Undermine roadway rock burst prediction and three-dimensional model visualization system based on Internet of Things

Abstract

The invention belongs to the technical field of computer internet of things, and particularly relates to an under-mine roadway rock burst prediction and three-dimensional model visualization system based on the internet of things. Comprising a data processing module, an analysis module, a machine learning training module and a model verification prediction module. The data processing module is used for preprocessing and normalizing the sensor data; the data analysis module fully extracts features; the machine learning training module is used for training prediction and optimization by using an algorithm; and the model verification prediction module evaluates the model and outputs a result. The system collects data by using the Internet of Things, predicts rock burst by combining with a machine learning algorithm, and displays a roadway three-dimensional model and changes thereof on a computer in real time. The problem that a traditional monitoring method cannot comprehensively monitor and predict in real time is solved, rock burst can be prevented in advance, and the safety of undermine roadways is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer Internet of Things, and in particular relates to an Internet of Things-based underground mine tunnel rock burst prediction and three-dimensional model visualization system. Background Art

[0002] With the deepening of mining, the stability of underground tunnels is increasingly affected by stress changes and ground pressure, especially in stress concentration areas, where rock burst poses a major threat to the safety of personnel and equipment. Traditional underground monitoring methods rely solely on manual detection or single-point sensor data, which cannot monitor the state of tunnels in real time and comprehensively, let alone predict possible risks in the future. The existing technology lacks an effective response mechanism to prevent and deal with sudden rock bursts. In order to solve this problem, the present invention proposes a system based on the Internet of Things, which collects underground sensor data, combines machine learning algorithms to predict rock bursts, and displays a three-dimensional tunnel model and its future changes on a computer in real time, helping managers to detect problems in advance and take preventive measures. Summary of the invention

[0003] In view of the above-mentioned technical problems, the present invention proposes an underground mine tunnel rock burst prediction and three-dimensional model visualization system based on the Internet of Things.

[0004] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is, including a data processing module, a data analysis module, a machine learning training module and a model verification and prediction module; the data processing module: receives real-time data from the Internet of Things sensor and pre-processes and normalizes it to provide standardized input for the prediction model; the data analysis module: performs full feature extraction through hysteresis feature extraction, statistical features and differential features; the machine learning training module: based on the training data, uses a machine learning algorithm to train and predict, outputs stress change trends and probability of rock burst, and continuously optimizes accuracy; the model verification and prediction module: after the training is completed, uses an independent test set to evaluate the model, and further optimizes the classification boundary through a support vector machine to output the final prediction result.

[0005] Preferably, the data processing module implements data processing as follows:

[0006] Step 1: Time series data, where the data collected by stress, vibration, and temperature sensors are all used as time series input; the label data is the historical rock burst events;

[0007] Step 2: According to the time series characteristics of the data, the data is divided into different time windows. In each time window, the local mean and local standard deviation of the data in the window are calculated. Based on the dynamically determined local mean μ local and the local standard deviation σ localThe normalized formula is Where X is the original input feature and X′ is the normalized feature;

[0008] Step 2: Process the time series data through a sliding window mechanism to ensure that the model can fully capture the time dependency. The selection of the window length T is optimized through experiments, and the step length δt is also tuned. The formula is: Windows(t) = {X t-T+1 ,X t-T+2 ,...,X t}, where X t is the input data at time t;

[0009] Step 3: For unbalanced data, we randomly select some negative samples to make the number of positive and negative samples close to equal, and use SMOTE to generate synthetic positive samples. In addition, we assign higher weights to positive samples to improve the model's recognition ability on positive samples. The weight ω is calculated by the formula: where n total is the total number of samples, n positine is the number of positive samples.

[0010] Preferably, the specific implementation of feature extraction in the data analysis module is as follows:

[0011] Step 1: Extract hysteresis features. Extract hysteresis features from time series data to capture the impact of historical data on the current moment. For each sensor X, the time lag feature is X(t-τ)τ=1,2,3...n, where τ represents the lag time step and n is the number of lag steps.

[0012] Step 2: Statistical features: extract statistical features of the data in the window, including mean, standard deviation, maximum and minimum values. For window W t The statistical characteristic S is calculated as follows: S = {mean(W t ),std(W t ),max(W t ),min(W t )}, where mean(W t ) is the mean of the data in the window, std(W t ) is the standard deviation of the data in the window, max(W t ) is the maximum value of the data in the window, min(W t ) is the minimum value of the data in the window;

[0013] Step 3: Differential features. In order to capture the changing trend of data, differential features are introduced. The differential calculation formula is: ΔX t =X t -X t-1, where X t is the input data at time t, X t-1 is the input data at time t-1, ΔX t It is the difference between the current moment and the previous moment.

[0014] Preferably, the implementation steps of the machine learning training module are:

[0015] Step 1: First, use logistic regression to model the probability of rock burst events. The model input is the preprocessed time series features, and the output is the probability of rock burst events. The decision function of the logistic regression model is: Where P(y t -1X t ) is the given input X t When event y t The probability of occurrence, y t is the target value at the tth moment, 1 indicates that the event occurs, 0 indicates that it does not occur, X t is the input feature vector at the tth moment, β0 is the intercept term, and X1β1 is the weight of the model;

[0016] Step 2: To prevent the model from overfitting, L2 regularization is added, and the loss function is: Where N is the number of samples, y i is the predicted probability of the ith sample, λ is the regularization strength, is the jth coefficient of logistic regression;

[0017] Step 3: Based on logistic regression, support vector machine is further used to perform secondary classification on the prediction results of the model. The decision function of support vector machine is: Where f(X) is the output of the decision function of the support vector machine, α i represents the weight of the support vector, y i is the class label of the support vector, X i is the current vector, X is the current input feature, K(X i ,X) is used to measure X i The similarity with X, b is the bias term;

[0018] Step 4: Next, perform model training and use batch gradient descent to optimize the model. Use small batches of data to update the parameters each time. For the logistic regression model, the gradient update formula is: where β j is the jth coefficient of logistic regression, η is the learning rate, L is the loss function, is the loss function with respect to β j The partial derivative of

[0019] Step 5: Then perform dynamic learning adjustment and introduce a dynamic learning rate adjustment strategy. Initially, use a larger learning rate to accelerate training, and then gradually reduce the learning rate. The formula is: where η t is the learning rate at the tth iteration, η0 is the initial learning rate, α is the decay rate of the learning rate, and t is the number of iterations;

[0020] Step 6. Finally, perform early stopping monitoring on the training set. When the verification loss no longer decreases, terminate the training early to avoid overfitting of the model on the training set.

[0021] Preferably, the specific operations of the model verification and prediction module are:

[0022] Step 1: After training, use an independent test set to evaluate the model and give high weights ω to false negatives FN , giving lower weights ω to false positives FP , the weights of true positive TP and true negative TN are set to 1, then the weighted accuracy formula is: Among them, TP is the true positive example, which indicates the number of samples correctly predicted as positive by the model; TN is the true negative example, which indicates the number of samples correctly predicted as negative by the model; FP is the false positive example, which indicates the number of samples incorrectly predicted as positive by the model; FN is the false negative example, which indicates the number of samples incorrectly predicted as negative by the model;

[0023] Step 2: For rock burst events, the probability output by the logistic regression model is used as the risk level of the rock burst event, and a certain threshold value P>0.5 is set to determine whether a rock burst event will occur;

[0024] Step 3: For the output of the prediction model, the classification boundary is further optimized through the support vector machine to output the final prediction result. The final model can dynamically predict the risk of rock burst based on the real-time collected IoT data and display the changes of the underground 3D model on the computer.

[0025] Compared with the prior art, the advantages and positive effects of the present invention are that, in data processing, a dynamic window is used to calculate the local mean and standard deviation for normalization, which can adapt to data changes and process unbalanced data, thereby improving data quality. In terms of feature extraction, hysteresis, statistics, and differential features are combined to reflect data characteristics from different angles, thereby improving prediction accuracy. During model training, regularization and secondary classification are performed after logistic regression modeling, and a variety of training methods are used to avoid overfitting, thereby improving generalization capabilities. In terms of evaluation indicators, the weights of different error types are taken into account to evaluate the model more accurately, which meets actual needs and improves its effectiveness in preventing rock burst. DETAILED DESCRIPTION

[0026] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0027] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0028] Embodiment, the purpose of the present invention is to provide a system for predicting rock burst and three-dimensional model visualization of underground mine tunnels based on Internet of Things technology. The system collects data such as ground pressure, vibration, stress, displacement, etc. in the tunnel in real time through the Internet of Things, uses machine learning algorithms to predict the probability of future rock burst, and displays the current state and future change trend of the three-dimensional model of the tunnel in real time on a computer to achieve early prevention of rock burst. The system includes a data processing module, a data analysis module, a machine learning training module, and a model verification and prediction module.

[0029] In the data processing module, the data collected by stress, vibration, and temperature sensors are input as time series; the label data is the historical rock burst events; according to the time series characteristics of the data, the data is divided into different time windows, and in each time window, the local mean and local standard deviation of the data in the window are calculated. Based on the dynamically determined local mean μ local and the local standard deviation σ local The normalized formula is Where X is the original input feature, and X′ is the normalized feature. The time series data is processed through a sliding window mechanism to ensure that the model can fully capture the time dependency. The choice of window length T is optimized through experiments, and the step length δt is also tuned. The formula is: Windows(t) = {X t-T+1 ,X t-T+2 ,...,X t}, where X t is the input data at time t; for unbalanced data, we randomly select some negative samples to make the number of positive and negative samples close to equal, and use SMOTE to generate synthetic positive samples. In addition, we assign higher weights to positive samples to improve the recognition ability of the model on positive samples. The weight ω is calculated by the formula: where n total is the total number of samples, n positine is the number of positive samples.

[0030] In the data analysis module, it is considered that feature selection will have a great impact on the prediction accuracy of the model. In the present invention, the system adopts a combination of lag features, statistical features and differential features. The lag features capture historical influences and allow the model to consider time causality; statistical features reflect the overall situation and help the model understand the characteristics of the data; differential features show changing trends and key factors are highlighted. The combination of the three comprehensively improves the prediction accuracy of the model. Lag feature extraction, extracting lag features from time series data to capture the impact of historical data on the current moment. For each sensor X, the time lag feature is X(t-τ)τ=1,2,3...n, where τ represents the lag time step and n is the number of lag steps; statistical features, extracting statistical features of the data in the window, including mean, standard deviation, maximum and minimum values. For window W t The statistical characteristic S is calculated as follows: S = {mean(W t ),std(W t ),max(W t ),min(W t )}, where mean(W t ) is the mean of the data in the window, std(W t ) is the standard deviation of the data in the window, max(W t ) is the maximum value of the data in the window, min(W t ) is the minimum value of the data in the window; differential features, in order to capture the changing trend of data, differential features are introduced, and the differential calculation formula is: ΔX t =X t -X t-1 , where X t is the input data at time t, X t-1 is the input data at time t-1, ΔX t It is the difference between the current moment and the previous moment.

[0031] In order to model the probability of rock burst events, the system analyzes and models the time series features, and reasonably links the input features with the probability of rock burst events. Its decision function is clear and intuitive, and has high computational efficiency, providing a reliable basis for subsequent predictions. Logistic regression is used to model the probability of rock burst events. The model input is the preprocessed time series features, and the output is the probability of rock burst events. The decision function of the logistic regression model is: Where P(y t -1X t ) is the given input X t When event y t The probability of occurrence, y t is the target value at the tth moment, 1 indicates that the event occurs, 0 indicates that it does not occur, X tis the input feature vector at the tth moment, β0 is the intercept term, and X1β1 is the weight of the model; secondly, to prevent the model from overfitting, L2 regularization is added, and the loss function is: Where N is the number of samples, y i is the predicted probability of the ith sample, λ is the regularization strength, is the jth coefficient of logistic regression; based on logistic regression, support vector machine is further used to perform secondary classification on the prediction results of the model. The decision function of support vector machine is: Where f(X) is the output of the decision function of the support vector machine, α i represents the weight of the support vector, y i is the class label of the support vector, X i is the current vector, X is the current input feature, K(X i ,X) is used to measure X i and X, b is the bias term; next, the model is trained and the model is optimized using batch gradient descent. Each time, small batches of data are used to update the parameters. For the logistic regression model, the gradient update formula is: where β j is the jth coefficient of logistic regression, η is the learning rate, L is the loss function, is the loss function with respect to β j The partial derivative of ; Then dynamic learning adjustment is performed, and a dynamic learning rate adjustment strategy is introduced. A larger learning rate is used to accelerate training at the beginning, and then the learning rate is gradually reduced. The formula is: where η t is the learning rate at the tth iteration, η0 is the initial learning rate, α is the decay rate of the learning rate, and t is the number of iterations. Finally, early stopping monitoring is performed on the training set. When the verification loss no longer decreases, the training is terminated in advance to avoid overfitting of the model on the training set.

[0032] Finally, in order to ensure the accuracy of the model's prediction results, the system uses an independent test set to perform an evaluation using improved evaluation criteria. Common indicators treat all prediction results equally, while the present invention assigns high weights to false negative examples and lower weights to false positive examples, taking into account the differences in the impact of different error types on practical applications. In the prediction of rock burst in underground mine tunnels, the consequences of false negative errors are serious. This weighted approach highlights their impact and can more accurately evaluate model performance. The updated indicators are more in line with actual application needs, allowing R&D personnel to pay more attention to key error types, which helps to improve the effectiveness of the model in preventing rock burst. The specific operation is to evaluate the model using an independent test set after training, assigning a high weight ω to false negative examples. FN , giving lower weights ω to false positives FP , the weights of true positive TP and true negative TN are set to 1, then the weighted accuracy formula is: Among them, TP is a true positive, which indicates the number of samples correctly predicted as positive by the model; TN is a true negative, which indicates the number of samples correctly predicted as negative by the model; FP is a false positive, which indicates the number of samples incorrectly predicted as positive by the model; and FN is a false negative, which indicates the number of samples incorrectly predicted as negative by the model; for rock burst events, the probability output by the logistic regression model is used as the risk level of rock burst events, and a certain threshold value P>0.5 is set to determine whether a rock burst event will occur; for the output of the prediction model, the support vector machine is used to further optimize the classification boundary and output the final prediction result. The final model can dynamically predict the rock burst risk based on the real-time collected IoT data, and display the changes of the 3D model of the mine on the computer. The application of this system can effectively improve the accuracy and timeliness of rock burst prediction in underground mine tunnels, and provide important support for the safety of mining production.

[0033] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. The underground mine tunnel rock burst prediction and three-dimensional model visualization system based on the Internet of Things is characterized by: It includes a data processing module, a data analysis module, a machine learning training module, and a model verification and prediction module; the data processing module receives real-time data from IoT sensors and performs preprocessing and normalization to provide standardized input for the prediction model; the data analysis module performs full feature extraction through hysteresis feature extraction, statistical features, and differential features; The machine learning training module: based on the training data, uses the machine learning algorithm to train the prediction, outputs the stress change trend and the probability of rock burst, and continuously optimizes the accuracy; the model verification and prediction module: after the training is completed, uses an independent test set to evaluate the model, and further optimizes the classification boundary through the support vector machine to output the final prediction result.

2. The system for predicting rock burst and three-dimensional model visualization of underground mine tunnels based on the Internet of Things according to claim 1 is characterized in that: The implementation of data processing in the data processing module is as follows: Step 1: Time series data, where the data collected by stress, vibration, and temperature sensors are all used as time series input; the label data is the historical rock burst events; Step 2: According to the time series characteristics of the data, the data is divided into different time windows. In each time window, the local mean and local standard deviation of the data in the window are calculated. Based on the dynamically determined local mean μ local and the local standard deviation σ local The normalized formula is Where X is the original input feature and X′ is the normalized feature; Step 2: Process the time series data through a sliding window mechanism to ensure that the model can fully capture the time dependency. The selection of the window length T is optimized through experiments, and the step length δt is also tuned. The formula is: Windows(t) = {X t-T+1 ,X t-T+2 ,...,X t }, where X t is the input data at time t; Step 3: For unbalanced data, we randomly select some negative samples to make the number of positive and negative samples close to equal, and use SMOTE to generate synthetic positive samples. In addition, we assign higher weights to positive samples to improve the model's recognition ability on positive samples. The weight ω is calculated by the formula: Where n total is the total number of samples, n positine is the number of positive samples.

3. The system for predicting rock burst and three-dimensional model visualization of underground mine tunnels based on the Internet of Things according to claim 1 is characterized in that: The specific implementation of feature extraction in the data analysis module is as follows: Step 1: Extract hysteresis features. Extract hysteresis features from time series data to capture the impact of historical data on the current moment. For each sensor X, the time lag feature is X(t-τ)τ=1,2,3...n, where τ represents the lag time step and n is the number of lag steps. Step 2: Statistical features: extract statistical features of the data in the window, including mean, standard deviation, maximum and minimum values. For window W t The statistical characteristic S is calculated as follows: S = {mean(W t ),std(W t ),max(W t ),min(W t )}, where mean(W t ) is the mean of the data in the window, std(W t ) is the standard deviation of the data in the window, max(W t ) is the maximum value of the data in the window, min(W t ) is the minimum value of the data in the window; Step 3: Differential features. In order to capture the changing trend of data, differential features are introduced. The differential calculation formula is: ΔX t =X t -X t-1 , where X t is the input data at time t, X t-1 is the input data at time t-1, ΔX t It is the difference between the current moment and the previous moment.

4. The system for predicting rock burst and three-dimensional model visualization of underground mine tunnels based on the Internet of Things according to claim 1 is characterized in that: The implementation steps of the machine learning training module are: Step 1: First, use logistic regression to model the probability of rock burst events. The model input is the preprocessed time series features, and the output is the probability of rock burst events. The decision function of the logistic regression model is: Where P(y t -1|X t ) is the given input X t When event y t The probability of occurrence, y t is the target value at the tth moment, 1 indicates that the event occurs, 0 indicates that it does not occur, X t is the input feature vector at the tth moment, β0 is the intercept term, and X1β1 is the weight of the model; Step 2: To prevent the model from overfitting, L2 regularization is added, and the loss function is: Where N is the number of samples, y i is the predicted probability of the ith sample, λ is the regularization strength, is the jth coefficient of logistic regression; Step 3: Based on logistic regression, support vector machine is further used to perform secondary classification on the prediction results of the model. The decision function of support vector machine is: Where f(X) is the output of the decision function of the support vector machine, α i represents the weight of the support vector, y i is the class label of the support vector, X i is the current vector, X is the current input feature, K(X i ,X) is used to measure X i The similarity with X, b is the bias term; Step 4: Next, perform model training and use batch gradient descent to optimize the model. Use small batches of data to update the parameters each time. For the logistic regression model, the gradient update formula is: where β j is the jth coefficient of logistic regression, η is the learning rate, L is the loss function, is the loss function with respect to β j The partial derivative of Step 5: Then perform dynamic learning adjustment and introduce a dynamic learning rate adjustment strategy. Initially, use a larger learning rate to accelerate training, and then gradually reduce the learning rate. The formula is: where η t is the learning rate at the tth iteration, η0 is the initial learning rate, α is the decay rate of the learning rate, and t is the number of iterations; Step 6. Finally, perform early stopping monitoring on the training set. When the verification loss no longer decreases, terminate the training early to avoid overfitting of the model on the training set.

5. The system for predicting rock burst and three-dimensional model visualization of underground mine tunnels based on the Internet of Things according to claim 1 is characterized in that: The specific operations of the model verification and prediction module are: Step 1: After training, use an independent test set to evaluate the model and give high weights ω to false negatives FN , giving lower weights ω to false positives FP , the weights of true positive TP and true negative TN are set to 1, then the weighted accuracy formula is: Among them, TP is the true positive example, which indicates the number of samples correctly predicted as positive by the model; TN is the true negative example, which indicates the number of samples correctly predicted as negative by the model; FP is the false positive example, which indicates the number of samples incorrectly predicted as positive by the model; FN is the false negative example, which indicates the number of samples incorrectly predicted as negative by the model; Step 2: For rock burst events, the probability output by the logistic regression model is used as the risk level of the rock burst event, and a certain threshold value P>0.5 is set to determine whether a rock burst event will occur; Step 3: For the output of the prediction model, the classification boundary is further optimized through the support vector machine to output the final prediction result. The final model can dynamically predict the risk of rock burst based on the real-time collected IoT data and display the changes of the underground 3D model on the computer.

Images