Mine potential safety hazard early warning method and system based on big data

Abstract

The invention provides a mine potential safety hazard early warning method and system based on big data, and relates to the technical field of landslide early warning, and the method comprises the steps: constructing a space coordinate system, collecting the position data of a monitoring target, and collecting a depth information sequence before slope instability; then, a prediction model is trained through the depth information sequence. In the data processing stage, depth information of a monitored target is calculated, and the data is input into the trained model to predict the slope instability probability. And according to a comparison result of the prediction probability and a preset threshold value, whether data acquisition is continued or an early warning signal is sent is determined, so that the accuracy of early warning of the potential safety hazard of the mine is improved.

Description

Technical Field

[0001] The present application relates to the technical field of landslide early warning, and in particular to a method and system for early warning of mine safety hazards based on big data. Background Art

[0002] In the mining process, the stability of the slope is one of the key factors for safe production. However, due to complex geological conditions, frequent mining activities and the influence of natural factors, the mine slopes are often at risk of instability. Once the slope is unstable, it may lead to serious safety accidents such as landslides and collapses, posing a huge threat to the safety of life and property.

[0003] Traditional methods for monitoring the stability of mine slopes mainly rely on manual inspections and simple instrument measurements. These methods have many limitations, such as low monitoring frequency, limited coverage, and insufficient data accuracy, which will reduce the timeliness and accuracy of mine safety hazard warnings. Summary of the invention

[0004] In order to improve the accuracy of mine safety hazard early warning, the present application provides a mine safety hazard early warning method and system based on big data.

[0005] In the first aspect, the present application provides a mine safety hazard early warning method based on big data, which adopts the following technical solution: A mine safety hazard early warning method based on big data includes the following steps: First acquisition: construct a spatial coordinate system, collect the position data of the monitoring target in the spatial coordinate system, and record it as the first data; collect the depth information sequence corresponding to different slopes before they become unstable; First, modeling and training: establish a prediction model, use the deep information sequence to train the prediction model, and obtain the trained prediction model; Data processing: calculating the depth information of the monitoring target based on the first data, and recording it as the second data; Prediction: Input the second data into the trained prediction model to obtain the probability of slope instability; First judgment: judging whether the probability is less than a preset threshold, if so, executing the first collection step; if not, executing the early warning step; Early warning: Send out early warning signals.

[0006] The present application first constructs a spatial coordinate system and collects the position data of the monitoring target. By constructing the spatial coordinate system, the three-dimensional position information of the monitoring target can be accurately obtained, which is helpful for subsequent data processing and prediction. The present application also collects the depth information sequence corresponding to different slope instability. The depth information sequence can reflect the characteristic changes before the slope instability and provide valuable training data for the prediction model. After that, a prediction model is established, and the depth information sequence is used as training data to train the prediction model, which helps the prediction model learn the characteristic change law when the slope is unstable. After that, the depth information of the monitoring target is calculated based on the position data, and the depth information is input into the trained prediction model to obtain the probability of slope instability. The prediction model can quickly give the probability of slope instability according to the input depth information, so as to take measures in time. The present application also determines whether the predicted probability is less than the preset threshold. If so, continue to monitor, if not, send out an early warning signal. The present application trains the prediction model by using depth information, so that the prediction model can use the currently collected position data to predict the probability of slope instability in the future, and then can send out an early warning in time to reduce the potential hazards and losses caused by slope instability.

[0007] Optionally, after the step of performing data processing and before the step of performing prediction, the method further includes: Second acquisition: obtaining n depth information of the monitored target before the current moment, recorded as first depth information, where n is equal to the number of samples in the depth information sequence minus one; Information integration: integrating the n first depth information and the second data into a second data sequence; First calculation: normalize the kth depth information sequence and the second data sequence, and calculate the Euclidean distance between the normalized kth depth information sequence and the normalized second data sequence; Statistics: Counting the number of the Euclidean distances less than a preset Euclidean distance threshold, and recording it as the third data; Second judgment: judging whether the third data is greater than a preset quantity threshold, if so, executing the prediction step; if not, executing the iteration step; Iteration: taking the k+1th depth information sequence as the new kth depth information sequence, and executing the first calculation step until a preset stop condition is met.

[0008] The present application obtains n depth information of the monitoring target before the current moment, and integrates n first depth information and second data into a second data sequence, then normalizes the kth depth information sequence and the second data sequence, and calculates the Euclidean distance between the kth depth information sequence after normalization and the second data sequence after normalization. By calculating the Euclidean distance, the similarity between different depth information sequences and the second data sequence can be quantified, and then the number of Euclidean distances less than a preset Euclidean distance threshold is counted to quantify the similarity between the second data sequence and the kth depth information sequence, and then the number of depth information sequences with a high similarity to the second data sequence is counted to obtain the third data, and then it is determined whether the third data is greater than the preset number threshold to realize automatic judgment of the similarity of the data sequence, thereby deciding whether to execute the prediction step, thereby improving the accuracy of the prediction. The present application can determine whether the second data is suitable for prediction by calculating the Euclidean distance and counting the number of similar samples, and only predicts when the data similarity is high, so as to improve the accuracy of the prediction.

[0009] Optionally, after performing the second judgment step and before performing the prediction step, the method further includes: Construct a matrix: Construct a distance matrix, and the calculation model of the element in the i-th row and j-th column of the distance matrix is ​​as follows: ; in, is the element in the i-th row and j-th column of the distance matrix; is the Euclidean distance between the i-th sample in the k-th depth information sequence and the j-th sample in the second data sequence; First acquisition: using a dynamic path planning algorithm to find the shortest path through the distance matrix, and calculating the sum of all elements included in the shortest path, recorded as fourth data; The third judgment: judge whether the fourth data is greater than a preset threshold value, if so, execute the iteration step; if not, execute the prediction step.

[0010] The present application constructs a distance matrix that can intuitively display the cumulative distances between samples in different depth information sequences, providing a basis for subsequent shortest path search. Subsequently, the present application uses a dynamic path planning algorithm to find the shortest path and obtain the shortest path through the distance matrix to evaluate the overall similarity and correlation between the second data sequence and the kth depth information sequence. Subsequently, the sum of all elements included in the shortest path is calculated to time-align the second data sequence with the kth depth information sequence. Subsequently, it is determined whether the fourth data is greater than a preset threshold value to automatically determine the overall similarity between the second data sequence and the kth depth information sequence, thereby improving the accuracy of the prediction.

[0011] Optionally, after executing the third judgment step and before executing the prediction step, the method further includes: Second acquisition: obtaining the timestamp of the first data, deleting the year information in the timestamp, and recording the remaining part of the timestamp as time m; Third acquisition: respectively obtain the historical position data of the monitoring target at time m every year, and the offset corresponding to each historical position data; Second calculation: calculate the average value of the historical position data, recorded as the fifth data; calculate the average value of the offset, recorded as the sixth data; Third calculation: calculating the difference between the fifth data and the sixth data, recorded as the first difference; First adjustment: inputting the first difference and the first data into the FOPID controller to obtain an adjustment amount of the first data; Fourth calculation: performing a subtraction operation on the first data and the adjustment amount to obtain a calculation result, and using the calculation result as the new first data.

[0012] The present application obtains the timestamp of the first data and deletes the year information, retaining only the month, date and time part (and recording this part as the m moment), so that the historical position data obtained subsequently can be aligned with the current data in the time dimension, but not affected by the year difference, so that the historical behavior pattern at the same time point can be analyzed. Subsequently, the historical position data of the monitoring target at the m moment of each year and its corresponding offset are obtained to reveal the position change law of the target at this time point, as well as possible anomalies or trends. Subsequently, the average value of the historical position data (i.e., the fifth data) and the average value of the offset (i.e., the sixth data) are calculated. By calculating the average value, the abnormal fluctuations of individual data can be smoothed out to obtain a more representative historical position and offset benchmark. Subsequently, the difference between the fifth data and the sixth data (i.e., the first difference) is calculated, and the deviation of the fifth data can be corrected by the sixth data. Subsequently, the first difference and the first data are input into the FOPID controller to obtain the adjustment amount of the first data. The FOPID controller (fractional proportional-integral-differential controller) can calculate a suitable adjustment amount based on the input data to correct the deviation or error in the current data and improve the accuracy of the data. Subsequently, the first data is subtracted from the adjustment amount to obtain a new first data. The present application realizes the correction of the original first data by adopting the above-mentioned scheme, so that the new first data is closer to the expected position of the target at time m, providing a more accurate data basis for subsequent prediction steps, and helping to improve the accuracy of the prediction, especially when processing data with periodic or seasonal changes.

[0013] Optionally, after performing the first adjustment step and before performing the fourth calculation step, the method further includes: Second modeling and training: establishing an adjustment model, and training the adjustment model using the fifth data and the sixth data to obtain a trained adjustment model; Second prediction: input the first data into the trained adjustment model to obtain the predicted adjustment amount; Fifth calculation: calculating the difference between the adjustment amount and the predicted adjustment amount, recorded as the second difference; Fourth judgment: judging whether the second difference is less than a preset difference threshold, if so, executing the fourth calculation step; if not, executing the feedback step; Feedback: Send feedback signals to operation and maintenance personnel.

[0014] The present application establishes a regulation model and trains the model using the fifth data (the average value of the historical position data) and the sixth data (the average value of the offset). Through training, the model can learn the relationship between the historical position and the offset. Subsequently, the current first data is input into the trained regulation model to obtain the predicted regulation amount. Subsequently, the difference between the actual regulation amount (the regulation amount output from the FOPID controller) and the predicted regulation amount is calculated and recorded as the second difference. By comparing the actual and predicted regulation amounts, the prediction accuracy of the model can be evaluated. Subsequently, it is determined whether the second difference is less than the preset difference threshold. If so, the fourth calculation step is continued; if not, a feedback signal is sent to the operation and maintenance personnel, indicating that the difference between the predicted regulation amount and the actual regulation amount exceeds the acceptable range, so that the operation and maintenance personnel can promptly understand the prediction performance problems of the FOPID controller and take appropriate measures to correct them, thereby maintaining the accuracy and reliability of the entire data processing process.

[0015] Optionally, after executing the fourth judgment step and before executing the feedback step, the method further includes: Build model: Build a reinforcement learning model; Setting state: setting the state space of the reinforcement learning model, the state space including: the difference between the adjustment amount output by the FOPID controller and the predicted adjustment amount output by the adjustment model; setting the action space of the reinforcement learning model, the action space including: the parameter adjustment strategy of the FOPID controller; Setting an agent: Setting an agent, wherein the agent is used to select a parameter adjustment strategy of the FOPID controller according to a current state of the FOPID controller; First adjustment: inputting the second difference into the reinforcement learning model, obtaining a parameter adjustment strategy corresponding to the second difference, and adjusting the current parameters of the FOPID controller according to the parameter adjustment strategy corresponding to the second difference; Model update: Use the adjusted FOPID controller as the new controller and execute the first adjustment step to obtain a new adjustment amount; Fifth judgment: judge whether the new adjustment amount meets the expectation, if so, execute the fourth calculation step; if not, execute the feedback step.

[0016] This application introduces the basis of reinforcement learning by constructing a reinforcement learning model, and provides the necessary algorithm framework for the subsequent steps. Subsequently, the state space of the reinforcement learning model is defined, including the difference between the adjustment amount output by the FOPID controller and the predicted adjustment amount output by the adjustment model, and the action space of the reinforcement learning model is set, including the parameter adjustment strategy of the FOPID controller. By clarifying the state space and action space, the agent can clearly know how it should choose actions according to the current state. The choice of state space enables the model to pay attention to the difference between the adjustment amount and the predicted adjustment amount, while the action space allows the model to optimize this difference by adjusting the parameters of the FOPID controller. Subsequently, the agent is set to select the parameter adjustment strategy according to the current state of the FOPID controller, that is, to select the optimal action according to the current state. Subsequently, the second difference is input into the reinforcement learning model to obtain the corresponding parameter adjustment strategy, and the current parameters of the FOPID controller are adjusted according to the strategy. Subsequently, the adjusted FOPID controller is used as a new controller, and the first adjustment step is performed to obtain a new adjustment amount. Then, it is determined whether the new adjustment amount meets expectations. If the new adjustment amount meets expectations, the subsequent steps can be continued; if it does not meet expectations, the feedback step is executed, and the parameters of the reinforcement learning model or FOPID controller need to be further adjusted.

[0017] Optionally, after performing the first adjustment step and before performing the model update step, the method further includes: Fourth acquisition: acquiring training data, wherein the training data includes a difference between an adjustment amount output by the FOPID controller and a predicted adjustment amount output by the adjustment model, and parameters of the FOPID controller; Third, modeling and training: establishing an MLP model and using training data to train the MLP model to obtain a trained MLP model; Third prediction: Input the second difference into the trained MLP model to obtain the predicted parameters of the FOPID controller, which are recorded as the first parameters; Fifth acquisition: acquiring the parameter adjusted by the FOPID controller in the first adjustment step, recorded as the second parameter; Sixth judgment: judge whether the difference between the first parameter and the second parameter is less than the preset parameter difference threshold, if so, execute the model update step; if not, execute the feedback step.

[0018] The present application obtains training data, then establishes an MLP model, and uses the training data to train the MLP model to obtain a trained MLP model. By training the MLP model, the MLP model can learn the relationship between the difference between the adjustment amount and the predicted adjustment amount and the FOPID controller parameters, so that the MLP model can predict the optimal FOPID controller parameters based on the new difference in the future. Subsequently, the second difference (i.e., the difference between the current adjustment amount and the predicted adjustment amount) is input into the trained MLP model to obtain the predicted parameters of the FOPID controller, which are recorded as the first parameters. The present application adopts the above scheme to use the MLP model to predict the optimal FOPID controller parameters under the current situation, and then obtains the adjusted parameters of the FOPID controller in the first adjustment step, which are recorded as the second parameters, and determines whether the difference between the first parameter (the predicted parameter of the MLP model) and the second parameter (the actual adjusted parameter) is less than the preset parameter difference threshold. If the predicted parameters are close enough to the actual adjusted parameters (i.e., the difference between the two parameters is less than the preset threshold), it can be considered that the performance of the FOPID controller is in line with expectations, and the model update step can be continued; if the difference is too large, it indicates that the performance of the FOPID controller is not in line with expectations, and feedback steps need to be performed to further adjust the model or parameters, thereby improving the stability and reliability of the FOPID controller.

[0019] Optionally, after executing the fourth acquisition step and before executing the third modeling and training step, the method further includes: Seventh judgment: judging whether the number of samples in the training data meets the preset number threshold, if yes, executing the third modeling and training step; if no, executing the fourth modeling step; Fourth modeling: building a generative adversarial network; Generate data: Generate new training data using a generative adversarial network; Data update: The new training data and the training data obtained in the fourth step are used as new training data.

[0020] By adopting the above scheme, the present application can utilize the generative ability of the generative adversarial network to generate new training samples. These new samples will be used together with the existing training data to train the MLP model, thereby increasing the generalization ability and prediction accuracy of the model. Subsequently, the new training data and the training data in the fourth acquisition step are used as new training data to form a larger and more comprehensive training set. This updated training set will be used for subsequent MLP model training, which will help improve the performance and stability of the MLP model. By increasing the amount of training data, it can also reduce the problem of insufficient model training and improve the stability and reliability of the entire control system.

[0021] Optionally, after executing the fourth modeling step and before executing the data generating step, the method further includes: Fourth training: training the generative adversarial network using the training data obtained in the fourth step to obtain a trained generative adversarial network; In the step of generating data, the trained generative adversarial network is used to generate new training data.

[0022] This application uses the training data in the fourth acquisition step to train the generative adversarial network. Through training, the generator gradually learns how to generate samples similar to real data, and the discriminator learns how to more accurately identify real data and generated data. This process is an iterative process until the generator can generate new samples that are close enough to the real data and the discriminator has difficulty distinguishing them. Subsequently, in the data generation step (using the trained generative adversarial network), new training data is generated using the trained generative adversarial network. After the GAN has been fully trained, the generator has learned how to generate high-quality new samples that are similar to real data. These new samples can be used as additional training data to enhance the training effect of other machine learning models.

[0023] In the second aspect, the present application provides a mine safety hazard early warning system based on big data, which adopts the following technical solutions: A mine safety hazard early warning system based on big data, including: processor and memory, The memory stores program code; The processor executes the steps of the method when calling the program code in the memory.

[0024] In summary, the present application includes at least one of the following beneficial technical effects: 1. This application first constructs a spatial coordinate system and collects the position data of the monitoring target. By constructing the spatial coordinate system, the three-dimensional position information of the monitoring target can be accurately obtained, which is helpful for subsequent data processing and prediction. This application also collects the depth information sequence corresponding to different slope instability. The depth information sequence can reflect the characteristic changes before the slope instability, providing valuable training data for the prediction model.

[0025] 2. This application establishes a prediction model and uses the depth information sequence as training data to train the prediction model, which helps the prediction model learn the characteristic change law when the slope is unstable. Then, the depth information of the monitoring target is calculated based on the position data, and the depth information is input into the trained prediction model to obtain the probability of slope instability. The prediction model can quickly give the probability of slope instability based on the input depth information, so as to facilitate timely measures.

[0026] 3. The application also determines whether the predicted probability is less than a preset threshold. If so, monitoring continues; if not, an early warning signal is issued. The application trains the prediction model using depth information so that the prediction model can use the currently collected location data to predict the probability of future slope instability, and then issue an early warning in time to reduce the potential harm and losses caused by slope instability. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of Example 1 of the present application; Figure 2 is a flow chart of Example 2 of the present application; Figure 3 is a flow chart of Example 3 of the present application; Figure 4 This is a flow chart from S81 second modeling and training to S86 feedback in Example 4 of the present application; Figure 5 This is a flowchart from S8541 fourth acquisition to S8545 sixth judgment of Example 4 of the present application; Figure 6 This is a flow chart of Example 5 of the present application. DETAILED DESCRIPTION

[0028] The following combination Figures 1 to 6 This application is described in further detail.

[0029] Example 1: This example discloses a mine safety hazard early warning method based on big data, referring to Figure 1 , the method includes: S1 first acquisition, S2 first modeling and training, S3 data processing, S4 prediction, S5 first judgment and S6 early warning. First, a spatial coordinate system is constructed and the position data of the monitoring target is collected, as well as the depth information sequence before the slope becomes unstable. Subsequently, the prediction model is trained through the depth information sequence. The data processing stage calculates the depth information of the monitoring target, and the data is input into the trained model to predict the probability of slope instability. According to the comparison result between the predicted probability and the preset threshold, it is decided whether to continue data acquisition or issue an early warning signal. This embodiment includes the following steps: S1 is the first acquisition, which uses a camera to take regular or real-time photos of the monitoring target (i.e., the slope) to obtain image data.

[0030] A three-dimensional space coordinate system is established with the camera position as the coordinate origin, and the position data of the monitoring target (such as a specific point or area on the slope) is recorded. These position data are marked as first data.

[0031] For example, a GNSS receiver is deployed at the camera installation location to receive and record the camera's location data. This GNSS receiver continuously receives signals from multiple satellite systems such as GPS, Beidou, GLONASS, etc., and uses built-in algorithms to calculate the camera's longitude, latitude, and altitude information, which will be used as the origin data of the three-dimensional space coordinate system.

[0032] At the same time, additional GNSS receivers need to be deployed at specific points or key areas on the slope. These GNSS receivers will also receive signals from satellites and calculate their respective three-dimensional position data. These three-dimensional position data will be recorded in the database as the position data of the monitoring target and associated with the camera's position data (i.e., coordinate origin data) to obtain the coordinate value of the monitoring target in the three-dimensional space coordinate system.

[0033] Calculate the difference between the longitude of the monitoring target and the longitude of the camera, calculate the difference between the latitude of the monitoring target and the latitude of the camera, calculate the difference between the height of the monitoring target and the height of the camera, and use the three differences as the coordinate values ​​of the corresponding coordinate axes.

[0034] In this way, a three-dimensional space coordinate system with the camera position as the origin can be established, and a specific point or area on the slope can be accurately located in this coordinate system. These position data (first data) will be used for subsequent slope stability analysis. By comparing the changes in position data at different time points, the morphological changes of the slope can be discovered in time, thereby predicting the stability state of the slope.

[0035] For multiple known slope instability cases, their depth information sequences before instability are collected. These depth information reflects the morphological changes of the slope before instability.

[0036] S2 first models and trains, building a prediction model that can process deep information sequences, such as a deep learning model (such as a convolutional neural network CNN or a recurrent neural network RNN) or a traditional machine learning algorithm.

[0037] The prediction model is trained using the collected depth information sequences before different slope failures as training data, so that the prediction model can learn to identify the patterns or characteristics of slope instability from the depth information sequences.

[0038] S3 data processing, based on the first data (i.e., the position data of the monitoring target in the spatial coordinate system), calculates the depth information of the monitoring target in the image data. This information reflects the three-dimensional shape of the slope surface and is recorded as the second data.

[0039] S4 predicts, inputting the calculated depth information (second data) into the trained prediction model, and the trained prediction model outputs the probability of slope instability, thereby obtaining the probability of slope instability in the future under the condition of the second data.

[0040] S5 first judgment, preset a threshold value to judge the level of slope instability risk. If the predicted probability is less than this threshold, it is considered that the current slope state is relatively stable, and S1 first acquisition is continued to monitor the changes of the slope. If the probability is not less than the threshold, it is considered that the slope has a high risk of instability, and S6 warning needs to be executed.

[0041] S6 warning, issuing a warning signal.

[0042] This embodiment uses a camera to collect image data of the monitoring target and constructs a spatial coordinate system to obtain the position data of the monitoring target (i.e., the first data). By collecting depth information sequences before different slope instability, a prediction model is established and trained. Subsequently, the depth information of the monitoring target calculated based on the image data (i.e., the second data) is input into the trained model to predict the probability of slope instability. If the predicted probability exceeds the preset threshold, an early warning signal is issued; otherwise, data collection continues to continuously monitor the slope status.

[0043] Example 2: Reference Figure 2 The difference between this embodiment and embodiment 1 is that after executing S3 data processing and before executing S4 prediction, it also includes: S31 is a second acquisition, in which n depth information of the monitoring target before the current moment is obtained from previous data records, and the depth information of the monitoring target before the current moment is recorded as the first depth information.

[0044] To facilitate the calculation of subsequent data, n is equal to the number of samples in the depth information sequence minus one.

[0045] S32 information integration, integrating the n first depth information and the second data into a second data sequence, the integration method may be: arranging the n first depth information and the second data in chronological order, and recording the arranged sequence as the second data sequence; the integration method may also be: arranging the n first depth information and the second data in order of size (from large to small or from small to large), and recording the arranged sequence as the second data sequence.

[0046] S33 first calculation, normalizing the kth depth information sequence and the second data sequence, and calculating the Euclidean distance between the normalized kth depth information sequence and the normalized second data sequence, the calculation model is as follows: ; Where p=q, is the Euclidean distance between the kth depth information sequence after normalization and the second data sequence after normalization; is the qth sample in the second data sequence after normalization. is the kth depth information sequence after normalization. samples.

[0047] S34 performs statistics, setting a Euclidean distance threshold, and counting the number of all Euclidean distances smaller than the threshold in the first calculation in S33 as third data.

[0048] S35 is the second judgment, judging whether the third data is greater than a preset quantity threshold. If so, it indicates that there are enough sample pairs showing similarity, and S36 is executed to construct a matrix; if not, it indicates that the second data sequence is not similar enough to the current depth information sequence, and S39 needs to be executed for iteration.

[0049] S36 constructs a matrix, constructs a distance matrix, and the calculation model of the element in the i-th row and j-th column in the distance matrix is ​​as follows: ; in, is the element in the i-th row and j-th column of the distance matrix; is the Euclidean distance between the ith sample in the kth depth information sequence and the jth sample in the second data sequence.

[0050] S37 first obtains, using a dynamic path planning algorithm to find the shortest path from the starting point (first row and first column) to the end point (n+1th row and n+1th column) in the distance matrix, and calculates the sum of all elements on the path as the fourth data.

[0051] S38 is the third judgment, judging whether the fourth data is greater than a preset threshold, if so, executing S39 iteration; if not, executing S4 prediction.

[0052] S39 iterates, taking the k+1th depth information sequence as a new kth depth information sequence, and performing the first calculation in S33 until a preset stop condition is satisfied. The preset stop condition may be: the number of depth information sequences sufficiently similar to the second data sequence is greater than a preset number threshold, or all depth information sequences have been similarly calculated with the second data sequence.

[0053] The present embodiment is described in detail below with reference to specific cases.

[0054] Set the current time as T, the depth information at time T is 9.8m, and the depth information of the previous five time points needs to be obtained. These depth information are: T-1 moment: 10m, T-2 moment: 9.9m, T-3 moment: 9.9m, T-4 moment: 9.9m, T-5 moment: 9.9m.

[0055] Then the second data sequence is: {10, 9.9, 9.9, 9.9, 9.9, 9.8}, which becomes {1.73, 0, 0, 0, 0, -1.73} after normalization.

[0056] There are four depth information sequences: {18, 18, 18, 18, 17.8}, {5.3, 5.2, 5.1, 5, 4.9, 4.7}, {40, 40, 40, 39, 38, 37}, and {17, 17, 17, 16.9, 16.8, 16.7}.

[0057] The four depth information sequences after normalization are: {0.7023, 0.7023, 0.7023, 0.7023, 0.7023, −0.1395}, {1.0913, 0.6546, 0.2179, −0.2179, −0.6546, −1.0913}, {0.6198, 0.6198, 0.6198, −0.1240, −0.9300, −1.7400}, and {1.0, 1.0, 1.0, 0.0, −1.0, −2.0}.

[0058] Using the Euclidean distance calculation formula in the first calculation of S33, the Euclidean distance between {1.73, 0, 0, 0, 0, -1.73} and {0.7023, 0.7023, 0.7023, 0.7023, 0.7023, -0.1395} is calculated as follows: .

[0059] The Euclidean distance between {1.73, 0, 0, 0, 0, -1.73} and {1.0913, 0.6546, 0.2179, -0.2179, -0.6546, -1.0913} is: .

[0060] The Euclidean distance between {1.73, 0, 0, 0, 0, -1.73} and {0.6198, 0.6198, 0.6198, −0.1240, −0.9300, −1.7400} is: .

[0061] The Euclidean distance between {1.73, 0, 0, 0, 0, -1.73} and {1.0, 1.0, 1.0, 0.0, -1.0, -2.0} is: .

[0062] The Euclidean distance threshold is set to 1.9, and the third data is 3. For example, the preset number threshold is set to k / 2.

[0063] It is easy to know that in this embodiment, if 3 is greater than 4 / 2, S36 is executed to construct a matrix.

[0064] This embodiment is described by taking the first depth information sequence as an example, and the calculation method of other depth information sequences is the same as that of the first depth information sequence.

[0065] ; ; in, The distance matrix corresponding to the first depth information sequence.

[0066] The shortest path found by the dynamic path planning algorithm is: , the fourth data is 13.8804.

[0067] Assuming the preset threshold is 15, then 13.8804<15, that is, S4 prediction can be performed. Otherwise, S39 iteration is performed.

[0068] This embodiment calculates the Euclidean distance between the depth information sequence and the second data sequence to determine whether there is a sufficient depth information sequence similar to the second data sequence. If so, a distance matrix is ​​constructed to find out whether the depth information sequence can be aligned with the second data sequence. In particular, when the acquisition frequency of the depth information sequence and the second data sequence is inconsistent, this embodiment can use the distance matrix and the dynamic programming algorithm to explore an optimal alignment path so that the depth information sequence and the second data sequence can achieve the best match in time or space dimensions, and then map each data point in the second data sequence to the most similar position in the depth information sequence according to its position information in the distance matrix. This mapping not only considers the similarity of a single data point, but also comprehensively considers the coherence and consistency of the overall sequence, thereby improving the accuracy and reliability of the mapping result. By mapping the second data sequence to the depth information sequence, the accuracy of the slope instability probability in the S4 prediction step can be improved by using the change of depth information in the depth information sequence.

[0069] Embodiment 3: When monitoring target position data, due to periodic changes such as seasons, the collected position data will change periodically. Therefore, this embodiment sets a FOPID controller to adaptively calculate the offset of the position data to correct the periodic offset. Figure 3 The difference between this embodiment and embodiment 2 is that after executing the third judgment in S38 and before executing the prediction in S4, the following is further included: S71 is a second acquisition, in which the timestamp when the first data is collected is obtained, the year information in the timestamp is deleted, and the remaining part of the timestamp is recorded as time m.

[0070] For example, the time of the first data collection is 10:00:00 on January 1, 2025, then time m is 10:00:00 on January 1.

[0071] S72 third acquisition, for each past year, at the corresponding m time (i.e. 10:00:00 on January 1 of each year), search and record the historical location data of the monitoring target, the location data includes longitude, latitude and altitude, and the offset corresponding to each historical location data. The offset represents the deviation of the actual location data from the expected location data, and the deviation comes from the periodic offset of the location data caused by natural factors, and the natural factors can be: periodic freezing and thawing of soil.

[0072] S73: second calculation: average the position data of all years at time m to obtain an average historical position data as the fifth data. Calculate the average value of the offsets of all years at time m as the sixth data.

[0073] S74 is a third calculation, calculating a difference between the fifth data and the sixth data, and recording it as a first difference.

[0074] S75: first adjustment, inputting the first difference and the first data into a FOPID (fractional-order proportional-integral-differential) controller.

[0075] The FOPID controller can calculate the adjustment variable based on the input signal to optimize or adjust the system performance.

[0076] S76 is a fourth calculation, in which the first data is subtracted from the adjustment amount to obtain a calculation result, and the calculation result is used as the new first data.

[0077] This embodiment collects historical position data and offset at the same time in previous years, and calculates the average value of the two to provide reference data for the position data and offset at the current time, inputs the first data and the first difference into the FOPID controller to output the adjustment amount corresponding to the first data, and uses the adjustment amount to correct the first data to obtain the corrected first data (i.e., the new first data). This embodiment takes into account the periodic offset of the position data caused by external factors, realizes the correction of the current position data, improves the accuracy of the current position data, and further improves the accuracy of the early warning.

[0078] Example 4: Reference Figure 4 The difference between this embodiment and the third embodiment is that after performing the first adjustment in S75 and before performing the fourth calculation in S76, the following is further included: S81 second modeling and training, establishing an adjustment model, using the fifth data and the sixth data to train the adjustment model, and obtaining a trained adjustment model. During the training process, the sixth data is used as the true label of the fifth data to train the adjustment model, so that the trained adjustment model can predict the corresponding offset according to the input position data.

[0079] The adjustment model may adopt various neural network models, such as CNN, RNN or LSTM, etc., as long as the adjustment model is able to predict the corresponding offset according to the input position data after being trained with the fifth data and the sixth data.

[0080] S82 is a second prediction, in which the first data is input into the trained regulation model to obtain a predicted regulation amount, wherein the predicted regulation amount is obtained by the regulation model's understanding of historical data and analysis of current data.

[0081] S83 is a fifth calculation, which is to calculate the difference between the adjustment amount of the first data in the first adjustment of S75 and the predicted adjustment amount, and record it as a second difference. The second difference reflects the difference between the adjustment amount output by the FOPID controller and the expected adjustment amount.

[0082] S84 is the fourth judgment, judging whether the second difference is less than the preset difference threshold. If so, it means that the performance of the FOPID controller meets expectations, and then the fourth calculation of S76 is executed; if not, it means that the performance of the FOPID controller does not meet expectations, and S85 adaptive learning needs to be executed.

[0083] S85 adaptive learning, including S851 building a model, S852 setting a state, S853 setting an intelligent agent, S854 a first adjustment, S855 a model update and S856 a fifth judgment.

[0084] S851 builds a model and builds a reinforcement learning model.

[0085] S852 sets the state and defines the state space of the reinforcement learning model, wherein the state space includes: the difference between the adjustment amount output by the FOPID controller and the predicted adjustment amount output by the adjustment model; sets the action space of the reinforcement learning model, wherein the action space includes: the parameter adjustment strategy of the FOPID controller, such as the adjustment strategy of the gain, integral time, differential time, etc.

[0086] S853 sets an intelligent agent, sets an intelligent agent, and the intelligent agent is used to select a parameter adjustment strategy of the FOPID controller according to the current state of the FOPID controller.

[0087] S854: First adjustment: input the second difference into the reinforcement learning model, obtain the parameter adjustment strategy corresponding to the second difference, and adjust the current parameters of the FOPID controller according to the parameter adjustment strategy corresponding to the second difference.

[0088] S855 model update, the adjusted FOPID controller is used as the new controller, and S75 first adjustment is performed to obtain a new adjustment amount.

[0089] S856 is the fifth judgment, judging whether the new adjustment amount meets expectations. If so, executing S76 is the fourth calculation; if not, executing S86 is the feedback.

[0090] S86 feedback sends a feedback signal to the operation and maintenance personnel, providing them with information about the accuracy of the regulation model and the adjustment of the FOPID controller parameters.

[0091] This embodiment predicts the adjustment amount corresponding to the first data by setting the adjustment model, and obtains the predicted adjustment amount. Through the trained adjustment model, it can output accurate adjustment amount. The predicted adjustment amount and the adjustment amount output by the FOPID controller are subtracted to obtain the difference between the two, that is, the second difference. According to the second difference, it can be judged whether the adjustment amount output by the FOPID controller meets the expectations. If so, the fourth calculation step is executed, otherwise the adaptive learning step is executed to adjust the parameters of the FOPID controller to improve the calculation accuracy of the subsequent adjustment amount. If the adjustment amount output by the FOPID controller still does not meet the expectations after adjusting the parameters, the information about the model accuracy and the adjustment of the FOPID controller parameters is provided to the operation and maintenance personnel, so that the operation and maintenance personnel can adjust the FOPID controller in time. This embodiment further verifies the output of the FOPID controller to determine whether the performance of the FOPID controller meets expectations. If not, it is adjusted in time to improve the accuracy of the subsequent output results, thereby improving the accuracy of the early warning.

[0092] In other embodiments, reference Figure 5 , after executing S854 the first adjustment and before executing S855 the model update, further comprising: S8541 The fourth acquisition is to acquire training data, where the training data includes the difference between the adjustment amount output by the FOPID controller and the predicted adjustment amount output by the adjustment model, and the parameters of the FOPID controller.

[0093] S8542 The third step of modeling and training is to establish an MLP model, and train the MLP model using training data to obtain a trained MLP model. During the training process, the parameters of the FOPID controller are used as the true labels of the difference between the adjustment amount output by the FOPID controller and the predicted adjustment amount output by the adjustment model to train the MLP model, so that the trained MLP model can predict the corresponding parameters of the FOPID controller according to the difference between the adjustment amount output by the FOPID controller and the predicted adjustment amount output by the adjustment model.

[0094] S8543 The third prediction is to input the second difference into the trained MLP model to obtain the prediction parameters of the FOPID controller, which are recorded as the first parameters.

[0095] S8544 is the fifth acquisition, which obtains the parameters adjusted by the FOPID controller in the first adjustment of S854 and records them as the second parameters.

[0096] S8545 is the sixth judgment, judging whether the difference between the first parameter and the second parameter is less than the preset parameter difference threshold. If so, it means that the prediction of the MLP model is accurate and the parameter adjustment strategy is effective, and then S855 model update is executed; if not, S86 feedback is executed.

[0097] This embodiment first obtains training data including the difference between the adjustment amount output by the FOPID controller and the adjustment amount predicted by the adjustment model, as well as the corresponding FOPID controller parameters (as true labels). Next, an MLP model is established and trained using these training data so that the model can predict the parameters of the FOPID controller based on the adjustment amount difference. Then, the new adjustment amount difference is input into the trained MLP model to predict the parameters of the FOPID controller (first parameter). After that, the adjusted FOPID controller parameters (second parameter) are obtained. Finally, the difference between the two parameters is compared. If the difference is less than the preset parameter threshold, it is considered that the MLP model prediction is accurate and the parameter adjustment strategy is effective, and the model update step is executed; otherwise, the feedback step is executed to further optimize the model or adjustment strategy.

[0098] Example 5: Reference Figure 6 , after executing S8541 the fourth acquisition and before executing S8542 the third modeling and training, further comprising: S91 is the seventh judgment, judging whether the number of samples in the training data obtained in the fourth step S8541 meets the preset number threshold. If so, execute S8542 the third modeling and training; if not, execute S92 the fourth modeling.

[0099] S92 Fourth modeling, establishing a generative adversarial network.

[0100] S93 fourth training, using the training data obtained in S8541 fourth training to train the generative adversarial network, to obtain a trained generative adversarial network.

[0101] S94 generates data by using the trained generative adversarial network to generate new training data.

[0102] S95 data update, the new training data and the training data in the fourth acquisition in S8541 are used as new training data.

[0103] This embodiment determines the number of samples in the training data. When the number of samples is not sufficient, a generative adversarial network is used to supplement it to enrich the number of samples and improve the accuracy of the prediction results of the MLP model.

[0104] Embodiment 6: This embodiment discloses a mine safety hazard early warning system based on big data, the system comprising: processor and memory, The memory stores program code; The processor executes the steps of the method when calling the program code in the memory.

[0105] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A mine safety hazard early warning method based on big data, characterized in that: include: First acquisition: construct a spatial coordinate system and collect the position data of the monitoring target in the spatial coordinate system, which is recorded as the first data; Collect the depth information sequences corresponding to different slope failures; First, modeling and training: establish a prediction model, use the deep information sequence to train the prediction model, and obtain the trained prediction model; Data processing: calculating the depth information of the monitoring target based on the first data, and recording it as the second data; Prediction: Input the second data into the trained prediction model to obtain the probability of slope instability; First judgment: judging whether the probability is less than a preset threshold, if so, executing the first acquisition step; If not, follow the early warning steps; Early warning: Send out early warning signals.

2. The mine safety hazard early warning method based on big data according to claim 1 is characterized in that: After the data processing step and before the prediction step, it also includes: Second acquisition: obtaining n depth information of the monitored target before the current moment, recorded as first depth information, where n is equal to the number of samples in the depth information sequence minus one; Information integration: integrating the n first depth information and the second data into a second data sequence; First calculation: normalize the kth depth information sequence and the second data sequence, and calculate the Euclidean distance between the normalized kth depth information sequence and the normalized second data sequence; Statistics: Counting the number of the Euclidean distances less than a preset Euclidean distance threshold, and recording it as the third data; Second judgment: judging whether the third data is greater than a preset quantity threshold, if so, executing the prediction step; if not, executing the iteration step; Iteration: taking the k+1th depth information sequence as the new kth depth information sequence, and executing the first calculation step until a preset stop condition is met.

3. The mine safety hazard early warning method based on big data according to claim 2 is characterized in that: After executing the second judgment step and before executing the prediction step, the method further includes: Construct a matrix: Construct a distance matrix, and the calculation model of the element in the i-th row and j-th column of the distance matrix is ​​as follows: ； in, is the element in the i-th row and j-th column of the distance matrix; is the Euclidean distance between the i-th sample in the k-th depth information sequence and the j-th sample in the second data sequence; First acquisition: using a dynamic path planning algorithm to find the shortest path through the distance matrix, and calculating the sum of all elements included in the shortest path, recorded as fourth data; The third judgment: judge whether the fourth data is greater than a preset threshold value, if so, execute the iteration step; if not, execute the prediction step.

4. The mine safety hazard early warning method based on big data according to claim 3 is characterized in that: After executing the third judgment step and before executing the prediction step, the method further includes: Second acquisition: obtaining the timestamp of the first data, deleting the year information in the timestamp, and recording the remaining part of the timestamp as time m; Third acquisition: respectively obtain the historical position data of the monitoring target at time m every year, and the offset corresponding to each historical position data; Second calculation: calculate the average value of the historical position data, recorded as the fifth data; calculate the average value of the offset, recorded as the sixth data; Third calculation: calculating the difference between the fifth data and the sixth data, recorded as the first difference; First adjustment: inputting the first difference and the first data into the FOPID controller to obtain the adjustment amount of the first data; Fourth calculation: performing a subtraction operation on the first data and the adjustment amount to obtain a calculation result, and using the calculation result as the new first data.

5. The mine safety hazard early warning method based on big data according to claim 4 is characterized in that: After performing the first adjustment step and before performing the fourth calculation step, the method further includes: Second modeling and training: establishing an adjustment model, and training the adjustment model using the fifth data and the sixth data to obtain a trained adjustment model; Second prediction: input the first data into the trained adjustment model to obtain the predicted adjustment amount; Fifth calculation: calculating the difference between the adjustment amount and the predicted adjustment amount, recorded as the second difference; Fourth judgment: judging whether the second difference is less than a preset difference threshold, if so, executing the fourth calculation step; if not, executing the feedback step; Feedback: Send feedback signals to operation and maintenance personnel.

6. The mine safety hazard early warning method based on big data according to claim 5 is characterized in that: After executing the fourth determination step and before executing the feedback step, the method further includes: Build model: Build a reinforcement learning model; Setting state: setting the state space of the reinforcement learning model, the state space including: the difference between the adjustment amount output by the FOPID controller and the predicted adjustment amount output by the adjustment model; setting the action space of the reinforcement learning model, the action space including: the parameter adjustment strategy of the FOPID controller; Setting an agent: Setting an agent, wherein the agent is used to select a parameter adjustment strategy of the FOPID controller according to a current state of the FOPID controller; First adjustment: inputting the second difference into the reinforcement learning model, obtaining a parameter adjustment strategy corresponding to the second difference, and adjusting the current parameters of the FOPID controller according to the parameter adjustment strategy corresponding to the second difference; Model update: Use the adjusted FOPID controller as the new controller and execute the first adjustment step to obtain a new adjustment amount; Fifth judgment: judge whether the new adjustment amount meets the expectation, if so, execute the fourth calculation step; if not, execute the feedback step.

7. The mine safety hazard early warning method based on big data according to claim 6 is characterized in that: After executing the first adjustment step and before executing the model update step, the method further includes: Fourth acquisition: acquiring training data, wherein the training data includes a difference between an adjustment amount output by the FOPID controller and a predicted adjustment amount output by the adjustment model, and parameters of the FOPID controller; Third, modeling and training: establish an MLP model, and use the training data to train the MLP model to obtain the trained MLP model; Third prediction: Input the second difference into the trained MLP model to obtain the predicted parameters of the FOPID controller, which are recorded as the first parameters; Fifth acquisition: acquiring the parameter adjusted by the FOPID controller in the first adjustment step, recorded as the second parameter; Sixth judgment: judge whether the difference between the first parameter and the second parameter is less than the preset parameter difference threshold, if so, execute the model update step; if not, execute the feedback step.

8. The mine safety hazard early warning method based on big data according to claim 7 is characterized in that: After executing the fourth acquisition step and before executing the third modeling and training step, the method further includes: Seventh judgment: judging whether the number of samples in the training data meets the preset number threshold, if yes, executing the third modeling and training step; if no, executing the fourth modeling step; Fourth modeling: building a generative adversarial network; Generate data: Generate new training data using a generative adversarial network; Data update: The new training data and the training data obtained in the fourth step are used as new training data.

9. The mine safety hazard early warning method based on big data according to claim 8 is characterized in that: After executing the fourth modeling step and before executing the step of generating data, the method further includes: Fourth training: training the generative adversarial network using the training data obtained in the fourth step to obtain a trained generative adversarial network; In the step of generating data, the trained generative adversarial network is used to generate new training data.

10. A mine safety hazard early warning system based on big data, characterized in that: include: processor and memory, The memory stores program code; When the processor calls the program code in the memory, the steps of the method according to any one of claims 1 to 9 are executed.

Images