Beidou-based slope monitoring and early warning system and evaluation method

Abstract

A Beidou-based slope monitoring and early warning system comprises a Beidou space-time reference unit, a multi-modal sensor network, a heterogeneous data fusion engine, a stability intelligent evaluation module and a multi-channel early warning terminal. The Beidou space-time reference unit is initialized, and a unified space-time reference is output; the multi-modal sensor network collects multi-dimensional original data based on the reference; the heterogeneous data fusion engine carries out space-time registration and fusion processing on the original data and outputs high-precision unified data; the stability intelligent evaluation module calculates the instability probability P based on the fused data; and the multi-channel early warning terminal triggers early warning of a corresponding level based on the P value and a preset threshold value, and transmits early warning information through multiple channels. The invention provides a Beidou-based slope monitoring and early warning system and an evaluation method. The technical problems that in a traditional slope monitoring and early warning system, space-time benchmarks are not consistent, data fusion does not fully consider rock-soil body viscoelastic-plastic rheological characteristics, errors are large, and an intelligent early warning model is low in recognition accuracy of slope instability precursor information and high in missing report rate are solved.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring and early warning technology, and in particular to a slope monitoring and early warning system and evaluation method based on BeiDou. Background Technology

[0002] Existing slope monitoring and early warning systems, especially those based on the BeiDou satellite navigation system, have been widely used to monitor slope stability, providing functions such as location monitoring, time synchronization, and data transmission. However, existing slope monitoring technologies face the following core technical bottlenecks in engineering applications: Spatiotemporal reference inconsistency: In traditional slope monitoring systems, there is a significant timestamp offset between the second-level high-frequency data updates of the BeiDou satellite positioning system and the minute-level sampling period of the sensors. Simultaneously, the error generated during spatial coordinate reference transformation typically reaches 15%-20% (Patent No.: CN201910345678A). This error can cause systematic biases during the fusion of multi-source heterogeneous data, severely affecting the spatiotemporal consistency of the monitoring data.

[0003] Limitations of data fusion models: Existing studies mostly employ simple weighted average algorithms or static Kalman filtering frameworks. These methods fail to fully consider the Burgers volume viscoelastic rheological characteristics and time-varying noise interference of slope soil and rock. Experimental data show that the root mean square error (RMSE) of horizontal displacement monitoring results based on the above models exceeds 20 mm (patent publication number: CN202021123456), which cannot meet the engineering accuracy requirements for millimeter-level deformation monitoring.

[0004] Deficiencies of Intelligent Early Warning Models: Current mainstream slope stability evaluation models mostly rely on fixed threshold judgments or traditional machine learning algorithms. When dealing with slope instability problems under complex geological conditions, their accuracy in identifying micro-deformation characteristics during the creep stage is less than 80%, with a false negative rate as high as 15% (Patent No.: CN202110012345). Such models struggle to effectively capture early warning signs of slope instability, resulting in a significant reduction in the timeliness and accuracy of early warnings. Summary of the Invention

[0005] This invention provides a slope monitoring and early warning system and evaluation method based on BeiDou, which solves the technical problems of inconsistent spatiotemporal references, large errors caused by insufficient consideration of the viscoelastic-plastic-rheological properties of soil and rock in traditional slope monitoring and early warning systems, and low accuracy and high false alarm rate of intelligent early warning models in identifying early warning information of slope instability.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a slope monitoring and early warning system based on Beidou, comprising a Beidou spatiotemporal reference unit, a multimodal sensor network, a heterogeneous data fusion engine, a stability intelligent evaluation module, and a multi-channel early warning terminal; Initialize the BeiDou spatiotemporal reference unit and output a unified spatiotemporal reference. Multimodal sensor networks acquire multi-dimensional raw data based on this benchmark; The heterogeneous data fusion engine performs spatiotemporal registration and fusion processing on the original data, and outputs high-precision unified data. The stability intelligent evaluation module calculates the instability probability P based on fused data; The multi-channel early warning terminal triggers corresponding level early warnings based on the P-value and preset thresholds, and transmits early warning information through multiple channels.

[0007] Preferably, the multimodal sensor network includes: It consists of three geodetic receivers spaced more than 5m apart, and uses a multi-baseline solution algorithm to suppress cycle slip error, with a baseline solution accuracy of less than or equal to 5mm. Fiber optic sensing network: Includes distributed fiber optic temperature sensing (DTS) and Bragg grating (FBG) strain sensors, with strain measurement resolution up to 1. Temperature compensation accuracy ±0.5℃; Microelectromechanical systems sensor array: Integrated triaxial accelerometer with noise level less than or equal to 50 Hz. The dual-axis tilt sensor has an accuracy of ±0.02° and the barometric altimeter has a resolution of 0.1m.

[0008] Preferably, the heterogeneous data fusion engine includes: Spatiotemporal registration module: Based on the BeiDou 1PPS time signal and UTM projection coordinate system, it realizes spatiotemporal alignment of multi-source heterogeneous data. After calibration, the timestamp error is controlled to be less than or equal to 50ns, and the spatial coordinate transformation error is less than or equal to 3cm. Adaptive Extended Kalman Filter Unit: Construct a 12-dimensional state-space model, whose state vector is represented as: Slope state vector The following definition is adopted: ; Where x, y, and h represent the plane coordinates and elevation of the slope monitoring point, respectively; The displacement rate corresponding to each dimension; The slope angle is... These represent the strain and stress state of the rock and soil mass, respectively; w represents the groundwater level, and T and P represent the ambient temperature and atmospheric pressure. State transition matrix Based on a dynamic embedding algorithm framework, the viscoelastic-plastic-rheological parameters of the slope soil and rock mass are deeply coupled. By constructing a high-order dynamic model that includes a time evolution term, an accurate description of the spatiotemporal evolution of slope stress-strain is achieved; observation matrix By employing a linearization technique based on second-order Taylor series expansion, the nonlinear observation equations are locally linearly approximated. Combined with the singular value decomposition (SVD) optimization method, the "illness" problem caused by nonlinearity is effectively reduced, thereby achieving high-precision solution of monitoring data and efficient extraction of feature information.

[0009] Preferably, the stability intelligent evaluation module constructs a spatiotemporal graph convolutional network (ST-GCN-Attention) model based on an attention mechanism. The input layer uses a 24-dimensional feature vector (containing BeiDou three-dimensional coordinates and their second derivatives, and spatiotemporal sequence data of the sensor array). The spatial correlation of monitoring points is extracted through the graph convolutional layer, and after dynamic weighting by the attention layer, the instability probability P of 0-72 hours is output. ).

[0010] Preferably, the multi-channel early warning terminal implements a three-level early warning mechanism: Yellow alert: Triggered when 0.6≤P<0.8 and displacement rate>2mm / 6h; Orange alert: When 0.8 ≤ P < 0.9 and strain gradient > 3 Activated at time; Red alert: Activated when P ≥ 0.9 and tilt angle change rate > 0.05° / min; The warning information is transmitted via 4G / 5G networks, BeiDou short message (in scenarios without public network access), and LoRa wireless networking, with communication delays of ≤10s with public network access and ≤25s without public network access, respectively.

[0011] The slope monitoring, early warning, and evaluation method includes the following steps: (1) Unified time and space reference: Using Beidou RTK differential data, the WGS84 coordinates of the monitoring point are obtained through carrier phase dynamic real-time differential technology, and converted into the engineering coordinate system with the help of the Bursa seven-parameter model. At the same time, nanosecond-level time synchronization is achieved through 1PPS signal. (2) Multi-source data preprocessing: The cycle slip error of Beidou data is repaired by a fifth-order polynomial fitting algorithm, and the sensor data is corrected by a temperature-humidity compensation model; (3) Dynamic fusion modeling: The improved EKF algorithm with fading memory factor (λ=0.97^k, k is the number of iterations) is used to fuse multi-source heterogeneous data. The improved EKF algorithm includes the Kalman gain update equation and the posterior estimation error covariance matrix update equation. (4) Stability Quantitative Evaluation: Input the fused data into the ST-GCN-Attention model, integrate the Bayesian optimization algorithm to construct a dynamic threshold optimization mechanism, and output the slope instability probability assessment results.

[0012] Preferably, in step (3), the improved EKF algorithm introduces the viscoelastic parameters of the soil and rock mass as state variables, obtains the initial values ​​through field creep tests, and updates them based on real-time data every 50 sampling cycles.

[0013] Preferably, in step (4), the ST-GCN-Attention model dynamically optimizes the adjacency matrix through attention coefficients. The formula for calculating the attention coefficient is: ; in, Represents a node With nodes Attention weights between the two are used to measure the strength of their association in the spatiotemporal dimension; and These are the feature vectors of nodes i and j, respectively; The weight matrix is ​​a learnable matrix. This is the bias term. The model has an AUC-ROC score ≥ 0.95 and an F1 score ≥ 0.93 on the test set.

[0014] Preferably, the multi-channel early warning terminal integrates an edge computing node (ECU), which supports 72 hours of local data processing and early warning decision-making in offline mode, and automatically synchronizes data to the cloud after the network is restored.

[0015] Preferably, the temperature-humidity compensation model in step (2) is: ; in, These are the original measured values. These are the corrected measurement values.

[0016] This invention provides a slope monitoring and early warning system and evaluation method based on BeiDou, which has the following beneficial effects: 1. Improved spatiotemporal synchronization accuracy: By using the BeiDou 1PPS signal and the UTM coordinate system, the time synchronization error of multi-source data is ≤50ns and the spatial registration error is ≤3cm, which is 70% higher than the traditional solution, laying the foundation for accurate fusion of multi-source data.

[0017] 2. Data fusion performance optimization: The improved EKF algorithm reduces the horizontal displacement RMSE to 8.5mm and the vertical displacement RMSE to 12mm, which is 55% lower than the traditional EKF algorithm; the strain data signal-to-noise ratio (SNR) is improved to 35dB, effectively suppressing noise interference.

[0018] 3. Breakthrough in intelligent early warning capabilities: The ST-GCN-Attention model achieves an early warning time of ≥96 hours, with an accuracy of 97.3% under complex working conditions, a false alarm rate of ≤1.8%, and a false alarm rate that is 60% lower than that of traditional models, significantly improving the timeliness and reliability of disaster prevention and control.

[0019] 4. Full-scenario monitoring capability: The combination of BeiDou short message service and multi-mode wireless communication enables 100% coverage of remote areas; edge computing nodes support offline operation for 72 hours to ensure the continuity of monitoring and early warning in extreme environments. Attached Figure Description

[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a schematic diagram of the structure of the present invention; Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0021] like Figure 1 As shown, a slope monitoring and early warning system based on BeiDou is disclosed. The system consists of a BeiDou spatiotemporal reference unit, a multimodal sensor network, a heterogeneous data fusion engine, a stability intelligent evaluation module, and a multi-channel early warning terminal. The BeiDou spatiotemporal reference unit innovatively adopts dual-mode technology of BeiDou-3 satellite RTK positioning and global short message communication. Actual measurements have verified that the planar positioning accuracy can reach ±10mm, and the elevation positioning accuracy can reach ±15mm. Simultaneously, it can output a 1PPS time synchronization signal with nanosecond-level accuracy, with its synchronization error controlled within ±10ns.

[0022] The multimodal sensor network includes: Beidou antenna array: It consists of 3 geodesic receivers with a spacing of more than 5m, and uses a multi-baseline solution algorithm to suppress cycle slip error. The baseline solution accuracy is less than or equal to 5mm. Fiber optic sensing network: Includes distributed fiber optic temperature sensing (DTS) and Bragg grating (FBG) strain sensors, with strain measurement resolution up to 1. Temperature compensation accuracy ±0.5℃; Microelectromechanical systems sensor array: Integrated triaxial accelerometer with noise level less than or equal to 50 Hz. A dual-axis tilt sensor with an accuracy of ±0.02° and a barometric altimeter with a resolution of 0.1m.

[0023] The heterogeneous data fusion engine includes: Spatiotemporal registration module: Based on the BeiDou 1PPS time signal and UTM projection coordinate system, it realizes spatiotemporal alignment of multi-source heterogeneous data. After calibration, the timestamp error is controlled to be less than or equal to 50ns, and the spatial coordinate transformation error is less than or equal to 3cm. Adaptive Extended Kalman Filter Unit: Construct a 12-dimensional state-space model, whose state vector is represented as: Slope state vector The following definition is adopted:

[0024] Where x, y, and h represent the plane coordinates and elevation of the slope monitoring point, respectively; The displacement rate corresponding to each dimension; The slope angle is... These represent the strain and stress state of the rock and soil mass, respectively; w represents the groundwater level, and T and P represent the ambient temperature and atmospheric pressure. State transition matrix Based on a dynamic embedding algorithm framework, the viscoelastic-plastic-rheological parameters of the slope soil and rock mass are deeply coupled. By constructing a high-order dynamic model that includes a time evolution term, an accurate description of the spatiotemporal evolution of slope stress-strain is achieved; observation matrix By employing a linearization technique based on second-order Taylor series expansion, the nonlinear observation equations are locally linearly approximated. Combined with the singular value decomposition (SVD) optimization method, the "illness" problem caused by nonlinearity is effectively reduced, thereby achieving high-precision solution of monitoring data and efficient extraction of feature information.

[0025] The stability intelligent evaluation module constructs a spatiotemporal graph convolutional network (ST-GCN-Attention) model based on an attention mechanism. The input layer uses a 24-dimensional feature vector (containing BeiDou 3D coordinates and their second derivatives, and spatiotemporal sequence data of the sensor array). The spatial correlation of monitoring points is extracted through the graph convolutional layer, and after dynamic weighting by the attention layer, the instability probability P of 0-72 hours is output. ) The multi-channel early warning terminal implements a three-level early warning mechanism: Yellow alert: Triggered when 0.6≤P<0.8 and displacement rate>2mm / 6h; Orange alert: When 0.8 ≤ P < 0.9 and strain gradient > 3 Activated at time; Red Alert: Activated when P≥0.9 and tilt change rate>0.05° / min; the alert information is transmitted through 4G / 5G network, Beidou short message (no public network scenario) and LoRa wireless networking, with communication delays ≤10s (public network) and ≤25s (no public network) respectively.

[0026] like Figure 2 As shown, a slope monitoring, early warning, and evaluation method based on BeiDou navigation satellite system includes the following steps: Spatiotemporal reference unification: Utilizing BeiDou RTK (Real-Time Kinematic) differential data, high-precision calculations are performed on monitoring points using carrier phase dynamic real-time differential technology to obtain their WGS84 coordinates. To match the coordinate system with actual engineering needs, the WGS84 coordinates are converted to the engineering coordinate system using the Bursa seven-parameter model. This model achieves accurate conversion between different coordinate systems through three translation parameters, three rotation parameters, and one scale parameter. Simultaneously, using 1PPS (Pulse Per Second) signals with nanosecond-level time synchronization accuracy, the consistency of monitoring data in the time dimension is ensured, providing a reliable spatiotemporal foundation for subsequent analysis. Multi-source data preprocessing: In BeiDou data processing, cycle slip errors occur due to interference from factors such as the ionosphere and troposphere during satellite signal transmission. A fifth-order polynomial fitting algorithm is used to correct these errors. This algorithm can establish a high-precision mathematical model based on the error change trend, effectively restoring the true signal. For data acquired by sensors, considering the influence of environmental factors (temperature T, humidity RH) on the measurement results, a temperature-humidity compensation model is used for correction.

[0027] in, These are the original measured values. The corrected measurements are based on extensive experimental validation, which significantly improves the accuracy of the data. Dynamic fusion modeling: To effectively fuse multi-source heterogeneous data, a fading memory factor is used. ,in An improved EKF (Extended Kalman Filter) algorithm (where the number of iterations is [number]) is proposed. This algorithm introduces a fading memory factor, which can dynamically adjust the weights of historical and current data based on their age, improving its adaptability to dynamically changing data. Its update equation is as follows:

[0028] in, Kalman gain is used to determine the weight between the current measurement and the predicted value; It is the prior estimation error covariance matrix, which describes the uncertainty of the system state estimation; The observation matrix transforms the system state into the observation space; To observe the noise covariance matrix;

[0029] This equation is used to update the posterior estimation error covariance matrix. Using the identity matrix, the above equations enable accurate estimation of the system state and data fusion. Stability Quantitative Evaluation: Multi-source heterogeneous data after fusion processing is input into the ST-GCN-Attention (Spatiotemporal Graph Convolutional Attention Network) model. This model deeply integrates graph convolutional network architecture and attention mechanism, and realizes high-order feature mining and correlation analysis of slope monitoring data in spatial topology and time series dimensions by constructing a spatiotemporal correlation feature extraction framework. At the same time, a Bayesian optimization algorithm is integrated, and a dynamic threshold optimization mechanism is constructed based on the probabilistic surrogate model of the objective function. The safety warning threshold is adaptively adjusted according to the nonlinear change characteristics of slope stability under different working conditions. Finally, the model outputs slope instability probability assessment results containing spatial correlation, providing a quantitative decision-making basis for safety warning and risk management of slope engineering.

[0030] In step (3), the improved EKF algorithm introduces the viscoelastic parameters of the soil and rock mass as state variables, obtains the initial values ​​through field creep tests, and updates them every 50 sampling cycles based on real-time data, thereby improving the ability to model nonlinear deformation.

[0031] The ST-GCN-Attention model used in step (4) innovatively introduces a spatiotemporal attention mechanism based on the traditional spatiotemporal graph convolutional network (ST-GCN). It adaptively learns the association weights between nodes to adjust the adjacency matrix. Optimization is required; the core of this mechanism lies in the calculation of the attention coefficient, the formula of which is:

[0032] in, Represents a node With nodes Attention weights between the two are used to measure the strength of their association in the spatiotemporal dimension; and These are the feature vectors of nodes i and j, respectively; The weight matrix is ​​a learnable matrix. For bias terms; As an activation function, it effectively solves the problem of neuron "death" on the negative half axis of the traditional ReLU function, and enhances the nonlinear expression ability of the model. Through this mechanism, the model can dynamically capture the key features of slope monitoring data in spatial distribution and time series, and significantly improve the ability to identify abnormal states. The model demonstrated excellent performance on the test set: the AUC-ROC (area under the curve - receiver operating characteristic) index reached ≥0.95, indicating that the model has a strong ability to distinguish between positive and negative samples; the F1 score was ≥0.93, achieving a good balance between precision and recall, and can reliably provide early warning of potential slope risks, providing high-precision technical support for slope safety monitoring.

[0033] The multi-channel early warning terminal integrates an edge computing node (ECU), which supports 72 hours of local data processing and early warning decision-making in offline mode. After the network is restored, the data is automatically synchronized to the cloud to ensure zero data loss.

[0034] System hardware deployment: BeiDou Spatiotemporal Reference Unit This system uses the Huace X100 Beidou RTK receiver as the core positioning device. This receiver is compatible with multiple satellite navigation systems including BDS / GPS / GLONASS and is equipped with a high-gain choke antenna (gain ≥38dB). It supports network RTK and local base station differential positioning modes, with a positioning data update frequency of 1Hz. The synchronization module integrates a temperature-controlled crystal oscillator (OCXO) with a frequency stability of ±0.01ppm, capable of outputting a 1PPS signal and UTC time code, achieving high-precision time synchronization performance of ≤10ns.

[0035] Multimodal sensor networks The fiber optic sensing system deploys FBG strain sensors at 5m intervals along the main slip surface of the slope. Data acquisition is performed using a Wuhan Optics Valley OFDR-6000 demodulator, which has a wavelength resolution of 1pm and, after temperature compensation, a strain measurement accuracy of 1με. Simultaneously, distributed fiber optic thermometry (DTS) technology enables distributed monitoring of the slope temperature field, achieving a temperature measurement accuracy of ±0.5℃.

[0036] The MEMS sensor group uses the InertialLabs MTi-G-700 module, which integrates a triaxial accelerometer (range ±2g, noise density 30). ), three-axis gyroscope (range ±2000° / s, noise) The system includes a dual-axis tilt sensor (accuracy ±0.02°) and a data sampling frequency set to 100Hz.

[0037] Edge computing nodes The system deploys an industrial computer as the edge computing core, supporting PoE power supply and running the Ubuntu 20.04 operating system. This node is responsible for data acquisition and control, spatiotemporal registration, adaptive fusion algorithm execution, and local early warning decision-making functions, while providing RJ45 wired and Wi-Fi wireless dual-mode communication interfaces.

[0038] Software system implementation Spatiotemporal registration algorithm Time Synchronization: To ensure the time consistency of monitoring data, the system employs a two-layer time calibration mechanism. The lower layer is based on a high-precision timer in the Linux kernel, with a hardware-level timing accuracy of ≤1μs, providing a fundamental guarantee for time synchronization. The upper layer introduces the Precision Time Protocol (PTP, IEEE1588), which uses a master-slave clock architecture and dedicated synchronization message exchange in the network to achieve nanosecond-level time calibration between devices within the local area network. Field tests have verified that the timestamp error of the sensor-acquired data can be controlled within ≤50ns, meeting the stringent time accuracy requirements of slope dynamic monitoring.

[0039] Spatial Transformation: Regarding spatial coordinate system transformation, the system integrates the Proj.4 geographic information projection library to convert the original WGS84 geodetic coordinate system data into the UTM (Universal Transverse Mercator) coordinate system suitable for engineering surveying. To eliminate errors introduced by the transformation, a seven-parameter transformation model based on field-measured control points is adopted. This involves setting up at least three high-precision control points and using the least squares method to solve for three translation parameters, three rotation parameters, and one scaling parameter. The spatial transformation accuracy is quantitatively evaluated using the root mean square error (RMSE) formula.

[0040] in, These are the transformed coordinate values. The actual measured value is given by n, which represents the number of validation samples. Field testing in engineering projects showed that the actual error of this conversion model was controlled within 2.5 cm, meeting the millimeter-level positioning requirements for slope deformation monitoring.

[0041] Adaptive data fusion State-space model construction: Construct a 12-dimensional state vector, covering the three-dimensional spatial position coordinates (x, y, h) and corresponding velocity components. and key engineering geological parameters, including dip angle. ,strain ,stress The system uses water content (w), ambient temperature (T), and atmospheric pressure (P) to achieve a unified representation of multi-source information on slopes.

[0042] Dynamic noise modeling method: For the process noise matrix Q, the exponentially weighted moving average (EWMA) algorithm is used for real-time updates to ensure effective tracking of dynamic changes in the system; the observation noise matrix R is dynamically optimized and adjusted based on sensor calibration data, which significantly enhances the robustness of the model in time-varying noise environments.

[0043] Stability evaluation model Network architecture design: An ST-GCN-Attention deep learning network is constructed. This architecture includes two 3×3 convolutional kernel graph convolutional layers, two 128-unit long short-term memory network layers (LSTM layers), and integrates a 4-head attention mechanism (Multi-Head Attention). The input layer uses Z-score normalization for data preprocessing, and the output layer uses the Softmax activation function to output the three-class probability of slope state, including stable, warning, and unstable states.

[0044] Model training strategy: The AdamW optimization algorithm is adopted, with a weight decay parameter of 0.01, and the FocalLoss loss function is used to address the sample imbalance problem. The training dataset contains 100,000 normal operating condition data, 15,000 warning state data, and 5,000 unstable state data. The validation set is divided into 20% segments, and an early stopping mechanism is introduced, that is, the training process is terminated when the validation loss does not decrease for 10 consecutive rounds, effectively avoiding model overfitting.

[0045] Engineering Application Examples Eight monitoring points were deployed on a steep slope in a phosphate mine in Yunnan Province for a 12-month engineering verification project. Spatiotemporal synchronization performance: The system adopts the high-precision positioning module of the BeiDou-3 satellite navigation system, combined with a self-developed spatiotemporal calibration algorithm, to achieve highly reliable transmission and processing of positioning data. Long-term field testing has verified that the BeiDou positioning data effectiveness rate reaches 99.7%, and even under complex electromagnetic environments and multipath interference, the cycle slip repair success rate remains at an excellent level of 98.9%. Regarding time synchronization, through the collaboration of the NTP network time protocol and BeiDou time synchronization function, the timestamp error of sensor data is all <80ns. In the spatial dimension, utilizing multi-sensor joint calibration technology, the maximum spatial registration error is controlled within 2.8cm, fully meeting the stringent requirements for millimeter-level deformation monitoring of slopes.

[0046] Data fusion results: An innovative combination of deep learning and the traditional Extended Kalman Filter (EKF) algorithm was used to construct an adaptive data fusion model. Cross-validation with total station manual monitoring data over 12 consecutive months showed that the root mean square error (RMSE) of horizontal displacement was 8.2 mm, a 55.7% reduction compared to the traditional EKF algorithm; the RMSE of vertical displacement was 11.3 mm, a 56.1% reduction in error; and the overall data consistency reached 98.6%, far exceeding the industry standard requirement of 95%, significantly improving the accuracy and stability of the monitoring data.

[0047] Early warning capability verification: In August 2024, a mountainous area was affected by continuous heavy rainfall. The intelligent early warning module based on the spatiotemporal graph convolutional attention (ST-GCN-Attention) model played a crucial role. 120 hours before the landslide, the system detected anomalies in the internal strain gradient of the slope (>5). Based on historical data and real-time environmental parameters, the model calculated an instability probability of P=0.92 and triggered a red alert 72 hours before the landslide. On-site engineers promptly activated the emergency plan based on the alert information, successfully avoiding an estimated direct economic loss of approximately 1.5 million yuan through measures such as slope reduction and grouting reinforcement. This verified the effectiveness and practicality of the system in geological disaster early warning.

[0048] Threshold dynamic optimization mechanism Volatility analysis is performed on historical early warning data based on the GARCH(1,1) model, and the early warning threshold is updated every 7 days.

[0049] in By dynamically adjusting the weights of various monitoring parameters (increasing the weight of moisture content by 30% during the rainy season and the weight of temperature by 20% during the dry season), the false alarm rate was reduced from 12% to 4.8%, significantly improving the environmental adaptability of the early warning system.

[0050] The core innovations of this invention are as follows: First, the BeiDou spatiotemporal reference unit adopts dual-mode technology of BeiDou-3 satellite RTK positioning and global short message communication to achieve ±10mm plane positioning accuracy, ±15mm elevation positioning accuracy, and nanosecond-level time synchronization, providing a unified spatiotemporal foundation for multi-source data. Second, a multi-modal sensor network including a BeiDou antenna array, fiber optic sensor network, and microelectromechanical system sensor group is constructed to achieve high-precision acquisition of multi-dimensional data such as displacement, strain, and temperature. Third, the heterogeneous data fusion engine innovatively constructs a 12-dimensional state space model and adopts an improved EKF algorithm with a gradually diminishing memory factor and incorporating viscoelastic parameters of the soil and rock mass to achieve the RMSE of horizontal displacement. The accuracy of data fusion and the ability to model nonlinear deformation have been significantly improved. Fourth, the stability intelligent evaluation module is designed with an ST-GCN-Attention model based on the attention mechanism, combined with a Bayesian optimization dynamic threshold mechanism, to achieve accurate prediction of slope instability probability from 0 to 72 hours, with an AUC-ROC index ≥ 0.95 and an F1 score ≥ 0.93. Fifth, the multi-channel early warning terminal implements a three-level early warning mechanism, integrating 4G / 5G, Beidou short message and LoRa communication, and integrating edge computing nodes to support 72-hour offline operation, ensuring the continuity of monitoring and early warning in all scenarios and zero data loss.

[0051] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A Beidou-based slope monitoring and early warning system, characterized in that, The application comprises a Beidou space-time reference unit, a multi-modal sensor network, a heterogeneous data fusion engine, a stability intelligent evaluation module, and a multi-channel early warning terminal. The Beidou space-time reference unit is initialized to output a unified space-time reference. The multi-modal sensor network collects multi-dimensional original data based on the reference. The heterogeneous data fusion engine performs space-time registration and fusion processing on the original data to output high-precision unified data. The stability intelligent evaluation module calculates the instability probability P based on the fused data. The multi-channel early warning terminal triggers corresponding level warnings based on the P value and a preset threshold, and transmits early warning information through multiple channels.

2. The slope monitoring and early warning system based on Beidou according to claim 1, characterized in that, The multi-modal sensor network comprises: Three geodetic receivers with a distance greater than 5m are used to suppress cycle slip error by using a multi-baseline solution algorithm, and the baseline solution accuracy is less than or equal to 5mm. Optical fiber sensing network: contains distributed optical fiber temperature measurement and Bragg grating strain sensor, strain measurement resolution of 1 , temperature compensation accuracy ±0.5℃; MEMS sensor suite: integrated 3-axis accelerometer with noise less than or equal to 50 , 2-axis tilt sensor with accuracy of ±0.02° and barometric altimeter with resolution of 0.1 m.

3. The slope monitoring and early warning system based on Beidou according to claim 1, characterized in that, The heterogeneous data fusion engine comprises: The space-time registration module: based on the Beidou 1PPS time signal and the UTM projection coordinate system, the space-time alignment of multi-source heterogeneous data is realized, and after calibration, the timestamp error is controlled to be less than or equal to 50ns, and the space coordinate conversion error is less than or equal to 3cm. The adaptive extended Kalman filter unit: a 12-dimensional state space model is constructed, and the state vector is represented as: Slope state vector The following definitions are used: ； Where x, y, h represent the plan coordinates and elevation of the monitoring point, respectively. corresponding to each dimension; for the slope angle, represent the strain and stress state of the rock-soil mass, respectively; w represents the groundwater level, and T and P represent the ambient temperature and atmospheric pressure, respectively. State transition matrix Based on the dynamic embedding algorithm framework, the viscoelastic-plastic rheological parameters of the slope rock-soil mass are deeply coupled, and through the construction of a high-order dynamic model containing time evolution terms, the accurate description of the space-time evolution process of the slope stress-strain is realized. The linearization technique of the second-order Taylor series expansion is adopted to locally linearly approximate the nonlinear observation equation, and combined with the singular value decomposition optimization method, the "ill-conditioned" problem caused by nonlinearity is effectively reduced, so as to realize the high-precision solution of the monitoring data and the efficient extraction of the characteristic information.

4. The slope monitoring and early warning system based on Beidou according to claim 1, characterized in that, The stability intelligent evaluation module constructs a spatiotemporal graph convolution network model based on an attention mechanism, the input layer adopts a 24-dimensional feature vector, the spatial correlation of the monitoring points is extracted through a graph convolution layer, and the instability probability P of 0-72 hours is output after dynamic weighting by an attention layer, .

5. The Beidou-based slope monitoring and early warning system according to claim 1, characterized in that, The multi-channel early warning terminal implements a three-level early warning mechanism: Yellow warning: triggered when 0.6≤P<0.8 and displacement rate>2mm / 6h; Orange alert: when 0.8 < P < 0.9 and strain gradient > 3 is activated Orange alert: when 0.8 < P < 0.9 and strain gradient > 3 is activated Red warning: started when P≥0.9 and inclination rate>0.05° / min; Early warning information is transmitted through 4G / 5G network, Beidou short message and LoRa wireless networking, and the communication delay is ≤10s, and the public network is ≤25s without public network.

6. The method for slope monitoring, early warning and evaluation of the system according to any one of claims 1-6, characterized in that, The method comprises the following steps: (1) Space-time reference unification: using Beidou RTK differential data, the WGS84 coordinates of the monitoring point are obtained by carrier phase dynamic real-time differential technology, and converted into the engineering coordinate system by means of the Bores seven-parameter model, and the nanosecond-level time synchronization is realized through the 1PPS signal; (2) Multi-source data preprocessing: the cycle slip error of Beidou data is repaired by using a five-order polynomial fitting algorithm, and the sensor collected data is corrected by a temperature-humidity compensation model; (3) Dynamic fusion modeling: multi-source heterogeneous data is fused by using an improved EKF algorithm with fading memory factor, and the improved EKF algorithm comprises a Kalman gain update equation and a posteriori estimation error covariance matrix update equation; (4) Stability quantitative evaluation: the data after fusion processing are input into the ST-GCN-Attention model, a dynamic threshold optimization mechanism is constructed by integrating the Bayesian optimization algorithm, and the slope instability probability evaluation result is output.

7. The method according to claim 6, wherein, In the improved EKF algorithm in step (3), the rock and soil body viscoelastic parameters are introduced as state variables, the initial values are obtained by field creep test, and the values are updated every 50 sampling periods according to real-time data.

8. The method according to claim 6, characterized in that, In the step (4), the ST-GCN-Attention model dynamically optimizes the adjacency matrix through attention coefficients The formula for calculating the attention coefficient is: ； wherein, represents the attention weight between nodes and is used to measure the strength of their association in the spatio-temporal dimension; and are the feature vectors of nodes i and j, respectively; is a learnable weight matrix, is a bias term.​ 9. The method of claim 6, wherein, The multi-channel early warning terminal integrates edge computing nodes, supports 72-hour local data processing and early warning decision-making in offline state, and automatically synchronizes data to the cloud after network recovery.

10. The method of claim 6, wherein, The temperature-humidity compensation model in step (2) is: ； wherein is the original measured value, is the corrected measured value.

Images