A multivariable intelligent prediction method for mine rock burst
Through the TFT neural network model and differential evolution algorithm, a multivariate prediction model is constructed, which solves the problem of single data feature dependence in the existing technology, realizes multivariate prediction and early warning of mine impact ground pressure, and improves the prediction accuracy.
Patent Information
- Application Number
- CN202410667273.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-28
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-05-28
AI Technical Summary
The existing deep learning prediction methods for impact ground pressure are overly dependent on single data characteristics, resulting in poor prediction effects and accuracy, and lack of reasonable multivariable intelligent prediction methods for mine impact ground pressure.
Using the TFT neural network model combined with the differential evolution algorithm, a multivariate prediction model is constructed by obtaining microseismic monitoring data, including b-value, energy index, cumulative apparent volume and Schmidt number, multivariate prediction and forecasting are performed to improve impact ground pressure prediction Accuracy.
Multivariate prediction and early warning of impact ground pressure are realized, prediction accuracy is improved, the shortcomings of single parameter prediction are solved, and more scientific prediction methods are provided.
Smart Images

Figure CN118584542B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of rock burst prediction, and in particular relates to a multivariate intelligent prediction method for rock burst in mines. Background Art
[0002] The accuracy of rock burst prediction is of great significance to ensure the safe mining of coal mines and the personal and property safety of underground personnel. At present, the rock burst prediction method is still mainly based on traditional monitoring. With the integration of artificial intelligence methods and traditional mining problems, the continuous improvement of the intelligent rock mechanics theory system has provided a scientific and effective prediction method for rock burst. Introducing deep learning algorithms into rock burst prediction and applying perfect prediction algorithms for rock burst prediction has gradually become the current research method. However, the existing rock burst deep learning prediction methods mostly use a single parameter seismological b value, which is overly dependent on a single data feature, and has poor prediction effect and accuracy.
[0003] Therefore, there is a lack of a well-designed multivariate intelligent prediction method for mine rock burst pressure. The TFT neural network model is used to realize multivariate prediction and forecast of b value, energy index, cumulative apparent volume and Schmidt number, which improves the accuracy of rock burst pressure prediction. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a multivariate intelligent prediction method for mine rock burst pressure in view of the deficiencies in the above-mentioned prior art. The method has simple steps and reasonable design. The multivariate prediction and forecast of b value, energy index, cumulative apparent volume and Schmidt number are realized through the TFT neural network model, thereby improving the accuracy of rock burst pressure prediction.
[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a multivariate intelligent prediction method for rock burst in mines, characterized in that the method comprises the following steps:
[0006] Step 1: Obtain microseismic monitoring data:
[0007] Step 101, using a microseismic monitoring system to monitor the mine area to be monitored, and obtaining microseismic events per unit time and the microseismic magnitude and energy corresponding to each microseismic event; wherein the mine area to be monitored is a mine tunnel or goaf;
[0008] Step 102, using a computer to obtain a b value, an energy index, a cumulative apparent volume and a Schmidt number according to the microseismic events in a unit time and the microseismic magnitude and energy corresponding to each microseismic event;
[0009] Step 2: Use a computer to build a TFT neural network model;
[0010] Step 3: Get the setting parameters of the TFT neural network model:
[0011] Step 301, set the number of individuals, the maximum number of evolutionary iterations, the mutation factor, the crossover factor and the dimension of the individual; wherein the dimension of the individual is 6, and the 6 dimensions are the number of layers, the number of heads in the multi-head attention, the dimension of each head, the learning rate, the batch size, and the number of training rounds;
[0012] Step 302: Initialize individuals randomly, use a computer to use a differential evolution algorithm to iteratively update each individual until the maximum number of evolutionary iterations is met, and then obtain the optimal individual;
[0013] Step 303, using a computer to record the number of layers in the optimal individual, the number of heads in the multi-head attention, the dimension of each head, the learning rate, the batch size, and the number of training rounds as the optimal number of layers, the optimal number of heads, the optimal dimension of each head, the optimal learning rate, the optimal batch size, and the optimal number of training rounds, respectively;
[0014] Step 4: Training of TFT neural network model:
[0015] Step 401: using a computer to set the constructed TFT neural network model according to the optimal number of layers, the optimal number of heads, the optimal dimension of each head, and the optimal learning rate;
[0016] Step 402: according to the optimal batch size and the optimal number of training rounds, the b value, energy index, cumulative apparent volume and Schmidt number in step 102 are input into the TFT neural network model for training to obtain a trained TFT neural network model;
[0017] Step 5: predict the b value, energy index, cumulative apparent volume and Schmidt number at subsequent times according to the trained TFT neural network model;
[0018] Step 6: Predict and warn the rock burst based on the b-value, energy index, cumulative apparent volume and Schmidt number at subsequent times.
[0019] The above-mentioned multivariable intelligent prediction method for rock burst in a mine is characterized in that: step 102, the specific process is as follows:
[0020] Step 1021: Use a computer to classify the microseismic magnitudes corresponding to each microseismic event in a unit time into grades, and obtain the number of microseismic events N corresponding to the i-th grade. i The microseismic magnitude M corresponding to the i-th level i ; Wherein, i and m are both positive integers, and 1≤i≤m, and m represents the total number of grades;
[0021] Step 1022: Using a computer according to Get the b value;
[0022] Step 1023: Use a computer to sum the energy corresponding to each microseismic event in unit time to obtain E, and average the energy corresponding to each microseismic event in unit time under the same seismic moment to obtain And according to Get the energy index EI;
[0023] Step 1024: Use a computer to calculate the The apparent volume V is obtained; where M represents the seismic distance of the microseismic event, μ represents the shear modulus of the rock in the mine area to be monitored, and E1 represents the energy of the microseismic event; and the apparent volume of each microseismic event per unit time is summed to obtain the cumulative apparent volume V a ;
[0024] Step 1025: sum the energy corresponding to each microseismic event per unit time to obtain E, and then calculate the energy corresponding to each microseismic event per unit time according to The Schmidt number Sc between microseismic events is obtained, and the Schmidt number Sc between microseismic events per unit time is averaged to obtain the Schmidt number Sc a ; where ΔV represents the apparent volume change between two adjacent microseismic events, and Δt represents the time change between two adjacent microseismic events. represents the average time difference between two adjacent microseismic events, ρ represents the rock mass density of the mine area to be monitored, represents the average distance between two adjacent microseismic events, M s Represents the sum of seismic distances of microseismic events per unit time.
[0025] The above-mentioned multivariable intelligent prediction method for rock burst in a mine is characterized in that: step 402, the specific process is as follows:
[0026] Step 4021: record the b value, energy index, cumulative apparent volume and Schmidt number in step 102 from the current time t to the starting time as the time series data X at time t. t , ..., time series data X at time t-kΤ t-kΤ ; Among them, the time series data X at time t t is the b value, energy index, cumulative apparent volume and Schmidt number at time t and t; the time series data X at time t-kΤ t-kΤ are the b value, energy index, cumulative apparent volume and Schmidt number at time t-kΤ and t-kΤ; k represents the number of time series data input in the past, and k is a positive integer, and Τ represents the unit time;
[0027] Step 4022: Select the time series data X corresponding to the time t+T to the time t+τ×T t+T To time series data X t+τT ; Wherein, τ represents the prediction step length, and τ is a positive integer;
[0028] Step 4023, obtain the coordinates of the four vertices of the minimum circumscribed rectangular area of the mine area to be monitored from the surface, and use the coordinates of the four vertices as static metadata of the TFT neural network model; use a computer to convert the time series data X at time t into t , ..., time series data X at time t-kΤ t-kΤ As the past input of the TFT neural network model, the time t+T to t+τ×T is used as the known future input of the TFT neural network model; the time series data X t+T To time series data X t+τT As the true output value of the TFT neural network model, multiple sets of training data are obtained;
[0029] Step 4024: Use a computer to input any set of training data into the TFT neural network model for training until the training data is trained and one round of training is completed;
[0030] Step 4025, repeat step 4024 multiple times until the optimal number of training rounds is reached, and a trained TFT neural network model is obtained.
[0031] The above-mentioned multivariable intelligent prediction method for rock burst in a mine is characterized in that: Step 5, the specific process is as follows:
[0032] Step 501: predict the b value, energy index, cumulative apparent volume and Schmidt number at the subsequent time according to the trained TFT neural network model, obtain the energy index at the subsequent initial time from the energy index at the subsequent time and record it as EI0, and obtain the energy index EI from the energy index at the subsequent time s , and the energy index EI s Satisfy lg[EI s -EI0]<0.3, the corresponding subsequent moment is the starting moment of early warning judgment;
[0033] Step 502: Use a computer to obtain the b value, cumulative apparent volume and Schmidt number after the start time of the early warning judgment, and record them as the b value to be judged, the cumulative apparent volume to be judged and the Schmidt number to be judged; and calculate the b value b at the later time in two consecutive unit times. h and the value b at the previous moment q , the cumulative visual volume V to be judged at the next moment ah and the cumulative visual volume V to be determined at the previous moment aq , the Schmidt number Sc to be determined at the next moment ah and the Schmidt number Sc to be determined at the previous moment aq Make judgments;
[0034] Step 503: Use a computer to determine if b h -bq <-0.7, V ah -V aq >2 and Sc ah -Sc aq <-2, rock burst is predicted and an early warning is issued.
[0035] Compared with the prior art, the present invention has the following advantages:
[0036] 1. The method of the present invention has simple steps and reasonable design, and solves the problem of intelligent prediction of rock burst based on a single variable.
[0037] 2. The present invention constructs a TFT neural network model, and uses a differential evolution algorithm to obtain the optimal number of heads of the model, the optimal dimension of each head, and the optimal learning rate for setting, so as to further improve the performance of the prediction model.
[0038] 3. The present invention trains the TFT neural network model by inputting b-value, energy index, cumulative apparent volume and Schmidt number into the model, obtains the trained TFT neural network model and predicts the b-value, energy index, cumulative apparent volume and Schmidt number at subsequent moments, and finally performs intelligent prediction of rock burst pressure by comprehensively considering the b-value, energy index, cumulative apparent volume and Schmidt number at subsequent moments, and provides information support for subsequent rock burst pressure prediction and early warning through multivariate prediction.
[0039] In summary, the method of the present invention has simple steps and reasonable design. It realizes multivariate prediction and forecasting of b value, energy index, cumulative apparent volume and Schmidt number through the TFT neural network model, thereby improving the accuracy of rock burst prediction.
[0040] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 The figure is a flowchart of the method of the present invention. DETAILED DESCRIPTION
[0042] like Figure 1 As shown, the multivariate intelligent prediction method for mine rock burst of the present invention comprises the following steps:
[0043] Step 1: Obtain microseismic monitoring data:
[0044] Step 101, using a microseismic monitoring system to monitor the mine area to be monitored, and obtaining microseismic events per unit time and the microseismic magnitude and energy corresponding to each microseismic event; wherein the mine area to be monitored is a mine tunnel or goaf;
[0045] Step 102, using a computer to obtain a b value, an energy index, a cumulative apparent volume and a Schmidt number according to the microseismic events in a unit time and the microseismic magnitude and energy corresponding to each microseismic event;
[0046] Step 2: Use a computer to build a TFT neural network model;
[0047] Step 3: Get the setting parameters of the TFT neural network model:
[0048] Step 301, set the number of individuals, the maximum number of evolutionary iterations, the mutation factor, the crossover factor and the dimension of the individual; wherein the dimension of the individual is 6, and the 6 dimensions are the number of layers, the number of heads in the multi-head attention, the dimension of each head, the learning rate, the batch size, and the number of training rounds;
[0049] Step 302: Initialize individuals randomly, use a computer to use a differential evolution algorithm to iteratively update each individual until the maximum number of evolutionary iterations is met, and then obtain the optimal individual;
[0050] Step 303, using a computer to record the number of layers in the optimal individual, the number of heads in the multi-head attention, the dimension of each head, the learning rate, the batch size, and the number of training rounds as the optimal number of layers, the optimal number of heads, the optimal dimension of each head, the optimal learning rate, the optimal batch size, and the optimal number of training rounds, respectively;
[0051] Step 4: Training of TFT neural network model:
[0052] Step 401: using a computer to set the constructed TFT neural network model according to the optimal number of layers, the optimal number of heads, the optimal dimension of each head, and the optimal learning rate;
[0053] Step 402: according to the optimal batch size and the optimal number of training rounds, the b value, energy index, cumulative apparent volume and Schmidt number in step 102 are input into the TFT neural network model for training to obtain a trained TFT neural network model;
[0054] Step 5: predict the b value, energy index, cumulative apparent volume and Schmidt number at subsequent times according to the trained TFT neural network model;
[0055] Step 6: Predict and warn the rock burst based on the b-value, energy index, cumulative apparent volume and Schmidt number at subsequent times.
[0056] In this embodiment, step 102, the specific process is as follows:
[0057] Step 1021: Use a computer to classify the microseismic magnitudes corresponding to each microseismic event in a unit time into grades, and obtain the number of microseismic events N corresponding to the i-th grade. i The microseismic magnitude M corresponding to the i-th leveli ; Wherein, i and m are both positive integers, and 1≤i≤m, and m represents the total number of grades;
[0058] Step 1022: Using a computer according to Get the b value;
[0059] Step 1023: Use a computer to sum the energy corresponding to each microseismic event in unit time to obtain E, and average the energy corresponding to each microseismic event in unit time under the same seismic moment to obtain And according to Get the energy index EI;
[0060] Step 1024: Use a computer to calculate the The apparent volume V is obtained; where M represents the seismic distance of the microseismic event, μ represents the shear modulus of the rock in the mine area to be monitored, and E1 represents the energy of the microseismic event; and the apparent volume of each microseismic event per unit time is summed to obtain the cumulative apparent volume V a ;
[0061] Step 1025: sum the energy corresponding to each microseismic event per unit time to obtain E, and then calculate the energy corresponding to each microseismic event per unit time according to The Schmidt number Sc between microseismic events is obtained, and the Schmidt number Sc between microseismic events per unit time is averaged to obtain the Schmidt number Sc a ; where ΔV represents the apparent volume change between two adjacent microseismic events, and Δt represents the time change between two adjacent microseismic events. represents the average time difference between two adjacent microseismic events, ρ represents the rock mass density of the mine area to be monitored, represents the average distance between two adjacent microseismic events, M s Represents the sum of seismic distances of microseismic events per unit time.
[0062] In this embodiment, step 402, the specific process is as follows:
[0063] Step 4021: record the b value, energy index, cumulative apparent volume and Schmidt number in step 102 from the current time t to the starting time as the time series data X at time t. t , ..., time series data X at time t-kΤ t-kΤ ; Among them, the time series data X at time t t is the b value, energy index, cumulative apparent volume and Schmidt number at time t and t; the time series data X at time t-kΤ t-kΤ are the b value, energy index, cumulative apparent volume and Schmidt number at time t-kΤ and t-kΤ; k represents the number of time series data input in the past, and k is a positive integer, and Τ represents the unit time;
[0064] Step 4022: Select the time series data X corresponding to the time t+T to the time t+τ×T t+T To time series data X t+τT ; Wherein, τ represents the prediction step length, and τ is a positive integer;
[0065] Step 4023, obtain the coordinates of the four vertices of the minimum circumscribed rectangular area of the mine area to be monitored from the surface, and use the coordinates of the four vertices as static metadata of the TFT neural network model; use a computer to convert the time series data X at time t into t , ..., time series data X at time t-kΤ t-kΤ As the past input of the TFT neural network model, the time t+T to t+τ×T is used as the known future input of the TFT neural network model; the time series data X t+T To time series data X t+τT As the true output value of the TFT neural network model, multiple sets of training data are obtained;
[0066] Step 4024: Use a computer to input any set of training data into the TFT neural network model for training until the training data is trained and one round of training is completed;
[0067] Step 4025, repeat step 4024 multiple times until the optimal number of training rounds is reached, and a trained TFT neural network model is obtained.
[0068] In this embodiment, step five, the specific process is as follows:
[0069] Step 501: predict the b value, energy index, cumulative apparent volume and Schmidt number at the subsequent time according to the trained TFT neural network model, obtain the energy index at the subsequent initial time from the energy index at the subsequent time and record it as EI0, and obtain the energy index EI from the energy index at the subsequent time s , and the energy index EI s Satisfy lg[EI s -EI0]<0.3, the corresponding subsequent moment is the starting moment of early warning judgment;
[0070] Step 502: Use a computer to obtain the b value, cumulative apparent volume and Schmidt number after the start time of the early warning judgment, and record them as the b value to be judged, the cumulative apparent volume to be judged and the Schmidt number to be judged; and calculate the b value b at the later time in two consecutive unit times. h and the value b at the previous moment q , the cumulative visual volume V to be judged at the next moment ah and the cumulative visual volume V to be determined at the previous moment aq , the Schmidt number Sc to be determined at the next moment ah and the Schmidt number Sc to be determined at the previous momentaq Make judgments;
[0071] Step 503: Use a computer to determine if b h -b q <-0.7, V ah -V aq >2 and Sc ah -Sc aq <-2, rock burst is predicted and an early warning is issued.
[0072] In this embodiment, the value of the total number of levels m is as follows:
[0073] Step A: Obtain the minimum microseismic magnitude L from the microseismic magnitude corresponding to each microseismic event min and the maximum magnitude of micro-earthquakes L max ;
[0074] Step B: According to The total number of levels m is obtained; where l represents the magnitude interval, Indicates a round-up operation.
[0075] In this embodiment, in actual use, the value of the magnitude interval l is 0.2.
[0076] In this embodiment, in actual use, the micro-seismic magnitude corresponding to the first level is [L min , L min +0.2), the second level corresponds to the micro-earthquake magnitude [L min +0.2, L min +0.4), ..., the microseismic magnitude corresponding to the mth level is [L min +0.2(m-1),L min +0.2m).
[0077] In this embodiment, in actual use, the microseismic magnitude M corresponding to the i-th gear level is i The values of are as follows:
[0078] The first level corresponds to a microseismic magnitude M1 of L min , the microseismic magnitude M2 corresponding to the second level is L min +0.2, the third level corresponds to the micro-earthquake magnitude M3 of L max +0.4, the fourth level corresponds to the micro-earthquake magnitude M4 of L min +0.6, ..., the microseismic magnitude M corresponding to the mth level m For L min +0.2(m-1).
[0079] In this embodiment, TFT in the TFT neural network model is Temporal Fusion Transformer, and the TFT neural network model is a time series prediction neural network model.
[0080] In this embodiment, in actual use, the value of k is 20-50, the value range of unit time T is 1 min-24 h, and the value range of τ is 5-15.
[0081] In this embodiment, the maximum number of evolution iterations is 100-500.
[0082] In this embodiment, it is further preferred that the number of individuals in step 301 is 50, the maximum number of iterations is 100, the mutation factor is 0.9, and the crossover factor is 0.6.
[0083] In summary, the method of the present invention has simple steps and reasonable design. It realizes multivariate prediction and forecasting of b value, energy index, cumulative apparent volume and Schmidt number through the TFT neural network model, thereby improving the accuracy of rock burst prediction.
[0084] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent structural change made to the above embodiment based on the technical essence of the present invention still falls within the protection scope of the technical solution of the present invention.
Claims
1. A multivariate intelligent prediction method for mine rock burst, characterized in that: The method comprises the following steps: Step 1: Obtain microseismic monitoring data: Step 101, using a microseismic monitoring system to monitor the mine area to be monitored, and obtaining microseismic events per unit time and the microseismic magnitude and energy corresponding to each microseismic event; wherein the mine area to be monitored is a mine tunnel or goaf; Step 102, using a computer to obtain a b value, an energy index, a cumulative apparent volume and a Schmidt number according to the microseismic events in a unit time and the microseismic magnitude and energy corresponding to each microseismic event; Step 2: Use a computer to build a TFT neural network model; Step 3: Get the setting parameters of the TFT neural network model: Step 301, set the number of individuals, the maximum number of evolutionary iterations, the mutation factor, the crossover factor and the dimension of the individual; wherein the dimension of the individual is 6, and the 6 dimensions are the number of layers, the number of heads in the multi-head attention, the dimension of each head, the learning rate, the batch size, and the number of training rounds; Step 302: Initialize individuals randomly, use a computer to use a differential evolution algorithm to iteratively update each individual until the maximum number of evolutionary iterations is met, and then obtain the optimal individual; Step 303, using a computer to record the number of layers in the optimal individual, the number of heads in the multi-head attention, the dimension of each head, the learning rate, the batch size, and the number of training rounds as the optimal number of layers, the optimal number of heads, the optimal dimension of each head, the optimal learning rate, the optimal batch size, and the optimal number of training rounds, respectively; Step 4: Training of TFT neural network model: Step 401: using a computer to set the constructed TFT neural network model according to the optimal number of layers, the optimal number of heads, the optimal dimension of each head, and the optimal learning rate; Step 402: according to the optimal batch size and the optimal number of training rounds, the b value, energy index, cumulative apparent volume and Schmidt number in step 102 are input into the TFT neural network model for training to obtain a trained TFT neural network model; Step 5: predict the b value, energy index, cumulative apparent volume and Schmidt number at subsequent times according to the trained TFT neural network model; Step 6: Predict and warn the rock burst based on the b-value, energy index, cumulative apparent volume and Schmidt number at subsequent times.
2. A multivariate intelligent prediction method for mine rock burst according to claim 1, characterized in that: Step 102, the specific process is as follows: Step 1021: Use a computer to classify the microseismic magnitudes corresponding to each microseismic event in a unit time into grades, and obtain the number of microseismic events N corresponding to the i-th grade. i The microseismic magnitude M corresponding to the i-th level i ; Wherein, i and m are both positive integers, and 1≤i≤m, and m represents the total number of grades; Step 1022: Using a computer according to Get the b value; Step 1023: Use a computer to sum the energy corresponding to each microseismic event in unit time to obtain E, and average the energy corresponding to each microseismic event in unit time under the same seismic moment to obtain And according to Get the energy index EI; Step 1024: Use a computer to calculate the The apparent volume V is obtained; where M represents the seismic distance of the microseismic event, μ represents the shear modulus of the rock in the mine area to be monitored, and E1 represents the energy of the microseismic event; and the apparent volume of each microseismic event per unit time is summed to obtain the cumulative apparent volume V a ; Step 1025: sum the energy corresponding to each microseismic event per unit time to obtain E, and then calculate the energy corresponding to each microseismic event per unit time according to The Schmidt number Sc between microseismic events is obtained, and the Schmidt number Sc between microseismic events per unit time is averaged to obtain the Schmidt number Sc a ; where ΔV represents the apparent volume change between two adjacent microseismic events, and Δt represents the time change between two adjacent microseismic events. represents the average time difference between two adjacent microseismic events, ρ represents the rock mass density of the mine area to be monitored, represents the average distance between two adjacent microseismic events, M s Represents the sum of seismic distances of microseismic events per unit time.
3. A multivariate intelligent prediction method for mine rock burst according to claim 1, characterized in that: Step 402, the specific process is as follows: Step 4021: record the b value, energy index, cumulative apparent volume and Schmidt number in step 102 from the current time t to the starting time as the time series data X at time t. t , ..., time series data X at time t-kΤ t-kΤ ; Among them, the time series data X at time t t is the b value, energy index, cumulative apparent volume and Schmidt number at time t and t; the time series data X at time t-kΤ t-kΤ are the b value, energy index, cumulative apparent volume and Schmidt number at time t-kΤ and t-kΤ; k represents the number of time series data input in the past, and k is a positive integer, and Τ represents the unit time; Step 4022: Select the time series data X corresponding to the time t+T to the time t+τ×T t+T To time series data X t+τT ; Wherein, τ represents the prediction step length, and τ is a positive integer; Step 4023, obtain the coordinates of the four vertices of the minimum circumscribed rectangular area of the mine area to be monitored from the surface, and use the coordinates of the four vertices as the static metadata of the TFT neural network model; use a computer to convert the time series data X at time t into t , ..., time series data X at time t-kΤ t-kΤ As the past input of the TFT neural network model, the time t+T to t+τ×T is used as the known future input of the TFT neural network model; the time series data X t+T To time series data X t+τT As the true output value of the TFT neural network model, multiple sets of training data are obtained; Step 4024: using a computer to input the training data into the TFT neural network model according to the optimal batch size for training until the training data is trained and one round of training is completed; Step 4025, repeat step 4024 multiple times until the optimal number of training rounds is reached, and a trained TFT neural network model is obtained.
4. A multivariate intelligent prediction method for mine rock burst according to claim 3, characterized in that: Step 5: The specific process is as follows: Step 501: predict the b value, energy index, cumulative apparent volume and Schmidt number at the subsequent time according to the trained TFT neural network model, obtain the energy index at the subsequent initial time from the energy index at the subsequent time and record it as EI0, and obtain the energy index EI from the energy index at the subsequent time s , and the energy index EI s Satisfy lg[EI s -EI0]<0.3, the corresponding subsequent moment is the starting moment of early warning judgment; Step 502: Use a computer to obtain the b value, cumulative apparent volume and Schmidt number after the start time of the early warning judgment, and record them as the b value to be judged, the cumulative apparent volume to be judged and the Schmidt number to be judged; and calculate the b value b at the later time in two consecutive unit times. h and the value b at the previous moment q , the cumulative visual volume V to be judged at the next moment ah and the cumulative visual volume V to be determined at the previous moment aq , the Schmidt number Sc to be determined at the next moment ah and the Schmidt number Sc to be determined at the previous moment aq Make judgments; Step 503: Use a computer to determine if b h -b q <-0.7, V ah -V aq >2 and Sc ah -Sc aq <-2, rock burst is predicted and an early warning is issued.
Citation Information
Patent Citations
Rock mass instability prediction method and device based on micro-seismic monitoring
CN114895352A
Three-element quantitative prediction method for time-space intensity of rock burst mine micro-seismic event
CN115375025A
Intelligent long-term and short-term prediction and early warning method for micro-seismic frequency and energy of rock burst mine
CN115545272A
Rock burst prediction method based on comparison of different danger monitoring data
CN117849855A
Early warning method and device for rock burst in well mine, terminal and storage medium
CN118013842A
Cited By
Microseismic activity parameter time sequence feature fusion method
CN119917998A
A method for fusing the time series characteristics of microseismic activity parameters
CN119917998B
