A slope disaster early warning method and system

Abstract

The present invention relates to the field of geological disaster monitoring and early warning, and in particular to a slope disaster early warning method and system. The method comprises the following steps: establishing a spatiotemporal database, storing and managing pre-processed slope data from a variety of monitoring sources, and collating to obtain a slope disaster data set; extracting features from the slope disaster data set to obtain a key information set reflecting the characteristics of the slope disaster; analyzing the key information set to obtain a slope disaster value; based on the slope data in the spatiotemporal database, calculating the early warning threshold and real-time displacement prediction value of the slope, and fusing the real-time displacement prediction value and the slope disaster value to obtain a fusion value; based on the obtained fusion value and early warning threshold, evaluating the real-time stability of each monitoring section of the slope, and judging whether to issue an early warning signal and an alarm signal based on the evaluation result. The present invention realizes the accurate identification of slope disaster values ​​and potential risk areas, and improves the accuracy and reliability of the early warning system.

Description

Technical Field

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

[0002] Slope stability is one of the important issues faced in various types of engineering construction. Traditional slope monitoring methods mainly include manual inspection, geodetic method, GPS measurement method, etc. However, these methods have many significant defects. For example, manual inspection is not only inefficient, but also difficult to cover a large area of ​​slope area, and it is easy to miss. There is also a large degree of subjectivity and uncertainty in empirical judgment, and it is impossible to provide accurate and objective evaluation. Limited monitoring methods, such as single-type sensor monitoring, can often only obtain partial and one-sided information, and cannot fully and deeply understand the actual state of the slope, and it is difficult to achieve real-time monitoring; although the geodetic method has high accuracy, the observation period is long and is greatly affected by weather conditions; the GPS measurement method is expensive, and the signal is easily interfered under complex terrain conditions. In the fields of traffic road construction, water conservancy and hydropower engineering, mining, etc., slope instability may cause serious casualties and property losses. With the continuous expansion of the scale and complexity of engineering construction, the requirements for slope stability monitoring are also getting higher and higher. With the continuous advancement of technology, some new monitoring technologies such as laser scanning-based monitoring technology and Internet of Things-based monitoring technology have also been widely used in slope monitoring.

[0003] There are also a large number of scholars in my country who are committed to the research of landslide early warning technology. Fan Wenliang took daily rainfall and cumulative rainfall as control parameters and constructed a joint landslide probability model of the two; Li Xiaolan studied the overall structure and functional structure of GIS in the geological disaster information processing system, and developed a geological disaster information processing system software to realize the data association operation between the model evaluation and analysis system and the spatial database; Li Haifeng and others established a geological disaster information management system and database with functions such as data input, query, data statistics, graphic browsing, information output, and disaster emergency plan processing.

[0004] Patent CN116504032B proposed a landslide hazard monitoring and early warning method and system based on real-time 3D. Although some progress has been made in landslide monitoring, there is still room for improvement in the real-time and accuracy of slope disaster early warning. Patent CN118230534A proposed a disaster monitoring and early warning method based on satellite remote sensing data. Although it improves the efficiency of disaster early warning, the real-time and accuracy of slope disaster early warning still need to be strengthened.

[0005] In summary, in order to ensure the safety and smooth progress of engineering construction, it is urgent to develop a slope disaster early warning method and system. Summary of the invention

[0006] The present invention provides a slope disaster early warning method and system, aiming to solve the problem of how to ensure the real-time and accuracy of slope disaster early warning.

[0007] To achieve the above purpose, the following technical solution is adopted.

[0008] A slope disaster early warning method, comprising:

[0009] Establish a spatiotemporal database to store and manage pre-processed slope data from multiple monitoring sources and organize the slope disaster data set;

[0010] Extract features from the slope disaster dataset to obtain key information sets reflecting the characteristics of slope disasters;

[0011] Apply the pre-trained deep learning model to analyze the key information set after feature extraction to obtain the slope disaster value and potential risk area;

[0012] Based on the slope data in the spatiotemporal database, the warning threshold and real-time displacement prediction value of the slope are calculated, and the real-time displacement prediction value and the slope disaster value are fused to obtain the fusion value;

[0013] According to the obtained fusion value and warning threshold, the real-time stability status of each monitoring section of the slope is evaluated, and it is determined whether to issue a warning signal and an alarm signal based on the evaluation results.

[0014] Optionally, the step of establishing the spatiotemporal database specifically includes:

[0015] Collect raw data from multi-source slope monitoring data, including data from satellite remote sensing, UAV remote sensing, and ground sensor networks. The data from the ground sensor network includes displacement, stress, rainfall, geological structure, and slope morphology;

[0016] Preprocess the collected raw data. The preprocessing operations include data cleaning, denoising, outlier removal, data format conversion and normalization to obtain the preprocessed data set.

[0017] Integrate the preprocessed data sets into the spatiotemporal database, create indexes on the spatiotemporal grid for fast data retrieval and extraction, organize and manage data at different locations and times using a unified spatial representation framework, and obtain integrated data sets that can be used for further analysis;

[0018] In the spatiotemporal database, spatial analysis methods are applied to conduct correlation analysis on the temporal and spatial data in the integrated data set, analyze the spatiotemporal evolution of slope deformation, determine the diffusion range and impact area of ​​deformation, and obtain spatiotemporal analysis results for slope stability assessment;

[0019] By using multi-source monitoring data and the obtained spatiotemporal analysis results, a slope disaster data set that comprehensively reflects the slope status is formed through a fusion algorithm.

[0020] Optionally, the step of extracting features from the slope disaster data set to obtain a key information set reflecting the characteristics of the slope disaster includes:

[0021] The principal component analysis method is used to analyze the data of the slope disaster data set, extract the main components in the data, reflect the stress concentration and distribution pattern, and obtain the key feature set of geology and topography;

[0022] The key feature set was further processed using independent component analysis to extract the main information factors affecting slope disasters.

[0023] Wavelet transform data processing method is used to conduct time-frequency analysis on the slope disaster data set to identify the mutation points and trend changes in the displacement data;

[0024] Retrieve monitoring data related to the identified mutation points and trend changes from the spatiotemporal database to extract key information sets reflecting the characteristics of slope disasters;

[0025] The obtained key feature set is combined with the extracted key information set to form a complete information set reflecting the characteristics of slope disasters.

[0026] Optional, pre-training steps for deep learning models, including:

[0027] Based on the historical monitoring data in the spatiotemporal database, data preprocessing is performed, including normalization and outlier removal, to form a data set for training;

[0028] Divide the preprocessed dataset into training subset and validation subset;

[0029] Initialize at least one or more deep learning models including convolutional neural network, recurrent neural network and long short-term memory network;

[0030] Use the training subset to train the initialized deep learning model, and obtain the initial parameters of the model through forward propagation and loss function calculation; use the validation subset to verify the obtained initial model parameters, and adjust the model parameters through the back propagation algorithm to optimize the model performance; repeat this step until the model reaches the predetermined accuracy or loss threshold on the validation subset;

[0031] The trained and optimized model parameters are fixed to form a pre-trained deep learning model for slope disaster early warning analysis.

[0032] Optionally, a step of applying a pre-trained deep learning model to analyze the key information set after feature extraction to obtain the identification results of the slope disaster value and the potential risk area includes:

[0033] The key information set after feature extraction is input into the pre-trained deep learning model, including convolutional neural network, recurrent neural network and long short-term memory network, for feature recognition and analysis;

[0034] The convolutional neural network model is used to analyze the image features in the extracted data, identify the spatial patterns and abnormal features related to the slope disaster, and obtain the first analysis result;

[0035] The time series data are analyzed using the models of recurrent neural network and long short-term memory network to identify the deformation trend and potential disaster precursors in time and obtain the second analysis result;

[0036] The first analysis result and the second analysis result are fused, and the fused result is input into the random forest to obtain the comprehensive identification result of the slope disaster value and the potential risk area.

[0037] Optionally, based on the slope data in the spatiotemporal database, the warning threshold and real-time displacement prediction value of the slope are calculated, the real-time displacement prediction value and the slope disaster value are fused to obtain a fusion value, and the real-time stability state of each monitoring section of the slope is evaluated according to the obtained fusion value and warning threshold, and the steps of determining whether to issue a warning signal and an alarm signal according to the evaluation result specifically include:

[0038] Receiving slope hazard value from deep learning model;

[0039] Acquire the displacement monitoring data of the slope in the period of imminent landslide in the monitoring data of the spatiotemporal database, wherein the displacement monitoring data of the slope in the period of imminent landslide includes the real-time displacement information of multiple measuring points of the slope monitoring profile;

[0040] The federated Kalman filter method is used to process the real-time displacement information of multiple measuring points in the slope monitoring profile, and the real-time displacement prediction value is obtained by fusion.

[0041] The real-time displacement prediction value and the slope disaster value are fused to obtain the fusion value; the fusion value and the warning threshold are used to evaluate the real-time stability of each monitoring section of the slope;

[0042] Compare the evaluation results with the determined warning threshold to determine whether a warning signal needs to be issued;

[0043] When the evaluation results show that the stability of the slope exceeds the warning threshold, a preliminary warning signal is first issued, and then the danger level and scope of the disaster are automatically deduced according to the number of warning signals corresponding to different alarm signals;

[0044] Determine the level of the early warning signal based on the deduced danger level, and issue an alarm signal of the corresponding level based on the preset multiple safety level thresholds;

[0045] Send early warning and alarm signals in a timely manner through multiple channels;

[0046] Record detailed information of warning and alarm events, including time, location, catastrophic value, risk level, issuance time of alarm and warning signals, and response measures taken, for subsequent analysis and system optimization.

[0047] Optionally, in the step of calculating the warning threshold of the slope based on the slope data in the spatiotemporal database, the calculation step specifically includes:

[0048] Define the slope displacement state vector X(k), which represents the displacement state value of the monitoring point at time k, including the displacement x(k) and relative displacement velocity v(k);

[0049] According to the displacement state vector X(k) and the Gaussian white noise sequence W(k) of the displacement information, the state transfer matrix Φ and the input matrix Γ are used to calculate the displacement state value X(k+1) at the next moment through the expression X(k+1)=ΦX(k)+ΓW(k) to obtain the slope displacement information data;

[0050] The displacement state vector expression is Among them, x(k) represents the displacement of the monitoring point at time k, v(k) represents the relative displacement speed of the monitoring point at time k, T represents the displacement monitoring sampling time interval, and a(k) represents the acceleration of the measuring point from k to k+1;

[0051] The local filter is used to perform multi-source fusion processing on the slope displacement information data;

[0052] All slope displacement information data after multi-source fusion processing are distributed and processed using the following formula: Among them, β i represents the displacement information allocation factor, and i=1, 2, ..., n; n represents the number of local filters, which is determined according to the number of points on the monitoring surface; Q i (k) represents the variance of the Gaussian white noise sequence at time k; P(k) represents the global estimation error; represents the global optimal estimate;

[0053] The displacement state vector is updated according to the sampling period of the displacement information to obtain a new displacement state vector: In the formula, Represents the real-time displacement state prediction update result; represent The displacement gain matrix at time t; Represents the estimated value of displacement state; H i represents the feedback parameter of the local filter; represents the system state of the local filter;

[0054] The real-time displacement state prediction result is calculated by the formula The early warning threshold of the slope imminent sliding state is obtained by fusion.

[0055] A slope disaster early warning system, comprising:

[0056] Multiple monitoring sources, including remote sensing from satellites, drones, and ground sensor networks, including data on displacement, stress, rainfall, geological structure, and slope morphology;

[0057] Data transmission module, used to aggregate the data collected by the monitoring sources and transmit them through a combination of network and satellite communications;

[0058] Data acquisition module, used to establish a spatiotemporal database to store and manage slope data from multiple monitoring sources;

[0059] The data processing and analysis module is used to apply the pre-trained deep learning model to analyze the pre-processed data set in the spatiotemporal database to obtain the slope disaster value and potential risk area;

[0060] The dynamic early warning module is used to calculate the early warning threshold and real-time displacement prediction value of the slope based on the slope data in the spatiotemporal database, fuse the real-time displacement prediction value and the slope disaster value to obtain a fusion value, and evaluate the real-time stability of each monitoring section of the slope according to the obtained fusion value and early warning threshold, and determine whether to issue an early warning signal and an alarm signal based on the evaluation results.

[0061] Optionally, the dynamic warning module includes a warning threshold model and a slope function model;

[0062] The slope function model is used to process the real-time displacement information of multiple measuring points from each monitoring section of the slope by using the federal Kalman filter method, fuse the real-time displacement prediction value, and use the obtained real-time displacement prediction value to fuse the received slope disaster value to obtain a fusion value;

[0063] The warning threshold model is used to calculate the warning threshold based on the historical monitoring data and geological analysis in the spatiotemporal database, receive the fusion value of the slope function model, use the fusion value and the warning threshold to evaluate the real-time stability of each monitoring section of the slope, and determine whether it is necessary to issue a warning signal and an alarm signal.

[0064] Optionally, also include,

[0065] The 3D model demonstration module is used to build and display a refined 3D slope model, visually demonstrating the current state, historical changes, and predicted changes of the slope;

[0066] The geological disaster simulation module is used to simulate the development process of geological disasters based on the three-dimensional slope model, including the degree of disaster risk and the scope of disaster impact;

[0067] The remote control module is used to receive real-time data from multiple monitoring sources through a combination of network and satellite communications, and to update the three-dimensional slope model in real time.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] The present invention uses a combination of multiple sensors, including satellite remote sensing, drone remote sensing and ground sensor networks, to comprehensively obtain various physical parameters of the slope, such as displacement, stress, rainfall, geological structure and slope morphology. By establishing a spatiotemporal database and applying a deep learning model, accurate identification of slope disaster values ​​and potential risk areas is achieved, improving the accuracy and reliability of the early warning system.

[0070] By integrating multi-source monitoring data and spatiotemporal analysis methods, a comprehensive data set that fully reflects the state of the slope is formed, which provides rich features for the deep learning model and enhances the comprehensiveness and accuracy of the early warning analysis.

[0071] By using advanced feature extraction techniques such as principal component analysis and independent component analysis, key information is effectively extracted from multi-source monitoring data, providing accurate feature input for the deep learning model, further improving the predictive capability of the early warning system.

[0072] By analyzing the key information set after feature extraction through a pre-trained deep learning model, accurate identification of slope disaster values ​​and potential risk areas can be achieved, thereby improving the response speed and accuracy of the early warning system.

[0073] Dynamically adjusting the warning threshold and automatically issuing warning signals enable the warning system to respond flexibly based on real-time monitoring data and forecast results, providing timely warning information for disaster prevention and control, and effectively reducing the occurrence of disaster losses.

[0074] The federal Kalman filtering method and deep learning model fusion technology are used to improve the fusion accuracy of slope displacement information, enabling the early warning system to more accurately judge the stability and potential risks of the slope, thereby improving the accuracy of the early warning.

[0075] Through integrated platform management and intuitive demonstration of three-dimensional models, intuitive display and disaster simulation of massive geological disaster data are achieved, which greatly improves the decision-making efficiency of emergency management departments and the effectiveness of disaster prevention and control work, and enhances the ability to respond quickly to disasters.

[0076] By combining multiple sensors, various physical parameters of the slope are fully acquired, which improves the accuracy and reliability of monitoring. The efficient transmission method of the data acquisition and transmission module ensures the real-time and stability of the data, and ensures accurate data transmission even in complex terrain and harsh environments.

[0077] The recognition model and finite element analysis of deep learning algorithms have improved the accuracy of disaster simulation, prediction and early warning, and can more accurately assess the stability of slopes, providing a scientific basis for disaster prevention and control, and effectively improving the practicality and effectiveness of the early warning system. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 The figure is a control flow diagram of an embodiment of a slope disaster early warning method of the present invention.

[0079] Figure 2 The present invention is a schematic diagram of a slope disaster early warning system module. DETAILED DESCRIPTION

[0080] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0081] The following detailed description is an exemplary description, which is intended to provide further detailed description of the present invention. Unless otherwise specified, all technical terms used in the present invention have the same meaning as those generally understood by those skilled in the art to which the present application belongs. The terms used in the present invention are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention.

[0082] Example 1

[0083] like Figure 1As shown in the figure, a slope disaster early warning method uses spatiotemporal database technology to establish a spatiotemporal database of slopes. For slopes with different characteristics, corresponding spatiotemporal grid data systems are set up respectively to facilitate efficient data storage. By establishing indexes on the spatiotemporal grid, data can be quickly retrieved and extracted. At the same time, the unified spatial representation enables monitoring data at different locations to be organized and managed under a common framework to facilitate spatial analysis and comparison, improve the positioning accuracy and monitoring of real-time monitoring, and associate multidimensional data, and associate time and space data, so that the system can better understand the spatiotemporal evolution of slope deformation. Support spatial analysis and assist spatial analysis methods, such as spatial interpolation and buffer zone analysis, which can analyze the spatial relationship and pattern between grid data. For example, determine the diffusion range and impact area of ​​deformation, and provide strong support for slope stability assessment and prediction. This system, combined with spatiotemporal grid technology, can more easily be integrated with other related geographic information, geological data or data processed by different models, and can compensate for model errors or anomalies to improve the stability and robustness of data fusion. The system also includes an information acquisition module, which is used to detect slope data through multiple detection sources, including satellite remote sensing (combined with ground-based synthetic aperture radar technology) and drone remote sensing, to perform fully automatic intelligent monitoring of the slope, and to transmit the stable signal to the data preprocessing module, which transmits the preprocessed data to the spatiotemporal database. The system is also equipped with an analysis module to retrieve the disaster state information of the slope from the spatiotemporal database, and regionally analyze the slope type based on the collected disaster state information of the slope, and then generate the corresponding sensor layout plan. The layout plan output by this system should include the layout location map, sensor parameter table, monitoring frequency and other contents, to provide guidance and basis for the implementation of slope monitoring.

[0084] UAV remote sensing technology includes but is not limited to optical cameras, multispectral cameras, thermal infrared cameras, lidar, etc. for collecting slope data signals.

[0085] Satellite remote sensing technologies include but are not limited to optical sensors, infrared scanners, high-resolution cameras, synthetic aperture radar (SAR), microwave radiometers, imaging spectrometers, hyperspectral sensors, etc.

[0086] According to the corresponding deployment plan, the drone is equipped with the corresponding type of sensor, and the drone is used for flight deployment. After the deployment is completed, the sensor is activated and the parameters are set through wireless communication technology: high-precision displacement sensors, stress and strain sensors, rain gauges, geological radars, etc. are used during the deployment task. The displacement, stress, rainfall, geological structure, slope morphology and other key position parameters of the slope are monitored in an all-round and real-time manner, so as to obtain rich and comprehensive multi-source monitoring data.

[0087] Specifically, the data monitored by the multi-source monitoring sensors are pre-processed by the data preprocessing module and then input into the spatiotemporal database. The data processing and analysis module performs secondary processing on the satellite remote sensing data, UAV remote sensing data and sensor data input into it, and analyzes the data through the analysis module.

[0088] Example 2

[0089] Data preprocessing includes data cleaning, denoising, outlier removal, data format conversion, normalization, etc.

[0090] Median filtering, mean filtering and other methods are used to denoise the data to reduce the interference of noise on subsequent analysis. By combining statistical analysis and expert experience, obvious outliers are eliminated to ensure the reliability of the data.

[0091] The collected monitoring data from multiple monitoring sources can be subjected to a series of rigorous secondary processing, including but not limited to data analysis and visualization and extraction of key features. The data processing module includes but is not limited to the configuration of TBC (Trimble Business Center) and Geomagic Studio data processing tools.

[0092] TBC (Trimble Business Center) processes data from GNSS (Global Navigation Satellite System) to monitor the disaster status of the slope, analyze the geological information of the slope and output the disaster point information where the slope may suffer a disaster.

[0093] Geomagic Studio is used to process the 3D laser scanning data sent back by the laser scanning equipment carried by drones, and realize operations such as point cloud data splicing, denoising, and modeling. It can accurately reflect the surface morphology changes of the slope and provide detailed data support for analyzing the displacement and deformation of the slope.

[0094] Feature extraction: Use principal component analysis (PCA) to extract geological information, topographic information, and meteorological information in slope disasters. Analyze the main components in multiple stress monitoring data to reflect the concentration and distribution patterns of stress. Extract the main information factors that affect the disaster through independent component analysis (ICA), such as the increase in rainfall and the saturation of rock and soil, which leads to slope disasters. Use wavelet change data processing to analyze the pre-processed data in the spatiotemporal database, such as the time-frequency characteristics of displacement data, to find the mutation points and trend changes of displacement changes, so as to extract key information that can significantly reflect the characteristics of slope disasters from the processed massive monitoring data retrieved from the spatiotemporal database.

[0095] In the data processing and analysis module, a recognition model based on the deep learning algorithm is built, and the data processed by the principal component analysis (PCA) and the independent component analysis (ICA) are input into the recognition model of the deep learning algorithm. In this process, the distributor in the data processing module inputs the data into the convolutional neural network (CNN), recurrent neural network (RNN) and long short-term memory network (LSTM) models respectively. After the data from the spatiotemporal grid is processed, it is analyzed by the convolutional neural network (CNN), recurrent neural network (RNN) and long short-term memory network (LSTM), and the output data is fused, and then the fused features are input into the random forest (RF) to make the final deduction prediction. When training the model, the historical monitoring data of the spatiotemporal database is fully utilized, and combined with the actual slope disaster cases, the model is repeatedly trained and optimized, so as to establish an efficient and accurate slope disaster recognition model so that it can accurately identify the precursors and development trends of slope disasters.

[0096] In a preferred embodiment, the data input to the convolutional neural model is the image feature or matrix representation of the slope, represented by X in the shape of Where n is the number of samples; h and w are the height and width of the image, respectively, and c is the number of channels (e.g. different types of sensor data, terrain features, etc. can serve as different channels).

[0097] The convolutional neural network structure includes:

[0098] Suppose the output of the lth convolutional layer is , and its calculation formula is: Among them, * represents the convolution operation, is the convolution kernel weight matrix of layer l, is the bias term, Is an activation function, such as the ReLU function .

[0099] Pooling layer:

[0100] Usually, a pooling layer is used after the convolutional layer for downsampling to reduce the number of parameters and the amount of computation. Suppose the output of the pooling layer is , can be calculated by maximum pooling or average pooling, such as maximum pooling: , where s is the pooling window size.

[0101] Fully connected layer

[0102] After passing through multiple convolutional layers and pooling layers, the fully connected layer is connected to convert the feature map into a one-dimensional vector. Let the output of the fully connected layer be F, and the calculation formula is ,in is the output of the last convolutional layer or pooling layer, WF and b F are the weights and biases of the fully connected layer respectively.

[0103] Prediction output and loss function

[0104] Prediction Output

[0105] The last layer is usually the output layer. For slope disaster prediction, it can be a binary classification problem (disaster occurs or not), and the output is a probability value y, which can be obtained through the sigmoid function: .

[0106] Loss Function

[0107] Use a suitable loss function to measure the difference between the predicted value and the true value. For a binary classification problem, you can use the binary cross entropy loss function: ,in is the true label value (0 or 1).

[0108] Model Training

[0109] Optimization Algorithm

[0110] Use an optimization algorithm to minimize the loss function and update the weights and biases of the network. Common optimization algorithms include stochastic gradient descent (SGD), Adam, etc. For example, when using the Adam optimization algorithm, the update rule is: , where t is the number of iterations, α is the learning rate, and m t and v t It is the first-order moment estimate and second-order moment estimate in the Adam algorithm, and e is a very small constant.

[0111] Training process

[0112] The training data is input into the convolutional neural network, the prediction value is calculated by forward propagation, and then the loss is calculated according to the loss function. The gradient of the loss to each weight and bias is calculated by the back propagation algorithm, and finally the weight and bias are updated using the optimization algorithm. This process is repeated until the loss converges or the predetermined number of iterations is reached.

[0113] In a preferred embodiment, a model based on a recurrent neural network (RNN) in slope disaster prediction is:

[0114] Time series data organization: organize sensor data into a time series format, with the data at each time point as a sample. For example, for a slope monitoring system containing n sensors, the data at time point t can be represented as an n-dimensional vector X t .

[0115] Recurrent Neural Network Architecture

[0116] Input layer: receives time series data as input. The number of nodes in the input layer is equal to the number of sensors.

[0117] Hidden layer: usually composed of multiple recurrent neurons. Recurrent neurons have a memory function and can remember previous input information to better process time series data.

[0118] Output layer: Output the prediction results of slope disaster. For binary classification problems (disaster occurs or not), the output layer can be a single node, using the sigmoid activation function to output a probability value.

[0119] Computation of recurrent neurons

[0120] Simple Recurrent Neuron (SRN): At each time step t, the output h of a simple recurrent neuron is t From the current input X t and the hidden state h at the previous time step t-1 Joint decision. The calculation formula is: , where U and W are weight matrices, b is the bias term, is an activation function, such as the tanh function or the ReLU function.

[0121] Long Short-Term Memory Network (LSTM), LSTM unit consists of input gate, forget gate, output gate and cell state. At each time step t, the calculation process of LSTM unit is as follows:

[0122] Forget Gate: , determines what information is forgotten from the cell state.

[0123] Input Gate: , determines the information that should be added to the cell state.

[0124] Cell status update: , calculate new candidate cell states. , update the cell state.

[0125] Output Gate: , determines what information should be output from the cell state. , output the hidden state.

[0126] in, Represents the value of the forget gate at time step t, between 0 and 1, Indicates that the sigmoid function maps the input value to between 0 and 1. represents the weight matrix of the forget gate, X t Represents the input at the current moment, represents the weight matrix of the hidden layer, ht-1 Indicates the hidden state at the previous moment, represents the forget gate bias, represents the value of the input gate at time step t, represents the weight matrix of the input gate, represents the weight matrix of the hidden layer, Represents the input gate bias. represents the cell state at time step t, Candidate cell states.

[0127] Model Training

[0128] Loss function: Choose a suitable loss function to measure the difference between the model's prediction and the true label. For binary classification problems, you can use the binary cross entropy loss function.

[0129] Optimization algorithm: Use an optimization algorithm to minimize the loss function and update the parameters of the model. Common optimization algorithms include stochastic gradient descent (SGD), Adam, etc.

[0130] Training process: Input the sorted time series data into the recurrent neural network, calculate the prediction results through forward propagation, then calculate the loss according to the loss function, and then calculate the gradient of the loss to the model parameters through the back propagation algorithm, and finally use the optimization algorithm to update the parameters. Repeat this process until the loss converges or the predetermined number of training times is reached.

[0131] Model evaluation and application

[0132] Model evaluation: Use test data to evaluate the trained model and calculate evaluation indicators such as accuracy, recall, F1 value, etc. to determine the performance of the model.

[0133] Model application: Apply the trained model to the actual slope disaster prediction, input sensor data in real time, and the model outputs the probability prediction results of slope disasters. According to the prediction results, corresponding early warning and prevention measures can be taken.

[0134] It should be noted that in practical applications, data preprocessing, feature selection, hyperparameter adjustment and other operations are also required to improve the performance and generalization ability of the model. At the same time, combined with other technologies such as data fusion and model integration, the accuracy and reliability of slope disaster prediction can be further improved.

[0135] In a preferred embodiment, the early warning threshold model collects displacement monitoring data of the slope during the period of imminent landslide and fuses the data to generate an early warning threshold through the following model.

[0136] The displacement space state expression of the warning threshold model is: Where: X(k) is the displacement state value at time k; W(k) and V(k) are independent Gaussian white noise sequences representing displacement information; their variances are Q and R respectively; Z(k) is the monitoring value of the slope by the sensor; Φ represents the state transfer matrix; Γ represents the input matrix, and H is the feedback parameter of the filter. The displacement space state of the monitoring point can be characterized by the slope displacement and displacement velocity. Therefore, the displacement and displacement velocity are used as state vectors to obtain the expression: Where x(k) represents the displacement of the monitoring point at time k, mm; v(k) represents the relative displacement velocity of the monitoring point at time k, mm / s; T represents the displacement monitoring sampling time interval s; a(k) represents the acceleration of the measuring point from k to k+1, mm / s 2 In this regard, this paper combines local filters to perform multi-source fusion processing on the above slope displacement information data. The distribution processing of all slope displacement monitoring information data is shown in the following formula:

[0137] Among them, β i represents the displacement information allocation factor, and i=1, 2, ..., n; n represents the number of local filters; it is determined according to the number of points on the monitoring surface, Q i (k) represents the variance of the Gaussian white noise sequence at time k; P(k) represents the global estimation error; represents the global optimal estimate. The state vector is updated according to the sampling period of the displacement information to obtain a new state vector. The specific update result is formula (4):

[0138] In the formula, Represents the real-time displacement state prediction update result; represent The displacement gain matrix at time Represents the estimated displacement state value H i represents the feedback parameter of the local filter, Represents the system state of the local filter.

[0139] According to the formula The real-time displacement state prediction update results of each filter are fused, and the fused data can be used as the displacement warning threshold of the slope slip state. Where P (K + 1) is the error covariance matrix.

[0140] The displacement information collected by the sensor is input into the deep learning algorithm recognition model and the warning threshold model through a signal splitter. The real-time displacement prediction value obtained by the warning model and the prediction value obtained by the deep learning algorithm recognition model are assigned different weights and fused through a weighted average algorithm. When the fused value is higher than the warning threshold, the system issues a warning.

[0141] Example 3

[0142] The system's precise risk identification technology. According to the slope information of each specific project, the warning threshold setting is optimized; the system automatically identifies the linkage characteristics between the monitoring points through the long short-term memory network (LSTM) model that has been trained to learn the dynamic relationship between the monitoring points, and automatically determines the warning level according to the changes in the monitoring parameters and the number of alarms at the monitoring points, triggering the corresponding level of warning. In the present invention, different monitoring parameters and alarm quantity indicators are set for different geological characteristics of slopes and the surrounding environment of the slopes. In a preferred embodiment, when there are buildings in the surrounding environment and the slope is soil, when the displacement corresponding to the monitoring parameter reaches 20-30 mm on the day, and the number of warnings reaches 8 times, a second-level alarm message is issued.

[0143] When a disaster alarm is issued, the preset alarm threshold is multiple safety levels. When the output result exceeds the preset scientific and reasonable alarm threshold, different ranges of forecasts and warnings are immediately issued through multiple channels for different safety levels. Accurate and reliable forecast and warning information is promptly issued through SMS, email, sound and light alarms, etc., to notify relevant personnel to quickly take corresponding preventive and response measures.

[0144] Based on the three-dimensional deduction and early warning technology, a refined identification, prediction and early warning model for slope disasters is established, and massive geological disaster data can be directly displayed in three dimensions. The data can be integrated in multiple dimensions, such as geospatial data, time series data, socio-economic data, etc., so as to conduct a comprehensive assessment of risks.

[0145] The corresponding corresponding ones are "tailored" to improve the rationality, accuracy and effectiveness of the current prediction model and make the slope risk analysis more comprehensive.

[0146] The data processing process of the data processing module of this system relies on big data, cloud computing and satellite Internet technology to ensure that the model has sufficient computing power, analysis and early warning capabilities when dealing with the rapid growth of various data.

[0147] Example 4

[0148] like Figure 2 As shown, a slope disaster early warning system includes:

[0149] Multiple monitoring sources, including data from satellite remote sensing, UAV remote sensing and ground sensor networks. The ground sensor network data includes displacement, stress, rainfall, geological structure and slope morphology; it also includes displacement sensors, stress sensors, rainfall sensors, tilt sensors, etc., which are used to collect various physical parameters of the slope in real time.

[0150] Data transmission module, used to aggregate the data collected by the monitoring sources and transmit them through a combination of network and satellite communications;

[0151] Data acquisition module, used to establish a spatiotemporal database to store and manage slope data from multiple monitoring sources;

[0152] The data processing and analysis module is used to apply the pre-trained deep learning model to analyze the pre-processed data set in the spatiotemporal database to obtain the identification results of slope disaster values ​​and potential risk areas; the specific data processing and analysis module processes and analyzes the collected data, and uses advanced algorithms and models, including finite element analysis, neural networks, etc., to accurately evaluate the stability of the slope.

[0153] The dynamic early warning module is used to calculate the early warning threshold and real-time displacement prediction value of the slope based on the slope data in the spatiotemporal database, fuse the real-time displacement prediction value and the slope disaster value to obtain a fusion value, and evaluate the real-time stability of each monitoring section of the slope according to the obtained fusion value and early warning threshold, and determine whether to issue an early warning signal and an alarm signal based on the evaluation results.

[0154] Specifically, the dynamic warning module determines the dynamic graded warning threshold of displacement of each profile based on the monitoring profile of the slope, taking into account the layout time of each measuring point and the dynamic change of slope height during slope excavation. By establishing numerical models of each monitoring profile, simulating the step-by-step excavation process, calculating the displacement value of each measuring point under different mining heights and graded warning safety factors, integrating the graded warning displacement values ​​of all measuring points at the same monitoring profile at the same mining height, taking the average value as the graded warning threshold of slope displacement, and obtaining the dynamic graded warning threshold of displacement of all monitoring profiles at different slope excavation heights through curve fitting, matching the corresponding danger level and issuing corresponding warning information according to the corresponding level. When the fusion value of the predicted slope displacement state is greater than the warning threshold of the corresponding level, a warning signal is issued in time to notify relevant personnel to take corresponding measures.

[0155] Specifically, the dynamic warning module includes a warning threshold model and a slope function model.

[0156] The slope function model is used to process the real-time displacement information of multiple measuring points from each monitoring section of the slope by using the federal Kalman filter method, fuse the real-time displacement prediction value, and use the obtained real-time displacement prediction value to fuse the received slope disaster value to obtain a fusion value;

[0157] The warning threshold model is used to calculate the warning threshold based on the historical monitoring data and geological analysis in the spatiotemporal database, receive the fusion value of the slope function model, use the fusion value and the warning threshold to evaluate the real-time stability of each monitoring section of the slope, and determine whether it is necessary to issue a warning signal and an alarm signal.

[0158] As a preferred example of the above example, it also includes a three-dimensional model demonstration module, which is used to construct and display a refined three-dimensional slope model, and intuitively demonstrate the current state, historical changes and predicted changes of the slope; a geological disaster simulation module, which is used to simulate and display the development process of geological disasters based on the three-dimensional slope model, including the degree of disaster risk and the scope of disaster impact; a remote management and control module, which is used to receive real-time data from multiple monitoring sources through a combination of network and satellite communications, and update the three-dimensional slope model in real time.

[0159] The present invention adopts a combination of multiple sensors to comprehensively acquire various physical parameters of the slope and improve the accuracy and reliability of monitoring.

[0160] The data transmission module adopts a transmission method that combines 4G network communication and Ethernet lines to ensure the stability and real-time performance of data transmission. Even in complex terrain and harsh environmental conditions, it can ensure accurate data transmission.

[0161] The data processing and analysis module uses the recognition model of deep learning algorithms to improve the accuracy of disaster simulation, prediction and early warning, and can more accurately evaluate the stability of the slope. Through finite element analysis, the stress and strain state of the slope under different working conditions can be simulated; the neural network model can perform intelligent prediction and analysis of monitoring data.

[0162] The early warning module adopts the method of warning first and then alarming, which buys precious time for relevant personnel to take measures and effectively reduces the occurrence of disaster losses.

[0163] Design of system logic model based on data fusion. Integrate the characteristics of each type of slope and the data in the spatiotemporal data warehouse, calculate the intrinsic functional relationship between monitoring parameters and time and inducing factors, establish the underlying early warning framework logic of the sub-model, and then build a comprehensive early warning model system.

[0164] Intelligent spatiotemporal data can sense subtle changes at the millimeter level. The construction of the spatiotemporal database uses cutting-edge monitoring technologies such as big data, satellite navigation, and drones to achieve accurate, real-time, and unmanned operation of slope data monitoring. The data accuracy reaches the millimeter level, and the system will automatically increase the monitoring frequency in severe weather conditions.

[0165] Dynamic warning threshold, slope risk judgment in seconds. Based on the initial warning value, the warning threshold of each subsystem is dynamically optimized according to the state of each slope, making the system's judgment of slope risk more accurate and effectively avoiding false alarms, misreporting, and missed reports.

[0166] This system uses an integrated platform for management and control, which can realize intuitive demonstration of three-dimensional models and visualize geological disaster simulation. It can also realize remote control, intelligent analysis and disaster warning of the system. The refined three-dimensional slope model can more intuitively display massive geological disaster data; geological disaster simulation can visualize the risk level of geological disasters and the scope of disaster impact, which greatly helps the geological disaster prevention and control work of emergency management departments.

[0167] The technical solution of the present invention achieves real-time monitoring data accuracy of millimeter level, which improves the accuracy of slope monitoring. The relative error of monitoring data of slope monitoring equipment can be reduced to less than 5%. The monitoring frequency of the early warning system and the frequency of system algorithm model analysis results reach second-level, and the accuracy of early warning information release is above 95%.

[0168] Specifically, among the sensors deployed in this system, the displacement sensor can accurately measure the displacement changes on the surface and inside of the slope and capture small displacement trends in a timely manner.

[0169] Stress-strain sensors can monitor the stress distribution and changes of slope rock or soil and identify potential stress concentration areas.

[0170] The rain gauge records rainfall in real time, providing data support for analyzing the impact of rainfall on slope stability.

[0171] Geological radar is used to detect the geological structure and potential weak layers, cracks, etc. inside the slope.

[0172] In the slope monitoring of a certain expressway, the high-precision slope monitoring system of the present invention is adopted.

[0173] The deployment scheme obtained by remote sensing technology in this system generally includes placing sensors for slope monitoring at the top and bottom of the slope, near the potential sliding surface, and in weak geological structures: such as faults, cracks, weak interlayers, etc. When determining the potential sliding surface, the approximate location of the potential sliding surface is determined through geological surveys, and then the sensors are placed on the slopes near it. Displacement sensors, stress sensors, and rainfall sensors are arranged. The displacement sensor uses a high-precision laser displacement sensor and is installed at key parts of the slope, such as the top, foot, and near the potential sliding surface to monitor the displacement changes of the slope surface in real time. The stress sensor uses a vibrating string stress sensor or a light grating stress sensor, which is buried at different depths inside the slope to measure the stress state of the rock and soil. The rainfall sensor is installed in an open area near the slope to accurately record the rainfall.

[0174] The data processing and analysis module uses the finite element analysis algorithm to establish a numerical model of the slope. The displacement and stress data collected by the sensor are input into the model as boundary conditions to calculate the stress and strain distribution of the slope. At the same time, the convolutional neural network (CNN), recurrent neural network (RNN) and long short-term memory network (LSTM) models are used to learn and train historical monitoring data to predict the future deformation trend of the slope.

[0175] The dynamic warning module can set three levels of warning thresholds. When the displacement changes between 105-180mm, a third-level alarm is issued, and a primary warning is issued; when it is between 180-240mm, a second-level warning is issued; when the change is between 240-840mm, it reaches a serious danger value and a high-level warning is issued. After receiving the warning information, relevant staff can take corresponding measures in a timely manner, such as strengthening inspections, setting up warning signs, taking reinforcement measures, etc., to ensure the safe operation of the highway.

[0176] Further, in a highway slope monitoring, the sensors of the monitoring system provided by the present invention include: a high-precision GPS displacement sensor, installed on the slope, to achieve large-scale displacement monitoring of the slope surface. A fiber grating stress sensor, which has the characteristics of strong anti-interference ability and high precision, is embedded in the rock mass of the slope. A rainfall sensor is installed on the side of the slope close to the highway to accurately measure the rainfall.

[0177] The data transmission module uses a combination of 5G network and satellite communication. In areas with 5G network coverage on the slope, data is transmitted using the 5G network, while in remote areas, data is transmitted to the data processing center via satellite communication.

[0178] The data processing and analysis module uses a deep learning-based algorithm to automatically analyze and extract features from a large amount of collected data. It can quickly identify abnormal patterns and trends in slope deformation, providing a more accurate basis for assessing slope stability.

[0179] The dynamic warning module sets different levels of warning thresholds according to the actual conditions of the slope and safety standards. When the monitoring data reaches the warning threshold, warning information is sent to managers and operators through SMS, email, and on-site alarms to ensure that personnel can evacuate the dangerous area in time to avoid accidents.

[0180] Specifically, the high-precision slope monitoring system of the present invention is used in the monitoring of the dam slope of a water conservancy project. The displacement sensor adopts a measuring robot, which can automatically perform high-precision displacement measurement on multiple monitoring points. The stress sensor adopts a piezoelectric stress sensor and is installed in the key structural parts of the dam slope. The rainfall sensor is installed at a suitable position around the dam to accurately monitor the rainfall.

[0181] The three-dimensional model demonstration module and the geological disaster simulation module simulate the slope stability under different working conditions by establishing a three-dimensional numerical model of the dam slope; at the same time, they conduct statistical analysis on long-term monitoring data to find out the laws and trends of slope deformation.

[0182] The collected data was initially cleaned and denoised to remove outliers caused by equipment failure or external interference. Then, the principal component analysis method was used to extract features from the processed data, focusing on the long-term trend of displacement and the concentrated area of ​​stress.

[0183] Based on the extracted features, a model combining a neural network recognition model and a random forest (RF) was trained. The training data included monitoring data from the past two years and data related to a small rock collapse. During the training process, the model parameters were continuously adjusted to improve the accuracy of the model.

[0184] In real-time monitoring, when newly collected data is input into the model, if the risk value predicted by the model exceeds the preset threshold, the system immediately issues an early warning to relevant departments and passing vehicles through text messages and sound and light alarms on the roadside. During a period of continuous heavy rainfall, the system successfully predicted the possibility of a large-scale collapse of the slope two days in advance, and took timely traffic control and emergency response measures to avoid casualties and major property losses.

[0185] Specifically, consider a soil slope located in an urban construction area surrounded by residential buildings and important infrastructure. On this slope, displacement sensors, pore water pressure sensors, drone remote sensing equipment, and automatic rain gauges are installed. The displacement sensor collects data four times a day, the pore water pressure sensor collects data every two hours, the drone takes aerial photos once a week, and the rain gauge records rainfall in real time.

[0186] The collected data were first formatted and normalized for subsequent analysis. Then, the time-frequency characteristics of the displacement data were extracted by wavelet transform, and the key components were extracted from the pore water pressure data by independent component analysis.

[0187] When real-time monitoring data is input into the model, if the prediction results show that there is a risk of disaster and it exceeds the warning threshold of the corresponding level, the system will immediately send warning information to relevant departments and residents via email and set up warning signs in the surrounding area. After a heavy rain, the system issued a warning one day in advance, allowing relevant personnel to evacuate residents in time and protect life and property safety.

[0188] In practical applications, this system can be flexibly adjusted and optimized according to specific slope types, geological conditions, monitoring requirements, etc. In different implementation scenarios, the slope disaster early warning method of the present invention has demonstrated excellent performance and practical value, and can play an important role in protecting people's lives and property safety and social stability. The slope disaster early warning method and system of the present invention can play a good role in different types of engineering slope monitoring, providing an effective technical means to ensure engineering safety.

[0189] It is known from common technical knowledge that the present invention can be implemented by other embodiments that do not deviate from its spirit or essential features. Therefore, the above disclosed embodiments are only illustrative in all respects and are not exclusive. All changes within the scope of the present invention or within the scope equivalent to the present invention are included in the present invention.

Claims

1. A slope disaster early warning method, characterized in that: include, Establish a spatiotemporal database to store and manage pre-processed slope data from multiple monitoring sources and organize the slope disaster data set; The steps of establishing the spatiotemporal database specifically include: Collect raw data from multi-source slope monitoring data, including data from satellite remote sensing, UAV remote sensing, and ground sensor networks. The data from the ground sensor network includes displacement, stress, rainfall, geological structure, and slope morphology; Preprocess the collected raw data. The preprocessing operations include data cleaning, denoising, outlier removal, data format conversion and normalization to obtain the preprocessed data set. Integrate the preprocessed data sets into the spatiotemporal database, create indexes on the spatiotemporal grid for fast data retrieval and extraction, organize and manage data at different locations and times using a unified spatial representation framework, and obtain integrated data sets that can be used for further analysis; In the spatiotemporal database, spatial analysis methods are applied to conduct correlation analysis on the temporal and spatial data in the integrated data set, analyze the spatiotemporal evolution of slope deformation, determine the diffusion range and impact area of ​​deformation, and obtain spatiotemporal analysis results for slope stability assessment; Using multi-source monitoring data and the obtained spatiotemporal analysis results, a slope disaster data set that fully reflects the slope status is formed through a fusion algorithm; Extract features from the slope disaster dataset to obtain key information sets reflecting the characteristics of the slope disaster. Specifically, use the principal component analysis method to analyze the data of the slope disaster dataset, extract the main components in the data, reflect the stress concentration and distribution pattern, and obtain the key feature sets of geology and topography; The key feature set was further processed using independent component analysis to extract the main information factors affecting slope disasters. Wavelet transform data processing method is used to conduct time-frequency analysis on the slope disaster data set to identify the mutation points and trend changes in the displacement data; Retrieve monitoring data related to the identified mutation points and trend changes from the spatiotemporal database to extract key information sets reflecting the characteristics of slope disasters; The obtained key feature set is combined with the extracted key information set to form a complete information set reflecting the characteristics of slope disasters; Apply the pre-trained deep learning model to analyze the key information set after feature extraction to obtain the slope disaster value and potential risk area; Based on the slope data in the spatiotemporal database, the warning threshold and real-time displacement prediction value of the slope are calculated, and the real-time displacement prediction value and the slope disaster value are fused to obtain the fusion value; According to the obtained fusion value and warning threshold, the real-time stability status of each monitoring section of the slope is evaluated, and it is determined whether to issue a warning signal and an alarm signal based on the evaluation results.

2. A slope disaster early warning method according to claim 1, characterized in that: The pre-training steps of deep learning models include: Based on the historical monitoring data in the spatiotemporal database, data preprocessing is performed, including normalization and outlier removal, to form a data set for training; Divide the preprocessed data set into a training subset and a validation subset; initialize at least one or more deep learning models including a convolutional neural network, a recurrent neural network, and a long short-term memory network; Use the training subset to train the initialized deep learning model, and obtain the initial parameters of the model through forward propagation and loss function calculation; use the validation subset to verify the obtained initial model parameters, and adjust the model parameters through the back propagation algorithm to optimize the model performance; repeat this step until the model reaches the predetermined accuracy or loss threshold on the validation subset; The trained and optimized model parameters are fixed to form a pre-trained deep learning model for slope disaster early warning analysis.

3. A slope disaster early warning method according to claim 1, characterized in that: The steps of applying the pre-trained deep learning model to analyze the key information set after feature extraction to obtain the identification results of the slope disaster value and potential risk area include: The key information set after feature extraction is input into the pre-trained deep learning model, including convolutional neural network, recurrent neural network and long short-term memory network, for feature recognition and analysis; The convolutional neural network model is used to analyze the image features in the extracted data, identify the spatial patterns and abnormal features related to the slope disaster, and obtain the first analysis result; The time series data are analyzed using the recurrent neural network and long short-term memory network models to identify the deformation trend and potential disaster precursors over time, and obtain the second analysis result; The first analysis result and the second analysis result are fused, and the fused result is input into the random forest to obtain the comprehensive identification result of the slope disaster value and the potential risk area.

4. A slope disaster early warning method according to claim 1, characterized in that: Based on the slope data in the spatiotemporal database, the warning threshold and real-time displacement prediction value of the slope are calculated, the real-time displacement prediction value and the slope disaster value are fused to obtain a fusion value, and the real-time stability state of each monitoring section of the slope is evaluated according to the obtained fusion value and warning threshold. The steps of determining whether to issue a warning signal and an alarm signal according to the evaluation results specifically include: Receive the slope disaster value from the deep learning model; obtain the displacement monitoring data of the slope in the period of imminent landslide from the spatiotemporal database monitoring data, wherein the displacement monitoring data of the slope in the period of imminent landslide includes the real-time displacement information of multiple measuring points of the slope monitoring profile; The federated Kalman filter method is used to process the real-time displacement information of multiple measuring points in the slope monitoring profile, and the real-time displacement prediction value is obtained by fusion; the real-time displacement prediction value and the slope disaster value are fused to obtain the fusion value; The real-time stability status of each monitoring section of the slope is evaluated using the fusion value and the warning threshold; the evaluation result is compared with the determined warning threshold to determine whether a warning signal needs to be issued; When the evaluation results show that the stability of the slope exceeds the warning threshold, a preliminary warning signal is first issued, and then the danger level and scope of the disaster are automatically deduced according to the number of warning signals corresponding to different alarm signals; Determine the level of the early warning signal based on the deduced danger level, and issue an alarm signal of the corresponding level based on the preset multiple safety level thresholds; Send early warning and alarm signals in a timely manner through multiple channels; Record detailed information of warning and alarm events, including time, location, catastrophic value, risk level, issuance time of alarm and warning signals, and response measures taken, for subsequent analysis and system optimization.

5. A slope disaster early warning method according to claim 4, characterized in that: In the step of calculating the early warning threshold of the slope based on the slope data in the spatiotemporal database, the calculation step specifically includes: Define the slope displacement state vector X(k), which represents the displacement state value of the monitoring point at time k, including the displacement x(k) and relative displacement velocity v(k); According to the displacement state vector X(k) and the Gaussian white noise sequence W(k) of the displacement information, the state transfer matrix Φ and the input matrix Γ are used to calculate the displacement state value X(k+1) at the next moment through the expression X(k+1)=ΦX(k)+ΓW(k) to obtain the slope displacement information data; The displacement state vector expression is Among them, x(k) represents the displacement of the monitoring point at time k, v(k) represents the relative displacement speed of the monitoring point at time k, T represents the displacement monitoring sampling time interval, and a(k) represents the acceleration of the measuring point from k to k+1; The local filter is used to perform multi-source fusion processing on the slope displacement information data; All slope displacement information data after multi-source fusion processing are distributed and processed using the following formula: Among them, β i represents the displacement information allocation factor, and i=1, 2, ..., n; n represents the number of local filters, which is determined according to the number of points on the monitoring surface; Q i (k) represents the variance of the Gaussian white noise sequence at time k; P(k) represents the global estimation error; Represents the global optimal estimate; the displacement state vector is updated according to the sampling period of the displacement information to obtain a new displacement state vector: In the formula, Represents the real-time displacement state prediction update result; represent The displacement gain matrix at time t; Represents the estimated value of displacement state; H i represents the feedback parameter of the local filter; represents the system state of the local filter; The real-time displacement state prediction result is calculated by the formula The early warning threshold of the slope imminent sliding state is obtained by fusion.

6. A slope disaster early warning system, based on a slope disaster early warning method according to any one of claims 1 to 5, characterized in that: include, Multiple monitoring sources, including remote sensing from satellites, drones, and ground sensor networks, including data on displacement, stress, rainfall, geological structure, and slope morphology; Data transmission module, used to aggregate the data collected by the monitoring sources and transmit them through a combination of network and satellite communications; Data acquisition module, used to establish a spatiotemporal database to store and manage slope data from multiple monitoring sources; The data processing and analysis module is used to apply the pre-trained deep learning model to analyze the pre-processed data set in the spatiotemporal database to obtain the slope disaster value and potential risk area; The dynamic early warning module is used to calculate the early warning threshold and real-time displacement prediction value of the slope based on the slope data in the spatiotemporal database, fuse the real-time displacement prediction value and the slope disaster value to obtain a fusion value, and evaluate the real-time stability of each monitoring section of the slope according to the obtained fusion value and early warning threshold, and determine whether to issue an early warning signal and an alarm signal based on the evaluation results.

7. A slope disaster early warning system according to claim 6, characterized in that: The dynamic early warning module includes an early warning threshold model and a slope function model; The slope function model is used to process the real-time displacement information of multiple measuring points from each monitoring section of the slope by using the federal Kalman filter method, fuse the real-time displacement prediction value, and use the obtained real-time displacement prediction value to fuse the received slope disaster value to obtain a fusion value; The warning threshold model is used to calculate the warning threshold based on the historical monitoring data and geological analysis in the spatiotemporal database, receive the fusion value of the slope function model, use the fusion value and the warning threshold to evaluate the real-time stability of each monitoring section of the slope, and determine whether it is necessary to issue a warning signal and an alarm signal.

8. A slope disaster early warning system according to claim 6, characterized in that: Also includes, The 3D model demonstration module is used to build and display a refined 3D slope model, visually demonstrating the current state, historical changes, and predicted changes of the slope; The geological disaster simulation module is used to simulate the development process of geological disasters based on the three-dimensional slope model, including the degree of disaster risk and the scope of disaster impact; The remote control module is used to receive real-time data from multiple monitoring sources through a combination of network and satellite communications, and to update the three-dimensional slope model in real time.

Images