Geological disaster hidden danger monitoring and early warning system based on data analysis

Abstract

The invention discloses a geological disaster hidden danger monitoring and early warning system based on data analysis, and belongs to the technical field of natural disaster prevention and control. A geological disaster hidden danger monitoring and early warning system based on data analysis comprises a data acquisition module used for acquiring geological environment parameters in real time; the data transmission module adopts a mixed transmission protocol of a satellite internet and a ground network; the data analysis module is used for constructing a hybrid analysis model in combination with a physical model and a machine learning algorithm model; and the early warning module triggers a multi-stage early warning mechanism according to the analysis result. The geological disaster hidden danger monitoring and early warning system based on data analysis effectively solves the problem of low transmission rate of the satellite internet in the prior art through a hybrid transmission protocol of the satellite internet and the ground network and supporting edge calculation and dynamic adjustment of a data compression strategy; and due to the application of edge calculation, the data can be locally preprocessed and analyzed at the network edge node, the data transmission quantity is reduced, and the transmission efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of natural disaster prevention and control technology, and more specifically, to a geological disaster hazard monitoring and early warning system based on data analysis. Background Technology

[0002] Geological disasters are a significant threat to human life, property, and ecological environment stability. They are diverse in type, encompassing landslides, debris flows, earthquakes, and other forms. Once these disasters occur, they often cause severe casualties and huge economic losses, while also inflicting irreparable damage on the ecological environment. Therefore, effective monitoring and early warning of geological disasters have become a crucial aspect of geological disaster prevention and control.

[0003] In recent years, with the continuous development of science and technology, advanced technologies such as the Internet of Things, satellite remote sensing, and artificial intelligence have been gradually applied to the field of geological disaster monitoring, bringing new opportunities to geological disaster monitoring work; however, existing technologies still have some core defects in practical applications.

[0004] In terms of data transmission, although satellite internet technology has made remote data transmission possible, the problem of low data transmission rate remains prominent. According to the 2025 Satellite Internet Technology Report, satellite communication rates are low in some remote areas and high-altitude areas. Geological disaster monitoring data often has the characteristics of high-frequency sampling and large capacity. Low-speed transmission directly leads to the delay in the transmission of early warning information, which cannot play an early warning role in a timely manner. Summary of the Invention

[0005] The purpose of this invention is to provide a geological disaster hazard monitoring and early warning system based on data analysis, so as to solve the problems mentioned in the background art above:

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A data analysis-based geological hazard monitoring and early warning system includes a data acquisition module for real-time acquisition of geological environmental parameters. By deploying a multi-source sensor network, it collects and records data on surface displacement, changes in surface cracks, rainfall, soil moisture, and dynamic changes in groundwater.

[0008] The data transmission module uses a hybrid transmission protocol combining satellite internet and terrestrial networks to transmit the collected geological environmental parameters and dynamically adjusts the compression strategy of the transmitted data based on the real-time rate of the satellite internet.

[0009] The data analysis module uses a hybrid analysis model, which combines physical models and machine learning algorithms, to analyze the received geological environmental parameters and generate analysis results based on preset thresholds.

[0010] The early warning module is used to issue early warning information to the region, trigger multi-level early warning mechanisms based on analysis results, and release early warning information through multiple channels.

[0011] Preferably, the data cleaning and preprocessing unit employs a constraint mechanism to ensure data integrity and handles data loss, drift, and noise anomalies;

[0012] A GIS single-map visualization unit is used to display three-dimensional earth scenes and spatial analysis;

[0013] The system has a self-learning function, continuously receiving new data to optimize the hybrid analysis model and automatically updating the warning thresholds.

[0014] Preferably, the data transmission module supports edge computing. By deploying computing resources at network edge nodes, it performs local preprocessing and analysis on the collected data and dynamically adjusts the data compression strategy according to the real-time fluctuations in satellite internet speed. When the satellite internet speed is below 1Mbps, the collected surface displacement, surface crack, and rainfall data are processed using wavelet transform algorithm, and the compression ratio is increased to 10:1. Soil moisture and groundwater dynamic data are stored in the edge node SSD for caching. When the satellite internet speed is not less than 1Mbps, the collected geological data are processed using lossless LZMA algorithm, and the compression ratio is adjusted to 2:1. The backlog queue at the edge node is automatically triggered for uploading, and non-critical data cached during the low-speed phase is re-uploaded in a "first-in, first-out" order.

[0015] Preferably, the physical model is used to simulate geological processes. By deploying multi-source sensors, high-precision surface displacement data, surface crack change data, and groundwater dynamic change data are collected and input into the physical model in real time. After receiving the collected data, the physical model simulates the normal stress, shear stress, and stress gradient of the surface and generates stress time history curves; it simulates the three-dimensional displacement, displacement rate, and deformation acceleration of cracks and generates deformation time series curves; it simulates the depth, length, and width of the underground sliding surface and the internal friction angle and generates evolution time series diagrams; and it obtains a stability coefficient through the coordinated changes of the three factors. When the coefficient is greater than 1, it is judged as stable; when the coefficient is less than 1, it is judged as unstable, and this coefficient serves as the core basis for early warning.

[0016] The stability coefficient is calculated using a model based on the ratio of the shear strength to the shear stress of the geological mass, and the formula is as follows:

[0017] Stability coefficient (K) = Anti-slip force (Fanti) / Slip force (Fslip),

[0018] in:

[0019] The anti-sliding force (F_resistance) is calculated by combining the internal friction angle (φ), sliding surface length (L), and geological body weight (G, which is obtained by multiplying the volume and density of the geological body by high-precision displacement data of the surface) of the underground sliding surface in the evolution time series diagram. The formula F_resistance = G × cosθ × tanφ + C × L is used, where θ is the sliding surface dip angle (obtained by fitting the sliding surface depth and length), and C is the geological body cohesion (corrected based on the average value of the stable segment of shear stress in the stress time history curve).

[0020] The calculation of sliding force (F_slip) is based on the vertical component (h) of the three-dimensional displacement of the crack and the length of the sliding surface (L) in the deformation time history curve, combined with the surface normal stress (σn, taken from the real-time value of the stress time history curve), and is calculated by the formula F_slip = G × sinθ + σn × L × sinθ, where the value of θ is consistent with that in the calculation of anti-slip force.

[0021] The stability coefficient (K) is calculated using the above formula. When K > 1, the geological environment is considered stable; when K < 1, the geological environment is considered unstable.

[0022] Machine learning algorithm model: used to optimize the parameters of the physical model. It extracts historical geological disaster datasets from the system storage unit, specifically including displacement sequences, ground fissure gauge monitoring sequences, rain gauge data, soil moisture data, groundwater level data, and intensity parameters at the time of the disaster within 72 hours before the disaster. The historical data is divided into training and validation sets, input into the machine learning algorithm model and feature extraction is performed. The optimized parameters output by the machine learning algorithm model are fed back to the physical model to dynamically correct the initial parameters in the physical model.

[0023] Preferably, the working process of the hybrid analysis model further includes:

[0024] Dynamic threshold model: Real-time rainfall intensity, soil saturation, surface displacement rate and crack propagation rate are obtained through rain gauges, soil moisture sensors, GNSS receivers and ground fissure meters in the data acquisition module. Combined with the basic threshold matrix built into the dynamic threshold model, the deviation between real-time data and basic thresholds is calculated. When the deviation exceeds ±15%, the threshold correction mechanism is automatically activated.

[0025] Deviation calculation formula:

[0026]

[0027] Preferably, the machine learning algorithm adopts an LSTM and CNN architecture, wherein:

[0028] The LSTM architecture is used to process geological data with time-series characteristics. It continuously collects surface displacement data through deployed displacement sensors to form a continuous displacement time series; it collects rainfall data through rain gauges to construct a time series of rainfall changes over time; and it collects groundwater dynamic change data through groundwater level sensors to form a time series of groundwater level changes. The collected geological data with time-series characteristics is transmitted to the system in real time. After receiving the time series data, the LSTM architecture performs standardization and normalization operations to capture long-term dependencies in the data.

[0029] The CNN architecture is used to process geological data with spatial characteristics. It collects surface crack change data at various monitoring points through crack gauges to form the spatial distribution and variation characteristics of surface cracks; it collects soil moisture data at different spatial points through soil moisture sensors to form the spatial distribution data of soil moisture; it integrates the collected geological data with spatial characteristics to construct a dataset with spatial distribution characteristics; after the CNN architecture obtains the spatial feature data, it performs spatial alignment and gridding on the data, and uses the convolution kernels of the convolutional layers to perform convolution operations on the gridded data to extract the spatial features from the data;

[0030] An integrated transfer learning mechanism is used to train the model in areas where geological disasters have occurred, using historical data collected in the region. When the model is applied to disaster prediction, general feature knowledge related to the geological environment in the trained model is extracted and transferred to the new disaster prediction model.

[0031] Preferably, the data cleaning and preprocessing unit adopts a constraint mechanism. By setting a numerical range for the sensor data, if the collected surface displacement, surface cracks, rainfall, soil moisture and groundwater dynamic data are detected to exceed the preset range, they are marked as abnormal. Data that does not exceed the range are marked as normal and preprocessed.

[0032] Normal data marked by the data cleaning unit is preprocessed. Linear interpolation is used for time series data, and data missing is filled by combining historical data and the mean of adjacent time periods. The Kolmogorov-Smirnov test is used to compare the changes in data distribution and calculate the difference between real-time data and historical patterns. Weighted average adjustment is used for drift data. High-frequency noise is removed by low-pass filter and low-frequency effective signals are retained.

[0033] Preferably, the GIS single-map visualization unit supports 3D Earth scene display and spatial analysis:

[0034] 3D Earth Scene Display: By integrating geographic information system technology, high-precision digital elevation models, geological structure maps, and disaster point distribution data are combined to generate a realistic 3D model, and geological data collected by sensors are connected in real time to update the 3D scene through texture mapping;

[0035] Spatial analysis: Multi-level buffer zones are delineated with the disaster point as the center. The potential impact range is assessed by combining the data analysis module. Multi-level data such as rainfall, lithology, and slope are superimposed to calculate the geological disaster susceptibility index. The displacement-time curve is fitted using the least squares method to predict the future deformation trend of the landslide body.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] 1) This invention effectively solves the problem of low transmission rate of satellite internet in the prior art by using a hybrid transmission protocol of satellite internet and terrestrial network, and supporting edge computing and dynamic adjustment of data compression strategy; the application of edge computing enables data to be preprocessed and analyzed locally at network edge nodes, reducing the amount of data transmission and improving transmission efficiency; and the method of dynamically adjusting the compression ratio according to the real-time fluctuation of satellite internet speed, combined with dynamic spectrum sharing technology, ensures that data transmission remains stable and efficient in remote areas, high altitudes and other areas with poor network conditions, avoiding untimely early warning due to transmission delay;

[0038] 2) By deploying a multi-source sensor network, encompassing various devices such as GNSS receivers, fissure meters, rain gauges, soil moisture sensors, and groundwater level sensors, geological environmental parameters can be collected comprehensively and in real time. These different types of sensors each perform their specific functions, monitoring key data such as high-precision surface displacement, changes in surface fissures, rainfall, soil moisture, and dynamic changes in groundwater. This provides rich and accurate raw data for subsequent analysis and early warning, ensuring the comprehensiveness and detail of monitoring and enabling staff to understand the development trend of potential geological hazards from multiple dimensions.

[0039] 3) This invention combines a physical model with a machine learning algorithm model, which not only preserves the mechanistic description of geological processes but also optimizes parameters through data-driven approaches. This significantly improves the model's adaptability to complex geological environments. The dynamic threshold model automatically adjusts the warning threshold based on real-time rainfall, soil moisture, and other environmental conditions, greatly improving the accuracy of warnings for debris flows induced by extreme rainfall. The transfer learning mechanism transfers model parameters trained in one region to a new region, reducing the need for most of the labeled data and improving the generalization ability of cross-regional disaster prediction, thus solving the "regional dependence" problem of deep learning models.

[0040] 4) This invention integrates the three-dimensional visualization function of a single GIS map, which integrates and displays information such as disaster risk areas, sensor distribution, and rescue forces, enabling decision-makers to make quick decisions on the disaster situation and reduce judgment time. Attached Figure Description

[0041] Figure 1 This is the overall flowchart of the data analysis-based geological hazard monitoring and early warning system;

[0042] Figure 2 This is a flowchart of the data acquisition and transmission module of this system;

[0043] Figure 3 This is the flowchart of the data analysis module of this system;

[0044] Figure 4 This is the self-learning flowchart for this system. Detailed Implementation

[0045] Please see Figure 1 - Figure 4 A geological hazard monitoring and early warning system based on data analysis includes a data acquisition module for real-time acquisition of geological environmental parameters. By deploying a multi-source sensor network, it collects and records data on surface displacement, changes in surface cracks, rainfall, soil moisture, and dynamic changes in groundwater. The multi-source sensor network includes a GNSS receiver, a ground fissure meter, a rain gauge, a soil moisture sensor, and a groundwater level sensor.

[0046] The data transmission module adopts a hybrid transmission protocol of satellite internet and terrestrial network (the terrestrial network uses 5G+CPE equipment, and Beidou-3 short message terminals are deployed in remote areas). It supports edge computing, which can reduce cloud dependence, reduce transmission latency, and dynamically adjust data compression strategies to adapt to fluctuations in satellite internet speed. At the same time, by optimizing the dynamic spectrum sharing technology in the satellite internet transmission protocol, and realizing dynamic preemption of spectrum resources by multiple users through dynamic spectrum sharing technology, the spectrum utilization rate is improved, and the data transmission bottleneck in remote areas is solved.

[0047] The data analysis module uses a hybrid analysis model constructed by combining physical models and machine learning algorithm models to analyze the received geological environmental parameters and generate analysis results based on preset thresholds. The hybrid analysis model includes a physical model, a machine learning algorithm model, and a dynamic threshold model. The physical model is used to simulate geological processes, the machine learning algorithm model is used to optimize the parameters of the physical model, and the dynamic threshold model adjusts the warning threshold based on real-time environmental conditions. The machine learning algorithm model uses LSTM and CNN architectures to process long-sequence geological data and integrates a transfer learning mechanism to improve cross-regional prediction capabilities.

[0048] The early warning module is used to issue early warning information to the region, trigger a multi-level early warning mechanism based on the analysis results, and issue early warning information through multiple channels. The multi-level early warning mechanism includes attention level, warning level, alert level and alarm level.

[0049] The data cleaning and preprocessing unit employs constraint mechanisms to ensure data integrity and handles data loss, drift, and noise anomalies.

[0050] A GIS single-map visualization unit is used to display three-dimensional earth scenes and spatial analysis;

[0051] The system has a self-learning function, continuously receiving new data to optimize the hybrid analysis model and automatically updating the warning thresholds.

[0052] In the data acquisition module, a GNSS receiver, model TrimbleNet R9, is used for high-precision surface displacement monitoring and is fixed to a bedrock outcrop or concrete pier. A ground fissure gauge is used to monitor changes in surface fissures; a vibrating wire fissure gauge (such as the VW-102 model) is used, buried every 15 meters along potential fissure zones, with the sensor axis perpendicular to the fissure direction and the fixed end embedded in the bedrock to a depth of ≥0.5 meters. Rain gauges and soil moisture sensors are used to collect rainfall and soil moisture data in real time, respectively. The rain gauge is a tipping bucket type, installed in an unobstructed open area with the rain collection port 1.2 meters above the ground. Soil moisture sensors (such as the TDR model) are buried at a depth of 80 centimeters, with one set deployed every 50 meters. A groundwater level sensor is used to monitor dynamic changes in groundwater; a borehole is drilled to the aquifer, a pressure-type water level gauge is inserted, the borehole opening is sealed with cement, and a built-in temperature compensation module eliminates the influence of water temperature.

[0053] The data transmission module supports edge computing. By deploying computing resources at network edge nodes, it performs local preprocessing and analysis on the collected data and dynamically adjusts the data compression strategy based on real-time fluctuations in satellite internet speed. When the satellite internet speed is below 1Mbps, the collected surface displacement, surface crack, and rainfall data are processed using wavelet transform algorithm, and the compression ratio is increased to 10:1. Soil moisture and groundwater dynamic data are cached in the edge node SSD. When the satellite internet speed is not less than 1Mbps, the collected geological data are processed using lossless LZMA algorithm, and the compression ratio is adjusted to 2:1. The backlog queue at the edge node is automatically triggered for uploading, and non-critical data cached during the low-speed phase is re-uploaded in a "first-in, first-out" order.

[0054] The working process of the hybrid analysis model is as follows:

[0055] Physical Model: Used to simulate geological processes. By deploying multi-source sensors, high-precision surface displacement data, surface crack change data, and groundwater dynamic change data are collected and input into the physical model in real time. After receiving the collected data, the physical model simulates the normal stress, shear stress, and stress gradient of the surface and generates stress time history curves; it simulates the three-dimensional displacement, displacement rate, and deformation acceleration of cracks and generates deformation time series curves; it simulates the depth, length, and width of the underground sliding surface and the internal friction angle and generates evolution time series diagrams; and it derives a stability coefficient through the coordinated changes of these three factors. When the coefficient is greater than 1, the system is considered stable; when the coefficient is less than 1, the system is considered unstable, and this coefficient serves as the core basis for early warning.

[0056] The stability coefficient is calculated using a model based on the ratio of the shear strength to the shear stress of the geological mass, and the formula is as follows:

[0057] Stability coefficient (K) = Anti-slip force (Fanti) / Slip force (Fslip),

[0058] in:

[0059] The anti-sliding force (F_resistance) is calculated by combining the internal friction angle (φ), sliding surface length (L), and geological body weight (G, which is obtained by multiplying the volume and density of the geological body by high-precision displacement data of the surface) of the underground sliding surface in the evolution time series diagram. The formula F_resistance = G × cosθ × tanφ + C × L is used, where θ is the sliding surface dip angle (obtained by fitting the sliding surface depth and length), and C is the geological body cohesion (corrected based on the average value of the stable segment of shear stress in the stress time history curve).

[0060] The sliding force (F_slip) is calculated based on the vertical component (h) of the three-dimensional displacement of the crack and the length of the sliding surface (L) in the deformation time history curve, combined with the surface normal stress (σn, taken from the real-time value of the stress time history curve), using the formula F_slip = G × sinθ + σn × L × sinθ, where the value of θ is consistent with that in the calculation of the anti-slip force.

[0061] The stability coefficient (K) is calculated using the above formula. When K > 1, the geological environment is considered stable; when K < 1, the geological environment is considered unstable.

[0062] Machine learning algorithm model: used to optimize the parameters of the physical model. It extracts historical geological disaster datasets from the system storage unit, specifically including displacement sequences, fissure gauge monitoring sequences, rain gauge data, soil moisture data, groundwater level data, and intensity parameters at the time of the disaster within 72 hours before the disaster. The historical data is divided into training and validation sets, input into the machine learning algorithm model and feature extraction is performed. The optimized parameters output by the machine learning algorithm model are fed back to the physical model to dynamically correct the initial parameters in the physical model.

[0063] Dynamic threshold model: Real-time rainfall intensity, soil saturation, surface displacement rate and crack propagation rate are obtained through rain gauges, soil moisture sensors, GNSS receivers and ground fissure meters in the data acquisition module. Combined with the basic threshold matrix built into the dynamic threshold model, the deviation between real-time data and basic thresholds is calculated. When the deviation exceeds ±15%, the threshold correction mechanism is automatically activated.

[0064] Deviation calculation formula:

[0065]

[0066] The machine learning algorithm model uses LSTM and CNN architectures, where:

[0067] The LSTM architecture is used to process geological data with time-series characteristics and can capture long-term dependencies in the data;

[0068] Specifically, surface displacement data is continuously collected by deployed displacement sensors to form a continuous displacement time series; rainfall data is collected by rain gauges to construct time series data on rainfall changes over time; groundwater level data is collected by groundwater level sensors to form a time series data on groundwater level changes; the collected geological data with time series characteristics are transmitted to the system in real time, and the LSTM architecture receives the time series data, performs standardization and normalization operations, and captures long-term dependencies in the data;

[0069] CNN architecture is used to process geological data with spatial features and can extract spatial features from the data;

[0070] Specifically, surface crack change data are collected at various monitoring points using crack gauges to form the spatial distribution and variation characteristics of surface cracks; soil moisture data are collected at different spatial points using soil moisture sensors to form spatial distribution data of soil moisture; the collected geological data with spatial characteristics are integrated to construct a dataset with spatial distribution characteristics; after the CNN architecture obtains the spatial feature data, the data is spatially aligned and gridded; the convolution kernels of the convolutional layer are used to perform convolution operations on the gridded data to extract the spatial features from the data.

[0071] The integrated transfer learning mechanism transfers knowledge learned in one type of disaster to the prediction of other types of disasters, thereby improving cross-regional prediction capabilities (transferring the parameters of the LSTM model trained in region A to the model in region B, fine-tuning only the output layer, and reducing the amount of training data from 2000 sets to 400 sets).

[0072] The LSTM architecture is responsible for processing the time-series features of geological data, capturing the long-term dependencies of parameters such as surface displacement and groundwater level over time through memory units; the CNN architecture is responsible for extracting the spatial features of geological data, identifying the spatial correlation between different monitoring points through convolutional layers; the two achieve joint modeling through a feature fusion layer: the temporal feature vector output by LSTM and the spatial feature map output by CNN are concatenated in the middle layer of the model, and combined with the "spatiotemporal coupling" characteristics of geological disasters, a comprehensive feature with both temporal trends and spatial distribution patterns is generated, and then the prediction result is output through a fully connected layer. At the same time, the model integrates a transfer learning mechanism, which transfers the spatiotemporal feature extraction logic trained in one region to other regions, and improves the cross-regional prediction capability by fine-tuning to adapt to different geological environments.

[0073] The specific content of the multi-level early warning mechanism is as follows:

[0074] Alert Level: Remind people in the surrounding area to stay away from the disaster area. Send text messages through the Alibaba Cloud SMS platform to mobile phone users within 5 kilometers of the risk point.

[0075] Warning level: Patrol personnel intervene to warn the public;

[0076] Alert Level: A cordon is set up, and professional personnel conduct an on-site investigation;

[0077] Alarm Level: Notify staff and construction departments in the property area to activate the emergency plan (activate the emergency broadcast system, play voice messages, and send a stop-work order to the construction unit).

[0078] The data cleaning and preprocessing unit adopts a constraint mechanism. By setting a numerical range for sensor data, if the collected surface displacement, surface cracks, rainfall, soil moisture and groundwater dynamic data are detected to exceed the preset range, they are marked as abnormal. Data that does not exceed the range are marked as normal and preprocessed.

[0079] Normal data marked by the data cleaning unit is preprocessed. Linear interpolation is used for time series data, and data missing is filled by combining historical data and the mean of adjacent time periods. The Kolmogorov-Smirnov test is used to compare the changes in data distribution and calculate the difference between real-time data and historical patterns. Weighted average adjustment is used for drift data. High-frequency noise is removed by low-pass filter and low-frequency effective signals are retained.

[0080] The GIS One-Map Visualization Unit integrates geographic information system technology to combine geological disaster-related data with geospatial information, displaying the location, extent, and severity of potential geological disaster sites in a three-dimensional Earth scene. It also supports spatial analysis functions to analyze the impact range of geological disasters and predict their development trends.

[0081] Specifically, SuperMap iDesktop3D is used to construct the 3D scene, and the data includes:

[0082] Topographic data: ASTERGDEM 30m resolution digital elevation model;

[0083] Disaster data: Real-time risk areas;

[0084] Sensor distribution: point feature layer, labeled with icons according to type;

[0085] The self-learning function establishes a data feedback mechanism to continuously receive new monitoring data and early warning results, optimize and update the hybrid analysis model, automatically adjust the model's parameters and structure, and automatically update the early warning thresholds based on new data and actual conditions, thereby improving the system's monitoring and early warning capabilities.

[0086] Dynamic spectrum sharing technology in satellite internet transmission protocols dynamically allocates spectrum resources by monitoring the spectrum usage of satellite internet in real time, enabling multi-user spectrum resource sharing, thereby improving spectrum utilization and solving the data transmission bottleneck problem in remote areas.

[0087] Steps for using this invention:

[0088] I. System Deployment Phase

[0089] Based on the geological structural characteristics of the monitoring area, such as fault distribution and slope orientation, a multi-source sensor network is deployed at potential disaster hazard points, specifically including...

[0090] The system includes a GNSS receiver (for acquiring high-precision displacement data of the ground surface), a ground fission meter (for monitoring changes in surface fissures), a rain gauge (for recording rainfall), a soil moisture sensor (for measuring soil saturation), and a groundwater level sensor (for monitoring dynamic changes in groundwater). Edge computing devices and SSD storage units are configured at network edge nodes to ensure that the communication distance with the sensors is ≤500 meters, guaranteeing real-time data transmission to the edge nodes.

[0091] Establish a hybrid transmission link combining satellite internet and terrestrial networks, and preset transmission strategy parameters in the system backend:

[0092] Set the satellite internet speed threshold (1Mbps), configure the switching trigger conditions for wavelet transform compression algorithm (compression ratio 10:1) and lossless LZMA algorithm (compression ratio 2:1); and configure edge node data caching rules to clarify the classification and storage logic of "critical data" (surface displacement, crack change, rainfall) and "non-critical data" (soil moisture, groundwater dynamics);

[0093] Activate the GIS One Map Visualization Unit, import the high-precision digital elevation model (DEM), geological structure map and historical disaster point data of the monitoring area, and generate a basic model of the three-dimensional earth scene;

[0094] In the data analysis module, input the initial parameters of the physical model (such as the friction angle and cohesion of the soil and rock), and load the historical geological disaster dataset (including monitoring data of the 72 hours before the disaster in the past 10 years and disaster intensity parameters); in the dynamic threshold model, configure the basic threshold matrix and preset the initial thresholds such as rainfall intensity and surface displacement rate.

[0095] II. Data Acquisition and Preprocessing Stage

[0096] The sensor network initiates data acquisition at a preset sampling frequency and transmits the raw data to the edge nodes. The edge computing device monitors the satellite internet transmission rate in real time and dynamically executes data processing strategies: when the rate is <1Mbps: critical data is compressed using a wavelet transform algorithm and transmitted first, while non-critical data is temporarily stored in the edge node's SSD cache; when the rate is ≥1Mbps: all data is compressed using a lossless LZMA algorithm and transmitted, while triggering a "first-in-first-out" queue to re-transmit the cached non-critical data.

[0097] The data cleaning and preprocessing unit performs constraint mechanism verification on the received data: for data such as surface displacement and crack changes, it compares them with preset numerical ranges. Data that exceeds the range is marked as abnormal data (temporarily stored in the abnormal database for manual review), and data that does not exceed the range is marked as normal data.

[0098] Preprocessing of normal data: Linear interpolation is used to fill missing values ​​in the time series by combining the mean of historical data from adjacent time periods; the distribution difference between real-time data and historical patterns is calculated using the Kolmogorov-Smirnov test, and drift data is adjusted using a weighted average method; high-frequency interference signals are filtered out using a low-pass filter, retaining low-frequency effective data; III. Data Analysis and Model Calculation Stage

[0099] Physical model calculation: Input the preprocessed real-time data (surface displacement, crack changes, groundwater dynamics) into the physical model to simulate the following parameters and generate visualization curves: normal stress, shear stress, and stress gradient of the surface (generating stress time history curves); three-dimensional displacement, displacement rate, and deformation acceleration of cracks (generating deformation time series curves); depth, length, width, and internal friction angle of the underground sliding surface (generating evolution time series diagrams); calculate the stability coefficient based on the above parameters, and mark the state as "unstable" when the coefficient is <1.

[0100] Machine learning model optimization:

[0101] Historical geological disaster datasets were extracted from the system storage unit and divided into a training set (70%) and a validation set (30%). The datasets were then input into a hybrid LSTM and CNN architecture: LSTM processed time series data (such as displacement sequences and rainfall time series data) to capture long-term dependencies; CNN processed spatial feature data (such as spatial distribution of cracks and spatial distribution of soil moisture) and extracted spatial features through convolutional layers; the model output optimized parameters, which were fed back to the physical model to dynamically correct its initial parameters.

[0102] Dynamic threshold adjustment and early warning condition judgment

[0103] The dynamic threshold model acquires data such as rainfall intensity, soil saturation, surface displacement rate, and crack propagation rate in real time, and calculates the deviation from the basic threshold.

[0104] Deviation calculation formula:

[0105]

[0106] When the deviation exceeds ±15%, the threshold correction mechanism is automatically activated to update the basic threshold matrix;

[0107] By combining the stability coefficient output by the physical model and the correction results of the dynamic threshold model, it is determined whether the early warning conditions are met.

[0108] IV. Early Warning Issuance and Visual Display Phase

[0109] The early warning module triggers corresponding levels of warnings based on the analysis results (e.g., blue warning: stability coefficient 0.8-1.0; red warning: stability coefficient <0.6), and releases the warnings through multiple channels.

[0110] Send early warning information to residents within the monitoring area via SMS and WeChat public account, including the type of disaster, the expected scope of impact, and evacuation advice;

[0111] Send early warning reports containing detailed data (including stress curves, deformation trend diagrams, etc.) to local emergency management departments;

[0112] In the GIS single-map visualization unit, the warning area is highlighted, and the hazard level and key monitoring point data are marked;

[0113] A single GIS map visualization unit updates a 3D Earth scene in real time: Real-time monitoring data is overlaid onto a base model via texture mapping, intuitively displaying the spatial distribution of potential disaster sites; spatial analysis is then conducted.

[0114] Three-level buffer zones of 50m, 100m, and 200m are delineated centered on the warning point to assess the potential impact range; multiple layers of data such as rainfall, lithology, and slope are superimposed to calculate the geological hazard susceptibility index; the least squares method is used to fit the displacement-time curve to predict the future deformation trend of landslides and other hazard bodies, providing a basis for emergency decision-making.

[0115] V. System Maintenance and Iteration Phase

[0116] The system automatically stores all monitoring data, analysis results, and early warning records in the database and backs them up regularly; it extracts newly generated historical data every quarter to supplement the training set of the machine learning model and retrains the model to optimize the accuracy of parameter prediction.

[0117] Monthly inspections are conducted on the sensor network to calibrate equipment accuracy. Based on seasonal changes (such as before the rainy season) and changes in geological conditions (such as slope conditions after heavy rainfall), the basic threshold matrix of the dynamic threshold model is manually reviewed and adjusted to ensure the accuracy of early warnings.

[0118] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A geological hazard monitoring and early warning system based on data analysis, characterized in that, include: The data acquisition module is used to collect geological environmental parameters in real time. By deploying a multi-source sensor network, it collects and records data on surface displacement, changes in surface cracks, rainfall, soil moisture, and dynamic changes in groundwater. The data transmission module uses a hybrid transmission protocol combining satellite internet and terrestrial networks to transmit the collected geological environmental parameters and dynamically adjusts the compression strategy of the transmitted data based on the real-time rate of the satellite internet. The data analysis module uses a hybrid analysis model, which combines physical models and machine learning algorithms, to analyze the received geological environmental parameters and generate analysis results based on preset thresholds. The early warning module is used to issue early warning information to the region, trigger multi-level early warning mechanisms based on analysis results, and release early warning information through multiple channels.

2. The geological hazard monitoring and early warning system based on data analysis according to claim 1, characterized in that, Also includes: The data cleaning and preprocessing unit employs constraint mechanisms to ensure data integrity and handles data loss, drift, and noise anomalies. A GIS single-map visualization unit is used to display three-dimensional earth scenes and spatial analysis.

3. The geological hazard monitoring and early warning system based on data analysis according to claim 1, characterized in that, The data transmission module supports edge computing. By deploying computing resources at network edge nodes, it performs local preprocessing and analysis on the collected data and dynamically adjusts the data compression strategy according to the real-time fluctuations of the satellite Internet speed. When the satellite Internet speed is lower than 1Mbps, the collected surface displacement, surface crack, and rainfall data are processed using wavelet transform algorithm and the compression ratio is increased to 10:

1. The dynamic data of soil moisture and groundwater are stored in the edge node SSD for caching. When the satellite internet speed is no less than 1Mbps, the collected geological data will be processed using the lossless LZMA algorithm, and the compression ratio will be adjusted to 2:

1. The backlog queue of edge nodes will be automatically triggered for uploading, and non-critical data cached during the low-speed phase will be re-uploaded in a "first-in, first-out" order.

4. The geological hazard monitoring and early warning system based on data analysis according to claim 1, characterized in that, The working process of the hybrid analysis model is as follows: Physical model: used to simulate geological processes. By deploying multi-source sensors, high-precision displacement data of the surface, surface crack change data, and groundwater dynamic change data are collected and input into the physical model in real time. After receiving the collected data, the physical model simulates the normal stress, shear stress, and stress gradient of the surface and generates stress time history curves. The system simulates the three-dimensional displacement, displacement rate, and deformation acceleration of cracks and generates deformation time-series curves; it also simulates the depth, length, width, and internal friction angle of underground sliding surfaces and generates evolution time-series diagrams; and it derives a stability coefficient through the coordinated changes of these three parameters. When the coefficient is greater than 1, the system is considered stable; when the coefficient is less than 1, the system is considered unstable, and this coefficient serves as the core basis for early warning. Machine learning algorithm model: used to optimize the parameters of the physical model. It extracts historical geological disaster datasets from the system storage unit, specifically including displacement sequences, ground fissure gauge monitoring sequences, rain gauge data, soil moisture data, groundwater level data, and intensity parameters at the time of the disaster within 72 hours before the disaster. The historical data is divided into training and validation sets, input into the machine learning algorithm model and feature extraction is performed. The optimized parameters output by the machine learning algorithm model are fed back to the physical model to dynamically correct the initial parameters in the physical model.

5. The geological hazard monitoring and early warning system based on data analysis according to claim 4, characterized in that, The working process of the hybrid analysis model also includes: Dynamic threshold model: Real-time rainfall intensity, soil saturation, surface displacement rate and crack propagation rate are obtained through rain gauges, soil moisture sensors, GNSS receivers and ground fissure meters in the data acquisition module. Combined with the basic threshold matrix built into the dynamic threshold model, the deviation between real-time data and basic thresholds is calculated. When the deviation exceeds ±15%, the threshold correction mechanism is automatically activated. Deviation calculation formula:

6. The geological hazard monitoring and early warning system based on data analysis according to claim 4, characterized in that, The machine learning algorithm employs an LSTM and CNN architecture, wherein: The LSTM architecture is used to process geological data with time-series characteristics. It continuously collects surface displacement data through deployed displacement sensors to form a continuous displacement time series; it collects rainfall data through rain gauges to construct a time series of rainfall changes over time; and it collects groundwater dynamic change data through groundwater level sensors to form a time series of groundwater level changes. The collected geological data with time-series characteristics is transmitted to the system in real time. After receiving the time series data, the LSTM architecture performs standardization and normalization operations to capture long-term dependencies in the data. The CNN architecture is used to process geological data with spatial characteristics. It collects surface crack change data at various monitoring points through crack gauges to form the spatial distribution and variation characteristics of surface cracks; it collects soil moisture data at different spatial points through soil moisture sensors to form the spatial distribution data of soil moisture; it integrates the collected geological data with spatial characteristics to construct a dataset with spatial distribution characteristics; after the CNN architecture obtains the spatial feature data, it performs spatial alignment and gridding on the data, and uses the convolution kernels of the convolutional layers to perform convolution operations on the gridded data to extract the spatial features from the data; An integrated transfer learning mechanism is used to train the model in areas where geological disasters have occurred, using historical data collected in the region. When the model is applied to disaster prediction, general feature knowledge related to the geological environment in the trained model is extracted and transferred to the new disaster prediction model.

7. The geological hazard monitoring and early warning system based on data analysis according to claim 2, characterized in that, The data cleaning and preprocessing unit adopts a constraint mechanism. By setting a numerical range for sensor data, if the collected surface displacement, surface cracks, rainfall, soil moisture and groundwater dynamic data are detected to exceed the preset range, they are marked as abnormal. Data that does not exceed the range are marked as normal and preprocessed. Normal data marked by the data cleaning unit is preprocessed. Linear interpolation is used for time series data, and combined with historical data, the mean of data from adjacent time periods is filled to handle missing data. By comparing changes in data distribution and calculating the difference between real-time data and historical patterns, a weighted average adjustment is applied to the drift data; high-frequency noise is removed by a low-pass filter, while retaining the effective low-frequency signal.

8. The geological hazard monitoring and early warning system based on data analysis according to claim 2, characterized in that, The GIS single-map visualization unit supports 3D Earth scene display and spatial analysis: 3D Earth Scene Display: By integrating geographic information system technology, high-precision digital elevation models, geological structure maps, and disaster point distribution data are combined to generate a realistic 3D model, and geological data collected by sensors are connected in real time to update the 3D scene through texture mapping; Spatial analysis: Multi-level buffer zones are delineated with the disaster point as the center. The potential impact range is assessed by combining the data analysis module. Multi-level data such as rainfall, lithology, and slope are superimposed to calculate the geological disaster susceptibility index. The displacement-time curve is fitted using the least squares method to predict the future deformation trend of the landslide body.

Images