Monitoring and early warning system for geotechnical engineering

Abstract

The invention discloses a monitoring and early warning system for geotechnical engineering, and relates to the technical field of geological disaster early warning. A monitoring and early warning system for geotechnical engineering comprises a data acquisition module, a data processing module, a model construction module and a disaster early warning module. According to the method, the displacement and the shear stress of the rock-soil body are predicted through the displacement-hydrological-geological coupling prediction model, so that the comprehensive influence of the mechanical state of the rock-soil body and environmental factors on geotechnical engineering disasters can be more comprehensively reflected, and the coupling effect of various monitoring data can be comprehensively considered; therefore, prediction errors and uncertainty caused by data missing or model singleness can be effectively reduced, hydrological data are processed through causal convolution, the long-term dependency relationship among the hydrological data can be effectively captured, the relevance and hysteresis among the water level, the rainfall and the pore water pressure can be reflected, and the prediction accuracy of the hydrological data is improved. Therefore, the accuracy of model prediction can be remarkably improved, and the early warning accuracy of the monitoring and early warning system is also improved.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster warning, and particularly to a monitoring and warning system for geotechnical engineering. Background Art

[0002] Geotechnical engineering is a branch of civil engineering, mainly studying the behavioral characteristics of natural or artificially modified rock and soil materials and their applications in engineering construction. It covers multiple fields such as geological exploration, foundation treatment, slope stability analysis, and tunnel and underground space development. Due to the characteristics of rock and soil masses such as inhomogeneity, anisotropy, and time effect, their behaviors are often complex and changeable and prone to forming geotechnical engineering disasters, and there are various types of disasters in geotechnical engineering.

[0003] The existing monitoring and warning systems for geotechnical engineering mainly first use devices such as sensors to monitor the key parameters of rock and soil in real time, such as displacement, stress, and seepage pressure, etc., and then use machine learning models to predict the occurrence of geotechnical engineering disasters and take corresponding measures. However, this warning method only considers the relationship between the monitoring data and geotechnical engineering disasters such as landslides, ignoring the correlation and coupling effects between the monitoring data. For example, the triggering of a landslide may be caused by a chain reaction of "rainwater infiltration leading to an increase in pore water pressure, then causing a decrease in the shear strength of the rock and soil, and finally resulting in a sudden displacement change", thus making the prediction accuracy poor. And when the existing machine learning models predict and output results, they may learn wrong rules due to the presence of noise and deviation in the input data, and the output results do not conform to the actual situation and physical laws, thus affecting the rationality of the prediction.

[0004] Based on the above situation, the present invention proposes a monitoring and warning system for geotechnical engineering with accurate and reasonable warning. Summary of the Invention

[0005] In order to overcome the shortcomings that the existing monitoring and warning systems for geotechnical engineering only consider the relationship between the monitoring data and geotechnical engineering disasters such as landslides, ignoring the correlation and coupling effects between the monitoring data, thus making the prediction accuracy poor, and when outputting the prediction results, they may learn wrong rules due to the presence of noise and deviation in the input data, and the output results do not conform to the actual situation and physical laws, thus affecting the rationality of the prediction, the present invention proposes a monitoring and warning system for geotechnical engineering with accurate and reasonable warning.

[0006] The monitoring and warning system for geotechnical engineering includes: A data acquisition module, configured to obtain the historical displacement data and historical hydrological data of the rock and soil body to be predicted through sensors, and obtain the historical geological data of the rock and soil body through a geological exploration report, wherein the historical geological data includes static geological data and dynamic geological data; A data processing module for preprocessing the acquired historical displacement data, historical hydrological data, and dynamic geological data to obtain preprocessed historical displacement data, historical hydrological data, and dynamic geological data; A model construction module for constructing a displacement-hydrology-geology coupling prediction model, and incorporating the Mohr-Coulomb criterion as a regularization term into the loss function of the model during model training, calculating the prediction error of the model through the loss function, and iteratively optimizing the model through backpropagation and parameter update; A disaster warning module for inputting the static geological data, preprocessed historical displacement data, historical hydrological data, and dynamic geological data of the rock and soil mass to be predicted into the trained displacement-hydrology-geology joint prediction model to obtain a prediction result, where the prediction result includes the displacement curve of the rock and soil mass in the next 72 hours and shear stress curve , classifying the rock and soil mass into different risk levels according to the prediction result, and initiating alarms at corresponding levels.

[0007] As a preferred aspect of the invention, the historical displacement data is the displacement time series of the rock and soil mass, the historical hydrological data includes an instantaneous rainfall intensity time series, a groundwater level height time series, and a pore water pressure time series, and the static geological data includes the type, slope, vegetation coverage, fracture development index, cohesion of the rock and soil mass and internal friction angle , and the dynamic geological data is the normal stress time series and shear stress time series of the rock and soil mass. time series of the rock and soil mass.

[0008] As a preferred aspect of the invention, the displacement-hydrology-geology coupling prediction model includes three input layers, three feature extraction layers, a spatio-temporal fusion layer, and two joint prediction layers. Among them, the three input layers are respectively: a time series input layer for receiving historical displacement data and dynamic geological data; a hydrological series input layer for receiving historical hydrological data; a geological parameter input layer for receiving static geological data; among the three feature extraction layers are respectively: an LSTM encoder layer for feature extraction of the time series received by the time series input layer; a temporal convolutional network layer for local feature extraction of the time series received by the hydrological series input layer; a fully connected embedding layer for mapping the high-dimensional geological information received by the geological parameter input layer into a low-dimensional feature vector; the spatio-temporal fusion layer is used to splice the outputs of the three feature extraction layers into a feature vector, and perform weighted fusion on the spliced feature vector through an attention mechanism to obtain a final fused feature vector; among the two joint prediction layers are respectively: a displacement prediction layer for outputting the displacement curve in the next 72 hours ; a shear stress prediction layer for outputting a shear stress curve for the next 72 hours .

[0009] As a preferred aspect of the invention, the training steps of the displacement-hydrogeology-geology coupling prediction model are specifically as follows: Data preparation and preprocessing, obtaining historical displacement data, historical hydrogeological data, and historical geological data of the sample rock and soil mass, preprocessing all the obtained data, and dividing all the preprocessed data into a training set, a validation set, and a test set according to a ratio; Forward propagation and loss calculation, inputting the training set into the displacement-hydrogeology-geology coupling prediction model, and obtaining a predicted displacement curve and a predicted shear stress curve , according to the predicted displacement curve and the actual displacement curve to obtain a displacement loss term , according to the predicted shear stress curve and the ultimate shear stress curve to obtain a physical law loss term based on the Mohr-Coulomb criterion , calculating to obtain a total loss value , and the specific calculation formula is: , where is the regularization strength coefficient; Backward propagation and parameter update, calculating the gradients of the parameters of each layer backward from the total loss value through the chain rule, and adjusting the parameters using the gradient descent algorithm; Iterative optimization, dividing the training set into multiple small batches, repeating the previous two steps batch by batch, traversing the entire training set multiple times until the model converges or reaches the preset number of iterations; Verification and testing, periodically evaluating the losses and metrics of the model through the validation set to prevent overfitting, and evaluating the performance of this prediction model through the test set after training is completed.

[0010] As a preferred aspect of the invention, the specific steps for obtaining the displacement loss term are as follows: According to the preprocessed displacement time series to obtain the actual displacement curve ; Calculating to obtain the displacement loss term , and the specific calculation formula is: , where represents the number of data in the preprocessed displacement time series, represents the displacement in the preprocessed displacement An actual displacement value, indicating the predicted displacement value corresponding to the th actual displacement value.

[0011] As a preferred aspect of the invention, the specific steps for obtaining the physical law loss term are as follows: According to the cohesion , the internal friction angle and the preprocessed normal stress time series, and obtaining the ultimate shear stress curve through calculation . The specific calculation formula is: ; Obtaining the physical law loss term based on the Mohr-Coulomb criterion through calculation. The specific calculation formula is: , where represents the number of data in the preprocessed normal stress time series, represents the th predicted shear stress corresponding to the normal stress data, represents the th ultimate shear stress corresponding to the normal stress data, represents taking the greater value of and 0.

[0012] As a preferred aspect of the invention, the specific steps for dividing the rock and soil mass into different risk levels according to the prediction result and activating the corresponding level of alarm are as follows: Obtaining the displacement velocity curve of the rock and soil mass according to the displacement curve of the rock and soil mass in the next 72 hours ; ; Obtaining the maximum displacement velocity of the rock and soil mass within the next 72 hours according to the displacement velocity curve . Based on the preset early warning level division rule and the maximum displacement velocity , dividing the risk level of the rock and soil mass, activating the corresponding level of alarm according to the divided risk level, and suggesting corresponding response measures to be taken.

[0013] The present invention has the following advantages: 1. The present invention predicts the displacement and shear stress of rock and soil masses through a displacement-hydrogeology-geology coupling prediction model, which can not only more comprehensively reflect the comprehensive influence of the mechanical state of rock and soil masses and environmental factors on geotechnical engineering disasters, but also comprehensively consider the coupling effect of various monitoring data, thereby effectively reducing the prediction errors and uncertainties caused by data loss or single models, significantly improving the accuracy of disaster prediction. By processing hydrological data through causal convolution in the temporal convolutional network, it can not only effectively capture the long-term dependence relationships between hydrological data, but also reflect the correlation and hysteresis between water levels, rainfall, and pore water pressure, thus significantly improving the accuracy of model prediction and the warning accuracy of this monitoring and warning system.

[0014] 2. By incorporating the Mohr-Coulomb criterion as a regularization term into the loss function of the model, the present invention can physically constrain the output of the displacement-hydrogeology-geology coupling prediction model. Thus, it can not only directly avoid the results that conflict with geotechnical mechanics principles in the model output, ensuring that the model output conforms to physical laws and can be explained by mechanical theories, enhancing the physical credibility and interpretability of the prediction results, but also suppress the sensitivity of the model to noise or outliers in the training data, and balance the advantages of data-driven and physical laws, effectively preventing overfitting of complex models and significantly improving the physical rationality and generalization ability of the model, improving the warning rationality of this monitoring and warning system.

[0015] 3. By classifying the risk levels of rock and soil masses according to the warning level classification rules and activating corresponding-level alarms based on the classified risk levels, the present invention can achieve hierarchical warning. Thus, it can not only remind relevant personnel to take corresponding preventive measures in advance and reasonably allocate limited resources, improving the efficiency of warning and response, avoiding waste of resources, and effectively avoiding or reducing disaster losses, but also provide clear decision-making basis for decision-makers, enabling them to reasonably arrange various tasks such as monitoring, warning, and emergency disposal according to the current risk situation, optimizing resource allocation, improving management efficiency, and ensuring the scientificity and effectiveness of disaster prevention and mitigation work, enhancing the practicality of this monitoring and warning system. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic structural diagram of the monitoring and warning system for geotechnical engineering adopted in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] In order to enable those skilled in the art of the present technology to better understand the technical solutions in the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention.

[0018] Embodiment 1. A monitoring and warning system for geotechnical engineering, as Figure 1As shown in the figure, it includes: A data acquisition module, which is used to obtain the historical displacement data and historical hydrological data of the geotechnical body to be predicted through sensors, and obtain the historical geological data of the geotechnical body through a geological exploration report, where the historical geological data includes static geological data and dynamic geological data; A data processing module, which is used to preprocess the obtained historical displacement data, historical hydrological data and dynamic geological data to obtain the preprocessed historical displacement data, historical hydrological data and dynamic geological data; A model construction module, which is used to construct a displacement-hydrology-geology coupling prediction model, and when training the model, incorporate the Mohr-Coulomb criterion as a regularization term into the loss function of the displacement-hydrology-geology joint prediction model to obtain a loss function with physical constraints, calculate the prediction error of the model through the loss function with physical constraints, and perform iterative optimization on the model through backpropagation and parameter update; A disaster warning module, which is used to input the static geological data, preprocessed historical displacement data, historical hydrological data and dynamic geological data of the geotechnical body to be predicted into the trained displacement-hydrology-geology joint prediction model to obtain a prediction result, where the prediction result includes the displacement curve of the geotechnical body in the next 72 hours and shear stress curve , divide the geotechnical body into different risk levels according to the prediction result of the displacement-hydrology-geology joint prediction model, and activate alarms at corresponding levels.

[0019] The historical displacement data is a displacement time series composed of the historical millimeter-level displacement change of the geotechnical body The historical hydrological data includes an instantaneous rainfall intensity time series, a groundwater level height time series and a pore water pressure time series. The static geological data includes the type, slope, vegetation coverage, fracture development index, cohesion and internal friction angle of the geotechnical body, while the dynamic geological data is the normal stress time series and shear stress time series of the geotechnical body.

[0020] The specific steps for preprocessing the obtained historical displacement data, historical hydrological data and dynamic geological data are as follows: Time alignment: Unify the starting reference time points of all time series, unify the time frequencies of all time series, and fill in the missing values of the time series with low time frequencies at the new time points through interpolation methods such as linear interpolation or polynomial interpolation to ensure that all time series data have values at the same time points; Missing value handling: Handle the null or missing values in the time series. Use interpolation methods such as linear interpolation or polynomial interpolation to fill in the missing values, or directly use the mean or median of the time series to fill in the missing values to ensure data integrity and model stability; Outlier handling: Identify outliers in the time series that do not conform to the expected pattern through the Z-Score method or the IQR method. Avoid the negative impact of these outliers on model training by deleting the outliers and replacing them with the mean or median of the time series or using interpolation methods to repair the outliers; Data standardization: Standardize data with different dimensions through min-max standardization. The calculation formula for min-max standardization is: , where represents the original data value, represents the minimum value in the time series, represents the maximum value in the time series, represents the standardized data value.

[0021] The displacement-hydrogeology-coupled prediction model includes three input layers, three feature extraction layers, one spatio-temporal fusion layer, and two joint prediction layers. Among them, the three input layers are respectively: the time series input layer, which is used to receive the displacement time series, the normal stress time series, and the shear stress time series; the hydrological sequence input layer, which is used to receive the instantaneous rainfall intensity time series, the groundwater level height time series, and the pore water pressure time series; the geological parameter input layer, which is used to receive static geological data. Among the three feature extraction layers are respectively: the LSTM encoder layer, whose structure is a bidirectional LSTM network integrated with a self-attention mechanism, which is used to extract features from the time series received by the time series input layer through a bidirectional LSTM model integrated with a self-attention mechanism and output a 128-dimensional feature vector; the temporal convolutional network layer, which is used to perform local feature extraction on the time series received by the hydrological sequence input layer through dilated convolution and causal convolution in the temporal convolutional network and output a 64-dimensional feature matrix; the fully connected embedding layer, whose structure is a three-layer fully connected network, which is used to map the high-dimensional geological information received by the geological parameter input layer into a low-dimensional feature vector through a three-layer fully connected network and output a 64-dimensional feature vector. The spatio-temporal fusion layer is used to splice the outputs of the three feature extraction layers into a feature vector and perform weighted fusion on the spliced feature vector through an attention mechanism to obtain the final fused feature vector. Among the two joint prediction layers are respectively: the displacement prediction layer, whose structure is a 3-layer TCN network + 1-layer linear regression layer, which is used to perform temporal feature extraction on the fused feature vector through the temporal convolutional network and output the displacement curve for the next 72 hours through the linear regression layer ; Shear stress prediction layer, with a structure of 3-layer TCN network + 1-layer linear regression layer, used to extract temporal features of the fused feature vector through a temporal convolutional network and output the shear stress curve for the next 72 hours through the linear regression layer .

[0022] It should be noted that the above steps predict the displacement and shear stress of the rock and soil mass through the displacement-hydrogeology-geology coupling prediction model, which can not only more comprehensively reflect the mechanical state of the rock and soil mass and the comprehensive influence of environmental factors on geotechnical engineering disasters, but also comprehensively consider the coupling effect of various monitoring data, so as to effectively reduce the prediction errors and uncertainties caused by data loss or single model, and significantly improve the accuracy of disaster prediction. By processing the hydrogeological data through causal convolution in the temporal convolutional network, not only can the long-term dependence relationship between hydrogeological data be effectively captured, but also the correlation and hysteresis between water level, rainfall and pore water pressure can be reflected, thus significantly improving the accuracy of model prediction and the warning accuracy of this monitoring and warning system.

[0023] The training steps of the displacement-hydrogeology-geology coupling prediction model are specifically as follows: Data preparation and preprocessing: Obtain the historical displacement data, historical hydrogeological data and historical geological data of the sample rock and soil mass, perform time alignment, missing value processing, outlier processing and data standardization on all the obtained data, and divide all the preprocessed data into training set, validation set and test set according to the ratio of 70%, 15% and 15% respectively; Forward propagation and loss calculation: Input the sample data in the training set into the displacement-hydrogeology-geology coupling prediction model, and finally obtain the predicted displacement curve through various linear transformations and non-linear activations and the predicted shear stress curve , according to the predicted displacement curve and the actual displacement curve to obtain the displacement loss term , according to the predicted shear stress curve and the ultimate shear stress curve to obtain the physical law loss term based on the Mohr-Coulomb criterion , and calculate the total loss value through the calculation formula of the total loss , and the calculation formula is specifically: , where is the regularization strength coefficient, used to control the weight of the physical law loss term , and the value of needs to be adjusted through tuning parameter strategies such as grid search optimization; Backward propagation and parameter update: Through the chain rule, starting from the total loss value Calculate the gradients of the parameters for each layer in reverse, and use gradient descent algorithms such as SGD and Adam to adjust the parameters to minimize the total loss value. ; Iterative optimization: Divide the training set into multiple small batches, repeat the previous two steps batch by batch, and traverse the entire training set multiple times until the model converges or reaches the preset number of iterations. Verification and testing: Periodically evaluate the various losses and metrics of the displacement-hydrogeology-geology coupling prediction model through the sample data in the validation set to prevent overfitting. After training, evaluate the final performance of this prediction model through the sample data in the test set.

[0024] The specific steps for obtaining the displacement loss term based on the difference between the predicted displacement curve and the actual displacement curve are as follows: Obtain the actual displacement curve from the preprocessed displacement time series in the historical displacement data. Calculate the displacement loss term through the calculation formula of the mean square error. The specific calculation formula is: , where represents the number of data in the preprocessed displacement time series, represents the th actual displacement value in the preprocessed displacement time series, represents the predicted displacement value corresponding to the th actual displacement value.

[0025] The steps for obtaining the physical law loss term based on the Mohr-Coulomb criterion based on the difference between the predicted shear stress curve and the ultimate shear stress curve are as follows: Obtain the ultimate shear stress curve by calculating based on the cohesion , internal friction angle in the static geological data and the preprocessed normal stress time series in the dynamic geological data. The specific calculation formula is: ; Calculate the physical law loss term based on the Mohr-Coulomb criterion through the calculation formula of the regularization loss term. The specific calculation formula is: , where represents the preprocessed normal stress The number of data in the time series, denotes the th normal stress The predicted shear stress corresponding to the data, denotes the th normal stress The ultimate shear stress corresponding to the data, denotes taking the greater value between and 0, that is, if

[0026] It should be noted that the above steps incorporate the Mohr-Coulomb criterion as a regularization term into the loss function of the model, which can physically constrain the output of the displacement-hydrogeology-coupled prediction model. Thus, not only can it directly avoid the results that conflict with the principles of geotechnical mechanics in the model output, ensuring that the model output conforms to physical laws and can be explained by mechanical theories, enhancing the physical credibility and interpretability of the prediction results, but also it can suppress the sensitivity of the model to noise or outliers in the training data and balance the advantages of data-driven and physical laws, effectively preventing the overfitting of complex models and significantly improving the physical rationality and generalization ability of the model, and improving the warning rationality of this monitoring and warning system.

[0027] The specific steps of dividing the rock and soil mass into different risk levels according to the prediction results of the displacement-hydrogeology-joint prediction model and initiating corresponding-level alarms are as follows: Obtain the displacement velocity curve of the rock and soil mass according to the displacement curve of the rock and soil mass in the next 72 hours ; Based on the displacement velocity curve obtain the maximum displacement velocity of the rock and soil mass within the next 72 hours , and divide the risk level of the rock and soil mass based on the preset warning level division rules and the maximum displacement velocity . Initiate corresponding-level alarms according to the divided risk levels and recommend corresponding response measures.

[0028] The specific content of dividing the risk level of the rock and soil mass based on the preset warning level division rules and the maximum displacement velocity and initiating corresponding-level alarms according to the divided risk levels and recommending corresponding response measures is as follows: When , the warning level is low risk, it is recommended to strengthen monitoring and closely pay attention to the change trend of the monitoring data; When , the warning level is medium risk, it is recommended to increase the monitoring frequency and prepare emergency measures; When When the warning level is high risk, it is recommended to take emergency measures immediately and evacuate the personnel.

[0029] It should be noted that the above steps classify the risk levels of rock and soil masses through the warning level classification rules and activate the corresponding level of alarms based on the classified risk levels, which can achieve hierarchical early warning. Thus, it can not only remind relevant personnel to take corresponding preventive measures in advance and reasonably allocate limited resources to improve the efficiency of early warning and response, avoid waste of resources, and effectively avoid or reduce disaster losses, but also provide clear decision-making basis for decision-makers, enabling them to reasonably arrange various tasks such as monitoring, early warning, and emergency disposal according to the current risk situation, optimize resource allocation, improve management efficiency, ensure the scientificity and effectiveness of disaster prevention and mitigation work, and enhance the practicability of this monitoring and early warning system.

[0030] It should be understood that those of ordinary skill in the art can make improvements or changes according to the above description, and all such improvements and changes should fall within the protection scope of the appended claims of the present invention. The parts not described in detail in this specification belong to the prior art well-known to those of ordinary skill in the art.

Claims

1. A monitoring and early warning system for geotechnical engineering, characterized in that: Included are: The data acquisition module is used to obtain the historical displacement data and historical hydrological data of the rock and soil body to be predicted through sensors, and obtain the historical geological data of the rock and soil body through geological exploration reports, wherein the historical geological data includes static geological data and dynamic geological data; A data processing module is used to pre-process the acquired historical displacement data, historical hydrological data and dynamic geological data to obtain the pre-processed historical displacement data, historical hydrological data and dynamic geological data; The model building module is used to build a displacement-hydrology-geology coupling prediction model, and incorporate the Mohr-Coulomb criterion as a regularization term into the model's loss function during model training. The prediction error of the model is calculated through the loss function, and the model is iteratively optimized through back propagation and parameter updating. The disaster warning module is used to input the static geological data of the rock and soil body to be predicted, the pre-processed historical displacement data, the historical hydrological data and the dynamic geological data into the trained displacement-hydrological-geological joint prediction model to obtain the prediction results, where the prediction results include the displacement curve of the rock and soil body in the next 72 hours and shear stress curve , based on the prediction results, the rock and soil mass is divided into different risk levels, and the corresponding level of alarm is activated.

2. The monitoring and early warning system for geotechnical engineering according to claim 1, characterized in that: The historical displacement data is the displacement of the rock and soil body The historical hydrological data include instantaneous rainfall intensity time series, groundwater level height time series and pore water pressure time series, and the static geological data include rock and soil type, slope, vegetation coverage, fracture development index, cohesion and internal friction angle The dynamic geological data is the normal stress of the rock and soil body Time series and shear stress Time series.

3. The monitoring and early warning system for geotechnical engineering according to claim 1, characterized in that: The displacement-hydrology-geology coupling prediction model includes three input layers, three feature extraction layers, a spatiotemporal fusion layer and two joint prediction layers, wherein the three input layers are: a time series input layer for receiving historical displacement data and dynamic geological data; a hydrological sequence input layer for receiving historical hydrological data; a geological parameter input layer for receiving static geological data; wherein the three feature extraction layers are: an LSTM encoder layer for extracting features from the time series received by the time series input layer; a temporal convolutional network layer for extracting local features from the time series received by the hydrological sequence input layer; a fully connected embedding layer for mapping the high-dimensional geological information received by the geological parameter input layer into a low-dimensional feature vector; wherein the spatiotemporal fusion layer is used to splice the outputs of the three feature extraction layers into a feature vector, and weightedly fuse the spliced ​​feature vectors through an attention mechanism to obtain a final fused feature vector; wherein the two joint prediction layers are: a displacement prediction layer for outputting a displacement curve for the next 72 hours ; Shear stress prediction layer, used to output shear stress curves for the next 72 hours .

4. The monitoring and early warning system for geotechnical engineering according to claim 3 is characterized in that: The training steps of the displacement-hydrology-geology coupling prediction model are specifically as follows: Data preparation and preprocessing: obtaining historical displacement data, historical hydrological data, and historical geological data of sample rock and soil bodies, preprocessing all acquired data, and dividing all preprocessed data into training sets, validation sets, and test sets in proportion; Forward propagation and loss calculation, input the training set into the displacement-hydrology-geology coupling prediction model, and obtain the predicted displacement curve and predicted shear stress curve , according to the predicted displacement curve The actual displacement curve The difference between the two gives the displacement loss term , according to the predicted shear stress curve The ultimate shear stress curve The difference between them gives the physical law loss term based on the Mohr-Coulomb criterion , the total loss value is obtained by calculating , the calculation formula is: ,in is the regularization strength coefficient; Back propagation and parameter update, from the total loss value through the chain rule Reversely calculate the gradient of each layer's parameters and adjust the parameters using the gradient descent algorithm; Iterative optimization: divide the training set into multiple small batches, repeat the previous two steps batch by batch, and traverse the entire training set multiple times until the model converges or reaches the preset number of iterations; Verification and testing: The model's losses and indicators are periodically evaluated through the validation set to prevent overfitting. After training, the performance of the prediction model is evaluated through the test set.

5. The monitoring and early warning system for geotechnical engineering according to claim 4, characterized in that: Obtain the displacement loss term The specific steps are: According to the displacement after preprocessing Time series to obtain the actual displacement curve ; The displacement loss term is obtained by calculation , the calculation formula is: ,in Represents the displacement after preprocessing The number of data in the time series, Represents the displacement after preprocessing The time series The actual displacement value, Indicates The predicted displacement value corresponding to the actual displacement value.

6. The monitoring and early warning system for geotechnical engineering according to claim 4, characterized in that: Obtain the physical law loss term The specific steps are: According to cohesion , internal friction angle and the normal stress after pretreatment Time series and ultimate shear stress curve obtained by calculation , the calculation formula is: ; The physical law loss term based on the Mohr-Coulomb criterion is obtained by calculation , the calculation formula is: ,in represents the normal stress after pretreatment The number of data in the time series, Indicates Normal stress The predicted shear stress corresponding to the data, Indicates Normal stress The data corresponds to the ultimate shear stress, Indicates taking and 0, whichever is greater.

7. The monitoring and early warning system for geotechnical engineering according to claim 1, characterized in that: The specific steps of classifying the rock and soil mass into different risk levels according to the prediction results and activating the corresponding level of alarm are: According to the displacement curve of the rock and soil in the next 72 hours Obtain the displacement velocity curve of the rock and soil mass ; According to the displacement velocity curve Obtain the maximum displacement velocity of the rock mass within the next 72 hours , based on the preset warning level classification rules and maximum displacement speed The rock and soil bodies are classified into risk levels, and the corresponding level of alarm is activated according to the classified risk level, and corresponding response measures are recommended.

Images