Visual slope settlement monitoring and early warning method and platform

Abstract

The invention relates to the technical field of slope settlement monitoring and early warning, in particular to a visual slope settlement monitoring and early warning method and platform. The method comprises the following steps: acquiring slope settlement monitoring data; preprocessing the acquired slope settlement monitoring data; respectively constructing a multi-scale slope digital twinborn model and a hybrid intelligent prediction model; constructing a multi-level early warning index system; constructing a fuzzy neural network early warning model based on a multi-stage early warning index system through a multi-scale slope digital twinborn model and a hybrid intelligent prediction model; and performing visual slope settlement early warning by using the fuzzy neural network early warning model. According to the space-air-ground integrated monitoring network constructed by the invention, a satellite InSAR, an unmanned aerial vehicle LiDAR and distributed optical fiber sensing are fused, and full-scale monitoring from regional macroscopic deformation to slope surface microcracks and deep soil displacement is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of slope settlement monitoring and early warning, and in particular to a visual slope settlement monitoring and early warning method and system. BACKGROUND

[0002] In the current field of highway slope settlement monitoring, many advanced technologies have been applied. For example, satellite remote sensing technology can obtain regional macro slope deformation information due to its large-area synchronous observation capability, and can provide basic data for large-area slope stability evaluation. Internet of Things sensor technology can realize real-time collection and transmission of displacement, stress, strain and other parameters by deploying various sensors at key parts of the slope, and can provide real-time monitoring of the slope state. Optical fiber sensing technology plays an important role in slope deep displacement and strain monitoring due to its high precision and strong anti-interference capability.

[0003] However, the existing technology still has some shortcomings in practical application: Some traditional monitoring equipment, such as total station and level, is limited by measurement principle and environmental factors, and the measurement accuracy will be greatly reduced in complex terrain and bad weather, which cannot meet the monitoring demand of millimeter-level or even sub-millimeter-level accuracy of slope settlement. Moreover, these devices can only monitor a single point or a small area, and cannot fully cover large areas of the slope, resulting in monitoring blind spots and making it difficult to capture the overall deformation trend and local sudden deformation of the slope. For example, in mountainous highways, the measurement accuracy error of the total station can reach centimeters due to the large undulating terrain and poor visibility, and the monitoring points cannot be effectively arranged in complex terrain such as valleys, making it impossible to effectively monitor large areas.

[0004] Although the Internet of Things sensors can collect data in real time, they often face weak signals, interruptions and other problems in data transmission, especially in remote mountainous areas or areas with poor network signals, resulting in data transmission delays or losses, which seriously affect the real-time performance of monitoring. In addition, the data formats and frequencies collected by different types of sensors are different, and a large amount of time and computing power is required for format conversion, space-time registration and other operations during data fusion processing. Moreover, traditional data processing algorithms are slow and inefficient in processing massive monitoring data, making it difficult to complete data mining and analysis in a short time, and thus cannot provide accurate basis for slope settlement early warning in a timely manner. For example, in a highway slope monitoring project, due to unstable data transmission, some key data was missing during heavy rain, and data processing took several hours, missing the best early warning opportunity.

[0005] Existing slope settlement early warning models are mostly based on empirical formulas or simple mechanical models, and do not fully consider the coupling effect of complex geological conditions and various influencing factors of the slope, resulting in poor accuracy of early warning, and false positives and false negatives often occur. Moreover, these models usually set parameters and train for specific geological conditions and slope types, and when applied to different geological regions or new slope projects, they have poor adaptability and are difficult to accurately reflect the actual deformation state and development trend of the slope. For example, in a certain silty soil slope project, the early warning model constructed by the traditional limit equilibrium method often misreported under rainfall conditions due to insufficient consideration of the softening characteristics of silty soil when water is added and the influence of changes in pore water pressure, and failed to timely warn when actual settlement mutation occurred. SUMMARY

[0006] To solve the above-mentioned problems, the present application provides a visual slope settlement monitoring and early warning method and system.

[0007] In a first aspect, the present application provides a visual slope settlement monitoring and early warning method, which adopts the following technical solution: A visual slope settlement monitoring and early warning method, comprising: obtaining slope settlement monitoring data; preprocessing the obtained slope settlement monitoring data; respectively constructing a multi-scale slope digital twin model and a hybrid intelligent prediction model; constructing a multi-level early warning index system; constructing a fuzzy neural network early warning model based on the multi-scale slope digital twin model and the hybrid intelligent prediction model and the multi-level early warning index system; using the fuzzy neural network early warning model to visually monitor and early warn the slope settlement.

[0008] Further, the obtaining of the slope settlement monitoring data comprises constructing an aerospace-terrestrial integrated monitoring network, using Sentinel1 and TerraSARX satellite data to obtain regional deformation field by the SBASInSAR method, and using a UAV equipped with a visible light camera and a laser radar to realize slope surface crack identification; laying a distributed fiber strain monitoring system DTS, using the BOTDR method to lay a vibrating wire settlement gauge and a MEMS tilt sensor in a three-dimensional grid to form a three-dimensional monitoring network; and burying an inclinometer tube in a typical section, and embedding an optical fiber grating inclinometer inside; using a vibrating wire pore water pressure gauge to monitor pore water pressure at different depths.

[0009] Further, the obtaining of the slope settlement monitoring data further comprises: synchronously collecting data based on Beidou satellite timing through the space-ground integration monitoring network, to ensure nanosecond-level synchronization of multi-source data; transmitting low-rate data through a LoRaWAN method for short distances; transmitting high-frequency data through a 5G private network for medium distances, wherein the bandwidth is 100 MHz and the time delay is less than or equal to 10 ms; and taking a Beidou short message as an emergency backup channel for long distances.

[0010] Further, the preprocessing of the obtained slope settlement monitoring data comprises: correcting data errors through an error correction formula, removing random errors through 5-layer decomposition of db4 wavelets; unifying coordinates to a WGS84 coordinate system, solving a conversion matrix by using a seven-parameter method, aligning time series by using a dynamic time warping (DTW) algorithm, and realizing spatio-temporal registration; finally, fusing homogeneous sensor data by using Kalman filtering, calculating a fusion estimation value through a state equation and an observation equation, establishing a slope monitoring ontology model, defining entity classes, attribute classes and relationship classes, and realizing data semantic mapping through rules.

[0011] Further, the constructing of the multi-scale slope digital twin model and the hybrid intelligent prediction model respectively comprises: constructing the multi-scale slope digital twin model based on geometric modeling and physical modeling, wherein the macro scale of the geometric modeling is based on DEM and satellite images, the meso scale is based on unmanned aerial vehicle point clouds, and the micro scale is based on three-dimensional laser scanning to construct a detailed model; and the physical modeling respectively comprises a constitutive model, a seepage model and a coupling model, wherein the modified Cambridge model is used to calculate the yield function of the silt layer in the constitutive model, the seepage field equation is calculated based on Darcy's law in the seepage model, and the simultaneous equations are constructed by using the Biot consolidation theory in the coupling model.

[0012] Further, the constructing of the multi-scale slope digital twin model and the hybrid intelligent prediction model respectively further comprises: constructing the hybrid intelligent prediction model based on an improved LSTMGM (1, 1) CNN model, including a feature extraction layer, a trend prediction layer and a fusion layer, wherein the spatio-temporal features are extracted through CNN convolution and double-layer LSTM in the feature extraction layer; the GM (1, 1) in the trend prediction layer introduces a forgetting factor to dynamically adjust the weight of historical data; the fusion layer adopts a whale optimization algorithm and optimizes the weight based on an objective function, and finally carries a hidden Markov model to train the state transition matrix and the observation probability matrix by using the BaumWelch algorithm, to realize parameter training.

[0013] Further, the constructing of the multi-level early warning index system comprises: basic indexes and derived indexes, wherein the basic indexes comprise absolute settlement, settlement rate, acceleration, pore water pressure ratio and crack propagation rate, the derived indexes comprise a safety factor calculated based on slope parameters, a risk entropy calculated based on the basic indexes, and a mutation index calculated based on historical slope settlement data.

[0014] Further, the fuzzy neural network early warning model is constructed based on the multi-scale slope digital twin model and the hybrid intelligent prediction model and a multi-level early warning index system, and includes an input layer, a fuzzification layer, a rule layer, an inference layer and an output layer, wherein the input layer has 7 nodes, each node in the fuzzification layer corresponds to 5 Gaussian membership functions, the rule layer has 5 7th power fuzzy rules, a genetic algorithm is used for screening, the inference layer uses the Mamdani inference method and the min operator, and the output layer outputs the early warning level.

[0015] Further, the fuzzy neural network early warning model is constructed based on the multi-scale slope digital twin model and the hybrid intelligent prediction model and a multi-level early warning index system, and includes an input layer, a fuzzification layer, a rule layer, an inference layer and an output layer, wherein the input layer has 7 nodes, each node in the fuzzification layer corresponds to 5 Gaussian membership functions, the rule layer has 5 7th power fuzzy rules, a genetic algorithm is used for screening, the inference layer uses the Mamdani inference method and the min operator, and the output layer outputs the early warning level.

[0016] The second aspect is a visualized slope settlement monitoring and early warning platform, comprising: A data acquisition module configured to acquire slope settlement monitoring data; A preprocessing module configured to preprocess the acquired slope settlement monitoring data; A model construction module configured to construct a multi-scale slope digital twin model and a hybrid intelligent prediction model; A system construction module configured to construct a multi-level early warning index system; An early warning module configured to construct a fuzzy neural network early warning model based on the multi-scale slope digital twin model and the hybrid intelligent prediction model and the multi-level early warning index system; A visualization module configured to perform visualized slope settlement early warning using the fuzzy neural network early warning model.

[0017] The third aspect provides a computer readable storage medium, wherein a plurality of instructions are stored, the instructions are suitable for being loaded and executed by a processor of a terminal device to implement the visualized slope settlement monitoring and early warning method.

[0018] The fourth aspect provides a terminal device, comprising a processor and a computer readable storage medium, the processor is used to implement instructions, and the computer readable storage medium is used to store a plurality of instructions, the instructions are suitable for being loaded and executed by the processor to implement the visualized slope settlement monitoring and early warning method.

[0019] In summary, the present application has the following beneficial technical effects: The space-air-ground integrated monitoring network constructed by the application fuses satellite InSAR, unmanned aerial vehicle LiDAR and distributed optical fiber sensing, and realizes full-scale monitoring from regional macro deformation to surface micro cracks of the slope and displacement of deep soil. Compared with traditional single-point discrete monitoring, the monitoring point density is increased by 8 times, the coverage blind area is reduced by 90%, and the creep settlement caused by self-weight stress and rainfall infiltration in the silt layer can be captured in real time, thereby providing high-density and high-precision data support for slope stability analysis.

[0020] By means of Beidou nanosecond-level timing and Kalman filtering algorithm, the spatio-temporal accurate registration of multi-source heterogeneous data is realized, and the data fusion delay is less than or equal to 50 ms; by means of the LSTM-GM (1,1)-CNN hybrid intelligent model, the weight is optimized by means of the whale optimization algorithm, and the 72-hour settlement prediction error is less than or equal to 10%, which is 50% higher in accuracy than the traditional empirical formula. When the slope settlement rate exceeds 2 mm / d, the system automatically switches to high-frequency sampling once every 5 minutes, and transmits data through the 5G private network, and the whole process from collection to early warning generation is less than or equal to 10 minutes, which is 99% shorter in response time than the traditional manual processing, so that the settlement mutation early warning under sudden conditions such as heavy rain can be responded to in time.

[0021] Based on the modified Cambridge model and the Biot consolidation theory, a multi-physical field coupling digital twin model considering the porosity, water content and pore water pressure change of silt is constructed, and the dynamic deduction of the slope settlement process is realized through the UE5 engine, and the coincidence degree with the field measurement is greater than or equal to 85%. Meanwhile, 7 early warning indexes are fused by means of the fuzzy neural network, 78125 fuzzy rules are screened by means of the genetic algorithm, the early warning accuracy is improved from 65% of the traditional single index to more than 85%, the false positive rate is reduced by 40%, and the slope monitoring under different geological conditions can be self-adapted through transfer learning without repeated model training. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 FIG. 1 is a schematic diagram of a visual slope settlement monitoring and early warning method according to an embodiment of the application. DETAILED DESCRIPTION

[0023] The application will be further described in detail below with reference to the accompanying drawings.

[0024] Embodiment 1 Referring to Figure 1 The visual slope settlement monitoring and early warning method according to the embodiment comprises the following steps. I. Obtain slope settlement monitoring data 1.1 Space-air-ground integrated monitoring network A space-ground-air integrated monitoring network is constructed, using Sentinel1 and TerraSARX satellite data to obtain regional deformation field by SBAS InSAR method, and using unmanned aerial vehicle to carry visible light camera and laser radar to realize slope surface crack identification; distributed fiber strain monitoring system DTS is laid, adopting BOTDR method to lay vibrating wire type settlement meter and MEMS tilt sensor in three-dimensional grid to form a three-dimensional monitoring network; and inclinometer is buried in the typical section, with fiber grating inclinometer inside; vibrating wire type pore water pressure gauge is used to monitor pore water pressure at different depths.

[0025] 1.1.1 Space monitoring system Satellite remote sensing: fuse Sentinel1 SAR data (spatial resolution 5m x 20m, revisit period 12 days) and TerraSARX data (resolution 1m x 1m), use small baseline set interferometric measurement (SBAS InSAR) technology to obtain regional deformation field, and the monitoring accuracy reaches ±2mm / year.

[0026] Unmanned aerial vehicle inspection: deploy DJI M300RTK unmanned aerial vehicle, carry visible light camera (20 million pixels) and laser radar (point cloud density 100 points / m²), flight height 100m, flight line spacing 50m, realize automatic identification of slope surface crack (≥0.5mm).

[0027] 1.1.2 Ground monitoring system Fiber sensing network: distributed fiber strain monitoring system (DTS) is laid in K171+608~K171+913 section, using BOTDR technology, spatial resolution 1m, strain measurement accuracy ±2με, sampling frequency 1Hz, real-time monitoring of soil deformation along the way.

[0028] Intelligent sensor array: Vibrating wire type settlement meter (accuracy ±0.01mm) and MEMS tilt sensor (resolution 0.0005°) are laid in three-dimensional grid, horizontal spacing 8m, vertical spacing 5m, forming a 20x15x6 three-dimensional monitoring network.

[0029] Mini weather station is laid to monitor environmental parameters such as wind speed (accuracy ±0.5m / s), wind direction (accuracy ±3°), air temperature (accuracy ±0.5℃), etc.

[0030] 1.1.3 Underground monitoring system Inclinometer array: 7 inclinometer tubes are buried in the typical section (K179+771), with a depth of 60m, and a fiber grating inclinometer (accuracy ±0.02mm / m) is placed inside, with a sampling frequency of 1 time / 10 minutes, to monitor the deep horizontal displacement of soil.

[0031] Pore water pressure monitoring: The vibrating string pore water pressure gauge (range 0~2MPa, accuracy 0.1%FS) was used to monitor the changes of pore water pressure in real time, with the buried depth of 5m, 15m and 25m underground.

[0032] 1.2 Data acquisition and transmission protocol Through the space-earth integration monitoring network, synchronous data collection based on Beidou satellite timing ensures nanosecond-level synchronization of multi-source data; for short distances, LoRaWAN method is used to transmit low-rate data; for medium distances, 5G private network is used to transmit high-frequency data, with a bandwidth of 100MHz and a time delay of ≤10ms; for long distances, Beidou short message is used as an emergency backup channel.

[0033] Multi-source data synchronous acquisition: Where Tbase is the reference clock, is the clock offset of each sensor, which realizes nanosecond-level synchronization through Beidou satellite timing (accuracy ≤5ns).

[0034] Mixed transmission architecture: Short distance: LoRaWAN technology (transmission distance 3km, bandwidth 125kHz) is used to transmit low-rate sensor data; Medium distance: 5G private network (bandwidth 100MHz, time delay ≤10ms) is used to transmit high-frequency monitoring data; Long distance: Beidou short message (120 Chinese characters each time) is used as an emergency backup channel.

[0035] II. Data preprocessing and fusion layer Through the error correction formula, the data error is calibrated, and the random error is denoised by db4 wavelet 5-layer decomposition; and by unifying the coordinates to WGS84 coordinate system, the conversion matrix is solved by seven-parameter method, the time series is aligned by dynamic time warping DTW algorithm, the space-time registration is realized; finally, the Kalman filter is used to fuse the same type of sensor data, after calculating the fusion estimation value through the state equation and observation equation, the slope monitoring ontology model is established, the entity class, attribute class and relationship class are defined, and the data semantic mapping is realized by rules.

[0036] 2.1 Multi-source data cleaning algorithm 2.1.1 Error correction of different types of sensors System error correction: Where k0 is the zero bias error, k1 and k2 are the temperature coefficients, and T is the environmental temperature.

[0037] Random error filtering: wavelet transform denoising is used, db4 wavelet is selected for 5-layer decomposition, and the threshold function is: Where Adaptive threshold, σj: the standard deviation of the jth layer noise.

[0038] 2.1.2 Spatio-temporal data registration Coordinate system unification: all monitoring data are unified to WGS84 coordinate system, conversion formula: Where M is the rotation matrix, (X0, Y0, Z0) is the translation parameter, which is solved by seven-parameter method.

[0039] Time series alignment: dynamic time warping (DTW) algorithm is adopted, distance measure: Where α = 0.5, β = 0.3 are weight coefficients, vi, ai are settlement rate and acceleration.

[0040] 2.2 Multi-source data fusion model 2.2.1 Three-layer fusion architecture 1. Data layer fusion: Kalman filter is used to fuse homogeneous sensor data, state equation: Post-fusion estimation: 2. Feature layer fusion: based on D-S evidence theory, belief function: Where is the sensor reliability, is the evidence conflict coefficient.

[0041] 3. Decision layer fusion: fuzzy analytic hierarchy process (FAHP) is adopted, fuzzy judgment matrix: Weight vector is solved by fuzzy number sorting algorithm.

[0042] 2.2.2 Heterogeneous data semantic mapping Establish the ontology model of slope monitoring, define classes and relationships: Entity class: slope, monitoring point, sensor, stratum, supporting structure Attribute class: settlement, rate, acceleration, pore water pressure, rainfall Relationship class: located, monitored, belongs to, influenced Semantic mapping rules: Three, intelligent early warning model construction 3.1 Multi-scale slope digital twin model A multi-scale slope digital twin model is constructed based on geometric modeling and physical modeling, wherein the macro scale of geometric modeling is based on DEM and satellite images; the meso scale uses unmanned aerial vehicle point cloud; the micro scale uses three-dimensional laser scanning to construct a detailed model; the physical modeling includes constitutive model, seepage model and coupling model, wherein the modified Cambridge model is used to calculate the yield function for the silt layer in the constitutive model, the seepage field equation is calculated based on Darcy's law in the seepage model, and the Biot consolidation theory is used to construct the coupled equations in the coupling model.

[0043] 3.1.1 Geometric modeling Macro scale (1:1000): regional terrain model is constructed based on DEM data (resolution 1m) and satellite images (resolution 0.5m); Meso scale (1:200): slope surface model is generated using unmanned aerial vehicle point cloud (point spacing 0.1m); Micro scale (1:50): three-dimensional laser scanning (point spacing 0.01m) is used to construct crack, support structure and other detailed models.

[0044] 3.1.2 Physical modeling Constitutive model: the modified Cambridge model is used for the silt layer, and the yield function is: Where p' is the average effective stress, q is the deviatoric stress, M is the material constant, σ c ' is the pre-consolidation pressure.

[0045] Seepage model: based on Darcy's law, the seepage field equation is: Where k is the permeability coefficient, h is the water head, and Ss is the water storage rate.

[0046] Coupling model: Biot consolidation theory is used, and the coupling equation is: Where u is the pore water pressure, α is the Biot coefficient, and εv is the volume strain.

[0047] 3.2 Hybrid intelligent prediction model A hybrid intelligent prediction model is constructed based on an improved LSTMGM(1,1) CNN model, including a feature extraction layer, a trend prediction layer and a fusion layer, wherein the feature extraction layer extracts spatial and temporal features through CNN convolution and double-layer LSTM; the GM(1,1) of the trend prediction layer introduces a forgetting factor to dynamically adjust the weight of historical data; the fusion layer adopts a whale optimization algorithm and optimizes the weight based on an objective function, and finally carries a hidden Markov model, trains the state transition matrix and observation probability matrix through the BaumWelch algorithm, and realizes parameter training.

[0048] 3.2.1 Improved LSTMGM(1,1) CNN model 1. Feature extraction layer: CNN module: 1D CNN is adopted to extract spatial features, the convolution kernel size is 5, the step is 2, and the activation function is ReLU; LSTM module: double-layer LSTM (256 neurons each) is adopted to extract time series features, and dropout=0.3.

[0049] 2. Trend prediction layer: GM(1,1) model: a forgetting factor \beta (0.8-1.0) is introduced to dynamically adjust the weight of historical data: 3. Fusion layer: The weight optimization adopts a whale optimization algorithm (WOA), and the objective function is: 3.2.2 Settlement mutation identification algorithm Based on a hidden Markov model (HMM), the state transition matrix is: The observation probability matrix is: State definition: stable, gradual change, mutation, and parameters are trained through the BaumWelch algorithm.

[0050] Four, early warning generation and interactive visualization 4.1 Multi-level early warning index system Including basic indexes and derivative indexes, wherein the basic indexes include absolute settlement, settlement rate, acceleration, pore water pressure ratio and crack propagation rate, the derivative indexes include safety factor calculated based on slope parameters, risk entropy calculated based on basic indexes and mutation index calculated based on slope settlement history data.

[0051] 4.1.1 Basic index 1. Absolute settlement: s th = 20 mm 2. Settlement rate: v th = 5 mm / d 3. Acceleration: a th = 1 mm / d 2 4. Pore water pressure ratio: u th = 0.7 u 0 (u 0 is initial pore water pressure) 5. Crack propagation rate: l th = 2 mm / d 4.1.2 Derivative indicators 1. Safety factor: where c is cohesion, ψ is internal friction angle, L is the length of sliding surface, σ n is normal stress, W is the weight of sliding body, and a is the inclination angle of sliding surface.

[0052] 2. Risk entropy: where p i is the normalized weight of each early warning indicator.

[0053] 3. Mutation index: where is the rate of change, is the time interval, is the maximum historical settlement.

[0054] 4.2 Fuzzy neural network early warning model Through the multi-scale slope digital twin model and hybrid intelligent prediction model, and based on the multi-level early warning indicator system, a fuzzy neural network early warning model is constructed, including the construction of input layer, fuzzification layer, rule layer, reasoning layer and output layer. Among them, the input layer adopts 7 nodes, each node in the fuzzification layer corresponds to 5 Gaussian membership functions, the rule layer sets 5 7 fuzzy rules, and genetic algorithm is used for screening, Mamdani reasoning method is used in the reasoning layer, and the inclusion operator is min, and finally the output layer outputs the early warning level.

[0055] 4.2.1 Network architecture Input layer: 7 nodes (I1~I5, Fs, H) Fuzzification layer: each node corresponds to 5 Gaussian membership functions: where c i is the center value, is the width Rule layer: 5 7 = 78125 fuzzy rules, screened by genetic algorithm Inference layer: Mamdani inference method is adopted, and the implication operator is min Output layer: early warning level (0~1 continuous value) 4.2.2 Early warning decision rule Among them, the comprehensive early warning value R is obtained by weighted fuzzy reasoning.

[0056] 4.3 Immersive visualization interactive system Adopt multi-resolution model and physical light model for three-dimensional scene rendering, and through the construction of space-time cube and augmented reality technology for interactive early warning display, among them, the space-time cube constructs three-dimensional subsidence cloud picture with time as Z axis, supports time slicing, space profile analysis and early warning trajectory backtracking; augmented reality technology realizes the conversion between virtual model coordinates and real world coordinates through Hololens2 device.

[0057] 4.3.1 Three-dimensional scene rendering Multi-resolution model: Among them, Dview is the distance of the viewpoint, Dmax is the maximum visible distance, and LODmax=4 Physical light model: adopt PBR (physically based rendering) technology, BRDF equation: Among them, fr is the reflection distribution function, Li is the incident radiation rate, and ωi, ωo are the incident and outgoing directions.

[0058] 4.3.2 Interactive early warning display Space-time cube: constructs three-dimensional subsidence cloud picture with time as Z axis, supports: Time slicing (resolution 1 hour) Space profile analysis (cutting in any direction) Early warning trajectory backtracking (displaying early warning propagation path within 72 hours) Augmented reality (AR): through Hololens2 device, realizes: Virtual model coordinates = K[R|t] Real world coordinates, Among them, K is the camera intrinsic matrix, R and t are the rotation matrix and translation vector, which realizes the accurate superposition of virtual early warning information and real slope.

[0059] Embodiment 2 The embodiment provides a visual slope subsidence monitoring and early warning system, which comprises: The data acquisition module is configured to A computer readable storage medium, wherein a plurality of instructions are stored, the instructions are suitable for being loaded and executed by the processor of the terminal device, and the instructions are suitable for being loaded and executed by the processor of the terminal device.

[0060] A terminal device comprises a processor and a computer readable storage medium, the processor is used to realize instructions; the computer readable storage medium is used to store a plurality of instructions, the instructions are suitable for being loaded by the processor and executing the visualized slope settlement monitoring and early warning method.

[0061] The above are preferred embodiments of the present application, not limited to the protection scope of the present application, therefore: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A visualized slope settlement monitoring and early warning method, characterized in that, The method comprises the following steps: acquiring slope settlement monitoring data; preprocessing the acquired slope settlement monitoring data; respectively constructing a multi-scale slope digital twin model and a hybrid intelligent prediction model; constructing a multi-level early warning index system; constructing a fuzzy neural network early warning model through the multi-scale slope digital twin model and the hybrid intelligent prediction model, and based on the multi-level early warning index system; using the fuzzy neural network early warning model to realize visual slope settlement early warning.

2. The visualized slope settlement monitoring and early warning method according to claim 1, characterized in that, The acquiring of the slope settlement monitoring data comprises the following steps: constructing an integrated space-ground monitoring network, using Sentinel1 and TerraSARX satellite data to acquire regional deformation field through the SBASInSAR method, and using a UAV to carry a visible light camera and a laser radar to realize slope surface crack identification; laying a distributed fiber strain monitoring system DTS, adopting the BOTDR method to arrange a vibrating string type settlement gauge and a MEMS tilt sensor in a three-dimensional grid to form a three-dimensional monitoring network; and burying a inclinometer tube in a typical section, and embedding a fiber bragg grating inclinometer inside; and using a vibrating string type pore water pressure gauge to monitor pore water pressure at different depths.

3. The visualized slope settlement monitoring and early warning method according to claim 2, characterized in that, The acquiring of the slope settlement monitoring data further comprises the following steps: through the integrated space-ground monitoring network, based on Beidou satellite timing for synchronous data acquisition to ensure nanosecond level synchronization of multi-source data; for short distances, using the LoRaWAN method to transmit low rate data; for medium distances, using a 5G private network to transmit high frequency data, wherein the bandwidth is 100MHz and the time delay is ≤10ms; for long distances, using Beidou short message as an emergency backup channel.

4. The visualized slope settlement monitoring and early warning method according to claim 3, characterized in that, The preprocessing of the acquired slope settlement monitoring data comprises the following steps: through an error correction formula to calibrate data errors, and using db4 wavelet 5 layer decomposition to remove random errors; and through coordinate unification to WGS84 coordinate system, using seven parameter method to solve the conversion matrix, using dynamic time warping DTW algorithm to align the time series, realizing space-time registration; finally, using Kalman filter to fuse the same type of sensor data, calculating the fusion estimation value through the state equation and the observation equation, establishing a slope monitoring ontology model, defining entity class, attribute class and relationship class, and realizing data semantic mapping through rules.

5. The visualized slope settlement monitoring and early warning method according to claim 4, characterized in that, The respectively constructing of the multi-scale slope digital twin model and the hybrid intelligent prediction model comprises the following steps: based on geometric modeling and physical modeling to construct the multi-scale slope digital twin model, wherein the macro scale of geometric modeling is based on DEM and satellite image; the meso scale uses UAV point cloud; the micro scale uses three-dimensional laser scanning to construct a detail model; the physical modeling respectively includes a constitutive model, a seepage model and a coupling model, wherein the modified cambridge model is used to calculate the yield function of the silt layer in the constitutive model, the seepage field equation is calculated based on darcy's law in the seepage model, and the biot consolidation theory is used to construct the coupled equation in the coupling model.

6. The visualized slope settlement monitoring and early warning method according to claim 5, characterized in that, The respective construction of the multi-scale slope digital twin model and the hybrid intelligent prediction model also includes constructing a hybrid intelligent prediction model based on an improved LSTMGM(1,1)CNN model, which includes a feature extraction layer, a trend prediction layer, and a fusion layer. The feature extraction layer extracts spatiotemporal features through CNN convolution and double-layer LSTM. The GM(1,1) of the trend prediction layer introduces a forgetting factor to dynamically adjust the historical data weight. The fusion layer adopts a whale optimization algorithm and optimizes the weight based on an objective function. Finally, a hidden Markov model is carried, and the state transition matrix and observation probability matrix are trained through the BaumWelch algorithm to realize parameter training.

7. The visualized slope settlement monitoring and early warning method according to claim 6, characterized in that, The construction of the multi-level early warning index system includes basic indicators and derived indicators. The basic indicators include absolute settlement, settlement rate, acceleration, pore water pressure ratio, and crack propagation rate. The derived indicators include the safety factor calculated based on slope parameters, the risk entropy calculated based on basic indicators, and the mutation index calculated based on slope settlement historical data.

8. The visualized slope settlement monitoring and early warning method according to claim 7, characterized in that, The fuzzy neural network early warning model is constructed based on the multi-scale slope digital twin model and the hybrid intelligent prediction model and the multi-level early warning index system. It includes the construction of an input layer, a fuzzification layer, a rule layer, an inference layer, and an output layer. The input layer has 7 nodes. In the fuzzification layer, each node corresponds to 5 Gaussian membership functions. The rule layer sets 5 fuzzy rules to the 7th power and uses a genetic algorithm for screening. The inference layer uses the Mamdani inference method with the min operator. Finally, the output layer outputs the early warning level.

9. The visualized slope settlement monitoring and early warning method according to claim 8, characterized in that, The fuzzy neural network early warning model is used for visual slope settlement early warning. It includes using a multi-resolution model and a physical lighting model for three-dimensional scene rendering and using a spatiotemporal cube and augmented reality technology for interactive early warning display. The spatiotemporal cube constructs a three-dimensional settlement cloud chart with time as the Z-axis, supporting time slicing, spatial profile analysis, and early warning trajectory backtracking. The augmented reality technology realizes the conversion of virtual model coordinates and real world coordinates through a Hololens2 device.

10. A visualized slope settlement monitoring and early warning platform, characterized in that, It includes: A data acquisition module configured to acquire slope settlement monitoring data; A preprocessing module configured to preprocess the acquired slope settlement monitoring data; A model construction module configured to construct a multi-scale slope digital twin model and a hybrid intelligent prediction model; A system construction module configured to construct a multi-level early warning index system; An early warning module configured to construct a fuzzy neural network early warning model based on the multi-scale slope digital twin model and the hybrid intelligent prediction model and the multi-level early warning index system; A visualization module configured to use the fuzzy neural network early warning model for visual slope settlement early warning.

Images