A system for monitoring geological hazard cracks

Abstract

The application relates to the technical field of geological monitoring, and discloses a system for monitoring geological disaster cracks, which realizes dynamic monitoring of crack states and intelligent early warning of disaster risks through the functions of crack multi-point acquisition, edge computing identification and remote trend analysis. The edge computing terminal extracts strain-displacement features and identifies stress concentration areas, dynamically triggers communication reporting based on risk levels, effectively reduces communication load, and based on the remote monitoring platform, historical and current data are fused to construct a crack trend model, realize identification and early warning response of the crack development trend, so that the system has the advantages of fast response speed, high data utilization rate, strong energy efficiency control and high risk identification precision.

Description

Technical Field

[0001] This invention relates to the field of geological monitoring technology, and in particular to a system for monitoring cracks in geological hazards. Background Technology

[0002] In the field of geological disaster monitoring and prevention, cracks are an important precursor to disasters such as landslides, collapses, and ground subsidence. The accuracy and response capability of their monitoring directly affect the timeliness and reliability of disaster early warning. Traditional geological crack monitoring methods mostly rely on manual fixed-point inspections, crack gauge measurements, or single-point sensor data collection. These methods suffer from problems such as long monitoring cycles, delayed response, and insufficient spatial coverage, making it difficult to meet the real-time needs for understanding minute deformations and dynamic evolution trends in complex geological environments. In recent years, solutions based on multi-point deployed sensor arrays for real-time data preprocessing and risk identification have gradually emerged. However, these solutions still have limitations, including insufficient crack evolution trend analysis capabilities, untimely communication response in high-risk scenarios, and limited equipment power consumption. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide a system for monitoring cracks in geological disasters, so as to solve the technical problem that the existing technology for monitoring cracks in geological disasters is insufficient in its ability to analyze the minute deformation and dynamic evolution trend of cracks in complex geological environments.

[0004] This invention discloses a system for monitoring geological disaster cracks, comprising a crack sensing terminal, an edge computing terminal, a main control communication module, and a remote monitoring platform; wherein,

[0005] The crack sensing terminal includes strain sensors and displacement sensors set at multiple measuring points around the crack, used to collect raw data of the crack area;

[0006] The edge computing terminal is used to perform data cleaning, abnormal fluctuation detection, strain-displacement feature extraction and stress concentration area determination on the raw data, and outputs monitoring feature indicators and risk level information after performing the above operations.

[0007] The main control communication module is used to determine whether to trigger the communication reporting mechanism based on the monitoring characteristic indicators and risk level information. When it is determined to be yes, the monitoring characteristic indicators, risk level information and raw data are transmitted to the remote monitoring platform.

[0008] The remote monitoring platform is equipped with a crack trend model, which is used to perform crack status assessment based on data transmitted by the main control communication module, and to provide disaster early warning based on the output of the crack trend model.

[0009] Furthermore, the edge computing terminal includes a data preprocessing unit, an abnormal fluctuation detection unit, a feature extraction unit, and a stress concentration identification unit.

[0010] Furthermore, the data preprocessing unit is used to perform filtering, noise reduction, and drift correction operations on the original data to obtain the first processed data;

[0011] The abnormal fluctuation detection unit is used to perform window sliding statistical analysis on the first processed data, identify short-term abnormal peaks and abrupt jump segments, and form a fluctuation abnormal marker sequence.

[0012] The feature extraction unit is used to extract strain-displacement features based on the first processed data and the fluctuation anomaly marker sequence within a preset time window, and to construct monitoring feature indicators based on the strain-displacement features; the strain-displacement features include at least strain rate, displacement abrupt amplitude, and stress increment trend.

[0013] The stress concentration identification unit is used to analyze the evolution trend of strain-displacement characteristics of multiple measuring points based on a spatiotemporal correlation model. When a consistent trend and / or synchronous abrupt change are determined, it is identified as a stress concentration area and the corresponding risk level label is output.

[0014] Furthermore, the spatiotemporal correlation model is constructed through a spatiotemporal graph convolutional network and trained based on the first historical crack monitoring data; wherein, the first historical crack monitoring data includes the spatial distribution coordinates of the measuring points, time-series strain-displacement data, and corresponding stress anomaly records.

[0015] Furthermore, the specific process of identifying stress concentration regions through a spatiotemporal correlation model includes:

[0016] Within the target sliding time window, the spatial proximity factor and time series similarity factor are determined based on the strain-displacement characteristics of multiple measuring points, and the spatiotemporal linkage weight matrix representing stress evolution is calculated based on the spatial proximity factor and time series similarity factor.

[0017] Based on the spatiotemporal linkage weight matrix, a subset of highly linked measurement points is identified. If the subset of highly linked measurement points meets the spatial correlation strength threshold, a candidate set of potential linkage regions is formed.

[0018] The strain-displacement characteristic variation trend of the test points in the candidate set of potential linkage regions is locally fitted and clustered. When there is a consistent trend and / or synchronous abrupt change, it is identified as a stress concentration region.

[0019] Furthermore, after identifying the stress concentration area, a risk value is calculated based on the comprehensive score of the monitoring characteristic indicators in the area, and the calculated risk value is mapped to a risk level label according to the set risk level classification threshold; the risk level label includes three levels: low risk, medium risk and high risk.

[0020] The main control communication module determines the risk level change trend based on the risk level label changes within multiple consecutive time windows, and automatically triggers the communication reporting mechanism when the risk level is high or when the medium risk level continues to rise.

[0021] Furthermore, the main control communication module includes a frequency control reporting unit and a data selection unit; among which,

[0022] The reporting frequency control unit is used to dynamically adjust the communication reporting frequency based on the risk level label and the trend of risk level changes;

[0023] The data selection unit is used to determine the scope of communication data content based on the risk level.

[0024] Furthermore, the edge computing terminal is a low-power embedded device that supports communication via LoRa or Narrowband Internet of Things (NB-IoT) in areas without public network coverage, and has local caching and periodic data synchronization capabilities.

[0025] The edge computing terminal also includes an energy management unit, which is used to adjust the wake-up cycle of the edge computing terminal and the processing frequency of edge computing tasks based on power supply status, data traffic requirements and risk level.

[0026] Furthermore, the remote monitoring platform includes a crack trend modeling unit and a condition assessment unit; wherein,

[0027] The crack trend modeling unit adopts a dual-channel modeling structure. It learns evolutionary patterns based on second-historical crack monitoring data through a historical data channel, and captures real-time trends based on current monitoring data through a current data channel. Feature alignment and joint modeling are performed at the fusion layer, ultimately outputting crack development trend type, predicted crack propagation rate, and estimated potential impact area. The second-historical crack monitoring data includes historical monitoring characteristic index sequences, historical risk level labels, historical raw data of monitoring points, spatial distribution information of monitoring points, topographic and geomorphological change information, and historical geological disaster records. The current monitoring data includes monitoring characteristic indicators, risk level information, and corresponding raw data of monitoring points transmitted by the main control communication module.

[0028] The condition assessment module is used to identify crack conditions and assess risk response levels based on the output of the crack trend model.

[0029] Furthermore, the remote monitoring platform also includes a disaster early warning unit;

[0030] The disaster early warning unit is used to match the crack status and risk response level output by the status assessment module with the preset disaster level response rules and output the corresponding disaster early warning level.

[0031] The disaster warning level is pushed to the emergency response terminal through a remote communication interface.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] This invention achieves dynamic monitoring of crack conditions and intelligent early warning of disaster risks by integrating multi-point crack monitoring, edge computing identification, and remote trend analysis. It extracts strain-displacement features and identifies stress concentration areas through an edge computing terminal, dynamically triggering communication reporting based on risk levels, effectively reducing communication load. Furthermore, it constructs a crack trend model based on historical and current data from a remote monitoring platform, enabling the identification and early warning response of crack development trends. This results in a system with advantages such as fast response speed, high data utilization, strong energy efficiency control, and high risk identification accuracy. Attached Figure Description

[0034] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and constitute a part of this application, do not limit the scope of the invention. In the drawings:

[0035] Figure 1 This is a schematic diagram of a system for monitoring cracks in geological disasters, as disclosed in Embodiment 1 of the present invention. Detailed Implementation

[0036] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0037] Example 1

[0038] This invention discloses a system for monitoring cracks in geological hazards. Please refer to [link / reference]. Figure 1 , Figure 1 This is a schematic diagram of a system for monitoring geological disaster cracks, as disclosed in an embodiment of the present invention. The system includes a crack sensing terminal, an edge computing terminal, a main control communication module, and a remote monitoring platform; wherein,

[0039] The crack sensing terminal includes strain sensors and displacement sensors set at multiple measuring points around the crack, used to collect raw data of the crack area;

[0040] The edge computing terminal is used to perform data cleaning, abnormal fluctuation detection, strain-displacement feature extraction and stress concentration area determination on the raw data, and outputs monitoring feature indicators and risk level information after performing the above operations.

[0041] The main control communication module is used to determine whether to trigger the communication reporting mechanism based on the monitoring characteristic indicators and risk level information. When it is determined to be yes, the monitoring characteristic indicators, risk level information and raw data are transmitted to the remote monitoring platform.

[0042] The remote monitoring platform is equipped with a crack trend model, which is used to perform crack status assessment based on data transmitted by the main control communication module, and to provide disaster early warning based on the output of the crack trend model.

[0043] Furthermore, the edge computing terminal includes a data preprocessing unit, an abnormal fluctuation detection unit, a feature extraction unit, and a stress concentration identification unit.

[0044] Furthermore, the data preprocessing unit is used to perform filtering, noise reduction, and drift correction operations on the original data to obtain the first processed data;

[0045] The abnormal fluctuation detection unit is used to perform window sliding statistical analysis on the first processed data to identify short-term abnormal peaks and abrupt jump segments, and form a fluctuation abnormal label sequence.

[0046] The feature extraction unit is used to extract strain-displacement features based on the first processed data and the fluctuation anomaly marker sequence within a preset time window, and to construct monitoring feature indicators based on the strain-displacement features; the strain-displacement features include at least strain rate, displacement abrupt change amplitude and stress increment trend.

[0047] In this embodiment of the invention, the abnormal fluctuation detection unit first performs a statistical analysis operation based on a sliding time window on the first processed data output by the data preprocessing unit. Specifically, a sliding window of fixed length is set, and within each window interval, the mean, standard deviation, and slope change rate of the strain and displacement sequences are calculated. By comparing with the historical steady-state data of that window, abnormal peaks (such as exceeding three times the standard deviation of the mean) or slope abrupt changes (such as the slope difference between two consecutive windows exceeding a set threshold) are identified. For each window identified as abnormal, its start and end times and the type of abnormality (such as jump, spike, drift, etc.) are recorded to form a fluctuation anomaly marking sequence, which marks the segments in the data where nonlinear disturbances exist.

[0048] Based on the aforementioned labeled sequence and the first processed data, the feature extraction unit extracts strain-displacement features such as strain rate, displacement abrupt change amplitude, and stress increment trend. Specifically, the strain rate is calculated as the derivative or sliding difference of the strain value within a preset time window, reflecting the degree of deformation per unit time in the crack region; the displacement abrupt change amplitude represents the maximum jump value of displacement change extracted within each fluctuating anomaly labeled segment, characterizing local rapid propagation behavior; and the stress increment trend is established based on a differential model of the strain-displacement coupling relationship, fitting the slope of the stress estimation results over continuous time periods to determine whether the stress is in a state of continuous accumulation or release.

[0049] Building upon this foundation, a set of monitoring characteristic indicators is further constructed to uniformly quantify and characterize the deformation evolution state of cracks. Specifically, these indicators may include strain rate indicators, maximum displacement abrupt change values, stress increment trend fitting slope, anomaly density coefficient (the number of anomalous marker segments appearing within a unit time window), and fluctuation consistency coefficient (anomaly overlap index of multiple measuring points within the same time window). These monitoring characteristic indicators can serve as input parameters for determining risk level labels and can also be used by subsequent remote monitoring platforms for crack condition assessment.

[0050] The stress concentration identification unit is used to analyze the evolution trend of strain-displacement characteristics of multiple measuring points based on a spatiotemporal correlation model. When a consistent trend and / or synchronous abrupt change are determined, it is identified as a stress concentration area and the corresponding risk level label is output.

[0051] Furthermore, the spatiotemporal correlation model is constructed through a spatiotemporal graph convolutional network and trained based on the first historical crack monitoring data; wherein, the first historical crack monitoring data includes the spatial distribution coordinates of the measuring points, time series strain-displacement data, and corresponding stress anomaly records.

[0052] Specifically, in this embodiment of the invention, the spatio-temporal correlation model is constructed using a spatio-temporal graph convolutional network (ST-GCN) based on graph neural networks. The purpose is to effectively capture the spatial correlation and temporal evolution characteristics between multiple monitoring points in the crack region, and to achieve intelligent identification of potential stress concentration areas.

[0053] The model was trained based on a first historical crack monitoring dataset, which consists of data collected and stored during the long-term deployment of crack sensing terminals. This dataset includes spatial distribution coordinates of measuring points, time-series strain-displacement data, and stress anomaly records. The spatial distribution coordinates of the measuring points are the geographical location information of each strain / displacement sensor (usually absolute / relative coordinates in a two-dimensional or three-dimensional coordinate system). This information can be obtained by the sensors during initial deployment through total station measurements, GPS calibration, or manual mapping. After initial system configuration, this information is recorded in the database and automatically retrieved. The time-series strain-displacement data refers to the strain and displacement data sequence recorded by each measuring point within a continuous monitoring period. The stress anomaly records include time windows historically identified as "stress concentration" or "accelerated crack propagation," and the corresponding set of measuring points, which serve as labels for the training samples.

[0054] In the model construction process, an initial graph structure is first constructed based on the spatial distribution information of all measuring points. Each node in the graph represents a monitoring measuring point, and the edge weights are defined based on spatial proximity. Simultaneously, the corresponding historical strain-displacement time series is loaded onto each node as a temporal feature input.

[0055] After construction, the ST-GCN model is trained using supervised learning. The core training process is as follows:

[0056] Input data structuring: For each time window, construct a node input matrix X∈R^{N×F×T} with a graph structure, where N is the number of measurement points, F is the input feature dimension of each measurement point (such as strain rate, displacement jump value, etc.), and T is the time length;

[0057] Graph Convolution Processing: Extract spatially relevant features at each time step through graph convolution (GCN) operations;

[0058] Temporal convolution processing: Modeling the extracted spatial features in the temporal dimension;

[0059] Supervised output layer: Outputs the potential "stress concentration" probability of each node under each time window. The training objective is to minimize the cross-entropy loss between the predicted probability and the historical stress anomaly record.

[0060] Spatiotemporal weight fusion mechanism: During training, the model's ability to perceive synchronous mutations and consistent evolutionary trends is enhanced by learning a dynamic weight matrix that fuses spatial proximity weights and temporal similarity weights.

[0061] After training, the model is deployed to an edge computing terminal as part of the stress concentration identification unit.

[0062] As a further preferred embodiment, when constructing the initial graph structure, not only is the spatial adjacency relationship defined based on the Euclidean distance between the measuring points, but the geomorphic label to which the measuring points belong is also introduced as an auxiliary criterion. When two measuring points are located in the same fault zone, slip zone, or geomorphic unit, their edge weights are increased to reflect the coupling in the geological structure, forming a structure-enhanced graph.

[0063] In actual operation, the spatiotemporal correlation model takes the strain-displacement characteristic indicators of multiple measuring points within the current time sliding window as input, combines the graph structure and parameter weights formed during the training phase, outputs the potential "high-linkage measuring point subset" in the current monitoring area, and identifies stress concentration areas by combining spatial and temporal threshold judgment logic, providing a basis for subsequent risk level assessment.

[0064] Furthermore, the specific process of identifying stress concentration regions through a spatiotemporal correlation model includes:

[0065] Within the target sliding time window, the spatial proximity factor and time series similarity factor are determined based on the strain-displacement characteristics of multiple measuring points, and the spatiotemporal linkage weight matrix representing stress evolution is calculated based on the spatial proximity factor and time series similarity factor.

[0066] Based on the spatiotemporal linkage weight matrix, a subset of highly linked measurement points is identified. If the subset of highly linked measurement points meets the spatial correlation strength threshold, a candidate set of potential linkage regions is formed.

[0067] The strain-displacement characteristic variation trend of the test points in the candidate set of potential linkage regions is locally fitted and clustered. When there is a consistent trend and / or synchronous abrupt change, it is identified as a stress concentration region.

[0068] To accurately determine the potential stress concentration trend in geological disaster fissure areas, this embodiment proposes a stress concentration area identification mechanism based on a spatiotemporal correlation model. Its core lies in combining the strain-displacement response characteristics of multiple measuring points in the time dimension with their proximity in the spatial layout to construct a graph structure model with dynamic spatiotemporal weight distribution, which is used to capture concentrated risk areas in the fissure area that may experience linked ruptures.

[0069] Specifically, within the target sliding time window, strain-displacement characteristics such as strain rate, displacement abrupt change amplitude, and stress increment trend are first extracted for each monitoring point. Based on these characteristics, a spatial proximity factor is constructed, whose value is normalized according to the geographical coordinate distance between the monitoring points and expressed in Gaussian kernel function form as follows:

[0070]

[0071] Among them, W s(i,j) is the spatial proximity factor between measurement point i and measurement point j; d i,j σ is the Euclidean spatial distance between measurement point i and measurement point j; σ is the spatial distance normalization smoothing factor, which determines the similarity decay rate and is an empirical value or determined through cross-validation.

[0072] A time series similarity factor is constructed, whose value is evaluated based on the similarity of feature sequences within a time window:

[0073]

[0074] Among them, W t (i,j) is the time series similarity factor between measurement point i and measurement point j; X i X j Let i be the strain-displacement eigenvector of measuring point i and measuring point j within the current time window; ||·|| is the Euclidean norm of the vector.

[0075] By weighting and fusing the two factors mentioned above, a final spatiotemporal linkage weight matrix is ​​formed. This matrix reflects the "coupling strength" between measurement point pairs within the current window. Its form is as follows:

[0076] W(i,j)=α·W s (i,j)+β·W t (i,j)

[0077] Where α and β are weighting coefficients, satisfying α+β=1.

[0078] Furthermore, measurement point pairs with edge weights higher than a set linkage threshold are extracted from the spatiotemporal linkage weight matrix to further identify highly coupled measurement point subsets. If the measurement points in this subset meet the association strength threshold in terms of spatial layout (e.g., the average spatial distance is less than a set distance), then this subset is determined to constitute a potential linkage region candidate set. Local polynomial fitting is performed on the strain-displacement characteristic change trend of each measurement point in each candidate subset, and the fitting residual and trend slope are extracted as clustering criteria. Hierarchical clustering or density clustering is used to cluster the measurement points to determine the trend consistency and abrupt change synchronicity. If there are highly consistent trend clusters in the clustering results, and / or multiple measurement points show abrupt change synchronicity in the same time window, the region is determined to be a stress concentration region.

[0079] This invention significantly enhances the ability of a geological disaster crack monitoring system to perceive the interconnected changes of multiple monitoring points by establishing a stress concentration area identification mechanism based on a spatiotemporal correlation model. By introducing graph structure modeling, each monitoring point is treated as a node, and the coupling strength is dynamically calculated based on their spatial and temporal correlations, achieving high-precision identification of potential interconnected areas. Simultaneously, the stress concentration areas output by this model provide reliable and structured prior inputs for the scheduling strategy of the main control communication module, crack trend modeling of the remote platform, and disaster early warning. This serves as a core support for the entire system's algorithm logic, effectively enhancing the system's judgment capability and response efficiency.

[0080] Furthermore, after identifying the stress concentration area, a risk value is calculated based on the comprehensive score of the monitoring characteristic indicators in the area, and the calculated risk value is mapped to a risk level label according to the set risk level classification threshold; the risk level label includes three levels: low risk, medium risk and high risk.

[0081] The main control communication module determines the risk level change trend based on the risk level label changes within multiple consecutive time windows, and automatically triggers the communication reporting mechanism when the risk level is high or when the medium risk level continues to rise.

[0082] Specifically, after identifying the stress concentration in the crack area, this invention further introduces a risk level labeling mechanism to achieve a quantitative expression of the potential disaster status and a classification of early warning levels. This mechanism is based on a comprehensive scoring of multiple monitoring characteristic indicators within the stress concentration area to form a risk value representing the current risk level. Specifically, core features such as strain rate, maximum displacement abrupt change, slope of stress increment trend fitting, abnormal density coefficient, and fluctuation consistency coefficient are first summarized and normalized across different measuring points to ensure the comparability of various data.

[0083] Subsequently, each indicator is weighted according to preset weighting coefficients to obtain a risk score characterizing the overall stability of the current fracture area. This score serves as the basis for risk level classification and is compared with a set risk level threshold to map to three levels: low risk, medium risk, or high risk. The classification criteria can be adjusted and optimized based on historical geological disaster samples or expert experience, and are not specifically limited in this embodiment of the invention.

[0084] In addition, in order to achieve dynamic perception of risk trends, the risk level labels within multiple consecutive time windows are analyzed to identify high-concern situations such as continuous rise in medium-risk levels and sustained maintenance of high-risk levels. Based on this, it is determined whether there is a significant trend of risk level change. When the judgment result shows that the risk level is rapidly rising or the current level has reached a high-risk state, the communication reporting mechanism of the main control communication module is automatically triggered.

[0085] After triggering the reporting mechanism, the main control communication module transmits the current monitoring characteristic indicators, risk level labels, risk scores, and related raw data to the remote monitoring platform, providing a detailed and effective data foundation for subsequent crack trend analysis, disaster response decision-making, and emergency dispatch. This mechanism achieves efficient linkage from edge intelligent judgment to remote centralized assessment, significantly improving the system's response capability and handling efficiency to potential geological disaster threats.

[0086] Furthermore, the main control communication module includes a frequency control reporting unit and a data selection unit; among which,

[0087] The reporting frequency control unit is used to dynamically adjust the communication reporting frequency based on the risk level label and the trend of risk level changes;

[0088] The data selection unit is used to determine the scope of communication data content based on the risk level.

[0089] Specifically, in order to improve communication efficiency and response strategy flexibility in the process of geological disaster monitoring, the main control communication module in this embodiment of the invention is further provided with a reporting frequency control unit and a data selection unit to achieve fine control of communication reporting behavior and on-demand data transmission.

[0090] The reporting frequency control unit dynamically adjusts the communication reporting frequency based on the current risk level label and risk level change trend of the crack area. Specifically, when the risk level is identified as low and is stable or declining, the default communication cycle is set to a longer time interval, with only periodic status synchronization performed. When the risk level is identified as medium and shows a continuous upward trend, or has reached a high risk level, the communication frequency is increased, switching to a high-frequency reporting mode to ensure that the remote monitoring platform can quickly receive key change information and respond. Simultaneously, to avoid wasting network resources and power, the system automatically reverts to a low-frequency reporting strategy after switching back to a low-risk state.

[0091] Understandably, during implementation, the main control communication module, based on the monitoring characteristic indicators, risk level information, and raw data as the basic reporting content, can dynamically expand the range of uploaded data content in high-risk or trend-sudden situations, according to the risk level and its changing trend output by the edge computing terminal and the strategy request of the remote monitoring platform. This includes, but is not limited to, fluctuation anomaly marker sequences, local raw data fragments of high-risk measurement points, feature window summary vectors of important stages, and evolutionary feature sequences.

[0092] The data selection unit intelligently determines the scope of the data content to be reported based on the risk level. At a low risk level, only summary monitoring indicators and feature window summaries for important stages are uploaded to reduce the processing load on the remote monitoring platform. At a medium risk level, synchronous transmission of some raw data fragments, fluctuation anomaly marker sequences, and feature window evolution sequence for important stages is added. When the risk level rises to high risk, complete monitoring indicators, risk level labels, feature window evolution sequence for important stages, and detailed raw data for key periods are reported to ensure the remote monitoring platform has comprehensive data input.

[0093] Through the above mechanism, the main control communication module realizes the intelligent control logic of "on-demand communication and hierarchical reporting", which not only improves communication efficiency and energy consumption management capabilities, but also ensures the timely acquisition and linkage response of critical disaster information. It is a key bridge for the efficient integration of edge intelligence and centralized processing.

[0094] Furthermore, the edge computing terminal is a low-power embedded device that supports communication via LoRa or Narrowband IoT in areas without public network coverage, and has local caching and periodic data synchronization capabilities.

[0095] The edge computing terminal also includes an energy management unit, which is used to adjust the wake-up cycle of the edge computing terminal and the processing frequency of edge computing tasks based on power supply status, data traffic requirements and risk level.

[0096] In this embodiment of the invention, the edge computing terminal is designed as a low-power embedded device to adapt to the complex environment of geological disaster monitoring sites, where operation is prolonged, power supply resources are limited, and network access is restricted. The terminal integrates core modules such as a low-power processor, a non-volatile memory unit, a communication module, and an energy management unit.

[0097] Specifically, to enhance communication adaptability, edge computing terminals support establishing low-power wide-area connections with the main control communication module or relay base station in areas without public network coverage or with unstable public network signals, via LoRa or Narrowband IoT communication protocols. LoRa communication is suitable for long-distance, low-rate, and intermittent data reporting scenarios, while NB-IoT is suitable for stable connections under narrowband channel conditions. Both communication methods can be flexibly configured according to the actual deployment environment. Furthermore, it also features local caching capabilities and a periodic data synchronization mechanism. That is, under conditions of poor communication or delays, intermittent data can be stored in local storage units and automatically uploaded in batches when the network becomes available, ensuring data integrity and continuity.

[0098] The energy management unit is responsible for dynamically adjusting the working status of the edge computing terminal based on parameters such as external power supply conditions (e.g., solar panel power supply voltage, battery remaining power), data processing intensity, and current risk level. Its control strategy mainly includes the following two aspects:

[0099] Wake-up cycle adjustment mechanism: The wake-up frequency of the device is automatically adjusted based on the risk level label. When the risk level is low, the terminal enters a dormant state with a long cycle, activating data acquisition and processing tasks only at periodic times to save power. When the risk level increases or an upward trend appears, the terminal will shorten the wake-up cycle, increase the data update frequency and response speed, and enhance the ability to capture disaster evolution.

[0100] Edge computing task processing frequency adjustment mechanism: Based on real-time power supply status and data traffic strategy, dynamically control the execution frequency and computation depth of computing tasks. For example, simplify the computation model for identifying abnormal fluctuations or delay the extraction process of certain non-critical features, and optimize the power consumption structure while ensuring data availability.

[0101] The aforementioned low-power embedded architecture and intelligent energy scheduling mechanism enable this invention to possess deployment flexibility, long-term stability, and adaptive energy efficiency control capabilities when facing field geological disaster monitoring scenarios. This is particularly beneficial in mountainous and forested areas far from municipal infrastructure, significantly enhancing its engineering practicality and continuous monitoring capabilities. Furthermore, by combining edge computing task localization execution strategies, intelligent processing is achieved while effectively avoiding the high energy consumption and high dependency of remote centralized computing architectures, providing a stable, energy-efficient, and sustainable solution for frontline perception and early warning judgment of sudden disasters.

[0102] Furthermore, the remote monitoring platform includes a crack trend modeling unit and a condition assessment unit; among which,

[0103] The crack trend modeling unit adopts a dual-channel modeling structure. It learns evolutionary patterns based on second-historical crack monitoring data through a historical data channel, and captures real-time trends based on current monitoring data through a current data channel. Feature alignment and joint modeling are performed at the fusion layer, ultimately outputting crack development trend type, predicted crack propagation rate, and estimated potential impact area. The second-historical crack monitoring data includes historical monitoring characteristic index sequences, historical risk level labels, historical raw data of monitoring points, spatial distribution information of monitoring points, topographic and geomorphological change information, and historical geological disaster records. The current monitoring data includes monitoring characteristic indicators, risk level information, and corresponding raw data of monitoring points transmitted by the main control communication module.

[0104] The condition assessment module is used to identify crack conditions and assess risk response levels based on the output of the crack trend model.

[0105] Furthermore, the remote monitoring platform also includes a disaster early warning unit;

[0106] The disaster early warning unit is used to match the crack status and risk response level output by the status assessment module with the preset disaster level response rules and output the corresponding disaster early warning level.

[0107] The disaster warning level is pushed to the emergency response terminal through a remote communication interface.

[0108] In this embodiment of the invention, the remote monitoring platform is the core analysis part of the system, mainly including a crack trend modeling unit and a status assessment unit, which are used to perform trend prediction and risk assessment operations on the data transmitted by the main control communication module, supporting the disaster early warning function of the overall system.

[0109] The core task of the fracture trend modeling unit is to build and train a fracture trend model, which is used to capture the changing patterns of geological fractures in time and space.

[0110] To enhance the model's expressive power and predictive accuracy, this invention employs a dual-channel modeling structure, processing currently reported data from the edge side and historical crack monitoring data maintained on the platform side, respectively, and performing deep feature fusion in the intermediate layer. The historical crack monitoring data includes, but is not limited to, historical monitoring feature index sequences, historical risk level labels, historical raw data of monitoring points, spatial distribution information of monitoring points, topographic and geomorphological change information, and recorded geological disaster occurrence data. This data constitutes a complete trend training sample for supervised training of the crack trend model.

[0111] In terms of model structure, the crack trend model employs an improved dual-channel spatiotemporal modeling network. Each channel encodes time-series data using a Long Short-Term Memory (LSTM) neural network structure to uncover nonlinear dynamic relationships at different time scales. Simultaneously, spatial location information of measurement points is constructed into a spatial topology map, and a spatial attention mechanism is introduced to enhance the features of measurement points with key spatial influence. In particular, topographic change information is encoded using raster topographic data and injected into the model as an auxiliary feature, further improving the ability to identify crack path evolution under complex terrain constraints.

[0112] It is understood that the crack trend model of this invention adopts a "pre-training + real-time inference" operating mechanism. That is, before the model goes online, it has been trained and its parameters optimized on a remote monitoring platform using a large amount of second historical crack monitoring data, forming a stable and mature model structure and weights. The purpose of introducing a dual-channel modeling architecture is that the historical channel and the current channel respectively undertake the functions of "historical knowledge injection" and "real-time perception and response" in the model. Dynamic coupling is achieved in the fusion layer through attention mechanisms or feature alignment strategies, enabling the model to quickly perceive the current situation based on existing experience and complete trend inference and response output. This structure avoids the problems of historical model lag and time-consuming real-time modeling in traditional models, ensuring the real-time performance and accuracy of the early warning system.

[0113] The raw data refers to the unprocessed time-series strain and displacement values ​​recorded by the crack sensing terminal sensors, forming an indispensable foundation for constructing the trend evolution model. It provides the underlying physical response signal and, in the process of building the trend evolution model, can be used to reconstruct fine-grained response coupling characteristics. This invention retains and inputs the raw data to introduce the raw signal through a branch in the dual-channel model for sequence feature extraction, ensuring that the model not only learns abstract statistical features but also possesses the ability to model real signal change patterns.

[0114] In actual deployment, the fracture trend model receives current monitoring characteristic indicators, risk level information, and raw data from measuring points uploaded by the main control communication module. Based on these inputs, the model generates three main outputs: first, the fracture development trend type, indicating whether the current fracture is in a stable, slowly expanding, rapidly expanding, or abruptly changing state; second, the predicted fracture propagation rate, a numerical estimate of the average rate of change of the fracture over a future time period; and third, the estimated potential impact area, outputting the spatial boundary or risk zone layer of the potentially affected area based on the distribution of measuring points, the direction of propagation, and historical geological data.

[0115] In a preferred embodiment of the present invention, the state assessment unit further analyzes the results output by the crack trend model and determines the current state and risk response level of the crack through a built-in state assessment rule set. This rule set is constructed based on multi-dimensional decision-making logic, comprehensively considering multiple dimensions of model output factors such as trend type, expansion rate, and affected area range, thereby improving the scientific rigor of the assessment.

[0116] Specifically, the state assessment unit first makes a preliminary classification judgment based on the crack development trend type output by the model. If the trend type is identified as "stable" or "slowly expanding", the current state is assumed to be low risk; if it is identified as "rapidly expanding" or "abruptly expanding", it enters the high-priority assessment path and initiates further expansion rate quantification and spatial impact analysis.

[0117] At the quantitative level, the condition assessment unit compares the predicted crack propagation rate output by the model with a preset risk threshold. If the predicted rate exceeds the set threshold (e.g., greater than a certain upper limit for safe crack growth rate), it is marked as a rate exceeding the limit; if the propagation rate does not exceed the limit but shows a significant increasing trend compared with the historical average rate, it is considered a rate surge signal. In the above situations, the crack state is automatically determined as "active period" or "critical stage".

[0118] Furthermore, combining the potential impact area estimation results output by the model, the status assessment unit further performs spatial cross-analysis. Based on built-in regional risk layers such as topographic information, location of important facilities, and residential area boundaries, it assesses whether the predicted crack development path covers or approaches high-value target areas. If the potential impact area spatially overlaps with these high-value areas, the status assessment level will be automatically upgraded by one level.

[0119] In actual operation, the condition assessment unit integrates the above judgment results to form the crack condition and risk response level, and triggers corresponding response strategies through the disaster early warning unit. For example, when the condition is identified as "critical stage" and the risk response level is "high", a red warning signal is immediately sent to the emergency response center, and it is recommended to initiate on-site verification or preventive measures.

[0120] The remote monitoring platform integrates historical and real-time data to construct a crack trend model with trend memory and spatial diffusion judgment capabilities. This model not only enables predictive assessment of future crack conditions but also provides visualized risk level outputs for emergency response systems. This solution effectively overcomes the limitations of traditional passive crack alarm systems, which can only identify excessive events and lack predictive capabilities. It represents a leap from "event triggering" to "trend early warning," significantly improving the intelligence and proactiveness of geological disaster prevention and control.

[0121] Finally, it should be noted that the system for monitoring geological disaster cracks disclosed in the embodiments of the present invention is only a preferred embodiment of the present invention and is only used to illustrate the technical solution of the present invention, not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A system for monitoring cracks in geological hazards, characterized in that, The system includes a crack sensing terminal, an edge computing terminal, a main control communication module, and a remote monitoring platform; wherein... The crack sensing terminal includes strain sensors and displacement sensors set at multiple measuring points around the crack, used to collect raw data of the crack area; The edge computing terminal is used to perform data cleaning, abnormal fluctuation detection, strain-displacement feature extraction, and stress concentration region determination on the raw data, and outputs monitoring feature indicators and risk level information after performing the above operations. The edge computing terminal includes a stress concentration identification unit, which analyzes the evolution trend of strain-displacement characteristics at multiple measuring points based on a spatiotemporal correlation model. If a consistent trend and / or synchronous abrupt change are determined, the region is identified as a stress concentration region, and a corresponding risk level label is output. The strain-displacement characteristics include strain rate, displacement abrupt change amplitude, and stress increment trend. The specific process of identifying stress concentration regions using a spatiotemporal correlation model includes: Within the target sliding time window, the spatial proximity factor and time series similarity factor are determined based on the strain-displacement characteristics of multiple measuring points, and the spatiotemporal linkage weight matrix representing stress evolution is calculated based on the spatial proximity factor and time series similarity factor. Based on the spatiotemporal linkage weight matrix, a subset of highly linked measurement points is identified. If the subset of highly linked measurement points meets the spatial correlation strength threshold, a candidate set of potential linkage regions is formed. Local fitting and clustering judgment are performed on the strain-displacement characteristic variation trend of the test points in the candidate set of potential linkage areas. When there is a consistent trend and / or synchronous abrupt change, it is identified as a stress concentration area. The main control communication module is used to determine whether to trigger the communication reporting mechanism based on the monitoring characteristic indicators and risk level information. When it is determined to be yes, the monitoring characteristic indicators, risk level information and raw data are transmitted to the remote monitoring platform. The remote monitoring platform is equipped with a crack trend model, which is used to perform crack status assessment based on data transmitted by the main control communication module, and to provide disaster early warning based on the output of the crack trend model.

2. The system for monitoring geological disaster cracks according to claim 1, characterized in that, The edge computing terminal also includes a data preprocessing unit, an abnormal fluctuation detection unit, and a feature extraction unit.

3. The system for monitoring geological disaster cracks according to claim 2, characterized in that, The data preprocessing unit is used to perform filtering, noise reduction, and drift correction operations on the original data to obtain the first processed data. The abnormal fluctuation detection unit is used to perform window sliding statistical analysis on the first processed data, identify short-term abnormal peaks and abrupt jump segments, and form a fluctuation abnormal marker sequence. The feature extraction unit is used to extract strain-displacement features based on the first processed data and the fluctuation anomaly marker sequence within a preset time window, and to construct monitoring feature indicators based on the strain-displacement features.

4. The system for monitoring geological disaster cracks according to claim 3, characterized in that, The spatiotemporal correlation model is constructed through a spatiotemporal graph convolutional network and trained based on the first historical crack monitoring data; wherein, the first historical crack monitoring data includes the spatial distribution coordinates of the measuring points, time series strain-displacement data, and corresponding stress anomaly records.

5. The system for monitoring geological hazard cracks according to any one of claims 3-4, characterized in that, After identifying the stress concentration area, the risk value is calculated based on the comprehensive score of the monitoring characteristic indicators in the area, and the calculated risk value is mapped to the risk level label according to the set risk level classification threshold. The risk level label includes three levels: low risk, medium risk, and high risk. The main control communication module determines the risk level change trend based on the risk level label changes within multiple consecutive time windows, and automatically triggers the communication reporting mechanism when the risk level is high or when the medium risk level continues to rise.

6. The system for monitoring geological disaster cracks according to claim 5, characterized in that, The main control communication module includes a frequency control reporting unit and a data selection unit; among which, The reporting frequency control unit is used to dynamically adjust the communication reporting frequency based on the risk level label and the trend of risk level changes; The data selection unit is used to determine the scope of communication data content based on the risk level.

7. The system for monitoring geological disaster cracks according to claim 1, characterized in that, The edge computing terminal is a low-power embedded device that supports communication via LoRa or Narrowband Internet of Things (NB-IoT) in areas without public network coverage, and has local caching and periodic data synchronization capabilities. The edge computing terminal also includes an energy management unit, which is used to adjust the wake-up cycle of the edge computing terminal and the processing frequency of edge computing tasks based on power supply status, data traffic requirements and risk level.

8. The system for monitoring geological disaster cracks according to claim 1, characterized in that, The remote monitoring platform includes a crack trend modeling unit and a condition assessment unit; wherein... The crack trend modeling unit adopts a dual-channel modeling structure. It learns evolutionary patterns based on second-historical crack monitoring data through a historical data channel, and captures real-time trends based on current monitoring data through a current data channel. Feature alignment and joint modeling are performed at the fusion layer, ultimately outputting crack development trend type, predicted crack propagation rate, and estimated potential impact area. The second-historical crack monitoring data includes historical monitoring characteristic index sequences, historical risk level labels, historical raw data of monitoring points, spatial distribution information of monitoring points, topographic and geomorphological change information, and historical geological disaster records. The current monitoring data includes monitoring characteristic indicators, risk level information, and corresponding raw data of monitoring points transmitted by the main control communication module. The condition assessment module is used to identify crack conditions and assess risk response levels based on the output of the crack trend model.

9. The system for monitoring geological disaster cracks according to claim 8, characterized in that, The remote monitoring platform also includes a disaster early warning unit; The disaster early warning unit is used to match the crack status and risk response level output by the status assessment module with the preset disaster level response rules and output the corresponding disaster early warning level. The disaster warning level is pushed to the emergency response terminal through a remote communication interface.

Images