CN114757266B - Construction method of rock burst prediction model driven by expert knowledge and data fusion - Google Patents
Construction method of rock burst prediction model driven by expert knowledge and data fusion Download PDFInfo
- Publication number
- CN114757266B CN114757266B CN202210306659.4A CN202210306659A CN114757266B CN 114757266 B CN114757266 B CN 114757266B CN 202210306659 A CN202210306659 A CN 202210306659A CN 114757266 B CN114757266 B CN 114757266B
- Authority
- CN
- China
- Prior art keywords
- data
- energy
- earthquake
- prediction model
- event
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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. analysis, for interpretation, for correction
- G01V1/288—Event detection in seismic signals, e.g. microseismics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
- G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Mining
Abstract
The invention discloses a construction method of a rock burst prediction model driven by expert knowledge and data fusion, which comprises the following steps: the method comprises the steps of ore seismic data acquisition, ore seismic data processing, feature display and implicit feature extraction and prediction model construction, wherein if the probability of a high-energy ore seismic event is greater than that of a non-high-energy ore seismic event, 1 is output, otherwise 0 is output, and therefore prediction of the rock burst high-energy event is achieved. The invention can comprehensively and accurately predict the occurrence probability of rock burst.
Description
Technical Field
The invention relates to a rock burst prediction model, in particular to a construction method of a rock burst prediction model driven by expert knowledge and data fusion, and belongs to the technical field of monitoring and early warning.
Background
Coal is a basic energy and an important raw material in China, and the coal industry is an important basic industry related to national economic pulse and energy safety. Rock burst is a coal dynamic disaster which seriously threatens the safe production of mines. The rock burst has the characteristics of sudden, rapid and violent damage, and has serious influence on the normal production of coal mines, metal mines, tunnels and the like, and huge economic loss and casualties are caused seriously.
Along with the increase of the mining depth of the mine, the danger of rock burst is gradually increased, and the mine which does not generate rock burst originally also starts to generate rock burst; originally, the mine with the occurrence of the rock burst has increasingly high impact strength and higher frequency. However, the complicated diversity of the time, place, area, source, etc. of occurrence of rock burst and the sudden nature of rock burst make the prediction work extremely difficult and complicated, and thus, become a worldwide problem to be solved urgently.
The traditional prediction methods mainly comprise the following two methods:
1) The mine earthquake is explained and described according to a physical method provided by researchers through an existing physical model, and meanwhile, the explicit characteristics of mine earthquake precursor extraction are researched through rock burst knowledge, so that the mine earthquake in a certain background can be truly described, and the mine earthquake is highly interpretable in a theoretical system. However, these artificially designed features may not fully utilize the information contained in the mine seismic sequence;
2) The implicit characteristics of the mine earthquake data are extracted by using a deep learning method to establish a corresponding model for prediction, and although the information contained in the mine earthquake sequence can be fully utilized, the interpretation capability of the mine earthquake sequence is weak in a theoretical system.
Both methods cannot comprehensively and accurately predict the occurrence probability of rock burst, so that how to comprehensively and accurately predict the occurrence probability of rock burst becomes a technical problem which needs to be solved at present.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a rock burst prediction model construction method driven by expert knowledge and data fusion, which can comprehensively and accurately predict the occurrence probability of rock burst.
In order to achieve the aim, the invention provides a construction method of a rock burst prediction model driven by expert knowledge and data fusion, which comprises the following steps of
1) Acquiring original data including mine earthquake time, mine earthquake energy and earthquake source coordinates through a micro-earthquake sensor;
2) Mine earthquake data processing: converting the original data into precursor mode sequence data for prediction model input, performing statistical analysis, calculating to obtain a daily maximum energy value and an average energy value, generating time sequence data with day as a minimum unit, designating the length of the precursor mode sequence, and generating a precursor mode sequence and a label thereof as massive mine earthquake data;
3) Characteristic extraction: including expert knowledge driven explicit feature extraction and data driven implicit feature extraction
Processing the mine earthquake indexes of the mining area by using a Principal Component Analysis (PCA) method according to the mine earthquake indexes of different mining areas to obtain the weight occupied by each mine earthquake index, and selecting a mine earthquake index combination to be used according to the actual needs of the mining area and the weight of the mine earthquake indexes, wherein the mine earthquake index combination is extracted explicit characteristics;
data-driven implicit feature extraction: inputting the mass mineral earthquake data in the step 2) into a deep convolutional neural network for implicit feature extraction;
4) Constructing a prediction model: taking the mass mineral earthquake data in the step 2) as a training data set sample to generate a prediction model, wherein the prediction model comprises a feature fusion module and a classification network module, performing deep fusion on explicit features and implicit features in the feature fusion module by using an attention mechanism to obtain fused features, inputting the features into the classification network module, processing the fused features by the classification network module to obtain the probability of whether a high-energy event exists, and outputting 1 if the probability of the high-energy mineral earthquake event exists is greater than the probability of the high-energy mineral earthquake event, otherwise outputting 0, thereby realizing the prediction of the high-energy event of the rock burst.
Compared with the prior art, the mining rock burst prediction method based on the deep fusion of the expert knowledge, namely the manually selected mining earthquake indexes (explicit characteristics) and the massive mining earthquake data (implicit characteristics) is combined deeply, and therefore the probability of occurrence of the rock burst is predicted comprehensively and accurately. In the training process, the prediction model can dynamically adjust the weight of each category in the calculation loss according to the distribution condition of batch samples and overall samples. In network training, a prediction model is calculated using an objective functionBy continuously updating the parameters in the neural network model, the loss of the prediction model on the training data set is minimized. In addition, in order to reduce the false alarm rate of the event caused by large energy in the rock burst prediction task, the invention adds the learning weight z of each category in the objective function 0 And z 1 By adjusting the learning weight of the large energy event, the prediction model can be more biased to the prediction of the large energy sample, so that the missing report rate of the large energy event is reduced, the problem of unbalanced data categories is effectively solved, the convergence speed of the model is effectively accelerated, and the prediction accuracy of the model is improved.
Drawings
FIG. 1 is a schematic view of the model structure of the present invention;
FIG. 2 is a process of mine seismic data processing of the present invention;
FIG. 3 is a schematic diagram of the deep learning convolutional neural network for implicit feature extraction according to the present invention;
FIG. 4 is a schematic diagram of fusion of explicit and implicit features in a predictive model of the present invention.
Detailed Description
The invention will be further explained with reference to the drawings.
According to the method, the data information of the area to be monitored is collected through the micro-seismic sensors, the micro-seismic sensors are installed on the periphery of the coal mining working face, then the collected data are transmitted to the upper computer of the ground control room, and the upper computer monitors the rock burst of the area to be detected by adopting the prediction model of the embodiment.
Referring to fig. 1, the invention relates to a construction method of a rock burst prediction model driven by expert knowledge and data fusion, which comprises the steps of
1) Acquiring original data including mine seismic time, mine seismic energy and seismic source coordinates through a micro-seismic sensor;
2) Mineral seismic data processing, see fig. 2 and 3:
converting the original data into precursor mode sequence data for prediction model input, performing statistical analysis, calculating to obtain a daily maximum energy value and an average energy value, generating time sequence data with day as a minimum unit, specifying the length of the precursor mode sequence, and generating a precursor mode sequence and a label thereof as massive mine earthquake data, wherein the specific method comprises the following steps:
2.1 ) counting the original data by a fixed time window to construct a time sequence data set, and assuming that the data record obtained by the calculation of the ith time window is m i The expression is as follows:
m i ＝＜id, e max , e mean , f＞ (1)
wherein id is the time window number, e max Maximum energy in a time window, e mean F is the frequency of the mineral shocks in the time window, so that when the data is divided into n time windows, the time window is traversed based on the formula (1), and the time series data set M = < M is obtained 0 ,m 1 ,m 2 ,.....,m n-1 ＞；
2.2 Constructing a precursor pattern sequence based on the time series data set M of step 2.1), assuming that the ith precursor pattern sequence is s i The expression is as follows:
s i ＝＜m i×j ,m i×j+1 ,m i×j+2 ,......,m i×j+p-1 ＞ (2)
wherein p is the precursor mode sequence length, j is the sampling step length, m i×j Sample data of step j for the ith time window, thus, n time windows and n>>On the premise of p, a precursor pattern sequence set S is generated based on formula (2), and the expression is as follows:
S＝＜s 0 ,s 1 ,s 2 ,,.....,s D-1 ＞ (3)
wherein D is the number of precursor mode sequences of the precursor mode sequence samples in the case that the prediction time range is N hours;
2.3 For subsequent prediction model training, a label set T corresponding to the precursor pattern sequence set S needs to be established, and the expression is as follows:
T＝＜t 0 ,t 1 ,t 2 ,,.....,t D-1 ＞ (4)
wherein, t i For a sequence s of precursor patterns i If a high energy event t is about to occur i =1, otherwise t i =0, the calculation method is as follows:
wherein e is i For a sequence s of precursor patterns i The maximum energy value in the future N hours, E is the energy threshold of a high-energy event, and the value of E can be set according to the actual requirement, such as E =5 × 10 4 J or E =1 × 10 5 J。
3) Feature extraction: including expert knowledge driven explicit feature extraction and data driven implicit feature extraction
3.1 Expert knowledge explicit feature extraction:
and processing the mine earthquake indexes of the mining area by using a Principal Component Analysis (PCA) method according to the mine earthquake indexes of different mining areas to obtain the weight of each mine earthquake index, and selecting a mine earthquake index combination to be used according to the actual needs of the mining area and the weight of the mine earthquake indexes. If the mine earthquake indexes of a certain mining area are a1, a2, a3 and a4, the four indexes are processed by using a principal component analysis method, and the weight of the mine earthquake indexes is obtained as follows: a1=1.4%, a2=2.5%, a3=50%, a4=46.1%, and from the viewpoint of data, the occupied weight of a1 and a2 is small, that is, the impact pressure on the prediction of the mine area is small, in order to improve the data operation, the use of a1 and a2 can be abandoned, and the combination of a3 and a4 is selected as the explicit feature of the mine area for constructing the prediction model;
the mine earthquake index is obtained by long-term research on rock burst through the existing rock burst knowledge by researchers, the mine earthquake index comprises comprehensive evaluation A (B) of factors such as microseismic activity level a, microseismic activity intensity B, frequency, energy and activity of microseismic time in a certain period, and lack earthquake B, and the expression of the indexes is as follows:
wherein m is the total number of grading levels, lgE i Is the energy level indicated in the ith gear, N i The total number of the microseisms of the ith gear level, the value a represents the level of the microseismic activity in a certain period of time, and the larger the value a is, the higher the frequency of the microseisms is; the b value represents the strength of the microseismic activity, and the smaller the b value is, the higher the possibility of occurrence of a large-energy event is;
in the formula, M k The value A (b) comprehensively considers factors such as frequency, energy, activity and the like of the microseismic events in a certain period of time for the energy level of the microseismic events, and can visually and quantitatively evaluate the microseismic activity;
in the formula, B is the lack of vibration,
is the average energy level, M, over a statistical period 0 Is the initial energy level, and n is the total number of microseismic events within the statistical time period. A lack of shock means that the mean energy level M in a certain time period is greater than the mean energy level in a longer time period->
Smaller, the region is more likely to have a large energy event to replenish this missing energy level.
The four indexes a, B, A (B) and B are selected from the mine earthquake index expert knowledge base, and the four indexes are selected because: and (3) processing the mine earthquake indexes of a plurality of mining areas by using a PCA method, and finding that each mining area comprises the four indexes, and the four indexes have larger weight in the indexes of each mining area, so that the accuracy of rock burst prediction in the later period is improved. Of course, all the mineral earthquake indexes in a certain mining area can be selected according to actual needs to be processed by the PCA method, the weight occupied by each mineral earthquake index is obtained, and then the required mineral earthquake indexes are selected according to the required total weight to form a mineral earthquake index set as the explicit characteristics.
Extracting implicit characteristics: implicit feature extraction using convolutional neural networks in deep learning
The convolutional neural network has a characteristic learning capability, and its artificial neuron can respond to a part of the surrounding units within the coverage range, and is composed of one or more convolutional layers and a top fully-connected layer, and also includes an associated weight and a pooling layer. This structure enables the convolutional neural network to utilize a two-dimensional structure of the input data. The convolution kernel parameter sharing in the hidden layer and the sparsity of interlayer connection enable the convolution neural network to realize hidden feature extraction with smaller calculation amount. Therefore, in the embodiment, the hidden features of the massive mineral seismic data are extracted by using the deep convolutional neural network, and the massive mineral seismic data (precursor sequence mode and labels thereof) in the step 2) are input into the deep convolutional neural network for implicit feature extraction; the deep convolution neural network in this embodiment is a 3-layer convolution, and includes 88 convolution kernels, and outputs 1000-dimensional implicit feature vectors, which are the implicit features to be extracted in this embodiment.
4) Constructing a prediction model: generating a prediction model by taking the mass mineral earthquake data (precursor sequence mode and labels thereof) in the step 2) as a training data set sample, wherein the prediction model comprises a feature fusion module and a classification network module, performing deep fusion on explicit feature extraction and implicit feature extraction by using an attention mechanism in the feature fusion module, and realizing classification by full-connection network fitting in the classification network module, so as to construct a rock burst high-energy event prediction model, and the specific method is as follows, referring to fig. 4:
4.1 Establishing an objective function by using the massive mineral earthquake data in the step 2) as a training data set:
wherein L is i Is the loss value of the ith precursor mode sequence, N is the number of precursor mode sequences, z 0 And z 1 Learning weights, w, for two classes respectively 0 And w 1 Sample distribution weights for class 0 and 1, respectively, y if the tag of the ith precursor mode sequence is a low energy event io ＝1，y i1 =0, otherwise y io ＝0，y i1 ＝1；p i0 For the predicted probability of observing a sample i as class 0, p i1 A predicted probability for observation sample i as class 1;
4.2 Network weighting w using a back propagation algorithm 0 、w 1 Minimizing the objective function loss, thereby generating a prediction model comprising a feature fusion module and a classification network module;
the method effectively solves the problem of unbalanced data categories, accelerates the convergence speed of the prediction model, and improves the prediction accuracy of the prediction model.
4.3 Depth fusion of explicit and implicit features using attention methods in a feature fusion module
Because the complexity and the heterogeneity of the explicit feature and the implicit feature cause that a simple weighted feature fusion method is not applicable, the embodiment uses the attention method in deep learning to realize the fusion of the explicit feature and the implicit feature and realize the weighting of each dimension in the explicit feature and the implicit feature, and the specific method is as follows:
4.3.1 Will display feature F e And implicit feature F i Merging to obtain an initial feature vector F s
Wherein the display feature vector and the implicit feature vector
Respectively d 1 And d 2 The initial feature vector->
Has a dimension of d 1 +d 2 (ii) a In the attention mechanism, the weight vectors for the implicit and explicit features are respectively designated ≦ ≦>
The calculation method is as follows:
wherein the content of the first and second substances,
for two learnable parameter matrices, H (x) is the activation function->
Weight vector V e And V i Corresponds to Fe and F in each dimension i A weight for each feature dimension;
4.3.2 Calculate the final fused feature vector F f Expressed as:
F f ＝[F e ⊙V e ，F j ⊙V i ] (14)
wherein, the all are Hadamard products;
4.4 ) the fusion feature vector F obtained in step 4.3.2 f The probability of the large-energy mineral earthquake event is obtained, if the probability of the large-energy mineral earthquake event is greater than the probability of the large-energy mineral earthquake event, 1 is output, otherwise 0 is output;
in some embodiments, the classification network module comprises a full connectivity layer and an activation functionIn deep learning, the fully-connected layer plays the role of a classifier and can map the learned distributed feature representation into a sample mark space, the fully-connected layer in the prediction model mainly comprises 2000 neurons and an activation function, and the classification network module fuses feature vectors F by the following method f And (3) processing:
fusing the feature vectors F f Inputting the data into a full connection layer, mapping the learned distributed feature representation into a sample mark space by the full connection layer, then carrying out normalization processing by using an activation function to obtain the probability of whether a high-energy event exists, if the probability of the high-energy mineral earthquake event exists is greater than the probability of the high-energy mineral earthquake event, outputting 1, otherwise, outputting 0. Such as: the probability of the high-energy mine earthquake event is 60%, the probability of the non-high-energy mine earthquake event is 40%, at the moment, the high-energy mine earthquake event is greater than the probability of the non-high-energy mine earthquake event, 1 is finally output, the probability of the high-energy mine earthquake event is shown to occur, and a worker can take corresponding measures according to a prediction result.
In the invention, the training data set sample is unbalanced because the occurrence frequency of the large-energy mine earthquake event is less, and the data labeled as the small-energy event is far more than the data of the large-energy event. If the traditional deep learning model training method is used, the problem that the prediction model is biased to a class with more learning samples during classification can be caused. In order to solve the problem, the embodiment uses the concept of "rescaling" for reference, and in the training process, the prediction model can dynamically adjust the weight of each class in calculating the loss according to the distribution condition of batch samples and overall samples. In network training, the loss value of the prediction model is calculated by using an objective function, and the loss of the prediction model on a training data set is minimized by continuously updating parameters in the neural network model. In addition, in order to reduce the false positive rate of the event caused by large energy in the rock burst prediction task, the learning weight z of each category is added to the objective function in the embodiment 0 And z 1 By adjusting the learning weight of the high-energy event, the prediction model can be more biased to the prediction of the high-energy sample, so that the missing report rate of the high-energy event is reduced, and the missing report rate of the high-energy event is effectively improvedThe problem of unbalanced data category effectively accelerates the convergence speed of the model and improves the prediction accuracy of the model.
Claims (5)
1. A rock burst prediction model construction method driven by expert knowledge and data fusion is characterized by comprising the following steps
1) Acquiring original data including mine seismic time, mine seismic energy and seismic source coordinates through a micro-seismic sensor;
2) And (3) mine earthquake data processing: converting the original data into precursor mode sequence data for prediction model input, performing statistical analysis, calculating to obtain a daily maximum energy value and an average energy value, generating time sequence data with day as a minimum unit, designating the length of the precursor mode sequence, and generating a precursor mode sequence and a label thereof as massive mine earthquake data;
3) Feature extraction: including expert knowledge driven explicit feature extraction and data driven implicit feature extraction
Processing the mine earthquake indexes of the mining area by using a principal component analysis method according to the mine earthquake indexes of different mining areas to obtain the weight occupied by each mine earthquake index, and selecting a mine earthquake index combination to be used according to the actual needs of the mining area and the weight of the mine earthquake indexes, wherein the mine earthquake index combination is extracted explicit characteristics;
data-driven implicit feature extraction: inputting the mass mineral earthquake data in the step 2) into a deep convolutional neural network for implicit feature extraction;
4) Constructing a prediction model: taking the mass mineral earthquake data in the step 2) as a training data set sample to generate a prediction model, wherein the prediction model comprises a feature fusion module and a classification network module, performing deep fusion on explicit features and implicit features in the feature fusion module by using an attention mechanism to obtain fused features, inputting the features into the classification network module, processing the fused features by the classification network module to obtain the probability of whether a high-energy event exists, and outputting 1 if the probability of the high-energy mineral earthquake event exists is greater than the probability of the high-energy mineral earthquake event, otherwise outputting 0, thereby realizing the prediction of the high-energy event of rock burst;
the step 4) comprises the following specific steps:
4.1 Taking the mass mineral earthquake data obtained in the step 2) as a training data set, and establishing an objective function:
wherein L is i Is the loss value of the ith precursor mode sequence, X is the number of precursor mode sequences, z 0 And z 1 Learning weights, w, for two classes respectively 0 And w 1 Sample distribution weights for class 0 and 1, respectively, y if the tag of the ith precursor mode sequence is a low energy event io ＝1，y i1 =0, otherwise y io ＝0，y i1 ＝1；p d0 For the predicted probability of observing that sample d is of class 0, p d1 A predicted probability that the observation sample d is of class 1;
4.2 Network weighting w using a back propagation algorithm 0 、w 1 Minimizing the objective function loss, thereby generating a prediction model, wherein the prediction model comprises a feature fusion module and a classification network module;
4.3 In the feature fusion module, an attention method is used to perform deep fusion on the explicit features and the implicit features, and the specific method is as follows:
4.3.1 Will display feature vector F e And implicit feature vector F i Merging to obtain an initial feature vector F s
F s ＝[F e ,F i ] (12)
Wherein the display feature vector and the implicit feature vector
Respectively d 1 And d 2 Initial feature vector
Has a dimension of d 1 +d 2 (ii) a In the attention mechanism, the weight vectors for implicit and explicit features are denoted as
The calculation method is as follows:
wherein the content of the first and second substances,
for two learnable parameter matrices, H (x) is an activation function >>
Weight vector V e And V i Corresponds to F in each dimension e And F i A weight for each feature dimension;
4.3.2 Calculate the final fused feature vector F f Expressed as:
F f ＝[F e ⊙V e ，F i ⊙V i ] (14)
wherein, the all are Hadamard products;
4) 4.4) fusing the feature vector F obtained in the step 4.3.2) f And inputting the data into a classification network module, processing the data by the classification network module to obtain the probability of the existence of the high-energy event, and outputting 1 if the probability of the existence of the high-energy mineral earthquake event is greater than the probability of the existence of the high-energy mineral earthquake event, otherwise outputting 0.
2. The method for constructing the rock burst prediction model driven by the fusion of expert knowledge and data according to claim 1, wherein the specific steps in the step 2) are as follows:
2.1 Is) fixed withCounting the original data in a timing window, constructing a time sequence data set, and assuming that the data obtained by the calculation of the ith time window is recorded as m i The expression is as follows:
m i ＝＜id,e max ,e mean ,f＞ (1)
wherein id is the time window number, e max Maximum energy in a time window, e mean F is the frequency of the mineral shocks in the time window, thus, when the data is divided into n time windows, the time window is traversed based on the formula (1), and the time series data set M = < M is obtained 0 ,m 1 ,m 2 ,.....,m n-1 ＞；
2.2 Constructing precursor mode sequence based on the time sequence data set M of the step 2.1), and assuming that the ith precursor mode sequence is s i The expression is as follows:
s i ＝＜m i×j ,m i×j+1 ,m i×j+ 2,......,m i×j+p-1 ＞ (2)
wherein p is the precursor mode sequence length, j is the sampling step length, m i×j Sample data of step size j for the ith time window, thus, n time windows and n>>On the premise of p, generating a precursor mode sequence set S based on the formula (2), wherein the expression is as follows:
S＝＜s 0 ,s 1 ,s 2 ,......,s D-1 ＞ (3)
d is the number of precursor mode sequences of the precursor mode sequence samples under the condition that the prediction time range is h hours;
2.3 A label set T corresponding to the precursor mode sequence set S is established, and the expression is as follows:
T＝＜t 0 ,t 1 ,t 2 ,......,t D-1 ＞ (4)
wherein, t i For a sequence s of precursor patterns i If a high energy event t is about to occur i =1, otherwise t i =0, the calculation method is as follows:
wherein e is i For a sequence s of precursor patterns i The maximum energy value in the h hours in the future, E is the energy threshold of a large energy event.
3. The method for constructing the rock burst prediction model driven by the fusion of expert knowledge and data according to claim 1, wherein the mineral earthquake indexes are obtained by long-term research on rock burst through the existing rock burst knowledge by researchers, the mineral earthquake indexes comprise a microseismic activity level a in a certain period, microseismic activity intensity B, frequency of microseismic time in a certain period, comprehensive evaluation A (B) of energy and activity factors, and lack earthquake B, and the expressions of the indexes are as follows:
wherein m is the total number of grading levels, lgE g Is the energy level shown in the g-th gear, N g The total number of microseisms of the g-th gear energy level;
in the formula, M k Is the microseismic event energy level;
in the formula, B is the lack of vibration,
is the average energy level, M, over a statistical period 0 V is the total number of microseismic events within the statistical time period.
4. The method for constructing the rock burst prediction model driven by the fusion of the expert knowledge and the data according to claim 1, wherein the deep convolutional neural network is a 3-layer convolution and comprises 88 convolution kernels, and an implicit feature vector with 1000 dimensions is output and is an implicit feature.
5. The expert knowledge and data fusion driven rock burst prediction model construction method according to claim 4, wherein the classification network module comprises a full connection layer consisting of 2000 neurons and an activation function; the classification network module fuses the feature vectors F by the following method f And (3) processing:
fusing the feature vectors F f Inputting the data into a full connection layer, mapping the learned distributed feature representation into a sample mark space by the full connection layer, then carrying out normalization processing by using an activation function to obtain the probability of whether a high-energy event exists, if the probability of the high-energy mineral earthquake event exists is greater than the probability of the high-energy mineral earthquake event, outputting 1, otherwise, outputting 0.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210306659.4A CN114757266B (en) | 2022-03-27 | 2022-03-27 | Construction method of rock burst prediction model driven by expert knowledge and data fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210306659.4A CN114757266B (en) | 2022-03-27 | 2022-03-27 | Construction method of rock burst prediction model driven by expert knowledge and data fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114757266A CN114757266A (en) | 2022-07-15 |
| CN114757266B true CN114757266B (en) | 2023-04-07 |
Family
ID=82326827
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210306659.4A Active CN114757266B (en) | 2022-03-27 | 2022-03-27 | Construction method of rock burst prediction model driven by expert knowledge and data fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114757266B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115963548B (en) * | 2023-01-16 | 2024-01-23 | 中国矿业大学 | Mine microseismic P wave arrival time pickup model construction method based on deduction learning |
| CN116127752B (en) * | 2023-02-07 | 2023-07-04 | 中国矿业大学 | Rock burst prediction method and system |
| CN116227714B (en) * | 2023-03-14 | 2023-10-27 | 西华大学 | Travel mode selection prediction and analysis method and system |
| CN116755143B (en) * | 2023-06-12 | 2023-11-07 | 中国矿业大学 | Mine microseism energy-frequency compensation method based on microseism system detection probability |
| CN116882302B (en) * | 2023-09-06 | 2023-11-21 | 煤炭科学研究总院有限公司 | Rock burst micro-macro precursor information identification method based on graph representation learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106777772A (en) * | 2017-01-09 | 2017-05-31 | 辽宁工程技术大学 | A kind of Rock Burst hazard prediction method based on coal or rock dynamic system |
| CN113469342A (en) * | 2021-07-08 | 2021-10-01 | 北京科技大学 | Rock burst early warning method based on deep learning microseismic monitoring data |
| CN113723595A (en) * | 2021-09-02 | 2021-11-30 | 重庆大学 | Coal mine rock burst intelligent early warning method based on microseismic event quantitative prediction |
| CN113962466A (en) * | 2021-10-27 | 2022-01-21 | 陕西正通煤业有限责任公司 | Dynamic prediction method for impact risk of rock burst mine fold area based on microseisms |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113779838B (en) * | 2021-09-13 | 2023-04-07 | 中国矿业大学 | Intelligent maintenance method for model and data fusion driven elevator spindle device |
| CN114219211A (en) * | 2021-11-12 | 2022-03-22 | 中国矿业大学 | Rock burst risk assessment method based on theory and data-driven fusion |
| CN113947197B (en) * | 2021-12-20 | 2022-04-26 | 煤炭科学研究总院有限公司 | Micro-seismic event risk prediction method considering rock burst precursor information based on CNN |
-
2022
- 2022-03-27 CN CN202210306659.4A patent/CN114757266B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106777772A (en) * | 2017-01-09 | 2017-05-31 | 辽宁工程技术大学 | A kind of Rock Burst hazard prediction method based on coal or rock dynamic system |
| CN113469342A (en) * | 2021-07-08 | 2021-10-01 | 北京科技大学 | Rock burst early warning method based on deep learning microseismic monitoring data |
| CN113723595A (en) * | 2021-09-02 | 2021-11-30 | 重庆大学 | Coal mine rock burst intelligent early warning method based on microseismic event quantitative prediction |
| CN113962466A (en) * | 2021-10-27 | 2022-01-21 | 陕西正通煤业有限责任公司 | Dynamic prediction method for impact risk of rock burst mine fold area based on microseisms |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114757266A (en) | 2022-07-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114757266B (en) | Construction method of rock burst prediction model driven by expert knowledge and data fusion | |
| Dong et al. | Discriminant models of blasts and seismic events in mine seismology | |
| Dong et al. | Discrimination of mine seismic events and blasts using the fisher classifier, naive bayesian classifier and logistic regression | |
| Yin et al. | Real-time prediction of rockburst intensity using an integrated CNN-Adam-BO algorithm based on microseismic data and its engineering application | |
| CN105260599A (en) | Rockburst dynamic prediction method based on BP neural network modeling | |
| CN113821977B (en) | Rock burst risk assessment system and method for TBM tunnel construction | |
| Niu et al. | Types and occurrence time of rockbursts in tunnel affected by geological conditions and drilling & blasting procedures | |
| Shukla et al. | Prediction and assessment of rock burst using various meta-heuristic approaches | |
| Qiu et al. | Short-term rockburst prediction in underground project: Insights from an explainable and interpretable ensemble learning model | |
| Ji et al. | Investigate contribution of multi-microseismic data to rockburst risk prediction using support vector machine with genetic algorithm | |
| CN116591777B (en) | Multi-field multi-source information fusion rock burst intelligent monitoring and early warning device and method | |
| Xu et al. | Accurate identification of microseismic waveforms based on an improved neural network model | |
| CN115983465A (en) | Rock burst time sequence prediction model construction method based on small sample learning | |
| CN116341901B (en) | Integrated evaluation method for landslide surface domain-monomer hazard early warning | |
| Saxena et al. | Comparative Analysis Of AI Regression And Classification Models For Predicting House Damages İn Nepal: Proposed Architectures And Techniques | |
| Gong et al. | Prediction and evaluation of coal mine coal bump based on improved deep neural network | |
| Zhao et al. | A hybrid recognition model of microseismic signals for underground mining based on CNN and LSTM networks | |
| CN115688251A (en) | Earthquake multi-occurrence-zone tunnel risk decision method and system based on deep learning | |
| Zhou et al. | Earthquake prediction model based on danger theory in artificial immunity | |
| Li et al. | A rockburst prediction model based on extreme learning machine with improved Harris Hawks optimization and its application | |
| Qiu et al. | Analysis and prediction of rockburst intensity using improved DS evidence theory based on multiple machine learning algorithms | |
| CN115966074B (en) | Machine learning-based shield cutter disc mud cake early warning method and system | |
| Li et al. | Waveform recognition and process interpretation of microseismic monitoring based on an improved LeNet5 convolutional neural network | |
| Wang et al. | An automatic recognition method of microseismic signals based on s transformation and improved Gaussian mixture model | |
| Qiu et al. | Risk prediction of coal and gas outburst based on abnormal gas concentration in blasting driving face |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |