intensity level, etc., the control factors and the monitoring status are dynamically integrated to form a real-time response strategy model. The model judges and ranks the applicability of the control factors through a rule engine or fuzzy control logic to ensure that the applied control operations are reasonable and have execution priority. For example, when a high-risk irreversible trend is identified and the current network is in a fully activated state, the response threshold will be automatically lowered and the sensitivity weight will be increased to prioritize the sensitivity of the early warning trigger; if the risk level is low and the monitoring data missing rate is high, the response threshold will be appropriately increased and the alarm cycle will be delayed to avoid false alarms.

[0135] To determine and rank the applicability of control factors using a rule engine or fuzzy control logic, a multidimensional input indicator system must be established, encompassing risk level, disturbance intensity, monitoring status integrity, and historical evolution patterns. Applicability conditions, trigger ranges, and priority rules must be set for each control factor. The rule engine is implemented based on a set of "condition-action" rules. For example, when the identified result is a high-confidence irreversible trend and the disturbance intensity is medium or above, the "response threshold reduction" action is executed with a high priority. Conversely, when the risk level is low and the monitoring node loss rate is high, the "delayed alarm window adjustment" action is executed with a medium priority. Fuzzy control logic is also introduced to convert these input indicators into fuzzy linguistic variables (such as "high risk," "medium disturbance," and "poor data integrity"). The fuzzy inference mechanism calculates the fitness and output activation of each control factor. The control factors are then weighted and ranked based on the scores, prioritizing the strategy combination with the highest fitness. This approach effectively resolves competition and conflict between control factors, enabling dynamic decision optimization and ensuring stable, reasonable, and prioritized response adjustments in complex and uncertain environments.

[0136] The resulting response and control strategies are then applied in real time to the early warning model's operational mechanisms, dynamically adjusting the model's parameter configuration and logical judgment paths. This process utilizes a closed-loop adaptive control approach, whereby the model locally optimizes its structure and response strategies after feedback from identification results to maintain sustained sensitivity to precursor signals and overall stability under different risk scenarios. Furthermore, this mechanism allows for the solidification of some strategy parameters in the form of rule templates, supporting manual intervention and emergency adjustments to enhance controllability and flexibility in the face of major disasters.

[0137] The core function of this step is to integrate the high-confidence risk identification results with the current operating status of the monitoring system, dynamically construct and adjust the response strategy of the early warning model, so that the system can still stably identify weak disaster precursor signals and make robust and accurate early warning decisions when facing complex and high-interference backgrounds. In the actual geological disaster monitoring process, due to frequent environmental disturbances, volatile signal characteristics, and different sensor states, fixed thresholds and static parameter configurations are often difficult to adapt to changing situations, which can easily lead to false alarms, missed alarms, or slow system responses. This step constructs a set of early warning response control factor systems. Based on the output results of the previous stage of high-confidence risk assessment, such as trend type, matching score, abnormal structural characteristics, and risk level, the response threshold and sensitivity weight of the model are dynamically set, and the real-time operating parameters such as node activity, data integrity, equipment status, and background interference level of the current monitoring network are comprehensively considered, thereby realizing adaptive adjustment of model parameters and strategy reconstruction. Specifically, when the identification results show a highly irreversible trend, the risk level is high, and the monitoring system is in good condition, the model will automatically lower the trigger threshold and enhance the sensitivity factor to respond in advance; conversely, in the case of low risk or incomplete monitoring data, the system will appropriately raise the response threshold to avoid misjudgment. The mechanism also introduces a rule engine and fuzzy control logic to judge the adaptability and prioritize the control factors, ensuring that the executed control operations are both practical and have dynamic optimization capabilities. In summary, this step not only improves the flexibility and adaptability of the early warning system in actual operation, but also enhances the robustness of the recognition of precursor signals through parameter cascade adjustment. It is an important guarantee link for achieving the triple goals of accurate early warning, efficient response, and error control.

[0138] The above-mentioned networked geological disaster monitoring and early warning method significantly improves the ability to identify geological disaster precursor signals and the response stability of the early warning system in high-interference environments, demonstrating significant practical benefits and engineering value. Starting from the signal source, the method constructs a high-frequency disturbance perception matrix and generates a disturbance feature index map. This enables active identification and spatial modeling of unnatural interference sources, preventing disturbance signals from masking precursor information. Disturbance propagation modeling and residual differential analysis effectively remove interference components and enhance the characteristics of weak abnormal signals. Furthermore, asymmetric trend identification and dynamic causal consistency analysis are combined to accurately characterize the hazard evolution path and determine its risk level. Finally, a response control factor is constructed to dynamically adjust model parameters based on monitoring status, achieving adaptive optimization of the early warning model. The overall solution forms a closed-loop mechanism from "interference identification" to "signal purification," from "trend extraction" to "risk inference," and from "outcome determination" to "strategy adjustment." This enables the early warning system to not only operate stably in interference environments but also possesses high sensitivity and robustness to disaster precursors, significantly improving the accuracy, reliability, and engineering adaptability of early warnings.

[0139] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0140] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

[0141] It should be noted that, in this document, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0142] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0143] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0144] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0145] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0146] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0147] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0148] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. A method for networked monitoring and early warning of geological disasters, characterized in that: The following steps are involved: S1 collects raw data from multiple types of monitoring equipment within the monitoring area, extracts high-amplitude mutation areas and frequency drift factors based on their time and frequency distribution, constructs a high-frequency disturbance perception matrix, and generates a disturbance feature index map to represent the spatial distribution of unnatural disturbance sources; S2, based on the disturbance feature index map and the spatial location of the monitoring equipment, analyzes the propagation trajectory of the disturbance signal, derives its spatial propagation boundary, constructs the dynamic disturbance impact area, and generates a disturbance coverage map; S3, based on the disturbance coverage map, establishes a disturbance propagation model, differentiates its output from the original monitoring signal, and constructs a signal residual map to enhance the weak abnormal signals of disaster precursors; S4, based on the signal residual map, identifies asymmetric abnormal trends in time series data, extracts continuous trend drift areas, local fluctuation heterogeneity areas, and time series asynchronous response points, and constructs the initial trend contour line of potential disaster evolution; S5, establish a causal consistency analysis mechanism to match the initial trend contour line with the historical disaster evolution path in time and space, determine whether it conforms to the reversible deformation or irreversible disaster evolution pattern, and generate high-confidence risk identification results; S6, based on the confidence risk identification results, constructs the early warning response control factor, and dynamically adjusts the response threshold and parameter sensitivity of the early warning model in combination with the monitoring status to achieve stable identification of disaster precursor signals and robust early warning output.

2. The method for networked monitoring and early warning of geological disasters according to claim 1, characterized in that: Step S1 includes: Deploy multiple types of monitoring equipment within the monitoring area, collect raw monitoring data, perform time alignment and frequency normalization processing, and construct the time-frequency response matrix of each monitoring node; Based on the time-frequency response matrix, the local maximum amplitude point, frequency change rate and energy concentration interval are extracted to form a feature vector set. The high-disturbance feature area is extracted through the clustering algorithm to construct a high-frequency disturbance perception matrix. A mapping relationship is established between the high-frequency disturbance perception matrix and the spatial coordinates of the monitoring equipment, and a disturbance feature index map is generated using interpolation modeling and weighting function to represent the spatial distribution of unnatural disturbance sources.

3. The method for networked monitoring and early warning of geological disasters according to claim 1, characterized in that: Step S2 includes: Obtain the disturbance feature index map and the spatial position of the monitoring equipment, and establish a mapping relationship between the disturbance center point and the disturbance weight; A spatial diffusion model is constructed based on the disturbance center point to deduce the propagation trajectory and spatial diffusion path of the disturbance signal; Gradient edge detection and probability contour extraction algorithms are used to identify disturbance propagation boundaries and construct disturbance impact range layers; The disturbance impact range layer is output as a disturbance coverage map, which is used as a spatial restriction template for subsequent abnormal signal screening.

4. The method for networked monitoring and early warning of geological disasters according to claim 1, characterized in that: Step S3 includes: Extract monitoring node data within the disturbance coverage map and construct a disturbance propagation model that integrates physical diffusion modeling and time series prediction; The disturbance propagation model is used to generate a disturbance response simulation sequence as the theoretical response of each node under the influence of the disturbance; Perform differential operation on the original monitoring signal of each monitoring node and the corresponding disturbance response simulation sequence to obtain the signal residual sequence; A signal residual graph is constructed based on the residual sequences of all monitoring nodes to enhance the identifiability and spatial aggregation characteristics of disaster precursor signals.

5. The method for networked monitoring and early warning of geological disasters according to claim 4, characterized in that: Building a disturbance propagation model includes: Based on the disturbance intensity, frequency drift pattern and propagation direction information of the disturbance center point in the disturbance coverage map, a physical model including the spatial diffusion behavior of the disturbance is established; Build a time series prediction model on the monitoring node to dynamically fit the disturbance response trend; The output of the physical model is used as the prior input of the time series model to achieve joint modeling of spatial attenuation and temporal evolution; The fusion modeling results generate a disturbance response simulation sequence of each monitoring node in the coverage area, which is used for differential analysis with the original monitoring signal.

6. The method for networked monitoring and early warning of geological disasters according to claim 1, characterized in that: Step S4 includes: Extract the residual signal time series of each monitoring node in the signal residual graph, identify the continuous trend drift area and mark the trend paragraph; Perform frequency decomposition and amplitude normalization on the local fluctuations in the residual signal to identify the fluctuation areas with heterogeneous frequency components and obvious amplitude mutations; Compare the response phase difference and mutation response time of each monitoring node in the same time period to extract the time series asynchronous response points; Combining trend drift areas, fluctuation heterogeneity areas and asynchronous response points, the initial trend contour line of the potential disaster evolution process is constructed.

7. The method for networked monitoring and early warning of geological disasters according to claim 1, characterized in that: Step S5 includes: Extract the structural characteristic parameters of the initial trend contour line and call the standard evolution path template classified by disaster type in the historical geological disaster database; The initial trend contour line and the historical evolution path are nonlinearly aligned in the time dimension using the dynamic time warping algorithm, and the response distribution similarity and path direction consistency are calculated in the spatial dimension; Establish a causal consistency judgment mechanism to determine whether the current trend is a reversible deformation mode or an irreversible disaster evolution mode based on trend matching, trend stage structure and response characteristics; Based on the discrimination results, a high-confidence risk identification result of geological hazards is generated, and the risk level, disaster type, impact range and response recommendation level are marked.

8. The method for networked monitoring and early warning of geological disasters according to claim 1, characterized in that: Step S6 specifically includes: Extract risk level labels, trend type characteristics, matching confidence scores, abnormal response structure types and spatial impact ranges from high-confidence risk identification results to construct a risk status parameter set; Based on the risk state parameter set, a set of early warning response control factors consisting of dynamic response threshold adjustment factor, model sensitivity enhancement factor, disturbance immunity weight factor and time series response window correction factor is constructed; Combined with the current monitoring status parameters, the applicability of the control factors is judged and ranked through the rule engine or fuzzy control logic to generate a real-time response strategy model; The response strategy model is applied to the early warning model in real time, and the parameter configuration and logical judgment path of the early warning model are dynamically adjusted to achieve closed-loop adaptive control of the early warning model.

Images