CN114692685A - System and method for predicting movement time-space law of overlying rock mass in mine goaf - Google Patents

System and method for predicting movement time-space law of overlying rock mass in mine goaf Download PDF

Info

Publication number
CN114692685A
CN114692685A CN202210311934.1A CN202210311934A CN114692685A CN 114692685 A CN114692685 A CN 114692685A CN 202210311934 A CN202210311934 A CN 202210311934A CN 114692685 A CN114692685 A CN 114692685A
Authority
CN
China
Prior art keywords
rock mass
mine goaf
time
space
law
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.)
Pending
Application number
CN202210311934.1A
Other languages
Chinese (zh)
Inventor
张俊文
赵利民
陈亚萍
钟帅
李积星
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Lanzhou Resources and Environment Voc Tech College
Original Assignee
Lanzhou Resources and Environment Voc Tech College
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Lanzhou Resources and Environment Voc Tech College filed Critical Lanzhou Resources and Environment Voc Tech College
Priority to CN202210311934.1A priority Critical patent/CN114692685A/en
Publication of CN114692685A publication Critical patent/CN114692685A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction

Abstract

The invention belongs to the technical field of mining and discloses a prediction method and a prediction system for the movement space-time law of overlying rock mass in a mine goaf. When the GMM-UBM model is trained, the data are screened according to the characteristics of phonemes and tones of the time-space law data of the overburden rock mass in the mine goaf, and a small representative amount of data is selected, so that the representation generalization of the model is ensured, the data operation amount is reduced, and the modeling efficiency is improved; firstly, a GMM-UBM model is trained, and then parameters of the GMM are adjusted by using an MAP adaptive algorithm, so that the problems that the sample size is small and the GMM model cannot be trained are solved, and the operation speed is accelerated.

Description

Mine goaf overlying rock mass movement space-time law prediction system and method
Technical Field
The invention belongs to the technical field of mining, and particularly relates to a method and a system for predicting the movement time-space law of an overlying rock mass in a mine goaf.
Background
At present, three-dimensional stress balance of surrounding rocks is damaged due to underground mining and tunnel excavation of a mine, overlying strata movement, surface subsidence and even collapse are easily caused, collapse of a goaf seriously threatens safety production, and meanwhile, adverse effects are caused on surface buildings and the environment. For the underground goaf overlying strata movement, mainly a rock stratum movement three-zone theory, a key layer theory, a supporting plate theory, a masonry beam theory and the like, a series of prediction methods for the overlying strata movement are formed on the basis, and a common probability integration method, a numerical simulation method, a similar model method, a random medium method, a neural network prediction and the like are adopted. At present, a plurality of research results aiming at the rock stratum movement of the coal mine achieve better practical application effects. The overburden moving subsidence of the mine has the sudden property, and most of the overburden moving subsidence is discontinuous subsidence, such as subsidence pit, barrel-shaped, tubular or funnel-shaped subsidence, and the subsidence of the coal mine different from the layered stratum has the slow degeneration. The potential safety hazard is large when the in-situ actual measurement or monitoring is carried out inside the overlying rock mass of the underground goaf of the mine, and the instability of the overlying rock mass can possibly collapse at any time to cause great threat to personnel and equipment. In the prior art, a method for realizing the prediction of the movement time-space law of the overlying rock mass of the mine goaf is unavailable, and the safe and accurate prediction of the movement of the overlying rock mass of the mine goaf cannot be realized.
Through the above analysis, the problems and defects of the prior art are as follows: in the prior art, a method for realizing the prediction of the movement time-space law of the overlying rock mass of the mine goaf is unavailable, and the safe and accurate prediction of the movement of the overlying rock mass of the mine goaf cannot be realized.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a method and a system for predicting the movement time-space law of an overlying rock mass in a mine goaf. With the development of deep learning and neural networks, the neural network model can obtain better effect under the condition of limited data samples, and the operation speed is also favorably ensured.
The invention identifies and classifies the equipment source based on the Gaussian super vector characteristics of the equipment information, and MFCC characteristics are firstly used in the process of extracting the flat Gaussian vector. Secondly, a UBM-GMM model is used as a reference to model a recording device channel, and then a MAP algorithm is used for self-adaptive adjustment, so that the speed of feature extraction is increased, the characteristic features of the features are improved, the data volume is reduced, and the defect that a large amount of data is needed for training a multivariate GMM model is overcome.
The invention is realized in such a way, and provides a method for predicting the movement time-space law of an overlying rock mass in a mine goaf, which comprises the following steps:
extracting MFCC characteristics of a time-space regular segment for training the movement of the overlying rock mass in the mine goaf to train a GMM-UBM model;
extracting MFCC characteristics based on specific mine goaf overburden rock mass movement space-time regular segments with abnormal information, and further adjusting GMM parameters;
and training the convolutional neural network by using the extracted features, and automatically identifying and classifying.
Further, the method for predicting the movement time-space law of the overlying rock mass in the mine goaf specifically comprises the following steps:
the method comprises the following steps: preprocessing signals of space-time laws for training movement of overlying rocks of the mine goaf to extract characteristic information;
step two: training a GMM-UBM model by utilizing the movement time-space law of the overburden rock mass of the trained mine goaf;
step three: preprocessing the moving time-space law signals of the overlying rock mass in the target mine goaf to extract characteristic information;
step four: calculating a GMM model specific to the moving time-space law signal of the overlying rock mass in the target mine goaf;
step five: extracting a Gaussian component;
step six: the convolutional neural network is trained and the model is tested.
Further, the first step specifically comprises:
step 1.1: screening signals for training the movement time-space law of the overlying rock mass in the mine goaf;
step 1.2: performing short-time Fourier transform on the trained moving time-space regular signals of the overlying rock mass of the mine goaf;
step 1.3: extracting characteristics of the frequency domain signals in the step 1.2;
the second step specifically comprises;
step 2.1: training a GMM-UBM model by using the characteristics obtained in the step 1.3;
the third step specifically comprises:
step 3.1: performing short-time Fourier transform on the moving time-space law of the overlying rock mass in the target mine goaf;
step 3.2: extracting characteristics of the frequency domain signals in the step 3.1;
the fourth step specifically comprises:
step 4.1: adjusting GMM model parameters of the features obtained in the step 3.2 through the GMM model trained in the step 1.3 and an MAP algorithm;
the fifth step specifically comprises:
step 5.1: extracting the mean parameter of each GMM trained in the step 4.1 as a characteristic signal;
the sixth step specifically comprises:
step 6.1: dividing the characteristic data into training data and testing data;
step 6.2: marking and classifying the characteristic data, and marking by using one-hot coding;
step 6.3: preprocessing the characteristic data;
step 6.4: training the constructed convolutional neural network by using the training data;
step 6.5: and performing test evaluation on the trained network by using the test data.
Further, the step one further comprises:
windowing and framing signals of the moving time-space law of the overlying rock mass of the trained mine goaf, and then performing short-time Fourier transform; the window length of the windowing is set to 256, the frame shift is 128, and then fourier transform is performed;
extracting characteristics of the medium-frequency domain signals, namely extracting MFCC characteristics of space-time regular signals of overburden rock mass movement of a mine goaf, selecting 12 coefficients and adding energy of F0, and simultaneously keeping first-order and second-order coefficients to obtain 39-dimensional data in total;
the second step further comprises: firstly, training a most basic recording GMM as a universal background model UBM; a GMM model with M gaussians and D-dimensional feature data is represented as follows:
Figure 100002_DEST_PATH_IMAGE001
；
wherein x is an input N x D dimension feature vector;
Figure 997622DEST_PATH_IMAGE002
，
Figure 100002_DEST_PATH_IMAGE003
is the weight of the Gaussian mixture model, and satisfies
Figure 331914DEST_PATH_IMAGE004
Finally obtaining the size of the parameter D x 1 dimension;
Figure 100002_DEST_PATH_IMAGE005
the mean value of each Gaussian component is M x D dimension;
Figure 754805DEST_PATH_IMAGE006
the variance of each Gaussian component is M x D dimension;
Figure 100002_DEST_PATH_IMAGE007
is the probability density of each Gaussian model and satisfies
Figure 938661DEST_PATH_IMAGE008
，
Figure 100002_DEST_PATH_IMAGE009
Is composed of
Figure 839621DEST_PATH_IMAGE010
The transpose of (a) is performed,
Figure 100002_DEST_PATH_IMAGE011
and
Figure 792534DEST_PATH_IMAGE012
is composed of
Figure 100002_DEST_PATH_IMAGE013
And (4) training the GMM-UBM model by using an EM algorithm.
Further, the fourth step further comprises: the adjustment of the mean matrix of the Gaussian mixture model specifically comprises the following steps:
Figure 884861DEST_PATH_IMAGE014
；
Figure 100002_DEST_PATH_IMAGE015
；
Figure 290435DEST_PATH_IMAGE016
；
Figure 100002_DEST_PATH_IMAGE017
；
Figure 260665DEST_PATH_IMAGE018
；
in the above formula
Figure 100002_DEST_PATH_IMAGE019
For adjusting parameters
Figure 68084DEST_PATH_IMAGE020
Satisfy the requirement of
Figure 100002_DEST_PATH_IMAGE021
To do so
Figure 832778DEST_PATH_IMAGE022
、
Figure 100002_DEST_PATH_IMAGE023
、
Figure 227112DEST_PATH_IMAGE024
Is an adaptive coefficient for adjusting the parameter, and
Figure 100002_DEST_PATH_IMAGE025
wherein
Figure 469875DEST_PATH_IMAGE026
Also adaptive coefficients;
the fifth step further comprises: extracting a Gaussian component;
and extracting the mean parameter of the mixed Gaussian model subjected to MAP self-adaptive adjustment to serve as characteristic data of the time-space law signal of the movement of the overlying rock mass of the mine goaf. The GMM model obtained after MAP self-adaptive training has three models, namely, the mean value, the variance and the weight, the size D of a mean matrix is L, D is the Gaussian quantity, L is the frame number, and finally the mean matrix is required to be converted into a one-bit matrix of { N1, N2, N3 … … };
after the mean parameter is extracted, the data needs to be normalized by the standard deviation, as shown in the following formula, where u is the mean of all sample data, and σ is the standard deviation of all sample data.
Figure 100002_DEST_PATH_IMAGE027
。
The invention also aims to provide a method for predicting the movement space-time law of the overlying rock mass in the mine goaf.
The invention also aims to provide a terminal, which at least carries a controller for realizing the prediction method of the movement space-time law of the overlying rock mass in the mine goaf.
Another object of the present invention is to provide a computer-readable storage medium, which includes instructions that, when executed on a computer, cause the computer to execute the method for predicting the movement spatiotemporal regularity of the overburden rock in the mine goaf.
The invention also aims to provide a system for predicting and controlling the movement space-time law of the overlying rock mass in the mine goaf, which comprises the following steps:
the characteristic information extraction module is used for preprocessing signals of the time-space law of the movement of the overlying rock mass of the training mine goaf to extract characteristic information;
the GMM-UBM model training module is used for training a GMM-UBM model by utilizing the movement time-space law of the overburden rock mass of the trained mine goaf;
the device comprises a target mine goaf overburden rock mass movement space-time law signal characteristic information extraction module, a pre-processing module and a data processing module, wherein the target mine goaf overburden rock mass movement space-time law signal characteristic information extraction module is used for pre-processing overburden rock mass movement space-time law signals of the target mine goaf to extract characteristic information;
the GMM model calculation module is used for calculating a GMM model specific to the movement time-space law signal of the overlying rock mass in the target mine goaf;
the Gaussian component extraction module is used for extracting a Gaussian component;
and the convolutional neural network training module is used for training the convolutional neural network and testing the model.
The invention also aims to provide the evidence obtaining equipment for the movement time-space law of the overlying rock mass in the mine goaf, wherein the evidence obtaining equipment for the movement time-space law of the overlying rock mass in the mine goaf is at least provided with the prediction control system for the movement time-space law of the overlying rock mass in the mine goaf.
In summary, the advantages and positive effects of the invention are:
(1) when the GMM-UBM model is trained, the data are screened according to the characteristics of the phonemes and the tones of the time-space law data of the overburden rock mass in the mine goaf, and a small representative amount of data is selected, so that the representation generalization of the model is ensured, the data computation amount is reduced, and the modeling efficiency is improved.
(2) The method firstly trains a GMM-UBM model, then adjusts the parameters of the GMM by using an MAP adaptive algorithm, overcomes the problems that the sample size is small and the GMM model cannot be trained, and simultaneously accelerates the operation speed.
(3) The method realizes the identification effect through the convolutional neural network, improves the identification accuracy, and finally obtains the identification precision of 92.1% in a closed set experiment of 21 equipment sources, while the identification precision of the existing mobile equipment source is only 82%.
(4) In order to improve the accuracy, in many current technical methods, characteristic information is extracted by using a non-mine goaf overburden rock mass movement space-time regular segment (a talking mine goaf overburden rock mass movement space-time regular segment) of a mine goaf overburden rock mass movement space-time regular segment, and then classification and identification are carried out, the non-mine goaf overburden rock mass movement space-time regular segment has no interference of other sounds, and only equipment abnormal information is reserved, so that the representation of an equipment source is higher, but in the actual application process, no more non-mine goaf overburden rock mass movement space-time regular segments can be used. The characteristics are extracted by using the time-space law section of the movement of the overburden rock mass in the whole mine goaf, and the generalization performance is improved.
Drawings
FIG. 1 is a flow chart of a method for predicting the movement spatiotemporal regularity of an overburden rock mass in a mine goaf, which is provided by the embodiment of the invention.
FIG. 2 is a schematic diagram of a prediction control system for the movement spatiotemporal regularity of an overburden rock mass in a mine goaf, according to an embodiment of the invention.
In the figure: 1. a characteristic information extraction module; 2. A GMM-UBM model training module; 3. the target mine goaf overlying rock mass movement space-time law signal characteristic information extraction module; 4. a GMM model calculation module; 5. a Gaussian component extraction module; 6. and a convolutional neural network training module.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention relates to an identification method based on a GMM-UBM general background model and a convolutional neural network. Firstly extracting MFCC characteristics of the overlying rock mass movement space-time law segment of the training mine goaf to train a GMM-UBM model, then extracting MFCC characteristics based on the specific overlying rock mass movement space-time law segment of the mine goaf with abnormal information, and further adjusting parameters of the GMM. And finally, the extracted features are used for training a convolutional neural network to meet the requirement of automatic identification and classification.
Referring to fig. 1, the method for predicting the movement spatiotemporal regularity of the overburden rock mass in the mine goaf provided by the embodiment of the invention specifically comprises the following steps:
s101, preprocessing signals of the space-time law of the movement of the overlying rock mass in the training mine goaf to extract characteristic information;
s102, training a GMM-UBM model by utilizing the movement time-space law of the overburden rock mass of the trained mine goaf;
s103, preprocessing the moving time-space law signals of the overlying rock mass in the target mine goaf to extract characteristic information;
s104, calculating a GMM model specific to the moving time-space regular signals of the overlying rock mass of the target mine goaf;
s105, extracting a Gaussian component;
and S106, training the convolutional neural network and testing the model.
The method for predicting the movement space-time law of the overlying rock mass in the mine goaf, provided by the embodiment of the invention, specifically comprises the following steps:
step 1: preprocessing signals of a space-time law for training the movement of an overlying rock mass in a mine goaf to extract characteristic information;
step 1.1: and (3) windowing and framing signals of the moving time-space law of the overlying rock mass in the trained mine goaf, and then performing short-time Fourier transform. In order to control the amount of data, the window length of windowing is set to 256, the frame shift is 128, and then fourier transform is performed;
step 1.2: for the frequency domain signal extraction features in the step 1.1, firstly, the MFCC features of the space-time regular signals of the overburden rock mass in the mine goaf are extracted, 12 coefficients are selected, the energy of F0 is added, and meanwhile, first-order and second-order coefficients are kept, so that 39-dimensional data are obtained in total.
Step 2: training a GMM-UBM model by utilizing the movement time-space law of the overburden rock mass of the trained mine goaf;
step 2.1: in the above, it has been explained that the data of the space-time law of the overburden rock mass movement in the mine goaf of 6min can only be used for training a GMM model with less than 64 gausses, and 128 gauss trees are used in the invention, so the sample size is far insufficient, and therefore a general background model needs to be trained first. A Gaussian Mixture Model (GMM) is a probability distribution model that is combined by a plurality of weighted gaussian models. In a natural situation, the distribution of data is generally distributed according to a gaussian model, however, one gaussian model cannot fit one multidimensional data, so that a plurality of gaussians can be used for weighting to represent the probability distribution of one data sample, and when the types of data are the same, the patterns of the probability models are approximately the same, and the overall response is reflected on parameters of the mean, the variance and the weight of the models. Therefore, the GMM model can reflect the probability distribution of certain data, and further can be used for identifying and classifying the features, and when the features are properly extracted, the GMM model can be used for identifying the equipment source of the sound recording. However, training a GMM model requires a large amount of data, and in reality, a large amount of data cannot be directly acquired, so that a most basic recording GMM model can be trained first as a Universal Background Model (UBM). A GMM model with M gaussians and D-dimensional feature data can be expressed as follows:
Figure 397379DEST_PATH_IMAGE001
；
wherein x is an input feature vector with dimension of N x D;
Figure 801816DEST_PATH_IMAGE002
，
Figure 181982DEST_PATH_IMAGE003
is the weight of the Gaussian mixture model, and satisfies
Figure 962856DEST_PATH_IMAGE004
The finally obtained parameter size is D x 1 dimension;
Figure 479288DEST_PATH_IMAGE005
the mean value of each Gaussian component is M x D dimension;
Figure 54626DEST_PATH_IMAGE006
the variance of each Gaussian component is M x D dimension;
Figure 444060DEST_PATH_IMAGE007
is the probability density of each Gaussian model and satisfies
Figure 28625DEST_PATH_IMAGE008
，
Figure 399564DEST_PATH_IMAGE009
Is composed of
Figure 145803DEST_PATH_IMAGE010
The transpose of (a) is performed,
Figure 969402DEST_PATH_IMAGE011
and
Figure 357658DEST_PATH_IMAGE012
is composed of
Figure 317524DEST_PATH_IMAGE013
Determinant and inverse matrix of (c). In the actual operation process, in order to reduce the number of parameters and increase the training rate, a diagonalized covariance matrix is generally used, and experiments prove that the diagonalized covariance matrix is sufficient to approximate.
The training process uses an EM algorithm, and the detailed steps are as follows:
1. initialization
Figure 500244DEST_PATH_IMAGE028
，
Figure DEST_PATH_IMAGE029
，
Figure 843763DEST_PATH_IMAGE030
And initializing a log-likelihood function.
2. Probability is estimated using current parameters:
Figure DEST_PATH_IMAGE031
；
3. updating the current parameters:
Figure 35710DEST_PATH_IMAGE032
；
Figure DEST_PATH_IMAGE033
；
Figure 381240DEST_PATH_IMAGE034
；
Figure DEST_PATH_IMAGE035
；
4. calculating a log-likelihood function:
Figure 266020DEST_PATH_IMAGE036
;
comparing the log-likelihood function or the limiting condition, if the condition is not satisfied, jumping to the step 2 to continue circulating until the condition is satisfied.
And step 3: preprocessing the moving time-space law signals of the overlying rock mass in the target mine goaf to extract characteristic information;
step 3.1: windowing and framing signals of the moving time-space law of the overlying rock mass in the mine goaf, preprocessing the signals, and then performing short-time Fourier transform. The preprocessing mode is synchronized with step 1, the window length is set to 256, and the frame shift is 128. Then carrying out short-time Fourier transform;
step 3.2: extracting characteristics of the frequency domain signals in the step 3.1, and selecting parameters in the same step 1.2;
and 4, step 4: calculating a GMM model specific to the moving time-space law signal of the overlying rock mass in the target mine goaf;
step 4.1: after GMM-UBM training is completed, only fine adjustment needs to be performed on the basis of an original general model, because the general background model reflects the most basic information of the time-space law of movement of overlying rocks of mine goafs of the recording equipment, and for target equipment, only a small amount of characteristics need to be used for training, and a unique relative difference between each model can be reflected. In this process, the MAP (maximum a posteriori) maximum posterior probability algorithm is needed to be used for adjustment.
However, in the research, the influence of the weight and the variance matrix on the mixed gaussian model is not very large, and the most important is the mean matrix of the mixed gaussian model, so in order to improve the training efficiency, only the adjustment of the mean matrix is considered in the application process.
Figure 798632DEST_PATH_IMAGE014
；
Figure 263112DEST_PATH_IMAGE015
；
Figure 197570DEST_PATH_IMAGE016
；
Figure 486206DEST_PATH_IMAGE017
；
Figure DEST_PATH_IMAGE037
；
In the above formula
Figure 771694DEST_PATH_IMAGE019
For adjusting parameters
Figure 39864DEST_PATH_IMAGE020
Satisfy the requirement of
Figure 94408DEST_PATH_IMAGE021
To do so
Figure 789832DEST_PATH_IMAGE022
、
Figure 297036DEST_PATH_IMAGE023
、
Figure 103318DEST_PATH_IMAGE024
Is an adaptive coefficient for adjusting the parameter, and
Figure 12368DEST_PATH_IMAGE025
wherein
Figure 144272DEST_PATH_IMAGE026
Also adaptive coefficients.
And 5: extracting a Gaussian component;
step 5.1: and extracting the mean parameter of the mixed Gaussian model subjected to MAP self-adaptive adjustment to serve as characteristic data of the time-space law signal of the movement of the overlying rock mass of the mine goaf. The GMM model obtained after MAP self-adaptive training has three values, namely, the mean value, the variance and the weight, the size D of the mean matrix is L, D is the Gaussian quantity, L is the frame number, and finally the mean matrix is required to be converted into a one-bit matrix of { N1, N2, N3 … … }.
Step 5.2: the mean parameter can not be directly used after being extracted, the data needs to be normalized, and the system adopts standard deviation standardization as shown in the following formula, wherein u is the mean value of all sample data, and sigma is the standard deviation of all sample data.
Figure 873194DEST_PATH_IMAGE027
。
Step 6: training a convolutional neural network and testing a model;
step 6.1: dividing the characteristic data into training data and testing data, dividing the training data into training samples and verification samples, using the training samples to train a convolutional neural network model, then using the verification data to check the quality of the model, and simultaneously adjusting the parameters of the network model, wherein the final testing data set is used for testing the quality of the final deep self-coding model;
step 6.2: the method comprises the steps that one-hot coding classification is carried out on a feature data set, in a classification model, a cross entropy mode is used in the final prediction, so that the probability is calculated, each coding mode is used for representing a class, automatic identification can be conveniently carried out to achieve the purpose of classification, and one-hot coding only has one bit being 1 for each class of hot coding, and the other bits are marked by 0;
step 6.3: for training data preprocessing, the feature data was constructed to fit the CNN model in a square data size using 0 padding, with the raw data being 128 × 39=4992 padded to 5041 and then changed to 71 × 71 square matrix.
Step 6.4: training the constructed convolutional neural network by using the training data;
the convolutional neural network is a convolution-based deep learning neural network. Compared with a fully-connected neural network, the method greatly reduces the number of parameters, so that the running speed of the neural network is greatly improved, and the neural network is more efficient in both training and inspection. However, in each layer of propagation from the input layer to the output layer, the neural network is a result obtained by performing high-layer extraction (deep extraction of features) on the original data features, and compared with the previous layer of data, the convolutional layer reduces the size of the data, increases the number of channels, performs deep extraction on the input data features, and obtains features with higher abstraction layer degree by performing more deep analysis on each small block of the original data.
Including an input layer in a convolutional neural network; a convolution layer; a pooling layer; a fully-connected layer; and (6) an output layer.
(one) input layer (or pooling layer) to the convolutional layer
In the process of transferring to the convolutional layer, the input layer or the pooling layer is a process of highly sampling a characteristic local area and increasing the number of channels. Is provided with a filter with a size of
Figure 250211DEST_PATH_IMAGE038
(where k is the number of channels,
Figure DEST_PATH_IMAGE039
also two-dimensional convolution kernel size) is passed to the convolution layer to form a node:
Figure 544926DEST_PATH_IMAGE040
；
wherein
Figure DEST_PATH_IMAGE041
Denotes the ith channel of the convolutional layer, see k ' denotes the k ' channels (k ' ≦ k) associated with the convolutional layer among the k channels of the input layer (or pooling layer),
Figure 113311DEST_PATH_IMAGE042
a convolution kernel representing a first requirement of the convolution layer, having k' convolution kernel matrices each having a size of
Figure 595108DEST_PATH_IMAGE039
。
Figure DEST_PATH_IMAGE043
The jth input (depending on the value of the input layer or pooling layer and the step size) g (i, j) of the input layer (or pooling layer) represents the specific value of the jth value of the ith channel.
(ii) convolution layer to pooling layer
The pooling layer is a correlated sampling operation performed to reduce the amount of parameters. For example, in one downsampling example, the following relationship is satisfied:
Figure 274351DEST_PATH_IMAGE044
；
wherein
Figure DEST_PATH_IMAGE045
Representing the j-th layer feature mapping after pooling the l-1 layer, while in the LeNet-5 model, the pooling layer related parameters are specialized,
Figure 423572DEST_PATH_IMAGE046
，
Figure DEST_PATH_IMAGE047
f is a linear function with a coefficient of 1, so the pooling process becomes relatively simple only
Figure 395814DEST_PATH_IMAGE048
Such pooling operation.
(III) all connected to the output layer
The layer mainly realizes the process of classifying or merging the output results, is a classification problem in a LeNet-5 model, forms a distribution column with the sum of 1 through a layer of Softmax layer, and judges the classification condition by forming a one-hot vector with the length of 10 through backward propagation and parameter adjustment.
In this system, the signature data size is 71 × 71, the constructed network is 8 layers, the convolution of the convolution kernel of 5 × 5 of 71 × 71 data is changed to 67 × 6, the pooling of the second step of 2 × 2 is changed to 34 × 6, the convolution of the convolution kernel of 5 × 5 is changed to 30 × 30 18 in the third step, the pooling of the kernel of 2 × 2 is changed to 15 × 18 in the fourth step, the convolution kernel of the fifth step of 5 × 5 is changed to 11 × 54, the pooling of the sixth step of 2 × 2 is changed to 6 × 54, the pooling of the seventh step of 6 × 6 is changed to 540 full connected layers of nodes, and the final connected layer is 21-dimensional data.
Step 6.5: and performing test evaluation on the trained network by using the test data.
After the model is trained, it is desirable to know the quality of the model, so that the model needs to be tested. To avoid the impact on the final accuracy of the model due to participation in the network structure training, the model is tested using additional data not participating in the training.
As shown in fig. 2, the prediction control system for the movement spatiotemporal regularity of the overburden rock mass in the mine goaf provided by the embodiment of the invention comprises:
the characteristic information extraction module 1 is used for preprocessing signals of the space-time law of movement of the overlying rock mass in the mine goaf to be trained and extracting characteristic information;
the GMM-UBM model training module 2 is used for training a GMM-UBM model by utilizing the movement time-space law of the overburden rock mass of the trained mine goaf;
the characteristic information extraction module 3 of the overburden rock mass movement time-space law signal of the target mine goaf is used for preprocessing the overburden rock mass movement time-space law signal of the target mine goaf to extract characteristic information;
the GMM model calculation module 4 is used for calculating a GMM model specific to the moving time-space law signal of the overlying rock mass of the target mine goaf;
a gaussian component extracting module 5, configured to extract a gaussian component;
and the convolutional neural network training module 6 is used for training the convolutional neural network and testing the model.
The invention is further described below in connection with the experiments.
In the experiment, a TIMIT and MOBIBPHONE database and a UBM training stage are used in the method, all training data of the TIMIT database are selected to extract MFCC characteristics to train a GMM model, a frame length is 256, a frame shift 128 is used to extract 12-order MFCC parameters plus first-order, second-order and zero-point parameters, and a universal background model UBM containing 128 Gaussian components is trained after normalization.
And a GMM model adjusting stage, selecting a data set of 21 devices (each device selects 96 mine goaf overlying rock mass movement space-time rule samples) of a MOBOPONE database training set, extracting 12-order MFCC parameters plus first-order, second-order and zero-point parameters by taking 256 frame lengths and frame movements 128 as frame lengths, respectively using the MFCC characteristics for training a GMM-UBM model to obtain 21 GMM models, and extracting the mean value of the 21 GMM models to obtain the Gaussian super vector of the training set. The test set operates in the same manner.
In the neural network training stage, 21 × 96 Gaussian supervectors obtained from the MOBOPONE training set are used as a training set, meanwhile, a one-hot coding form is adopted to code and mark 21 types of data, a CNN neural network is trained, and then the data of the test set is used for detection and evaluation.
Eight-layer networks (four convolutional layers, three pooling layers, and one full-link layer) were constructed in this experiment, respectively. Original data 0 is filled with 71 × 71, the convolution kernel step size of the first hidden layer 5 × 5 is 1 channel number, 6 is changed, and the dimension is 67 × 6; a second hidden layer 2 x 2 pooled core with a post-pooling dimension of 34 x 6; the third hidden layer 5 is characterized in that the step size of a convolution kernel is 1, the number of channels is changed from 6 to 18, and the dimension is 30-18; a fourth hidden layer 2 x 2 pooled nucleus with a post-pooling dimension of 15 x 18; the step size of a convolution kernel of a fifth hidden layer 5 by 5 is 1, the number of channels is changed from 18 to 54, and the dimension is 11 by 54; a sixth hidden layer 2 x 2 pooled nuclei with a post-pooling dimension of 6 x 54; the number of channels of the seventh hidden layer 6 x 6 convolution kernel with the step size of 1 is changed from 54 to 540; the eighth layer outputs 21 classification results by softmax.
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When used in whole or in part, can be implemented in a computer program product that includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, the procedures or functions according to the embodiments of the present invention are wholly or partially generated. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions may be transmitted from one website site, computer, server, or data center to another website site, computer, server, or data center via wire (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL), or wireless (e.g., infrared, wireless, microwave, etc.)). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that includes one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims (10)

1. A mine goaf overlying rock mass movement space-time law prediction method is characterized by comprising the following steps:
extracting MFCC characteristic training GMM-UBM models for training the time-space regular segments of the movement of the overlying rocks of the mine goaf;
extracting MFCC characteristics based on specific mine goaf overburden rock mass movement space-time regular segments with abnormal information, and adjusting GMM parameters;
and training the convolutional neural network by using the extracted features, and automatically identifying and classifying.
2. The method for predicting the movement spatiotemporal regularity of the overburden rock mass in the mine goaf according to claim 1, wherein the method for predicting the movement spatiotemporal regularity of the overburden rock mass in the mine goaf specifically comprises the following steps:
the method comprises the following steps: preprocessing signals of a space-time law for training the movement of an overlying rock mass in a mine goaf to extract characteristic information;
step two: training a GMM-UBM model by utilizing the movement time-space law of the overburden rock mass of the trained mine goaf;
step three: preprocessing the moving time-space law signals of the overlying rock mass in the target mine goaf to extract characteristic information;
step four: calculating a GMM model specific to the moving time-space law signal of the overlying rock mass in the target mine goaf;
step five: extracting a Gaussian component;
step six: the convolutional neural network is trained and the model is tested.
3. The mine goaf overburden rock mass movement spatiotemporal law prediction method according to claim 2, wherein the first step specifically comprises:
step 1.1: screening signals for training the movement time-space law of the overlying rock mass in the mine goaf;
step 1.2: performing short-time Fourier transform on the trained moving time-space regular signals of the overlying rock mass of the mine goaf;
step 1.3: extracting characteristics of the frequency domain signals in the step 1.2;
the second step specifically comprises;
step 2.1: training a GMM-UBM model by using the characteristics obtained in the step 1.3;
the third step specifically comprises:
step 3.1: performing short-time Fourier transform on the moving time-space law of the overlying rock mass in the target mine goaf;
step 3.2: extracting characteristics of the frequency domain signals in the step 3.1;
the fourth step specifically comprises:
step 4.1: adjusting GMM model parameters of the features obtained in the step 3.2 through the GMM model trained in the step 1.3 and an MAP algorithm;
the fifth step specifically comprises:
step 5.1: extracting the mean parameter of each GMM model trained in the step 4.1 as a characteristic signal;
the sixth step specifically comprises:
step 6.1: dividing the characteristic data into training data and testing data;
step 6.2: marking and classifying the characteristic data, and marking by using one-hot coding;
step 6.3: preprocessing the characteristic data;
step 6.4: training the constructed convolutional neural network by using the training data;
step 6.5: and performing test evaluation on the trained network by using the test data.
4. The mine goaf overburden rock mass movement spatiotemporal law prediction method according to claim 3, wherein the first step further comprises:
windowing and framing signals of the moving time-space law of the overlying rock mass of the trained mine goaf, and then performing short-time Fourier transform; the window length of the windowing is set to 256, the frame shift is 128, and then fourier transform is performed;
extracting characteristics of the medium-frequency domain signals, namely extracting MFCC characteristics of space-time regular signals of overburden rock mass movement of a mine goaf, selecting 12 coefficients and adding energy of F0, and simultaneously keeping first-order and second-order coefficients to obtain 39-dimensional data in total;
the second step further comprises: firstly, training a most basic recording GMM model as a universal background model UBM; a GMM model with M gaussians and D-dimensional feature data is represented as follows:
Figure DEST_PATH_IMAGE001
wherein x is an input feature vector with dimension of N x D;
Figure DEST_PATH_IMAGE002
，
Figure DEST_PATH_IMAGE003
is the weight of a Gaussian mixture model and satisfies
Figure DEST_PATH_IMAGE004
The finally obtained parameter size is D x 1 dimension;
Figure DEST_PATH_IMAGE005
the mean value of each Gaussian component is M x D dimension;
Figure DEST_PATH_IMAGE006
the variance of each Gaussian component is M x D dimension;
Figure DEST_PATH_IMAGE007
is the probability density of each Gaussian model and satisfies
Figure DEST_PATH_IMAGE008
，
Figure DEST_PATH_IMAGE009
Is composed of
Figure DEST_PATH_IMAGE010
The transpose of (a) is performed,
Figure DEST_PATH_IMAGE011
and
Figure DEST_PATH_IMAGE012
is composed of
Figure DEST_PATH_IMAGE013
And (4) training the GMM-UBM model by using an EM algorithm.
5. The mine goaf overburden rock mass movement spatiotemporal law prediction method according to claim 3, wherein the fourth step further comprises: the adjustment of the mean matrix of the Gaussian mixture model specifically comprises the following steps:
Figure DEST_PATH_IMAGE014
；
Figure DEST_PATH_IMAGE015
；
Figure DEST_PATH_IMAGE016
；
Figure DEST_PATH_IMAGE017
；
Figure DEST_PATH_IMAGE018
；
in the above formula
Figure DEST_PATH_IMAGE019
For adjusting parameters
Figure DEST_PATH_IMAGE020
Satisfy the requirement of
Figure DEST_PATH_IMAGE021
To do so
Figure DEST_PATH_IMAGE022
、
Figure DEST_PATH_IMAGE023
、
Figure DEST_PATH_IMAGE024
Is an adaptive coefficient for adjusting the parameter, and
Figure DEST_PATH_IMAGE025
wherein
Figure DEST_PATH_IMAGE026
Also adaptive coefficients;
the fifth step further comprises: extracting a Gaussian component;
extracting the mean parameter of the mixed Gaussian model subjected to MAP self-adaptive adjustment to serve as characteristic data of a time-space law signal of overburden rock mass movement of the mine goaf; the GMM model obtained after MAP self-adaptive training has three models, namely, the mean value, the variance and the weight, the size D of a mean matrix is L, D is the Gaussian quantity, L is the frame number, and finally the mean matrix is required to be converted into a one-bit matrix of { N1, N2, N3 … … };
after the mean parameter is extracted, normalization is performed on the data, and standard deviation normalization is adopted, wherein u is the mean of all sample data, and sigma is the standard deviation of all sample data;
Figure DEST_PATH_IMAGE027
。
6. a computer program for predicting the movement time-space law of an overlying rock mass in a mine goaf is characterized in that the computer program for predicting the movement time-space law of the overlying rock mass in the mine goaf realizes the method for predicting the movement time-space law of the overlying rock mass in the mine goaf according to any one of claims 1 to 5.
7. A terminal is characterized in that the terminal is at least provided with a controller for realizing the prediction method of the movement space-time law of the overlying rock mass in the mine goaf according to any one of claims 1 to 5.
8. A computer-readable storage medium comprising instructions which, when executed on a computer, cause the computer to perform the method of predicting the movement of the overburden of rock mass in a mine gob of any one of claims 1 to 5.
9. A mine goaf overburden rock mass movement space-time law prediction control system for realizing the mine goaf overburden rock mass movement space-time law prediction method of claim 1, wherein the mine goaf overburden rock mass movement space-time law prediction control system comprises:
the characteristic information extraction module is used for preprocessing signals of the time-space law of the movement of the overlying rock mass of the training mine goaf to extract characteristic information;
the GMM-UBM model training module is used for training a GMM-UBM model by utilizing the movement time-space law of the overburden rock mass of the trained mine goaf;
the target mine goaf overlying rock mass movement space-time law signal characteristic information extraction module is used for preprocessing the target mine goaf overlying rock mass movement space-time law signal to extract characteristic information;
the GMM model calculation module is used for calculating a GMM model with specific time-space regular signals of the movement of the overlying rock mass of the target mine goaf;
the Gaussian component extraction module is used for extracting a Gaussian component;
and the convolutional neural network training module is used for training the convolutional neural network and testing the model.
10. The mine goaf overburden rock mass movement space-time law evidence obtaining equipment is characterized in that the mine goaf overburden rock mass movement space-time law evidence obtaining equipment is at least provided with the mine goaf overburden rock mass movement space-time law prediction control system according to claim 9.
CN202210311934.1A 2022-03-28 2022-03-28 System and method for predicting movement time-space law of overlying rock mass in mine goaf Pending CN114692685A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210311934.1A CN114692685A (en) 2022-03-28 2022-03-28 System and method for predicting movement time-space law of overlying rock mass in mine goaf

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210311934.1A CN114692685A (en) 2022-03-28 2022-03-28 System and method for predicting movement time-space law of overlying rock mass in mine goaf

Publications (1)

Publication Number Publication Date
CN114692685A true CN114692685A (en) 2022-07-01

Family

ID=82141792

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210311934.1A Pending CN114692685A (en) 2022-03-28 2022-03-28 System and method for predicting movement time-space law of overlying rock mass in mine goaf

Country Status (1)

Country Link
CN (1) CN114692685A (en)

Similar Documents

Publication Publication Date Title
CN109993280B (en) Underwater sound source positioning method based on deep learning
CN110533631B (en) SAR image change detection method based on pyramid pooling twin network
CN110097755A (en) Freeway traffic flow amount state identification method based on deep neural network
CN108900346B (en) Wireless network flow prediction method based on LSTM network
CN106407649B (en) Microseismic signals based on time recurrent neural network then automatic pick method
CN112989708B (en) Well logging lithology identification method and system based on LSTM neural network
CN109800954B (en) Reservoir evaluation method based on logging data
MXPA06011012A (en) Genetic algorithm based selection of neural network ensemble for processing well logging data.
CN109378014A (en) A kind of mobile device source discrimination and system based on convolutional neural networks
CN108897975A (en) Coalbed gas logging air content prediction technique based on deepness belief network
CN111914478A (en) Comprehensive geological drilling well logging lithology identification method
CN115061219B (en) Fracture type reservoir prediction identification method and system based on petroleum and natural gas detection
CN113919219A (en) Stratum evaluation method and system based on logging big data
CN114723095A (en) Missing well logging curve prediction method and device
CN112836802A (en) Semi-supervised learning method, lithology prediction method and storage medium
CN104570109B (en) A kind of method of reservoir hydrocarbons prediction
CN111338002A (en) Porosity prediction method based on gated cyclic neural network model
CN114492521A (en) Intelligent lithology while drilling identification method and system based on acoustic vibration signals
CN114609667A (en) Geophysical oil gas prediction method, equipment and storage medium
CN114091333A (en) Shale gas content artificial intelligence prediction method based on machine learning
CN114114414A (en) Artificial intelligence prediction method for 'dessert' information of shale reservoir
CN114692685A (en) System and method for predicting movement time-space law of overlying rock mass in mine goaf
Kong et al. Dynamic multi-scale convolution for dialect identification
CN112464483B (en) Logging curve reconstruction method based on genetic neural network algorithm
CN116883709A (en) Carbonate fracture-cavity identification method and system based on channel attention mechanism

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