NL2032832B1 - Intelligent pre-warning method for coal mine rock burst based on quantitative prediction of microseismic event - Google Patents
Intelligent pre-warning method for coal mine rock burst based on quantitative prediction of microseismic event Download PDFInfo
- Publication number
- NL2032832B1 NL2032832B1 NL2032832A NL2032832A NL2032832B1 NL 2032832 B1 NL2032832 B1 NL 2032832B1 NL 2032832 A NL2032832 A NL 2032832A NL 2032832 A NL2032832 A NL 2032832A NL 2032832 B1 NL2032832 B1 NL 2032832B1
- Authority
- NL
- Netherlands
- Prior art keywords
- model
- msnet
- microseismic
- microseismic event
- data
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 17
- 239000003245 coal Substances 0.000 title claims abstract description 15
- 239000011435 rock Substances 0.000 title claims abstract description 15
- 238000012549 training Methods 0.000 claims description 33
- 239000013598 vector Substances 0.000 claims description 22
- 238000012360 testing method Methods 0.000 claims description 15
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 claims description 3
- 230000009172 bursting Effects 0.000 claims 1
- 238000013480 data collection Methods 0.000 claims 1
- 238000013527 convolutional neural network Methods 0.000 description 11
- 238000013528 artificial neural network Methods 0.000 description 10
- 230000000306 recurrent effect Effects 0.000 description 8
- 230000006870 function Effects 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000010276 construction Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 239000000523 sample Substances 0.000 description 3
- 239000000543 intermediate Substances 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000002045 lasting effect Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 229920000136 polysorbate Polymers 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000000243 solution Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
- G01V1/288—Event detection in seismic signals, e.g. microseismics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Environmental & Geological Engineering (AREA)
- Remote Sensing (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Business, Economics & Management (AREA)
- Geophysics (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Geology (AREA)
- Acoustics & Sound (AREA)
- Emergency Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides an intelligent pre—warning method for coal mine rock burst based on quantitative prediction of a microseismic event, which achieves integrated intelligent pre— warning based on data driving and reduces artificial subjective analysis. A microseismic event result predicted by a model is of great significance for pre—warning of rock burst risk.
Description
P1558 /NLpd
INTELLIGENT PRE-WARNING METHOD FOR COAL MINE ROCK BURST BASED ON
QUANTITATIVE PREDICTION OF MICROSEISMIC EVENT
The present invention relates to the technical field of pre- diction and pre-warning of underground rock burst disaster of a coal mine, specifically to an intelligent pre-warning method for coal mine rock burst based on quantitative prediction of a micro- seismic event.
At present, the research on prediction of rock burst time and areas mainly focuses on retrospective analysis of parameters such as energy, stress and microseismic in a disaster-pregnancy process after the occurrence of a disaster; relationships between the spa- tial-temporal evolution law of the various parameters and disaster start are studied, most of which are quantitative and regular de- scriptions.
The present invention adopts the following technical solu- tion.
An intelligent pre-warning method for coal mine rock burst based on quantitative prediction of a microseismic event includes the following steps:
Sl, collecting data: time, energy, three-dimensional space coordinate and waveform information, obtained by built-in data processing software of a coal mine microseismic monitoring system, of a microseismic event on a certain working face within a period of time are used as a data source of model pre-warning;
S2, building an MSNet model for acquiring a short-range and long-range microseismic event timing law: the MSNet model is di- vided into a linear path module and a nonlinear path module ac- cording to a data flow direction; the linear path module uses an autoregressive model to directly predict a future microseismic event sequence according to an input microseismic event sequence; the nonlinear path module achieves nonlinear mapping from the in- put microseismic event sequence to the future microseismic event sequence by means of a convolutional neural network, a recurrent neural network and a skip-LSTM network; a final prediction result of the MSNet model is a sum of a linear path result and a nonline- ar result; the convolutional neural network is specifically used for parsing a short-range dependency relationship between micro- seismic events within a single sequence length and a relationship among five attributes of the events such as XYZ coordinates of a spatial position, energy and a time shift; an output of the convo- lutional neural network also flows into the recurrent neural net- work and the skip-LSTM network; the recurrent neural network is specifically used for excavating a long-range dependency relation- ship between the microseismic event sequences; the skip-LSTM net- work is specifically used for solving the problem of a potential ultra-long-range dependency;
S3, dividing a data set: a microseismic event data set ac- quired underground is manually divided into a training set and a test set; the training set is used for training the MSNet model to enable the MSNet model to achieve a best fitting result, and the test set is used for testing the performance of the MSNet model;
S4, training the model: data of the training set is brought into the MSNet model; the model is run using a laboratory Personal
Computer (PC); coordinates, energy and time shifts of six micro- seismic events after 12 continuous predictions of microseismic events are used; in a training process, model parameters are up- dated using a small-batch gradient descent method; attributes of six continuous microseismic events output through model training are respectively regarded as six vectors with a length of 5; simi- larities between predicted values and real values of the attrib- utes are measured using the cosine similarity; the similarity is calculated using the following formula:
> 4 xB, ” A-B on cos (0) = rms = pe
EE Kep
Fe 2) +> (B)
Similarity= fad is where A; represents a real attribute vector of a microseismic event; B; represents a predicted attribute vector of the model; n represents the length of the vector; 35, developing an MSNet model-based pre-warning platform: three-dimensional geological data of a mine is acquired; and an underground three-dimensional geological precision model of the coal mine is built using Unity3D to display real-time microseismic event information and show a prediction result of the MSNet model.
In the present invention, positions and time where and when several microseismic events occur in the future and energy of the microseismic events are dynamically predicted using the continuous microseismic events on the working face, so as to quantitatively predict a future burst risk area; a corresponding intelligent rock burst pre-warning platform is developed to dynamically and in real time achieve quantitative prediction of a rock burst risk area on the working face; no personnel will subjectively participate in an analysis task in the whole model training process; a deep learning model has an enough accuracy to more accurately determine a posi- tion of a microseismic event; and a microseismic event result pre- dicted by the model is of great significance for pre-warning of rock burst risk.
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a schematic structural diagram of an MSNet model provided by the present invention.
FIG. 3 is an enlarged schematic diagram of a dotted box por- tion in FIG. 2.
FIG. 4 is a schematic diagram of construction of a train- ing/test sample provided in the present invention.
FIG. 5 is a schematic diagram of an MSNet model training data distribution box provided by the present invention.
Referring to FIG. 1 to FIG. 5, the present invention provides an intelligent pre-warning method for coal mine rock burst based on quantitative prediction of a microseismic event, including the following steps:
Sl, data is collected: time, energy, three-dimensional space coordinate and waveform information, obtained by built-in data processing software of a coal mine microseismic monitoring system, of a microseismic event on a certain working face within a period of time are used as a data source of model pre-warning. For exam- ple, 10,196 microseismic events monitored during the annual pro- gress of a certain mining working face in a coal mine in 2019 are selected as basic data for intelligent prediction.
S52, an MSNet model for acquiring a short-range and long-range microseismic event timing law is built: the MSNet model is divided into a linear path module and a nonlinear path module according to a data flow direction; the linear path module uses an autoregres- sive model to directly predict a future microseismic event se- quence according to an input microseismic event sequence; the non- linear path module achieves nonlinear mapping from the input mi- croseismic event sequence to the future microseismic event se- quence by means of a convolutional neural network, a recurrent neural network and a skip-LSTM network; a final prediction result of the MSNet model is a sum of a linear path result and a nonline- ar result; the convolutional neural network is specifically used for parsing a short-range dependency relationship between micro- seismic events within a single sequence length and a relationship among five attributes of the events such as XYZ coordinates of a spatial position, energy and a time shift; an output of the convo- lutional neural network also flows into the recurrent neural net- work and the skip-LSTM network; the recurrent neural network is specifically used for excavating a long-range dependency relation- ship between the microseismic event sequences; the skip-LSTM net- work is specifically used for sclving the problem of a potential ultra-long-range dependency.
There are two paths in the data flow direction of the MSNet model. A linear path uses the autoregressive model to directly predict the future microseismic event sequence according to the input microseismic event sequence; the nonlinear path achieves the nonlinear mapping from the input microseismic event sequence to 5 the future microseismic event sequence by means of the convolu- tional neural network, the recurrent neural network and the skip-
LSTM network; the final prediction result of the MSNet model is the sum of the linear path result and the nonlinear result, spe- cifically as shown in FIG. 2.
In the nonlinear path module, the data of the input micro- seismic event sequence first flows into the convolutional neural network which is used for parsing the short-range dependency rela- tionship between the microseismic events within the single se- quence length and the relationship among five attributes of the events, and the five attributes specifically include the X, Y and 7 coordinates of the spatial position, the energy and the time shift. The size of a convolution kernel in the convoluticnal neu- ral network is 6x5, and the number of convolution kernels is 32; an activation function of the convolutional neural network is
ReLU, specifically RelU(x)=max(0,x), where max represents finding the maximum value, and x represents an input value of the neural network; and the convolutional neural network can be implemented by using the existing convolutional neural network. An i*" convolu- tion kernel among the convolution kernels is operated by the fol- lowing formula: h; =ReLU{(W; *X+b;) Formula (1) where W; represents a convolution kernel matrix; X represents an input timing matrix; and b; represents bias.
The recurrent neural network in the nonlinear path module adopts a long short term memory (LSTM) network, which can better capture the long-range dependency relationship through a "gate mechanism" and also avoid the problem of gradient explosion. Re- ferring to FIG. 2 and FIG. 3, a calculation process of an LSTM da- ta flow direction includes: at time t, adding a hidden state hy, at time t-1 to an input x; at time t; calculating vectors of an input gate i., a forget gate ft and an output ot and an intermedi- ate state at this time; updating ct using it and f.; and finally calculating a hidden state h, at time t using ct and o,. This cal- culation process is circulated. The specific calculation process of the data flow direction of the LSTM network is as follows:
Cc, tanh
Oo, oa Co | x, = wv + Bb)
Z © Aa, , > 2
C, — J Ci on + LC, ° + A. Formula (3) he =o. tanhic.) where Ot; i., f., and x respectively represent the intermedi- ate state, an output gate vector, an input gate vector, a forget gate vector, and input parameters at time t; W and b are a model weight and bias of the LSTM network, which need to be updated dur- ing training; tanh is a hyperbolic tangent function; o represents an s function sigmoid; c¢..; represents a cell state at time t-1; € represents a cell state at time t; hy: represents the hidden time at time t-1 state; and h, represents the hidden state at time t.
The purpose of the skip-LSTM network in the nonlinear path module is to solve the problem of the potential ultra-long-range dependency, and a skip step length p is a model hyperparameter, which can be selected according to an actual task and optimized during the model training. After the skip step length p is intro- duced, the specific calculation process of the data flow direction of the skip-LSTM network is as follows: , tanh
Ela] + | = Ed + 5) í, o A,
J, o
C, — J. >p + LE, ’ : ’ Formula (3) hy = Ot tanh (Cc)
where p represents the skip step length.
For the autoregressive model used in the linear path module in the MSNet model, it is assumed that there is a linear relation- ship between the attribute vectors of all the microseismic events, a (k+t1)™ event is predicted by using the previous k microseismic events. The formula for predicting events by the autoregressive model in the present application is as follows:
Kk i=] Formula (4) wherein E, represents an attribute vector of an event to be predicted; W; and b; represent the model parameters; and k repre- sents the previous k microseismic events.
The loss function of the MSNet model is described as follows:
Predicted outputs of the MSNet model are attributes (a vector with a dimension of 6x5) of six continuous microseismic events in the future, so the loss function of the MSNet model defined in the present application is an average distance between two point se- quences with a length of 6 in a five-dimensional space, and an ex- pression of the loss function is as follows: 1 & N = SY —¥,)
J i=} Formula (5) where Y; represents a predicted microseismic event attribute oe vector (with a length of 5), and tx; represents a real microseis- mic event attribute vector.
S3, a data set is divided: a microseismic event data set ac- quired underground is manually divided into a training set and a test set; the training set is used for training the MSNet model to enable the MSNet model to achieve a best fitting result, and the test set is used for testing the performance of the MSNet model;
For example, in the data set of the 10,196 microseismic events collected in the aforementioned step Sl, 9,863 microseismic events from January to October are selected for model training, and 333 microseismic events in November and December are used for model testing. In order to use as the microseismic event data as many as possible, a step length between the input microseismic event se- quences of the input microseismic event sequences of the MSNet model is 1, and next six microseismic events are predicted by 12 microseismic events. According to the aforementioned training sam- ple construction rule, 9,846 training samples and 316 test samples can be obtained from the data set with the 10,196 microseismic events. FIG. 4 specifically shows a construction method of the training/test samples. 34, the model is trained: data of the training set is brought into the MSNet model; the model is run using a laboratory PC; co- ordinates, energy and time shifts of six microseismic events after 12 continuous predictions of microseismic events are used; and in a training process, model parameters are updated using a small- batch gradient descent method, and a batch size may be set to be 16. The five attributes input by the samples have an extremely large scale difference, so that during the model training, the at- tributes need to be standardized, and the values of all the five attributes are converted between 0 and 1; and the training process undergoes 100 iterations in total, lasting 545 s. The attributes of the six continuous microseismic events output through the model training are respectively regarded as six vectors with a length of 5; similarities between predicted values and real values of the attributes are measured using the cosine similarity; if the value of a similarity is closer to 1, it indicates that the two vectors are more similar, and the prediction accuracy is higher, thus ver- ifying a training result. The similarity is calculated using the following formula:
AB 2 A xB,
BEE IE ee EE ee
Similarity= feed gf Formula (6) where A; represents a real attribute vector of a microseismic event; B: represents a predicted attribute vector of the model; n represents the length of the vector.
By means of the calculation of the cosine similarities be-
tween the real values and the predicted values of the attributes of the 316 test samples over time, the model has the best predic- tion effect on the X coordinate and Y coordinate of a microseismic event; average cosine similarities of X and Y of the 316 test sam- ples are respectively 0.997 and 0.995; the prediction result of the Z coordinate of the microseismic event by the model is slight- ly poor, which is obviously poorer than the cosine similarities of
X and Y, and an average similarity of Z is 0.939.
FIG. 5 shows a diagram of a distribution box of the model training data (the 9,863 microseismic events from January to Octo- ber) after the five attributes are standardized. The rhombus in the figure represents an abnormal value of the data; the dot rep- resents an average value of the data; the upper and lower edges of the rectangle represent upper and lower quartiles of an attribute; and the long line represents a median of the data. 55, an MSNet model-based pre-warning platform is developed: after the model training is completed, online deployment is per- formed through an OpenVINO tool kit; and the microseismic event sequence acquired in real time is used to dynamically predict at- tributes of future events. Three-dimensional geological data of a mine is acquired; and an underground three-dimensional geological precision model of the coal mine is built using Unity3D to display real-time microseismic event information and show a prediction re- sult of the MSNet model. The position of the microseismic event is accurately embedded in the 3D geological precision model, which can represent a precise position of the event downhole, and the time information of the microseismic event is also consistent with the background time of the platform. The real-time microseismic event information displayed on the platform is from a built-in da- tabase of Arawin microseismic data processing software, and the
MSNet prediction result displayed on the platform is from an MSNet prediction result text file.
In actual operation of the intelligent rock burst pre-warning platform, the platform always keeps displaying the latest 12 con- tinuoous microseismic events, predicts the attributes of the next six microseismic events at the same time, and correspondingly dis- plays their positions. When an Aramis system monitors a latest mi-
croseismic event, the platform dynamically updates the displayed content (abandons the earliest microseismic event and displays the latest microseismic event), and simultaneously obtains a latest prediction result.
Real-time rock burst risk areas are divided ac- cording to the prediction results of the MSNet model.
A circle is drawn by taking the position of an event with the highest energy among the six predicted microseismic events as a circle center, and its edge passes through the position of the microseismic event farthest from the circle center.
The circle represents predicted dynamic risk of rock burst, and the risk gradually decreases along a radial direction.
Claims (1)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
NL2032832A NL2032832B1 (en) | 2022-08-23 | 2022-08-23 | Intelligent pre-warning method for coal mine rock burst based on quantitative prediction of microseismic event |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
NL2032832A NL2032832B1 (en) | 2022-08-23 | 2022-08-23 | Intelligent pre-warning method for coal mine rock burst based on quantitative prediction of microseismic event |
Publications (1)
Publication Number | Publication Date |
---|---|
NL2032832B1 true NL2032832B1 (en) | 2024-03-04 |
Family
ID=90106932
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
NL2032832A NL2032832B1 (en) | 2022-08-23 | 2022-08-23 | Intelligent pre-warning method for coal mine rock burst based on quantitative prediction of microseismic event |
Country Status (1)
Country | Link |
---|---|
NL (1) | NL2032832B1 (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN118133906A (en) * | 2024-05-10 | 2024-06-04 | 煤炭科学研究总院有限公司 | Training method and prediction method for coal mine rock burst strength grading prediction model |
-
2022
- 2022-08-23 NL NL2032832A patent/NL2032832B1/en active
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN118133906A (en) * | 2024-05-10 | 2024-06-04 | 煤炭科学研究总院有限公司 | Training method and prediction method for coal mine rock burst strength grading prediction model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113723595B (en) | Intelligent early warning method for rock burst of coal mine based on quantitative prediction of microseism event | |
Asadzadeh et al. | Sensor-based safety management | |
Baloyi et al. | The development of a mining method selection model through a detailed assessment of multi-criteria decision methods | |
Porwal et al. | Introduction to the special issue: mineral prospectivity analysis and quantitative resource estimation | |
Leng et al. | A hybrid data mining method for tunnel engineering based on real-time monitoring data from tunnel boring machines | |
Chen et al. | A quantitative pre-warning for coal burst hazardous zones in a deep coal mine based on the spatio-temporal forecast of microseismic events | |
Zhou et al. | Performance evaluation of rockburst prediction based on PSO-SVM, HHO-SVM, and MFO-SVM hybrid models | |
Li et al. | Stability prediction of hard rock pillar using support vector machine optimized by three metaheuristic algorithms | |
US11315228B2 (en) | System and method for mineral exploration | |
Wang et al. | AdaBoost-driven multi-parameter real-time warning of rock burst risk in coal mines | |
NL2032832B1 (en) | Intelligent pre-warning method for coal mine rock burst based on quantitative prediction of microseismic event | |
Gan et al. | A new spatial modeling method for 3D formation drillability field using fuzzy c-means clustering and random forest | |
Demirkan et al. | Evaluation of time series artificial intelligence models for real-time/near-real-time methane prediction in coal mines | |
Wu et al. | An improved fractal prediction model for forecasting mine slope deformation using GM (1, 1) | |
CN113283806A (en) | Enterprise information evaluation method and device, computer equipment and storage medium | |
Kant et al. | A review of approaches used for the selection of optimum stoping method in hard rock underground mine | |
Long et al. | Probability prediction method for rockburst intensity based on rough set and multidimensional cloud model uncertainty reasoning | |
CN106033126A (en) | Quantitative classification method of oil and gas unit exploration degree | |
CN113947309A (en) | Shield tunnel construction standard working hour measuring and calculating and scoring method based on big construction data | |
Elfes et al. | Extending the START framework: Computation of optimal capability development portfolios using a decision theory approach | |
Sulaiman et al. | A predictive model for the population growth of refugees in Asia: a multiple linear regression approach | |
Nanditha et al. | Optimized defect prediction model using statistical process control and Correlation-Based feature selection method | |
Al-Ansi et al. | Intelligent risk analysis of investment projects in the extractive industry | |
Novianto et al. | Implementation of Orange Data Mining to Predict Student Graduation on Time at Pringsewu Muhammadiyah University | |
Shnaydman | Industry drug development portfolio forecasting: productivity, risk, innovation, sustainability |