Rock burst and roof water inrush composite disaster four-field fusion monitoring and early warning method

Abstract

The invention discloses a rock burst and roof water inrush composite disaster four-field fusion monitoring and early warning method. The method comprises the following steps: firstly, determining each monitoring index and early warning value; during the recovery period, collecting each monitoring index value, screening out monitoring abnormal data and removing the monitoring abnormal data, and then supplementing missing data after removing to form a data set required by subsequent model training and verification; establishing a stress-vibration-energy-seepage four-field multi-source initial early warning model for analyzing the monitoring index values and performing initial early warning; meanwhile, the preliminary early warning result is combined with expert judgment, and the early warning model is optimized and adjusted; and finally, according to the optimized early warning model, judging whether a rock burst and roof water inrush composite disaster risk exists in the current working face stoping period or not, and determining the rock burst and roof water inrush composite disaster type according to a monitoring data change rule. According to the method, the type of the working face rock burst and roof water inrush composite disaster can be accurately judged, and data support is provided for accurate prevention and control of the rock burst and roof water inrush composite disaster.

Description

Technical Field

[0001] This invention relates to a four-field integrated monitoring and early warning method for combined rockburst and roof water inrush disasters, belonging to the technical field of monitoring and early warning of combined rockburst and roof water inrush disasters in coal mines. Background Technology

[0002] As coal mining intensity gradually increases and mining conditions become more complex, ground stress, osmotic pressure, and mining disturbances significantly increase. The extreme environment of high stress, strong disturbance, and high osmotic pressure coupling has drastically increased the risk of combined rockburst and roof water inrush disasters, seriously affecting coal mine safety in my country. Rockburst is a dynamic phenomenon characterized by the instantaneous release and rapid destruction of coal and rock masses around the mine face or working face under high stress; roof water inrush is a flood accident caused by confined water suddenly surging into the mine through roof fissures under the influence of mining. In the "three-high" environment of deep, high ground stress, high osmotic pressure, and strong mining disturbances, there is a close relationship between the two. However, the industry typically adopts a separate-control monitoring and early warning model for rockburst and roof water inrush, making it difficult to accurately predict the risk of combined disasters using existing methods, thus posing a serious threat to mine safety.

[0003] Currently, monitoring of rockbursts in mines mainly relies on microseismic monitoring systems, acoustic monitoring systems, stress monitoring, and drill cuttings methods, with the core objective of capturing stress, energy, and vibration information. Roof water hazard monitoring, on the other hand, depends on hydrological monitoring systems, such as those monitoring water pressure, volume, and temperature, with the core objective of capturing seepage information in the rock strata. Existing systems, due to the lack of data correlation, cannot perform spatiotemporal unified calibration, feature depth extraction, and intelligent fusion analysis of multi-source data. Furthermore, existing analytical techniques operate independently, failing to reveal the complete precursory patterns of complex disasters from the perspective of the coupled evolution of "stress-vibration-energy-seepage."

[0004] Therefore, the research direction required by this invention is to provide a new method for accurate monitoring and early warning of combined rockburst and roof water inrush disasters, so as to provide strong support for disaster prevention and control. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a four-field integrated monitoring and early warning method for combined rockburst and roof water inrush disasters, which can effectively solve the problems existing in the prior art.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: a four-field integrated monitoring and early warning method for combined rockburst and roof water inrush disasters, comprising the following steps:

[0007] Step 1: Based on the rockburst monitoring system and roof water inrush disaster monitoring system currently used in the mine, determine the monitoring indicators and early warning values ​​for each disaster;

[0008] Step 2: During the longwall mining period, continuously collect the monitoring index values ​​of each monitoring system, screen out abnormal data from each monitoring index value, and supplement the missing data after the removal.

[0009] Step 3: Based on the rockburst monitoring system and the roof water inrush disaster monitoring system, establish a multi-source initial early warning model with four fields: stress, vibration, energy, and seepage.

[0010] Step 4: After analyzing the monitoring indicator values ​​processed in Step 2 using the initial early warning model established in Step 3, output the preliminary early warning results;

[0011] Step 5: Verify the preliminary early warning results output in Step 4 by combining them with expert experience, and optimize and adjust the initial early warning model based on the verification results to form an optimized early warning model;

[0012] Step 6: Determine whether there is a risk of combined rockburst and roof water inrush during the current working face mining period based on the optimized early warning model.

[0013] Furthermore, the method for determining the monitoring indicators in step one is as follows: the monitoring indicators are determined based on the rockburst monitoring system and the roof water inrush monitoring system adopted by the mine, wherein rockburst monitoring includes stress field, vibration field and energy field monitoring; roof water inrush monitoring includes vibration field and seepage field monitoring;

[0014] The stress field monitoring includes stress monitoring, drill cuttings monitoring, and support resistance monitoring. The monitoring indicators corresponding to stress monitoring are stress value and stress increment, the monitoring indicators corresponding to drill cuttings monitoring are drill cuttings quantity, and the monitoring indicators corresponding to support resistance monitoring are support working resistance.

[0015] The vibration field monitoring includes microseismic monitoring, and the monitoring indicators corresponding to microseismic monitoring are daily total microseismic energy, daily average microseismic energy, daily maximum microseismic energy, and daily total microseismic frequency.

[0016] The energy field monitoring includes ground sound monitoring, and the monitoring indicators corresponding to ground sound monitoring are average number of events per class, average ground sound intensity per class, average number of events per hour, and average ground sound intensity per hour.

[0017] The seepage field monitoring includes hydrological information monitoring and drainage information monitoring. The monitoring indicators corresponding to hydrological information monitoring are borehole water level and borehole water temperature, and the monitoring indicators corresponding to drainage information monitoring are daily inflow and instantaneous inflow.

[0018] Furthermore, the process of determining the warning value in step one is as follows: the warning value of the monitoring indicators of each monitoring system is determined based on the warning value of the monitoring indicators of the adjacent working face and historical monitoring data. The warning value of the monitoring indicators of the adjacent working face is used as the initial value, and the initial value is adjusted based on the historical monitoring data of the adjacent working face to determine the warning value of the monitoring indicators of the current working face.

[0019] Furthermore, the process of filtering out abnormal data in step two is as follows:

[0020] ① For monitoring data acquired by active monitoring equipment in the monitoring system that performs real-time monitoring, such as stress monitoring, support resistance monitoring, hydrological information monitoring, and drainage information monitoring, when a certain monitoring data value exceeds 200% of the previous monitoring data value, the monitoring data is first marked as suspicious data and the next monitoring data is obtained; if the value of the next monitoring data does not exceed 150% of the value of the previous monitoring data of the suspicious data, the suspicious data is marked as abnormal data; otherwise, the suspicious data is considered normal monitoring data.

[0021] ② For monitoring data obtained by manual fixed-point monitoring in the monitoring system, such as drill cuttings monitoring, when a certain monitoring data value exceeds 200% of the previous monitoring data value, the monitoring data is marked as suspicious data. At the same time, the monitoring value is measured again near the monitoring point. If the current monitoring value does not exceed 150% of the previous monitoring data value of the suspicious data, the suspicious data is marked as abnormal data; otherwise, the suspicious data is considered as normal monitoring data.

[0022] ③ For monitoring data acquired by passive monitoring devices that perform real-time monitoring in the monitoring system, such as microseismic monitoring and ground sound monitoring, considering the randomness of their data monitoring, no abnormal data analysis is performed.

[0023] ④ Use the above methods to filter out and remove abnormal data.

[0024] Furthermore, in step two, the missing data after removal is supplemented as follows: The first 1000 missing data points are used as a dataset, which is then divided into a training set, a validation set, and a test set. A fully connected feedforward neural network is used, with a suitable network architecture designed and a model optimization algorithm selected. The number of training iterations and batch size are configured. The training set data is input into the model, and the predicted values ​​and validation loss are calculated. The gradient of the loss function for each weight is calculated using the backpropagation algorithm. An optimization algorithm is used to update the weights using these gradients to reduce the loss. The loss on the validation set is monitored; if the loss does not decrease after 50 consecutive iterations, training is automatically stopped, and the model weights are restored to the state where the validation loss is lowest. After training, the model is evaluated using the test set. If the evaluation is successful, the model is used to predict and supplement the missing data values.

[0025] Furthermore, the model establishment method in step three is as follows: based on the historical monitoring of the mine, the early warning values ​​of each monitoring indicator of each monitoring system are determined, and a graded early warning is carried out for each monitoring indicator. The location coordinates and time of the above-mentioned historical multi-source data are aligned, and the risks of rockburst, roof water inrush, and combined rockburst and roof water inrush disasters are judged at the hour, shift, and day levels. The subsequent monitoring data and early warning results are used as a comprehensive feature vector. The relationship between the multi-source data features and the early warning results is learned using a long short-term memory network model. Finally, a risk index is output, and the risk of rockburst, roof water inrush, and whether there is a combined rockburst and roof water inrush disaster risk are determined based on the risk index.

[0026] Furthermore, in step four, preliminary early warning is issued based on the analysis results. Specifically, during the working face mining period, the monitoring index values ​​processed in step two are input into the initial early warning model for analysis. Based on the rockburst risk index, roof water inrush risk index, and the comprehensive index of combined rockburst and roof water inrush disaster risk, the preliminary early warning results output by the model include: rockburst risk is categorized as no rockburst risk, weak rockburst risk, moderate rockburst risk, and strong rockburst risk; roof water inrush risk is categorized as no roof water inrush risk, weak roof water inrush risk, moderate roof water inrush risk, and strong roof water inrush risk; and combined rockburst and roof water inrush disaster risk is categorized as having combined rockburst and roof water inrush disaster risk and not having combined rockburst and roof water inrush disaster risk.

[0027] Furthermore, step five specifically involves comparing the preliminary early warning results from step four with the results determined through expert experience. If the results determined by the two are inconsistent, the results determined by expert experience are used as training data to retrain and optimize the initial early warning model, and the trained and optimized early warning model is used as the current disaster early warning model for the working face.

[0028] Furthermore, it also includes step seven: If step six determines that there is a risk of combined rockburst and roof water inrush disasters during the current working face mining period, then the risk type of combined rockburst and roof water inrush disasters is determined based on the changing patterns of the monitoring data subsequently acquired by each monitoring system. If the risk of rockburst disasters increases first, followed by an increase in the risk of roof water inrush disasters, then the risk type of combined rockburst and roof water inrush disasters is rockburst-water inrush type combined disaster; if the risk of roof water inrush disasters increases first, followed by an increase in the risk of rockburst disasters, then the risk type of combined rockburst and roof water inrush disasters is water inrush-rockburst type combined disaster; if the risk of rockburst disasters and the risk of roof water inrush disasters increase simultaneously, then the risk type of combined rockburst and roof water inrush disasters is coupled type combined disaster.

[0029] Compared with existing technologies, this invention first determines various monitoring indicators and early warning values ​​based on the mine rockburst and roof water inrush disaster monitoring system; during the working face mining period, the monitoring indicator values ​​of each monitoring system are collected, and abnormal monitoring data are screened out and removed. Then, the missing data after removal is supplemented to form the dataset required for subsequent model training and verification; based on the rockburst and roof water inrush disaster monitoring system, a four-field multi-source initial early warning model of "stress-vibration-energy-seepage" is established; the processed monitoring indicator values ​​during the working face mining period are analyzed and preliminary early warnings are issued; at the same time, the preliminary early warning results are compared with expert judgments, and the early warning model is optimized and adjusted according to whether it is consistent with the expert judgment results; finally, based on the optimized early warning model, it is determined whether there is a risk of combined rockburst and roof water inrush disasters during the current working face mining period; subsequently, the type of combined rockburst and roof water inrush disasters is determined according to the changing patterns of monitoring data between various systems. This method can dynamically adjust the early warning index values ​​for combined rockburst and roof water inrush disasters at the working face, accurately identify the types of combined rockburst and roof water inrush disasters, and truly realize the integrated monitoring and early warning of combined rockburst and roof water inrush disasters, providing support and guarantee for the precise prevention and control of combined rockburst and roof water inrush disasters. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the overall process of the present invention.

[0031] Figure 2 This is a schematic diagram of the deployment of the monitoring system for rockburst and roof water inrush disasters according to an embodiment of the present invention.

[0032] Figure 3 This is a schematic diagram of micro-vibration waveform calibration according to an embodiment of the present invention.

[0033] Figure 4 This is a schematic diagram of the monitoring and analysis process for abnormal data according to an embodiment of the present invention.

[0034] Figure 5 This is a schematic diagram of the missing data supplementation process according to an embodiment of the present invention.

[0035] Figure 6 This is a schematic diagram of monitoring indicators in an embodiment of the present invention.

[0036] Figure 7 This is a flowchart of the multi-source early warning process for four fields: stress, vibration, energy, and seepage, according to an embodiment of the present invention. Detailed Implementation

[0037] The present invention will be further described below.

[0038] like Figure 1 As shown, the present invention includes the following steps:

[0039] Step 1: Based on the current rockburst monitoring system and roof water inrush disaster monitoring system used in the mine, determine the monitoring indicators and early warning values ​​for each disaster, such as... Figure 6 As shown, the method for determining the monitoring indicators is as follows: The monitoring indicators are determined based on the rockburst monitoring system and roof water inrush monitoring system used in the mine. Rockburst monitoring includes stress field, vibration field, and energy field monitoring; roof water inrush monitoring includes vibration field and seepage field monitoring; the stress field monitoring includes stress monitoring, drill cuttings monitoring, and support resistance monitoring. The monitoring indicators corresponding to stress monitoring are stress value and stress increment; the monitoring indicators corresponding to drill cuttings monitoring are drill cuttings quantity; and the monitoring indicators corresponding to support resistance monitoring are support working resistance; the vibration field monitoring includes microseismic monitoring, and the monitoring indicators corresponding to microseismic monitoring are daily total microseismic energy, daily average microseismic energy, daily maximum microseismic energy, and daily total microseismic frequency; the energy field monitoring includes ground sound monitoring, and the monitoring indicators corresponding to ground sound monitoring are average number of events per shift, average ground sound intensity per shift, average number of events per hour, and average ground sound intensity per hour; the seepage field monitoring includes hydrological information monitoring and drainage information monitoring. The monitoring indicators corresponding to hydrological information monitoring are borehole water level and borehole water temperature; and the monitoring indicators corresponding to drainage information monitoring are daily water inflow and instantaneous water inflow.

[0040] In this embodiment, the monitoring system is arranged as follows: Figure 2 As shown, the microseismic monitoring system includes eight microseismic sensors: three sensors are installed in each of the two roadways of the working face, and two microseismic sensors are installed in the main roadway at the corresponding location of the working face; ground sound monitoring is installed at 30m, 130m, and 230m ahead of the working face in each of the two roadways; stress monitoring is installed within 300m ahead of the working face on both sides, with a set of monitoring points every 20m, each set containing two stress sensors, with shallow holes at 8m and deep holes at 14m; drill cuttings monitoring is installed within 100m ahead of the working face on both sides, with a monitoring point every 20m, monitoring at least three holes each time, with a monitoring interval of 1-3 days, and a drilling depth of 12m; support resistance monitoring points are installed along the hydraulic supports of the working face, with at least one measuring point installed on each hydraulic support; drainage information monitoring includes a pipeline flow meter installed on the horizontal section of the main drainage pipeline in the transport roadway, 50m from the outlet; hydrological information monitoring is installed in the middle of the working face.

[0041] The aforementioned microseismic monitoring uses microseismic sensors to receive seismic wave signals. The location and energy of the microseismic events can be determined through calculation, as shown in the following formula:

[0042] For a specific vibration signal monitored, based on the vibration waveforms acquired by each microseismic sensor, the starting point of the waveform is calibrated. Combined with the positions of each microseismic sensor, the location of the vibration can be determined using the following formula:

[0043]

[0044] In the formula: x, y, z represent the coordinates of the earthquake source; t represents the time of earthquake initiation; x i ,y i ,z i Represents the coordinates of the i-th microseismic sensor; t i The time it takes for the P-wave vibration signal to reach the i-th micro-seismic sensor is represented by v(x,y,z); the propagation speed of the P-wave is represented by w. i The weighting function represents the microseismic sensor observations; n represents the number of sensors labeled with P-waves; p represents a constant, which can be 1 or 2.

[0045] The above equation has four unknowns: x, y, z, and t. To solve this equation, at least four equations are needed to form a system of equations. Therefore, to ensure that the source signal of this earthquake is clearly received by at least four microseismic sensors, it is recommended to have more than six microseismic sensors covering the mining area to ensure high-precision positioning of the source.

[0046] Based on the waveform signal, the energy of this vibration can be calculated using the following formula:

[0047]

[0048] In the formula: E r Represents vibrational energy; ρ represents the density of the medium; v p v s Indicates the wave velocity of P-waves and S-waves; Indicates the radiation mode; t1 indicates that the average value of the radiation pattern is exceeded; t1 and t2 represent the onset and end times of the P-wave and S-wave, respectively.

[0049] In the above formula, all parameters for calculating vibration energy, except for the wave velocities of P-waves and S-waves, are solved by analyzing the vibration waveform signals. The wave velocities of P-waves and S-waves are determined based on field tests. After the microseismic monitoring system is installed on-site, at least three test points are determined within the coverage area of ​​the monitoring system. Drilling and blasting are performed at the test points. Simultaneously, the vibration energy is determined based on the amount of explosive used in the blasting. The calculation parameters are determined based on the vibration waveform and substituted into the above formula to back-calculate the wave velocities of P-waves and S-waves, thus finally determining the wave velocities of P-waves and S-waves.

[0050] In this embodiment, eight microseismic sensors are deployed: three sensors in each of the two roadways of the working face, and two sensors in the corresponding main roadway of the working face. This completely covers the mining area, ensuring that each vibration waveform can be clearly received by all eight sensors. The coordinates of the microseismic sensors are shown in Table 1. After each received waveform, the operator reads the waveform data stored in the data storage station at the intelligent data processing station and calibrates the P-wave and S-wave of the waveform. Figure 3 The waveform results of six of the microseismic sensors are shown.

[0051] Table 1 Coordinates of Microseismic Sensors

[0052] Micro-vibration sensor serial number X coordinate Y coordinate Z-coordinate 1 36614490 3865850 410 2 36614770 3865650 430 3 36614490 3865450 370 4 36614770 3865250 400 5 36614490 3865050 380 6 36614770 3864850 370 7 36614540 3863950 390 8 36614720 3863950 395

[0053] In this embodiment, blasting tests were conducted at locations of 300m, 700m, and 1100m in the working face transport roadway, and 500m, 900m, and 1300m in the return air roadway. Based on the amount of explosive used and other calculated parameters determined by the vibration waveform, the wave velocities of the P-wave and S-wave were 2400m / s and 980m / s, respectively. The microseismic monitoring system received a microseismic waveform as shown below. Figure 3 As shown, the coordinates of the microseismic event, calculated as (36614736, 385524, 420), and the energy of the microseismic event, are 1468 J.

[0054] The process for determining the warning value is as follows: the warning value of the monitoring indicators of each monitoring system is determined based on the warning value of the monitoring indicators of the adjacent working face and historical monitoring data. The warning value of the monitoring indicators of the adjacent working face is used as the initial value, and the initial value is adjusted based on the historical monitoring data of the adjacent working face to determine the warning value of the monitoring indicators of the current working face.

[0055] Step Two: During the longwall mining phase, continuously collect monitoring index values ​​from various monitoring systems, and filter out abnormal data from these values, such as... Figure 4 As shown, the specific process is as follows:

[0056] ① For monitoring data acquired by active monitoring equipment in the monitoring system that performs real-time monitoring, such as stress monitoring, support resistance monitoring, hydrological information monitoring, and drainage information monitoring, when a certain monitoring data value exceeds 200% of the previous monitoring data value, the monitoring data is first marked as suspicious data and the next monitoring data is obtained; if the value of the next monitoring data does not exceed 150% of the value of the previous monitoring data of the suspicious data, the suspicious data is marked as abnormal data; otherwise, the suspicious data is considered as normal monitoring data.

[0057] ② For monitoring data obtained by manual fixed-point monitoring in the monitoring system, such as drill cuttings monitoring, when a certain monitoring data value exceeds 200% of the previous monitoring data value, the monitoring data is marked as suspicious data. At the same time, the monitoring value is measured again near the monitoring point. If the current monitoring value does not exceed 150% of the previous monitoring data value of the suspicious data, the suspicious data is marked as abnormal data; otherwise, the suspicious data is considered as normal monitoring data.

[0058] ③ For monitoring data acquired by passive monitoring devices that perform real-time monitoring in the monitoring system, such as microseismic monitoring and ground sound monitoring, considering the randomness of their data monitoring, no abnormal data analysis is performed.

[0059] ④ Use the above methods to filter out and remove abnormal data.

[0060] In this embodiment, during a stress monitoring test, the previous stress value was 6.5 MPa, and the subsequent stress value was 15 MPa, exceeding 200% of the previous monitoring data value. This was marked as suspicious data. The next stress value decreased to 6.8 MPa, not exceeding 150% of the previous monitoring data value. Therefore, the suspicious data was marked as abnormal data and removed. In a drill cuttings monitoring test, the previous drill cuttings amount was 1.25 kg / m, and the subsequent drill cuttings amount was 2.7 kg / m, exceeding 200% of the previous monitoring data value. This was marked as suspicious data. Subsequently, another drill cuttings monitoring test was conducted near the borehole. This time, the drill cuttings amount was 2.9 kg / m, exceeding 150% of the previous monitoring data value. Therefore, this suspicious data was marked as normal monitoring data.

[0061] Next, the missing data after removal is supplemented, such as... Figure 5 As shown, the specific steps are as follows: Using the first 1000 missing data points as a dataset, the data is divided into training, validation, and test sets. A fully connected feedforward neural network is used, with a suitable network architecture designed and a model optimization algorithm selected. The number of training iterations and batch size are configured. The training set data is input into the model, and the predicted values ​​and validation loss are calculated. The gradient of the loss function for each weight is calculated using the backpropagation algorithm. An optimization algorithm is used to update the weights using these gradients to reduce the loss. The validation set loss is monitored; if the loss does not decrease after 50 consecutive iterations, training is automatically stopped, and the model weights are restored to the state where the validation loss was lowest. After training, the model is evaluated using the test set. If the evaluation is successful, the model is used to predict and supplement the missing data values.

[0062] In this embodiment, the first 1000 missing data points are used as a dataset. 70% of this dataset is used for direct model training, adjusting weight parameters; 15% is used to monitor model performance during training, adjusting hyperparameters (such as network structure) and for early stopping to prevent overfitting; and the remaining 15% is used to evaluate the model's generalization ability. A fully connected feedforward neural network architecture is selected, with three hidden layers and 64, 128, and 64 neurons respectively. The Adam optimization algorithm is chosen, which automatically adjusts the learning rate; the mean squared error function is selected as the loss function; the mean squared error metric is selected as the evaluation metric; the batch size is set to 32; and the number of iterations is 100.

[0063] Before data partitioning, data standardization preprocessing can speed up model computation. The data standardization formula is as follows:

[0064]

[0065] In the formula, Z is the standardized data matrix, X is the unstandardized data matrix, μ is the mean of the dataset, and σ is the standard deviation of the dataset.

[0066] The data is used to calculate the prediction result through a neural network, as shown in the following formula:

[0067] z [l] =W [l] a [l-1] +b [l] (4)

[0068] In the formula z [l] W represents the linear computation result of the l-th layer network. [l] Let a be the weight matrix of the l-th layer network. [l-1] For the output of layer l-1, b [l] is the bias vector of the l-th layer network.

[0069] The difference between the predicted value and the actual value, i.e., the prediction loss, can be calculated using the following formula:

[0070]

[0071] In the formula, m is the number of samples in the iterative batch. Let y be the predicted value of the i-th sample. (i) Let J(W,b) be the true value of the i-th sample, and let J(W,b) be the total loss that depends on all weights and biases.

[0072] After the model is trained, its performance is evaluated on the test set. Commonly used evaluation metrics include mean squared error and coefficient of determination.

[0073] The mean square error can be calculated using equation (6):

[0074]

[0075] The coefficient of determination can be calculated using equation (7):

[0076]

[0077] Step 3: Based on the rockburst monitoring system and the roof water inrush disaster monitoring system, establish a multi-source initial early warning model with four fields: stress, vibration, energy, and seepage. The specific method is as follows: Determine the early warning values ​​for each monitoring indicator of each monitoring system based on historical mine monitoring data. Implement graded early warnings for each monitoring indicator. Align the location coordinates and time of the aforementioned historical multi-source data. Assess the hourly, shift-level, and daily risks of rockburst, roof water inrush, and combined rockburst and roof water inrush disasters. Use subsequent monitoring data and early warning results as a comprehensive feature vector. Use a long short-term memory network model to learn the relationship between the multi-source data features and their early warning results. Finally, output a risk index. Based on this risk index, determine the risk of rockburst, the risk of roof water inrush, and whether there is a combined rockburst and roof water inrush disaster risk. The location coordinates and time alignment of the multi-source data are calculated using the following formula:

[0078] X T (t)=Ψ T (Φ(X(t))) (8)

[0079] In the formula, X T (t) represents the aligned data matrix, Ψ T Let Φ be the time alignment function, Φ be the spatial alignment function, and X(t) be the original multi-source data matrix.

[0080] The monitoring data and early warning results are combined into a single feature vector as follows:

[0081] F(t) = [L T (t),X T (t)](9)

[0082] In the formula, F(t) is the comprehensive eigenvector, and L... T (t) is the vector of monitoring and early warning levels for each indicator at time t.

[0083] like Figure 7 As shown, the computational process for learning the relationship between multi-source data features and their early warning results using a Long Short-Term Memory (LSTM) network model is as follows:

[0084] Input Gate:

[0085] it=σ(WxiF(t)+Whiht-1+b i (10)

[0086] Forgotten Gate:

[0087] ft=σ(WxfF(t)+Whfht-1+b f (11)

[0088] Status Update:

[0089]

[0090] Output gate:

[0091]

[0092] Risk index output:

[0093] RI(t)=σ(W o h t +b o (14)

[0094] Step 4: After analyzing the monitoring index values ​​processed in Step 2 using the initial early warning model established in Step 3, preliminary early warning results are output. Specifically, during the working face mining period, the monitoring index values ​​processed in Step 2 are input into the initial early warning model for analysis. Based on the rockburst risk index, roof water inrush risk index, and the comprehensive risk index of combined rockburst and roof water inrush disasters, the preliminary early warning results output by the model include rockburst risk as no rockburst risk, weak rockburst risk, moderate rockburst risk, and strong rockburst risk; and roof water inrush risk as no roof water inrush. Water risk, weak roof water inrush risk, medium roof water inrush risk, and strong roof water inrush risk; the combined risk of rockburst and roof water inrush is divided into combined risk of rockburst and roof water inrush and combined risk of no rockburst and roof water inrush, specifically: when the rockburst risk index RI(t)≤0.25, it is no rockburst risk; when 0.25<RI(t)≤0.50, it is weak rockburst risk; when 0.50<RI(t)≤0.75, it is medium rockburst risk; when 0.75<RI(t), it is strong rockburst risk.

[0095] When the roof water inrush risk index RI(t) ≤ 0.25, there is no roof water inrush risk; when 0.25 < RI(t) ≤ 0.50, there is a weak roof water inrush risk; when 0.50 < RI(t) ≤ 0.75, there is a moderate roof water inrush risk; and when 0.75 < RI(t), there is a strong roof water inrush risk.

[0096] When the risk index RI(t) of the combined disaster of rockburst and roof water inrush is ≤0.25, there is no risk of combined disaster of rockburst and roof water inrush; when 0.25 < RI(t), there is a risk of combined disaster of rockburst and roof water inrush.

[0097] In this embodiment, after the monitoring data is input into the four-field multi-source early warning model, the output rockburst risk index after learning by the long short-term memory network model is 0.57, which means moderate rockburst risk; the roof water inrush risk index is 0.42, which means weak roof water inrush risk; and the combined rockburst and roof water inrush disaster risk index is 0.67, which means there is a combined rockburst and roof water inrush disaster risk.

[0098] Step 5: Verify the preliminary early warning results output in Step 4 by combining them with expert experience, and optimize and adjust the initial early warning model based on the verification results to form an optimized early warning model. Specifically, compare the preliminary early warning results in Step 4 with the results determined by expert experience. If the two results are inconsistent, use the results determined by expert experience as training data to retrain and optimize the initial early warning model, and use the trained and optimized early warning model as the current disaster early warning model for the working face.

[0099] In this embodiment, the rockburst risk index predicted by the four-field multi-source early warning model is 0.57, indicating a moderate rockburst risk; the roof water inrush risk index is 0.42, indicating a weak roof water inrush risk; and the combined rockburst and roof water inrush disaster risk index is 0.67, indicating a combined rockburst and roof water inrush disaster risk. Then, expert experience is used to determine the actual disaster risk of the current working face. The working face has a moderate rockburst risk, a weak roof water inrush risk, and a combined rockburst and roof water inrush disaster risk. The two determinations are consistent, therefore, retraining the four-field multi-source early warning model is not required.

[0100] Step Six: Determine whether there is a risk of combined rockburst and roof water inrush during the current working face mining period based on the optimized early warning model; In this embodiment, the model's judgment of the combined disaster risk is consistent with the expert's judgment of the combined disaster risk, therefore it is determined that there is a risk of combined rockburst and roof water inrush during the current working face mining period.

[0101] Step 7: If Step 6 determines that there is a risk of combined rockburst and roof water inrush during the current working face mining period, then determine the risk type of combined rockburst and roof water inrush based on the changing patterns of the monitoring data subsequently acquired by each monitoring system. If the risk of rockburst increases first, followed by an increase in the risk of roof water inrush, then the risk type of combined rockburst and roof water inrush is rockburst-water inrush type combined disaster; if the risk of roof water inrush increases first, followed by an increase in the risk of rockburst, then the risk type of combined rockburst and roof water inrush is water inrush-rockburst type combined disaster; if the risks of both rockburst and roof water inrush increase simultaneously, then the risk type of combined rockburst and roof water inrush is coupled type combined disaster.

[0102] In this embodiment, step six determines that there is a combined risk of rockburst and roof water inrush during the current working face mining period. At the same time, according to the monitoring data of each system, the risk of rockburst at the working face first increases from no rockburst risk to moderate rockburst risk, and then the risk of roof water inrush increases from no water inrush risk to weak water inrush risk. Therefore, the change in the risk of rockburst leads to the change in the risk of roof water inrush. Thus, the combined risk type of rockburst and roof water inrush during the current working face mining period is rockburst-water inrush type combined disaster.

[0103] Through the above process, the early warning index values ​​can be dynamically adjusted for combined rockburst and roof water inrush disasters at the working face, accurately identifying the types of combined rockburst and roof water inrush disasters, and truly realizing integrated monitoring and early warning of combined rockburst and roof water inrush disasters, providing support and guarantee for the precise prevention and control of combined rockburst and roof water inrush disasters.

[0104] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A four-field integrated monitoring and early warning method for combined rockburst and roof water inrush disasters, characterized in that, Includes the following steps: Step 1: Based on the rockburst monitoring system and roof water inrush disaster monitoring system currently used in the mine, determine the monitoring indicators and early warning values ​​for each disaster; Step 2: During the longwall mining period, continuously collect the monitoring index values ​​of each monitoring system, screen out abnormal data from each monitoring index value, and supplement the missing data after the removal. Step 3: Based on the rockburst monitoring system and the roof water inrush disaster monitoring system, establish a multi-source initial early warning model with four fields: stress, vibration, energy, and seepage. Step 4: After analyzing the monitoring indicator values ​​processed in Step 2 using the initial early warning model established in Step 3, output the preliminary early warning results; Step 5: Verify the preliminary early warning results output in Step 4 by combining them with expert experience, and optimize and adjust the initial early warning model based on the verification results to form an optimized early warning model; Step 6: Determine whether there is a risk of combined rockburst and roof water inrush during the current working face mining period based on the optimized early warning model.

2. The four-field integrated monitoring and early warning method for combined rockburst and roof water inrush disasters according to claim 1, characterized in that, The method for determining the monitoring indicators in step one is as follows: the monitoring indicators are determined based on the rockburst monitoring system and the roof water inrush monitoring system used in the mine. Rockburst monitoring includes monitoring of stress field, vibration field and energy field; roof water inrush monitoring includes monitoring of vibration field and seepage field. The stress field monitoring includes stress monitoring, drill cuttings monitoring, and support resistance monitoring. The monitoring indicators corresponding to stress monitoring are stress value and stress increment, the monitoring indicators corresponding to drill cuttings monitoring are drill cuttings quantity, and the monitoring indicators corresponding to support resistance monitoring are support working resistance. The vibration field monitoring includes microseismic monitoring, and the monitoring indicators corresponding to microseismic monitoring are daily total microseismic energy, daily average microseismic energy, daily maximum microseismic energy, and daily total microseismic frequency. The energy field monitoring includes ground sound monitoring, and the monitoring indicators corresponding to ground sound monitoring are average number of events per class, average ground sound intensity per class, average number of events per hour, and average ground sound intensity per hour. The seepage field monitoring includes hydrological information monitoring and drainage information monitoring. The monitoring indicators corresponding to hydrological information monitoring are borehole water level and borehole water temperature, and the monitoring indicators corresponding to drainage information monitoring are daily inflow and instantaneous inflow.

3. The four-field integrated monitoring and early warning method for combined rockburst and roof water inrush disasters according to claim 1, characterized in that, The process of determining the warning value in step one is as follows: the warning value of the monitoring indicators of each monitoring system is determined based on the warning value of the monitoring indicators of the adjacent working face and historical monitoring data. The warning value of the monitoring indicators of the adjacent working face is used as the initial value, and the initial value is adjusted based on the historical monitoring data of the adjacent working face to determine the warning value of the monitoring indicators of the current working face.

4. The four-field integrated monitoring and early warning method for combined rockburst and roof water inrush disasters according to claim 1, characterized in that, The process of filtering out abnormal data in step two is as follows: ① For monitoring data acquired by active monitoring devices in the monitoring system, when a certain monitoring data value exceeds 200% of the previous monitoring data value, the monitoring data is first marked as suspicious data and the next monitoring data is acquired; if the value of the next monitoring data does not exceed 150% of the value of the previous monitoring data of the suspicious data, the suspicious data is marked as abnormal data; otherwise, the suspicious data is considered as normal monitoring data. ② For monitoring data obtained by manual fixed-point monitoring in the monitoring system, when a certain monitoring data value exceeds 200% of the previous monitoring data value, the monitoring data is marked as suspicious data. At the same time, the monitoring value is measured again near the monitoring point. If the current monitoring value does not exceed 150% of the previous monitoring data value of the suspicious data, the suspicious data is marked as abnormal data; otherwise, the suspicious data is considered as normal monitoring data. ③ For monitoring data acquired by passive monitoring devices that perform real-time monitoring in the monitoring system, considering the randomness of their data monitoring, no abnormal data analysis is performed. ④ Use the above methods to filter out and remove abnormal data.

5. The four-field integrated monitoring and early warning method for combined rockburst and roof water inrush disasters according to claim 1, characterized in that, Step two involves supplementing the missing data after removal. Specifically, the first 1000 missing data entries are used as a dataset, which is then divided into a training set, a validation set, and a test set. A fully connected feedforward neural network is used, with a suitable network architecture designed and a model optimization algorithm selected. The number of training iterations and batch size are configured. The training set data is input into the model, and the predicted values ​​and validation loss are calculated. The gradient of the loss function for each weight is calculated using the backpropagation algorithm. An optimization algorithm is used to update the weights using these gradients. The loss of the validation set is monitored. If the loss does not decrease for 50 consecutive iterations, training is automatically stopped, and the model weights are restored to the state where the validation loss is lowest. After training, the model is evaluated using a test set. If the evaluation is successful, the model is used to predict and supplement any data values ​​that need to be added.

6. The four-field integrated monitoring and early warning method for combined rockburst and roof water inrush disasters according to claim 1, characterized in that, The model establishment method in step three is as follows: Based on the historical monitoring of the mine, the early warning values ​​of each monitoring indicator of each monitoring system are determined. For each monitoring indicator, a graded early warning is issued. The location coordinates and time of the above-mentioned historical multi-source data are aligned. The risks of rockburst, roof water inrush, and combined rockburst and roof water inrush disasters are judged at the hour, shift, and day levels. The subsequent monitoring data and early warning results are used as a comprehensive feature vector. The relationship between the multi-source data features and the early warning results is learned using a long short-term memory network model. Finally, a risk index is output. Based on the risk index, the risks of rockburst, roof water inrush, and whether there is a combined rockburst and roof water inrush disaster risk are determined.

7. The four-field integrated monitoring and early warning method for combined rockburst and roof water inrush disasters according to claim 1, characterized in that, In step four, preliminary early warning is issued based on the analysis results. Specifically, during the working face mining period, the monitoring index values ​​processed in step two are input into the initial early warning model for analysis. Based on the rockburst risk index, roof water inrush risk index, and the comprehensive index of combined rockburst and roof water inrush disaster risk, the preliminary early warning results output by the model include: rockburst risk is categorized as no rockburst risk, weak rockburst risk, moderate rockburst risk, and strong rockburst risk; roof water inrush risk is categorized as no roof water inrush risk, weak roof water inrush risk, moderate roof water inrush risk, and strong roof water inrush risk; and combined rockburst and roof water inrush disaster risk is categorized as having combined rockburst and roof water inrush disaster risk and not having combined rockburst and roof water inrush disaster risk.

8. The four-field integrated monitoring and early warning method for combined rockburst and roof water inrush disasters according to claim 1, characterized in that, Specifically, step five involves comparing the preliminary early warning results from step four with the results determined through expert experience. If the results are inconsistent, the results determined by expert experience are used as training data to retrain and optimize the initial early warning model, and the optimized early warning model is used as the current disaster early warning model for the working face.

9. The four-field integrated monitoring and early warning method for combined rockburst and roof water inrush disasters according to claim 1, characterized in that, The process also includes step seven: If step six determines that there is a risk of combined rockburst and roof water inrush during the current working face mining period, then the risk type of combined rockburst and roof water inrush is determined based on the changing patterns of the monitoring data subsequently acquired by each monitoring system. If the risk of rockburst increases first, followed by an increase in the risk of roof water inrush, then the risk type of combined rockburst and roof water inrush is rockburst-water inrush type combined disaster; if the risk of roof water inrush increases first, followed by an increase in the risk of rockburst, then the risk type of combined rockburst and roof water inrush is water inrush-rockburst type combined disaster; if the risks of both rockburst and roof water inrush increase simultaneously, then the risk type of combined rockburst and roof water inrush is coupled type combined disaster.

Images