Slag disposal site slope stability monitoring system

Abstract

The invention discloses a waste slag site slope stability monitoring system, which comprises a data acquisition module which is responsible for acquiring deformation data of a slope in real time through various sensors deployed on the waste slag site slope and acquiring macroscopic information of the slope by using aerial images and digital elevation images; the data transmission module is responsible for transmitting the collected data to a data processing center in real time by using the Internet of Things technology; the data processing center comprises a data processing and analysis module, an early warning and decision support module and a user interaction module; and the data processing and analyzing module is responsible for processing and analyzing the transmitted data, building a slope stability model, evaluating the stability state of the slope, and generating some new and virtual unstable slope data through GAN to supplement a training set based on known unstable slope characteristics in the process of building the slope stability model.

Description

Technical Field

[0001] The invention relates to the technical field of safety monitoring, and in particular to a slope stability monitoring system for a waste dump. Background Art

[0002] A waste dump is a storage place for waste earth and stone generated by earth and stone excavation and tunnel development during construction. It is usually composed of loosely piled soil and stone mixture. This kind of slope has the characteristics of loose structure, low shear strength and easy deformation, so its stability is relatively poor. At the same time, since the formation and evolution process of the slope of the waste dump is complex and affected by many factors, such as rainfall, drainage facility maintenance, earthquake, human activities, etc., its stability monitoring is particularly important.

[0003] In the prior art, when processing and analyzing the monitored data, it may be encountered that the distribution of data of different categories is unbalanced. Under normal working conditions (i.e., special working conditions other than earthquakes, extreme disasters, wars, etc.), most slopes are stable, so the stable slope data collected in the data is often far more than the unstable slope data. Using these data for slope stability model prediction training will make the trained model pay more attention to the data of the dominant stable slopes, resulting in the model tending to judge the slopes as stable when predicting, and some slopes with poor stability may also be misjudged as stable. Such misjudgments will bring serious safety hazards to the lives and property of people downstream. To solve this problem, the present invention proposes a slope stability monitoring system for a waste dump. Summary of the invention

[0004] The present invention aims to solve the problem in the prior art that the prediction results of the slope stability prediction model are inaccurate due to the fact that the stable slope data is much more than the unstable slope data, and proposes a spoil dump slope stability monitoring system.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A spoil site slope stability monitoring system, comprising:

[0007] Data acquisition module: responsible for collecting slope deformation data in real time through various sensors (such as displacement sensors, stress sensors, inclination sensors, etc.) deployed on the slope of the waste dump, and obtaining macro information of the slope using aerial images and digital elevation images;

[0008] Data transmission module: responsible for using IoT technology to transmit the collected data to the data processing center in real time, including wired transmission and wireless transmission to ensure the real-time and reliability of the data;

[0009] The data processing center includes a data processing and analysis module, an early warning and decision support module, and a user interaction module;

[0010] Data processing and analysis module: responsible for processing and analyzing the transmitted data, evaluating the stability of the slope by building a slope stability model. In the process of building the slope stability model, based on the known unstable slope characteristics, some new and virtual unstable slope data are generated through GAN to supplement the training set;

[0011] Early warning and decision support module: responsible for setting the early warning threshold. The output result of the slope stability model is compared with the early warning threshold. Once the output result exceeds the threshold, the early warning mechanism will be automatically triggered, and the early warning information will be sent to relevant personnel through the user interaction module. At the same time, it provides decision support functions to provide a basis for slope reinforcement and management.

[0012] User interaction module: responsible for providing operation interface and visual display function. Users can view slope monitoring data, stability assessment results, early warning information and decision support in real time through the operation interface.

[0013] The above technical solution further includes:

[0014] Furthermore, in the data acquisition module, the horizontal and vertical displacements of the slope are monitored by displacement sensors to understand the overall deformation of the slope; the stress distribution inside the slope is monitored by stress sensors to help determine whether the slope is in a state of force balance. If the stress distribution is abnormal, it indicates that the slope is about to become unstable; the stability of the slope is comprehensively evaluated by combining the macro information and deformation data of the slope. For example, if the overall shape of the slope is steep, the slope is large, and the geological structure is complex, then its stability may be poor; conversely, if the overall shape of the slope is gentle, the slope is moderate, and the geological structure is simple, then its stability may be good.

[0015] Furthermore, the data processing and analysis module includes a data receiving unit, a data preprocessing unit, a data enhancement unit, a feature extraction unit, a model training and optimization unit, and the data receiving unit is responsible for receiving the slope deformation data and the macro information of the slope transmitted from the data transmission module, performing preprocessing operations such as cleaning, denoising, filtering, etc. on the received data to improve the accuracy and reliability of the data, and the processed data is passed to the subsequent feature extraction unit or data enhancement unit. The data enhancement unit generates new virtual data based on known unstable slope features using GAN, receives data from the data preprocessing unit, and passes the generated virtual data to the model training unit. The feature extraction unit extracts feature information using PCA, receives data from the data preprocessing unit, and passes the extracted features to the model training unit. The model training and optimization unit uses the received features and virtual data for training and optimization, and outputs the trained slope stability model. The stability assessment unit uses the trained slope stability model to perform stability assessment on the new slope deformation data.

[0016] Furthermore, the specific steps of the data enhancement unit generating new virtual data using GAN are as follows:

[0017] Data collection: Collect existing processed slope characteristic data, such as geological structure, topography, rainfall, soil moisture, etc.;

[0018] Feature extraction: Extract key features related to slope stability from the preprocessed data. These features may include slope gradient, height, soil type, groundwater level, etc.

[0019] GAN model training: Build a GAN model, including a generator and a discriminator. The task of the generator is to generate virtual data similar to real data, while the task of the discriminator is to distinguish between real data and virtual data. Through training, the generator can generate virtual data that is closer and closer to real data.

[0020] Virtual data generation: New virtual data are generated based on known unstable slope characteristics using the trained generator.

[0021] Furthermore, the key features related to slope stability were screened out through correlation analysis, and the Pearson correlation coefficient between each feature and slope stability was calculated. The Pearson correlation coefficient formula is:

[0022]

[0023] Among them, x i and i are the characteristic value and slope stability value of the i-th sample, respectively. and are the means of the characteristic values ​​and slope stability values, respectively, and n is the number of samples;

[0024] According to the value of Pearson's correlation coefficient, the correlation degree between each feature and slope stability is explained. The larger the absolute value of Pearson's correlation coefficient is, the higher the correlation degree between the feature and slope stability is. When |r|≈1, it means that there is a strong linear relationship between the feature and slope stability. When |r|≈0, it means that there is almost no linear relationship between the feature and slope stability.

[0025] Furthermore, the specific steps of the GAN model training are:

[0026] Construct the GAN model: Generator: Design a neural network structure whose input is a random noise vector and whose output is virtual data similar to the real slope feature data; Discriminator: Another neural network structure whose input is real data or virtual data generated by the generator, and whose output is the probability that the data is real data;

[0027] Define the loss function: Generator loss: measures the difference between the virtual data generated by the generator and the real data. The generator loss function is Where z represents a random noise vector, G(z) represents virtual data generated by the generator, and D(G(z)) represents the predicted probability of the discriminator for the virtual data generated by the generator; Discriminator loss: measures the ability of the discriminator to distinguish between real data and virtual data. The discriminator loss function is Among them, x represents the real data, and D(x) represents the predicted probability of the discriminator for the real data;

[0028] Training process: Initialize model parameters: assign random weights to the generator and discriminator; iterative training: train the discriminator: use real data and virtual data generated by the generator to train the discriminator so that it can accurately distinguish between the two; train the generator: use the feedback from the discriminator to train the generator so that the virtual data it generates is closer and closer to the real data; update weights: after each iteration, update the weights of the generator and discriminator according to the gradient of the loss function; through the iterative training process, continuously optimize the weights of the generator and discriminator so that the generator generates virtual data that is closer and closer to the real data.

[0029] Furthermore, the feature extraction unit extracts feature information using PCA and sets the processed data as M slope samples {X 1 ,X 2 ,...,X M ,},Each slope sample has N-dimensional slope features X i = Each slope feature X j Each has its own slope characteristic value;

[0030] First, all slope features are decentralized, that is, the mean is removed, and the average value of each slope feature is calculated. Then, for all slope samples, each slope feature is subtracted from its own mean, where the respective means are Find the covariance matrix after decentralization The diagonal lines are the variances of the slope features X1 and X2, and the off-diagonal lines are the covariances. The calculation formula for cov(X1,X1) is: Thus, the covariance matrix C of M slope samples under the N-dimensional slope characteristics is obtained;

[0031] After obtaining the covariance matrix, its eigenvalues ​​and their corresponding eigenvectors are calculated according to the characteristic equation Cμ=λμ, where λ is the eigenvalue and μ is its corresponding eigenvector. The largest first k eigenvalues ​​and corresponding eigenvectors are selected for projection. Projection is the process of dimensionality reduction, which reduces the original slope features from high dimensions to low dimensions. After dimensionality reduction, a large amount of redundant information is removed and at least 85% of the original information is retained. The processed slope features are then passed to the model training unit.

[0032] Furthermore, the model training and optimization unit uses the received features and virtual data to perform training and optimization, and outputs a trained slope stability model, including the following steps:

[0033] Model selection: According to the data characteristics and task requirements of the slope stability monitoring of the waste dump, the Transformer-based attention mechanism model is selected as the slope stability model;

[0034] Initialization parameters: Randomly initialize the model’s weights, biases and other parameters;

[0035] Data preparation: preprocessed features and virtual data as well as actual observations of slope stability are fed into the model;

[0036] Define the loss function: Select mean square error as the loss function. The mean square error formula is

[0037]

[0038] Among them, y i is the actual observed value, is the model prediction value, n is the number of samples;

[0039] Select optimization algorithm: Use Adam optimization algorithm for iterative training;

[0040] Iterative training: Calculate the loss through forward propagation, then update the model parameters through back propagation, and repeat this process until the loss function converges or reaches the preset number of iterations;

[0041] Evaluate the model: Use the validation set to evaluate the performance of the model, using accuracy as the evaluation metric;

[0042] Model tuning: Based on the evaluation results, adjust the model's hyperparameters or improve the model structure to improve the model's prediction accuracy. For example, we can increase the number of attention mechanism layers or adjust the learning rate to further optimize the model's performance.

[0043] Save model: save the trained slope stability model to the specified path or database, including the model's weight, bias and other parameters as well as the model's structural information;

[0044] Model deployment: Deploy the trained slope stability model to the stability assessment unit.

[0045] The present invention has the following beneficial effects:

[0046] In the present invention, based on the known unstable slope features, some new and virtual unstable slope data are generated by GAN to supplement the training set. The new and virtual unstable slope data can supplement the deficiency of real data, so that the model can be exposed to more slope features during the training process to ensure data balance. This helps the model learn a more comprehensive feature representation, thereby improving the accuracy of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a system architecture diagram of a waste dump slope stability monitoring system proposed in the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] like Figure 1 As shown, a spoil field slope stability monitoring system comprises:

[0050] Data acquisition module: responsible for collecting slope deformation data in real time through various sensors (such as displacement sensors, stress sensors, inclination sensors, etc.) deployed on the slope of the waste dump, and obtaining macro information of the slope using aerial images and digital elevation images;

[0051] Data transmission module: responsible for using IoT technology to transmit the collected data to the data processing center in real time, including wired transmission and wireless transmission to ensure the real-time and reliability of the data;

[0052] The data processing center includes data processing and analysis modules, early warning and decision support modules, and user interaction modules;

[0053] Data processing and analysis module: responsible for processing and analyzing the transmitted data, evaluating the stability of the slope by building a slope stability model. In the process of building the slope stability model, based on the known unstable slope characteristics, some new and virtual unstable slope data are generated through GAN to supplement the training set;

[0054] Early warning and decision support module: responsible for setting the early warning threshold. The output result of the slope stability model is compared with the early warning threshold. Once the output result exceeds the threshold, the early warning mechanism will be automatically triggered, and the early warning information will be sent to relevant personnel through the user interaction module. At the same time, it provides decision support functions to provide a basis for slope reinforcement and management.

[0055] User interaction module: responsible for providing operation interface and visual display function. Users can view slope monitoring data, stability assessment results, early warning information and decision support in real time through the operation interface.

[0056] In one embodiment, in the data acquisition module, the horizontal and vertical displacements of the slope are monitored by displacement sensors to understand the overall deformation of the slope; the stress distribution inside the slope is monitored by stress sensors to help determine whether the slope is in a state of force balance. If the stress distribution is abnormal, it indicates that the slope is about to become unstable; the stability of the slope is comprehensively evaluated by combining the macro information and deformation data of the slope. For example, if the overall shape of the slope is steep, the slope is large, and the geological structure is complex, then its stability may be poor; conversely, if the overall shape of the slope is gentle, the slope is moderate, and the geological structure is simple, then its stability may be good.

[0057] In one embodiment, the data processing and analysis module includes a data receiving unit, a data preprocessing unit, a data enhancement unit, a feature extraction unit, a model training and optimization unit, and the data receiving unit is responsible for receiving the slope deformation data and the macro information of the slope transmitted from the data transmission module, performing preprocessing operations such as cleaning, denoising, and filtering on the received data to improve the accuracy and reliability of the data, and the processed data is passed to the subsequent feature extraction unit or data enhancement unit. The data enhancement unit generates new virtual data using GAN based on known unstable slope features, receives data from the data preprocessing unit, and passes the generated virtual data to the model training unit. The feature extraction unit extracts feature information using PCA, receives data from the data preprocessing unit, and passes the extracted features to the model training unit. The model training and optimization unit uses the received features and virtual data for training and optimization, and outputs the trained slope stability model. The stability assessment unit uses the trained slope stability model to perform stability assessment on the new slope deformation data.

[0058] In one embodiment, the specific steps of the data enhancement unit generating new virtual data using GAN are as follows:

[0059] Data collection: Collect existing processed slope characteristic data, such as geological structure, topography, rainfall, soil moisture, etc.;

[0060] Feature extraction: Extract key features related to slope stability from the preprocessed data. These features may include slope gradient, height, soil type, groundwater level, etc.

[0061] GAN model training: Build a GAN model, including a generator and a discriminator. The task of the generator is to generate virtual data similar to real data, while the task of the discriminator is to distinguish between real data and virtual data. Through training, the generator can generate virtual data that is closer and closer to real data.

[0062] Virtual data generation: New virtual data are generated based on known unstable slope characteristics using the trained generator.

[0063] In one embodiment, the key features related to slope stability are screened out through correlation analysis, and the Pearson correlation coefficient between each feature and slope stability is calculated. The Pearson correlation coefficient formula is:

[0064]

[0065] Among them, x i and i are the characteristic value and slope stability value of the i-th sample, respectively. and are the means of the characteristic values ​​and slope stability values, respectively, and n is the number of samples;

[0066] According to the value of Pearson's correlation coefficient, the correlation degree between each feature and slope stability is explained. The larger the absolute value of Pearson's correlation coefficient is, the higher the correlation degree between the feature and slope stability is. When |r|≈1, it means that there is a strong linear relationship between the feature and slope stability. When |r|≈0, it means that there is almost no linear relationship between the feature and slope stability.

[0067] In one embodiment, the specific steps of GAN model training are:

[0068] Construct the GAN model: Generator: Design a neural network structure whose input is a random noise vector and whose output is virtual data similar to the real slope feature data; Discriminator: Another neural network structure whose input is real data or virtual data generated by the generator, and whose output is the probability that the data is real data;

[0069] Define the loss function: Generator loss: measures the difference between the virtual data generated by the generator and the real data. The generator loss function is Where z represents a random noise vector, G(z) represents virtual data generated by the generator, and D(G(z)) represents the predicted probability of the discriminator for the virtual data generated by the generator; Discriminator loss: measures the ability of the discriminator to distinguish between real data and virtual data. The discriminator loss function is Among them, x represents the real data, and D(x) represents the predicted probability of the discriminator for the real data;

[0070] Training process: Initialize model parameters: assign random weights to the generator and discriminator; iterative training: train the discriminator: use real data and virtual data generated by the generator to train the discriminator so that it can accurately distinguish between the two; train the generator: use the feedback from the discriminator to train the generator so that the virtual data it generates is closer and closer to the real data; update weights: after each iteration, update the weights of the generator and discriminator according to the gradient of the loss function; through the iterative training process, continuously optimize the weights of the generator and discriminator so that the generator generates virtual data that is closer and closer to the real data.

[0071] In one embodiment, the feature extraction unit uses PCA to extract feature information and sets the processed data as M slope samples {X 1 ,X 2 ,...,X M ,},Each slope sample has N-dimensional slope features X i = Each slope feature X j Each has its own slope characteristic value;

[0072] First, all slope features are decentralized, that is, the mean is removed, and the average value of each slope feature is calculated. Then, for all slope samples, each slope feature is subtracted from its own mean, where the respective means are Find the covariance matrix after decentralization The diagonal lines are the variances of the slope features X1 and X2, and the off-diagonal lines are the covariances. The calculation formula for cov(X1,X1) is: Thus, the covariance matrix C of M slope samples under the N-dimensional slope characteristics is obtained;

[0073] After obtaining the covariance matrix, its eigenvalues ​​and their corresponding eigenvectors are calculated according to the characteristic equation Cμ=λμ, where λ is the eigenvalue and μ is its corresponding eigenvector. The largest first k eigenvalues ​​and corresponding eigenvectors are selected for projection. Projection is the process of dimensionality reduction, which reduces the original slope features from high dimensions to low dimensions. After dimensionality reduction, a large amount of redundant information is removed and at least 85% of the original information is retained. The processed slope features are then passed to the model training unit.

[0074] In one embodiment, the model training and optimization unit uses the received features and virtual data to perform training and optimization, and outputs a trained slope stability model, including the following steps:

[0075] Model selection: According to the data characteristics and task requirements of the slope stability monitoring of the waste dump, the Transformer-based attention mechanism model is selected as the slope stability model;

[0076] Initialization parameters: Randomly initialize the model’s weights, biases and other parameters;

[0077] Data preparation: preprocessed features and virtual data as well as actual observations of slope stability are fed into the model;

[0078] Define the loss function: Select mean square error as the loss function. The mean square error formula is

[0079]

[0080] Among them, y is the actual observed value, is the model prediction value, n is the number of samples;

[0081] Select optimization algorithm: Use Adam optimization algorithm for iterative training;

[0082] Iterative training: Calculate the loss through forward propagation, then update the model parameters through back propagation, and repeat this process until the loss function converges or reaches the preset number of iterations;

[0083] Evaluate the model: Use the validation set to evaluate the performance of the model, using accuracy as the evaluation metric;

[0084] Model tuning: Based on the evaluation results, adjust the model's hyperparameters or improve the model structure to improve the model's prediction accuracy. For example, we can increase the number of attention mechanism layers or adjust the learning rate to further optimize the model's performance.

[0085] Save model: save the trained slope stability model to the specified path or database, including the model's weight, bias and other parameters as well as the model's structural information;

[0086] Model deployment: Deploy the trained slope stability model to the stability assessment unit.

[0087] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A spoil site slope stability monitoring system, characterized in that: include: Data acquisition module: responsible for collecting slope deformation data in real time through various sensors deployed on the slope of the waste dump, and obtaining macro information of the slope using aerial images and digital elevation images; Data transmission module: responsible for using IoT technology to transmit the collected data to the data processing center in real time, including wired transmission and wireless transmission; The data processing center includes a data processing and analysis module, an early warning and decision support module, and a user interaction module; Data processing and analysis module: responsible for processing and analyzing the transmitted data, evaluating the stability of the slope by building a slope stability model. In the process of building the slope stability model, based on the known unstable slope characteristics, some new and virtual unstable slope data are generated through GAN to supplement the training set; Early warning and decision support module: responsible for setting the early warning threshold. The output result of the slope stability model is compared with the early warning threshold. Once the output result exceeds the threshold, the early warning mechanism will be automatically triggered, and the early warning information will be sent to relevant personnel through the user interaction module. At the same time, it provides decision support functions to provide a basis for slope reinforcement and management. User interaction module: responsible for providing operation interface and visual display function. Users can view slope monitoring data, stability assessment results, early warning information and decision support in real time through the operation interface.

2. A spoil dump slope stability monitoring system according to claim 1, characterized in that: In the data acquisition module, the horizontal and vertical displacements of the slope are monitored by displacement sensors to understand the overall deformation of the slope; the stress distribution inside the slope is monitored by stress sensors to help determine whether the slope is in a state of force balance. If the stress distribution is abnormal, it indicates that the slope is about to become unstable; the stability of the slope is comprehensively evaluated by combining the macroscopic information and deformation data of the slope.

3. A spoil dump slope stability monitoring system according to claim 1, characterized in that: The data processing and analysis module includes a data receiving unit, a data preprocessing unit, a data enhancement unit, a feature extraction unit, a model training and optimization unit, and the data receiving unit is responsible for receiving the slope deformation data and the macro information of the slope transmitted from the data transmission module, performing preprocessing operations on the received data, and passing the processed data to the subsequent feature extraction unit or data enhancement unit. The data enhancement unit generates new virtual data based on known unstable slope features using GAN, receives data from the data preprocessing unit, and passes the generated virtual data to the model training unit. The feature extraction unit extracts feature information using PCA, receives data from the data preprocessing unit, and passes the extracted features to the model training unit. The model training and optimization unit performs training and optimization using the received features and virtual data, and outputs a trained slope stability model. The stability assessment unit performs stability assessment on new slope deformation data using the trained slope stability model.

4. A spoil dump slope stability monitoring system according to claim 3, characterized in that: The specific steps of the data enhancement unit generating new virtual data using GAN are as follows: Data collection: Collect the existing processed slope characteristic data; Feature extraction: extract key features related to slope stability from preprocessed data; GAN model training: Build a GAN model, including a generator and a discriminator. The task of the generator is to generate virtual data similar to real data, while the task of the discriminator is to distinguish between real data and virtual data. Through training, the generator can generate virtual data that is closer and closer to real data. Virtual data generation: New virtual data are generated based on known unstable slope characteristics using the trained generator.

5. A spoil dump slope stability monitoring system according to claim 4, characterized in that: The key features related to slope stability are screened out through correlation analysis, and the Pearson correlation coefficient between each feature and slope stability is calculated. The Pearson correlation coefficient formula is: Among them, x i and i are the characteristic value and slope stability value of the i-th sample, respectively. and are the means of the characteristic value and slope stability value, respectively, and n is the number of samples; According to the value of Pearson's correlation coefficient, the correlation degree between each feature and slope stability is explained. The larger the absolute value of Pearson's correlation coefficient is, the higher the correlation degree between the feature and slope stability is. When |r|≈1, it means that there is a strong linear relationship between the feature and slope stability. When |r|≈0, it means that there is almost no linear relationship between the feature and slope stability.

6. A spoil dump slope stability monitoring system according to claim 4, characterized in that: The specific steps of the GAN model training are: Construct the GAN model: Generator: Design a neural network structure whose input is a random noise vector and whose output is virtual data similar to the real slope feature data; Discriminator: Another neural network structure whose input is real data or virtual data generated by the generator, and whose output is the probability that the data is real data; Define the loss function: Generator loss: measures the difference between the virtual data generated by the generator and the real data. The generator loss function is L G =-E z～pz(z) [logD(G(z))], where z represents a random noise vector, G(z) represents virtual data generated by the generator, and D(G(z)) represents the predicted probability of the discriminator for the virtual data generated by the generator; Discriminator loss: measures the ability of the discriminator to distinguish between real data and virtual data. The discriminator loss function is Among them, x represents the real data, and D(x) represents the predicted probability of the discriminator for the real data; Training process: Initialize model parameters: assign random weights to the generator and discriminator; iterative training: train the discriminator: use real data and virtual data generated by the generator to train the discriminator so that it can accurately distinguish between the two; train the generator: use the feedback from the discriminator to train the generator so that the virtual data it generates is closer and closer to the real data; update weights: after each iteration, update the weights of the generator and discriminator according to the gradient of the loss function; through the iterative training process, continuously optimize the weights of the generator and discriminator so that the generator generates virtual data that is closer and closer to the real data.

7. A spoil dump slope stability monitoring system according to claim 3, characterized in that: The feature extraction unit uses PCA to extract feature information and sets the processed data as M slope samples {X 1 ,X 2 ,...,X M ,},Each slope sample has N-dimensional slope features Each slope feature X j Each has its own slope characteristic value; First, all slope features are decentralized, that is, the mean is removed, and the average value of each slope feature is calculated. Then, for all slope samples, each slope feature is subtracted from its own mean, where the respective means are Find the covariance matrix after decentralization The diagonal lines are the variances of the slope features X1 and X2, and the off-diagonal lines are the covariances. The calculation formula for cov(X1,X1) is: Thus, the covariance matrix C of M slope samples under the N-dimensional slope characteristics is obtained; After obtaining the covariance matrix, its eigenvalues ​​and their corresponding eigenvectors are calculated according to the characteristic equation Cμ=λμ, where λ is the eigenvalue and μ is its corresponding eigenvector. The largest first k eigenvalues ​​and corresponding eigenvectors are selected for projection. Projection is the process of dimensionality reduction, which reduces the original slope features from high dimensions to low dimensions. After dimensionality reduction, a large amount of redundant information is removed and at least 85% of the original information is retained. The processed slope features are then passed to the model training unit.

8. The system for monitoring the slope stability of a waste dump according to claim 3, characterized in that: The model training and optimization unit uses the received features and virtual data to perform training and optimization, and outputs a trained slope stability model, including the following steps: Model selection: According to the data characteristics and task requirements of the slope stability monitoring of the waste dump, the Transformer-based attention mechanism model is selected as the slope stability model; Initialization parameters: Randomly initialize the model’s weights, biases and other parameters; Data preparation: preprocessed features and virtual data as well as actual observations of slope stability are fed into the model; Define the loss function: Select mean square error as the loss function. The mean square error formula is Among them, y i is the actual observed value, is the model prediction value, n is the number of samples; Select optimization algorithm: Use Adam optimization algorithm for iterative training; Iterative training: Calculate the loss through forward propagation, then update the model parameters through back propagation, and repeat this process until the loss function converges or reaches the preset number of iterations; Evaluate the model: Use the validation set to evaluate the performance of the model, using accuracy as the evaluation metric; Model tuning: Adjust the model’s hyperparameters or improve the model structure based on the evaluation results; Save model: save the trained slope stability model to the specified path or database, including the model's weight, bias and other parameters as well as the model's structural information; Model deployment: Deploy the trained slope stability model to the stability assessment unit.

Images