CN110598796A - Machine learning-based CSAMT (continuous source mining) electrical characteristic enhancement and classification method for deep goaf - Google Patents
Machine learning-based CSAMT (continuous source mining) electrical characteristic enhancement and classification method for deep goaf Download PDFInfo
- Publication number
- CN110598796A CN110598796A CN201910877150.3A CN201910877150A CN110598796A CN 110598796 A CN110598796 A CN 110598796A CN 201910877150 A CN201910877150 A CN 201910877150A CN 110598796 A CN110598796 A CN 110598796A
- Authority
- CN
- China
- Prior art keywords
- convolution
- csamt
- goaf
- shallow
- resistivity
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- 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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
-
- 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
Abstract
The invention discloses a deep goaf CSAMT electrical characteristic enhancement and classification method based on machine learning, which comprises the following steps: I. initializing a convolution kernel capable of removing shallow anomaly and enhancing deep goaf characteristics aiming at the inverted CSAMT resistivity anomaly characteristics; II, performing convolution calculation by using the convolution kernel obtained in the step I to extract the shallow abnormal information and the deep goaf electrical abnormal information; III, calculating the shallow convolution resistivity characteristics and the deep goaf convolution resistivity characteristics obtained in the step II to obtain energy errors, and if the energy errors do not meet the precision requirement, returning to the step I to modify the initialized convolution kernel parameters; if the energy error meets the error requirement, shallow abnormal interference is removed, and the electrical characteristics of the deep goaf are enhanced; IV, performing kernel function-based fuzzy clustering analysis on the CSAMT electrical characteristics obtained in the step III; and V, according to the clustering result, identifying and classifying geological features of the goaf, the roadway and the like, and predicting the basic situation of the goaf.
Description
Technical Field
The invention relates to a machine learning-based CSAMT electrical characteristic enhancement and classification method for a deep goaf.
Background
Accidents such as mine earthquake, water inrush, toxic and harmful gas leakage and the like are easily induced in the goaf, and the safety of mine personnel and property is seriously threatened. At present, methods for detecting a gob mainly include a transient electromagnetic method, a high-density electrical method, a detection radar method, a Controlled Source Audio-frequency magnetotelluric method (CSAMT for short), and the like.
The CSAMT prediction method has the advantages of large exploration depth range, strong anti-interference capability, small high-resistance shielding effect, high resolution, high working efficiency and the like, and is widely applied to the fields of geothermal resource detection, bridge tunnel engineering, coal mine goaf prediction, hydrogeology, karst cave prediction, nonferrous metal deposit, petroleum exploration and development and the like.
In recent years, when the CSAMT is applied to goaf detection, although a good effect is obtained, since surface rivers, ponds, near-surface underground water collection areas and the like may cause high-low resistance abnormalities, the low-low resistance abnormalities of the low-shallow layer affect the inversion accuracy of the CSAMT and reduce the electrical characteristics of deep goafs, so that the goafs, roadways and the like cannot be accurately identified and classified, and the basic conditions of the goafs such as the positions, the distribution ranges, the scales and the like cannot be predicted.
Disclosure of Invention
The invention aims to provide a CSAMT electrical characteristic enhancement and classification method for a deep goaf based on machine learning, which is used for removing shallow abnormal information, enhancing the electrical characteristic of the deep goaf and realizing accurate prediction of the goaf.
In order to achieve the purpose, the invention adopts the following technical scheme:
the deep goaf CSAMT electrical characteristic enhancement and classification method based on machine learning comprises the following steps:
I. initializing a shallow abnormal convolution kernel and a deep goaf convolution kernel according to the CSAMT resistivity characteristics obtained by inversion;
the initialization process of the shallow abnormal convolution kernel is as follows:
firstly, determining the central position of low-resistance abnormal data of a certain shallow part from a CSAMT resistivity characteristic profile obtained by inversion; extracting data around the central position of the shallow low-resistance abnormal data to be used as a shallow abnormal convolution kernel;
the initialization process of the convolution kernel of the deep goaf is as follows:
firstly, determining the central position of electrical anomaly data of a certain deep goaf from a CSAMT resistivity characteristic profile obtained by inversion; extracting data around the center position of the electrical abnormal data of the deep goaf as a convolution kernel of the deep goaf;
II, taking the CSAMT resistivity characteristic profile obtained by inversion as input, respectively carrying out convolution calculation with each shallow abnormal convolution kernel, and then superposing each convolution result to obtain shallow convolution resistivity characteristics;
taking the CSAMT resistivity characteristic profile obtained by inversion as input, respectively carrying out convolution calculation with each deep goaf convolution kernel, and then carrying out superposition and averaging on each convolution result to obtain the deep goaf convolution resistivity characteristic;
removing shallow abnormal information and enhancing the electrical characteristics of the deep goaf, wherein the specific process is as follows:
firstly, adding the shallow convolution resistivity characteristic and the deep goaf convolution resistivity characteristic;
secondly, calculating the energy error of the sum obtained by adding the shallow convolution resistivity characteristic and the deep gob convolution resistivity characteristic and the CSAMT resistivity characteristic obtained by inversion;
judging whether the energy error meets the precision requirement:
if the precision requirement is not met, executing a step IV, and if the precision requirement is met, executing a step V;
fourthly, optimizing the size parameters of the shallow abnormal convolution kernel and the deep goaf convolution kernel, and returning to execute the step I;
performing difference calculation on the CSAMT resistivity characteristic obtained by inversion and the shallow convolution resistivity characteristic to obtain a deep goaf electrical characteristic, and adding the result to the deep goaf convolution resistivity characteristic obtained in the step (II) to obtain an enhanced deep goaf electrical characteristic;
IV, carrying out fuzzy C-means clustering analysis based on a nuclear space on the electrical characteristics of the enhanced deep goaf in the step III;
and V, identifying and classifying the goaf and the roadway according to the clustering result, and predicting the basic situation of the goaf.
Preferably, the shallow abnormal convolution kernel and the deep goaf convolution kernel are one-dimensional convolution kernels, two-dimensional convolution kernels or three-dimensional convolution kernels.
Preferably, in step II, shallow convolving the resistivity feature FcThe calculation formula of (a) is as follows:
wherein rho is CSAMT resistivity characteristics obtained by inversion;
is the ith1Shallow abnormal convolution kernel, n1The number of shallow anomalous convolution kernels;
in the step II, convolution resistivity characteristics F of deep goafdThe calculation formula of (a) is as follows:
wherein the content of the first and second substances,is jth1Convolution kernel of deep goaf, m1The number of convolution kernels for the deep goaf.
Preferably, in step (i):
the calculation formula of the sum y of the shallow convolution resistivity characteristic and the deep goaf convolution resistivity characteristic is as follows: y ═ Fc+Fd。
Preferably, in step two:
the expression of the sum y of the sum of the shallow convolution resistivity characteristic and the deep goaf convolution resistivity characteristic and the energy error of the CSAMT resistivity characteristic rho obtained by inversion is as follows:
wherein E represents an energy error, and R, S is the longitudinal and transverse point number of the CSAMT resistivity characteristics obtained by inversion respectively; y isrsDenotes a point of coordinate (r, s) in y, ρrsDenotes a point of coordinates (r, s) in ρ.
Preferably, in step (c):
the judgment formula of whether the energy error meets the precision requirement epsilon is as follows:ε is a constant greater than 0.
Preferably, in step (iv):
and optimizing the size parameters of the shallow abnormal convolution kernel and the deep goaf convolution kernel by using a gradient descent method.
Preferably, in step (v):
and performing difference calculation on the CSAMT resistivity characteristics obtained by inversion and the shallow convolution resistivity characteristics to obtain the electrical characteristics of the deep goaf according to the calculation formula: y is1＝ρ-Fc(ii) a Wherein, y1Representing the electrical characteristics of the deep goaf;
the calculation formula of the electrical characteristic x of the enhanced deep goaf is as follows: x ═ y1+Fd。
The invention has the following advantages:
as described above, the invention provides a method for enhancing and classifying electrical characteristics of CSAMT in a deep goaf based on machine learning, which is beneficial to reducing the influence of low-resistance anomaly of a shallow layer on the inversion accuracy of CSAMT, and simultaneously improving the electrical characteristics of the deep goaf, so that the goaf and a roadway can be more accurately identified and classified, and the basic situation of the goaf can be predicted.
Drawings
Fig. 1 is a flowchart of a deep goaf CSAMT electrical characteristic enhancement and classification method based on machine learning according to the present invention.
FIG. 2 is a CSAMT resistivity profile obtained after inversion in an embodiment of the invention.
Fig. 3 is a graph of clustering results obtained after kernel function-based fuzzy clustering analysis in the embodiment of the present invention.
Detailed Description
The invention is described in further detail below with reference to the following figures and detailed description:
as shown in fig. 1, the method for enhancing and classifying electrical characteristics of CSAMT in deep goaf based on machine learning includes the steps of:
I. and initializing convolution kernels capable of removing shallow anomaly and enhancing deep goaf characteristics, namely shallow anomaly convolution kernels and deep goaf convolution kernels, aiming at the CSAMT resistivity characteristics obtained by inversion.
The initialization process of the shallow abnormal convolution kernel is as follows:
firstly, determining the central position of low-resistance abnormal data of a certain shallow part from a CSAMT resistivity characteristic profile obtained by inversion; and extracting data around the central position of the shallow low-resistance abnormal data to be used as a shallow abnormal convolution kernel.
The extracted shallow abnormal convolution kernels are sequentially marked as CK1，…，CK n1，n1The number of shallow anomalous convolution kernels.
The initialization process of the convolution kernel of the deep goaf is as follows:
firstly, determining the central position of electrical anomaly data of a certain deep goaf from a CSAMT resistivity characteristic profile obtained by inversion; and extracting data around the center position of the electrical abnormal data of the deep goaf as a convolution kernel of the deep goaf.
The extracted deep goaf convolution kernels are sequentially marked as CK1，…，CK m1，m1The number of convolution kernels for the deep goaf.
The construction of the convolution kernel is crucial, the position and the size of the convolution kernel of the shallow abnormal and deep goaf can be represented, the identification effect of the goaf can be directly influenced, and the step I lays a foundation for the convolution calculation of the step II.
Given the differences in CSAMT data dimensions, the dimensions of convolution calculations are often relevant to solving different practical problems.
For two-dimensional CSAMT data, a one-dimensional convolution kernel or a two-dimensional convolution kernel can be designed by a shallow abnormal convolution kernel and a deep goaf convolution kernel; aiming at the three-dimensional CSAMT electrical method data, the three-dimensional convolution kernel can be designed besides the two convolution kernels.
The design process of each convolution kernel is as follows:
and (3) one-dimensional convolution kernel, which is a one-dimensional sequence, and the obtained one-dimensional convolution kernel is as follows:
k1＝w1(ρc,l) (1)
wherein, w1(ρcL) is expressed as ρcA one-dimensional convolution kernel of length l centered.
c is the center position of the convolution kernel (abnormal data), and c is (l + 1)/2.
l is the length of a one-dimensional convolution kernel, typically oddAnd (4) counting. RhocCSAMT resistivity characteristic representing the center position of the convolution kernel.
Assuming that the length of an abnormal region (shallow low-resistance abnormal region or deep goaf characteristic region) is ls, the length l of a designed convolution kernel needs to satisfy ls < l <2ls, and then convolution characteristics can be effectively extracted.
The two-dimensional convolution kernel is a two-dimensional sequence, and according to the characteristics of data, the obtained two-dimensional convolution kernel is as follows:
k2＝w2(ρc,l,w) (2)
wherein, w2(ρcL, w) is expressed as ρcAs a center, a two-dimensional convolution kernel of length l and width w, which are typically odd numbers.
Assuming that the length of an abnormal region (a shallow low-resistance abnormal region or a deep goaf characteristic region) is ls and the width is lw, the length l of a designed convolution kernel needs to satisfy ls < l <2ls, and the width w needs to satisfy lw < w <2lw, so that convolution characteristics can be effectively extracted.
c is the center position of the convolution kernel (abnormal data).
The three-dimensional convolution kernel is a three-dimensional sequence, the obtained three-dimensional convolution kernel is,
k3＝w3(ρc,l,w,h) (3)
wherein, w3(ρcL, w, h) is denoted by ρcAnd the three-dimensional convolution kernel is used as a center, and the length, the width and the height are l, w and h respectively.
l, w, h are typically odd numbers, and c is the center position of the convolution kernel (anomalous data).
Suppose the length of the anomaly region (shallow low-resistance anomaly or deep goaf feature region) is ls, the width is lw, and the height is lh.
The length l of the designed three-dimensional convolution kernel needs to satisfy ls < l <2ls, the width w needs to satisfy lw < w <2lw, and the height h needs to satisfy lh < h <2lh, so that the convolution characteristics can be effectively extracted.
And II, taking the CSAMT resistivity characteristic profile obtained by inversion as an input, respectively carrying out convolution calculation with each shallow abnormal convolution kernel, and then superposing each convolution result to obtain the shallow convolution resistivity characteristic.
The formula for performing convolution calculation on the input CSAMT data and the shallow abnormal convolution kernel is as follows:
wherein rho is CSAMT resistivity characteristics obtained by inversion,is the ith1A shallow abnormal convolution kernel of 1 ≦ i1≤n1。
Then each convolution result is processedCarrying out superposition averaging to obtain shallow convolution resistivity characteristics FcThe calculation formula is as follows:
and taking the CSAMT resistivity characteristic profile obtained by inversion as input, performing convolution calculation with each deep goaf convolution kernel respectively, and then superposing each convolution result to obtain the deep goaf convolution resistivity characteristic.
The formula for performing convolution calculation on the input CSAMT data and the convolution kernels of the deep goafs is as follows:
wherein rho is CSAMT resistivity characteristics obtained by inversion,is jth1A convolution kernel of deep goaf, j is not less than 11≤m1。
Then each convolution result is processedCarrying out superposition averaging to obtain the convolution resistivity characteristic F of the deep goafdThe calculation formula is as follows:
through the step II, the shallow abnormal information and the deep goaf electrical abnormal information can be extracted.
And III, removing the abnormal information of the shallow part and enhancing the electrical characteristics of the deep goaf. The specific process is as follows:
firstly, shallow part convolution resistivity characteristic FcConvolution resistivity feature with deep goaf FdAdd to get the sum y:
y＝Fc+Fd (8)
wherein the shallow part convolutes the resistivity characteristic FcUsed to characterize shallow low resistance anomalies.
Secondly, calculating the sum y obtained by adding the shallow convolution resistivity characteristic and the deep goaf convolution resistivity characteristic and the energy error E of the CSAMT resistivity characteristic rho obtained by inversion, wherein the calculation formula is as follows:
wherein E represents an energy error, and R, S is the longitudinal and transverse point number of the CSAMT resistivity characteristics obtained by inversion respectively; y isrsDenotes a point of coordinate (r, s) in y, ρrsDenotes a point of coordinates (r, s) in ρ.
Judging whether the energy error meets the precision requirement, wherein the judgment formula is as follows:
wherein ε is a constant greater than 0.
If the precision requirement is not met, namely the calculated energy error E is more than or equal to epsilon, executing a step (iv).
If the precision requirement is satisfied, that is, the calculated energy error E satisfies the above equation (10), the fifth step is executed.
The physical meaning of the energy error E is further explained here: the process from step I to step III is to separate the low-resistance abnormal information of the shallow part and the electrical characteristic information of the deep goaf from the original CSAMT data;
if the steps can accurately separate the electrical characteristic information of the shallow abnormal and deep goaf, the energy error E is smaller; if the electrical characteristic information of the shallow abnormal and deep goaf cannot be accurately separated in the steps, the energy error E is larger.
Therefore, the separation effect of the low-resistance abnormity of the shallow part and the electrical characteristics of the goaf of the deep part is judged by calculating the magnitude of the energy error.
And fourthly, optimizing the size parameters of the shallow abnormal convolution kernel and the deep goaf convolution kernel, and returning to execute the step I.
In this step (iv), the above parameters can be optimized, for example, by a gradient descent method.
For example: for a one-dimensional convolution kernel, the specific process of parameter optimization is as follows,
first, an energy error function is determined:
the objective of this embodiment is to minimize the value of the energy error function e (l), first using e (l) to calculate the partial derivative of l:
the length l of the one-dimensional convolution kernel is continuously updated by a gradient descent method, so that the energy error E is minimized:
wherein the content of the first and second substances,represents the energy error function E (l) to the partial derivative of l in ltWhere the subscript t denotes the number of iterations.
ltRepresenting the initial one-dimensional convolution kernel length, lt+1The updated one-dimensional convolution kernel length is represented, and eta represents the learning rate.
In the parameter optimization process of the one-dimensional convolution kernel, the length parameter of the one-dimensional convolution kernel is the size parameter of the convolution kernel.
The parameter optimization process of the two-dimensional convolution and the three-dimensional convolution is similar to the parameter optimization process of the one-dimensional convolution kernel.
The gradient descent of the two-dimensional convolution is to calculate partial derivatives of the energy error function E (l, w) in the length l and width w directions, and update the length l and the width w respectively through the gradient descent.
In the parameter optimization process of the two-dimensional convolution kernel, the length and width parameters of the two-dimensional convolution kernel are the size parameters of the convolution kernel.
The gradient descent of the three-dimensional convolution is realized by calculating partial derivatives of the energy error function E (l, w, h) in the directions of the length l, the width w and the height h and respectively completing the updating of the length l, the width w and the height h through the gradient descent.
In the parameter optimization process of the three-dimensional convolution kernel, the length, width and height parameters of the three-dimensional convolution kernel are the size parameters of the convolution kernel.
Fifthly, the CSAMT resistivity characteristic rho obtained by inversion and the shallow convolution resistivity characteristic FcDifference value calculation is carried out, and the electrical characteristic y of the deep goaf is obtained through the difference value calculation1The calculation formula is as follows:
y1＝ρ-Fc (14)
the result y1Convolution resistivity feature with deep goaf FdAnd adding to obtain the enhanced electrical characteristic x of the deep goaf:
x＝y1+Fd (15)
after the treatment, the abnormal interference of the shallow part can be removed, and the electrical characteristic of the deep goaf is improved.
And IV, carrying out fuzzy C-means clustering analysis based on a nuclear space on the electrical characteristics of the deep goaf enhanced in the step III.
The reason why the fuzzy C-means clustering based on the kernel space is selected in the embodiment is that:
due to the intrinsic defects of the clustering algorithms such as K-means and FCM, the traditional algorithm cannot perform clustering analysis on various data structures such as non-hypersphere data, data polluted by noise, data mixed by various mode prototypes, asymmetric data and the like.
The basic idea of the fuzzy C-means clustering algorithm (KFCM) based on the kernel space is as follows:
the CSAMT electrical data are mapped to a high-dimensional feature space through different kernel functions, so that features which are not displayed in the linear space are highlighted, the difference among the features is expanded, and clustering is performed in the high-dimensional feature space.
The specific process of the fuzzy C-means clustering algorithm (KFCM) based on the nuclear space comprises the following steps:
iv.1. obtaining CSAMT dataset as x ═ { x through step III1,x2,······,xn}; wherein:
xkfor data in CSAMT dataset x, k is 1,2, … …, n;
where KFCM divides the n vectors into p fuzzy groups, p representing the desired number of clusters.
Iv.2. generate initial membership matrix U and cluster center C ═ C1,c2,…,cp}。
FCM uses fuzzy division, and for each given point, the degree of the point belonging to each group is determined by using the membership U, where U is greater than or equal to 0 and less than or equal to 1, and since the membership U of different data points in the membership matrix U is independent, the constraint condition of the extremum is as follows:
IV.3 calculating the image phi (c) of the cluster center in the feature space from equation (17)i),i＝1,2,…,p；
Iv.4. calculate the objective function J, formula:
wherein:
Φ(xk) For a given data set X ═ X1,x2,…|,xnThe image in the corresponding nuclear space;
Φ(ci) For each cluster centre C ═ C1,c2,…,cpThe image in the corresponding nuclear space;
wherein, K (·) is the corresponding kernel function, a linear kernel function, a polynomial kernel function, a Gaussian kernel function, a Sigmoid kernel function and the like can be selected, and after a plurality of trial calculations, the Gaussian kernel function is found to be relatively stable and has relatively good generalization and convergence.
And calculating an objective function J, and stopping the algorithm if the initial set condition is met or the maximum iteration number is reached.
IV.5, the membership function and the clustering center in the feature space are respectively as follows:
substituting the formula (19) into the formula (20) can obtain a new membership matrix uikAs shown in equation (22):
and returning to execute the step IV.2.
The output of the fuzzy C-means clustering algorithm based on the kernel space is a fuzzy partition matrix of p clustering center point vectors and p center points, and the matrix represents the membership degree of each sample point belonging to each class.
According to the partition matrix, each point can be determined to belong to a goaf, a roadway or the like according to the maximum membership principle in the fuzzy set. The cluster center represents the average feature of each class, and is considered as a representative point of this class.
As can be seen from the algorithm steps, the whole process of the fuzzy C-means clustering algorithm based on the kernel space is as follows:
giving the cluster number p, the weighting index m and the sample data set x, initializing a membership matrix U randomly, calculating a cluster center C, calculating a new U according to the cluster center C, calculating the new C by the new U, and repeating the steps until an iteration termination condition is met.
And V, identifying and classifying the goaf and the roadway according to the clustering result, and predicting the basic situation of the goaf. Basic conditions of the goaf include the position, distribution range, scale and the like of the goaf.
After the processing of the method, the effective electrical characteristics of the CSAMT deep goaf are obviously enhanced, the boundary of the depicted goaf is clearer, and the detection precision of the testing area deep goaf is obviously improved.
Specific examples are given below to verify the effectiveness of the CSAMT electrical characteristic enhancement and classification method proposed by the present invention.
Fig. 2 is a CSAMT resistivity inversion cross-section diagram obtained originally, and it can be seen from fig. 2 that due to the influence of surface rivers, ponds, near-surface groundwater convergence regions and the like, false low-resistance anomalies are presented, and the actual geological conditions are covered.
In order to remove the low resistance abnormality of the near-surface and the low resistance abnormality of the protruding deep goaf, the roadway and the like are classified, and the original CSAMT resistivity inversion section diagram is processed in steps I to IV to obtain a fuzzy clustering result shown in figure 3.
As can be seen from fig. 3, the low resistance abnormality of the shallow part disappears, that is, the low resistance information of the shallow part is suppressed, the information of the deep goaf is enhanced, geological features such as goafs and roadways are identified and classified according to the clustering result, and basic conditions such as the position, the distribution range and the scale of the goaf are predicted.
Based on the clustering results, wells are placed near the point of the survey line 1950 for verification. The drilling result shows that in the drilling process within the range of 522.02-524.02m, the footage is extremely fast or the water level of an orifice suddenly disappears and the core is broken, the sampling rate is extremely low, a small amount of coal gangue is visible, and the fact that the drilling meets a goaf and the burial depth and the position of the actual goaf are matched with geophysical prospecting abnormity.
Therefore, the position and the distribution range of the goaf of the measuring area are further found out, and the goaf of the measuring area is detected with higher precision.
It should be understood, however, that the description herein of specific embodiments is not intended to limit the invention to the particular forms disclosed, but on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the invention as defined by the appended claims.
Claims (8)
1. A CSAMT electrical characteristic enhancement and classification method based on machine learning is characterized in that,
the method comprises the following steps:
I. initializing a shallow abnormal convolution kernel and a deep goaf convolution kernel according to the CSAMT resistivity characteristics obtained by inversion;
the initialization process of the shallow abnormal convolution kernel is as follows:
firstly, determining the central position of low-resistance abnormal data of a certain shallow part from a CSAMT resistivity characteristic profile obtained by inversion; extracting data around the central position of the shallow low-resistance abnormal data to be used as a shallow abnormal convolution kernel;
the initialization process of the convolution kernel of the deep goaf is as follows:
firstly, determining the central position of electrical anomaly data of a certain deep goaf from a CSAMT resistivity characteristic profile obtained by inversion; extracting data around the center position of the electrical abnormal data of the deep goaf as a convolution kernel of the deep goaf;
II, taking the CSAMT resistivity characteristic profile obtained by inversion as input, respectively carrying out convolution calculation with each shallow abnormal convolution kernel, and then superposing each convolution result to obtain shallow convolution resistivity characteristics;
taking the CSAMT resistivity characteristic profile obtained by inversion as input, respectively carrying out convolution calculation with each deep goaf convolution kernel, and then carrying out superposition and averaging on each convolution result to obtain the deep goaf convolution resistivity characteristic;
removing shallow abnormal information and enhancing the electrical characteristics of the deep goaf, wherein the specific process is as follows:
firstly, adding the shallow convolution resistivity characteristic and the deep goaf convolution resistivity characteristic;
secondly, calculating the energy error of the sum obtained by adding the shallow convolution resistivity characteristic and the deep gob convolution resistivity characteristic and the CSAMT resistivity characteristic obtained by inversion;
judging whether the energy error meets the precision requirement:
if the precision requirement is not met, executing a step IV, and if the precision requirement is met, executing a step V;
fourthly, optimizing the size parameters of the shallow abnormal convolution kernel and the deep goaf convolution kernel, and returning to execute the step I;
performing difference calculation on the CSAMT resistivity characteristics obtained by inversion and the shallow convolution resistivity characteristics to obtain deep goaf electrical characteristics; adding the result with the convolution resistivity characteristic of the deep goaf obtained in the step (II) to obtain an enhanced electrical characteristic of the deep goaf;
IV, carrying out fuzzy C-means clustering analysis based on a nuclear space on the electrical characteristics of the enhanced deep goaf in the step III;
and V, identifying and classifying the goaf and the roadway according to the clustering result, and predicting the basic situation of the goaf.
2. The CSAMT electrical characteristic enhancement and classification method of claim 1,
and the shallow abnormal convolution kernel and the deep goaf convolution kernel are one-dimensional convolution kernel, two-dimensional convolution kernel or three-dimensional convolution kernel.
3. The CSAMT electrical characteristic enhancement and classification method of claim 1,
in the step II, shallow convolution resistivity characteristic FcThe calculation formula of (a) is as follows:
wherein rho is CSAMT resistivity characteristics obtained by inversion;
is the ith1Shallow abnormal convolution kernel, n1The number of shallow anomalous convolution kernels;
in the step II, convolution resistivity characteristics F of deep goafdThe calculation formula of (a) is as follows:
wherein the content of the first and second substances,is jth1Convolution kernel of deep goaf, m1The number of convolution kernels for the deep goaf.
4. The CSAMT electrical characteristic enhancement and classification method of claim 3,
the steps are as follows:
the calculation formula of the sum y of the shallow convolution resistivity characteristic and the deep goaf convolution resistivity characteristic is as follows: y ═ Fc+Fd。
5. The CSAMT electrical characteristic enhancement and classification method of claim 4,
the step II comprises the following steps:
the expression of the sum y of the sum of the shallow convolution resistivity characteristic and the deep goaf convolution resistivity characteristic and the energy error of the CSAMT resistivity characteristic rho obtained by inversion is as follows:
wherein E represents an energy error, and R, S is the longitudinal and transverse point number of the CSAMT resistivity characteristics obtained by inversion respectively; y isrsDenotes a point of coordinate (r, s) in y, ρrsDenotes a point of coordinates (r, s) in ρ.
6. The CSAMT electrical characteristic enhancement and classification method of claim 5,
in the third step:
the judgment formula of whether the energy error meets the precision requirement epsilon is as follows:ε is a constant greater than 0.
7. The CSAMT electrical characteristic enhancement and classification method of claim 1,
in the fourth step:
and optimizing the size parameters of the shallow abnormal convolution kernel and the deep goaf convolution kernel by using a gradient descent method.
8. The CSAMT electrical characteristic enhancement and classification method of claim 6,
in the fifth step:
and performing difference calculation on the CSAMT resistivity characteristics obtained by inversion and the shallow convolution resistivity characteristics to obtain the electrical characteristics of the deep goaf according to the calculation formula: y is1＝ρ-Fc(ii) a Wherein, y1Indicating deep skyElectrically characterizing the region;
the calculation formula of the electrical characteristic x of the enhanced deep goaf is as follows: x ═ y1+Fd。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910877150.3A CN110598796B (en) | 2019-09-17 | 2019-09-17 | Machine learning-based CSAMT (continuous source mining) electrical characteristic enhancement and classification method for deep goaf |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910877150.3A CN110598796B (en) | 2019-09-17 | 2019-09-17 | Machine learning-based CSAMT (continuous source mining) electrical characteristic enhancement and classification method for deep goaf |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110598796A true CN110598796A (en) | 2019-12-20 |
| CN110598796B CN110598796B (en) | 2022-04-12 |
Family
ID=68860300
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910877150.3A Active CN110598796B (en) | 2019-09-17 | 2019-09-17 | Machine learning-based CSAMT (continuous source mining) electrical characteristic enhancement and classification method for deep goaf |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110598796B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111983718A (en) * | 2020-07-30 | 2020-11-24 | 中煤科工集团西安研究院有限公司 | Remote advanced detection method for directional drilling and tunneling working face |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180217284A1 (en) * | 2017-01-27 | 2018-08-02 | Saudi Arabian Oil Company | Virtual source redatuming using radiation pattern correction |
| US20190064389A1 (en) * | 2017-08-25 | 2019-02-28 | Huseyin Denli | Geophysical Inversion with Convolutional Neural Networks |
-
2019
- 2019-09-17 CN CN201910877150.3A patent/CN110598796B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180217284A1 (en) * | 2017-01-27 | 2018-08-02 | Saudi Arabian Oil Company | Virtual source redatuming using radiation pattern correction |
| US20190064389A1 (en) * | 2017-08-25 | 2019-02-28 | Huseyin Denli | Geophysical Inversion with Convolutional Neural Networks |
Non-Patent Citations (4)
| Title |
|---|
| 印兴耀等: "基于核空间的模糊聚类方法在储层预测中的应用", 《中国石油大学学报自然科学版》 * |
| 张冲等: "应用多级低通滤波法增强深部采空区有效电性特征-以可控源音频大地电磁法在潍日高速公路测区应用为例", 《科学技术与工程》 * |
| 林年添等: "地震油气储层的小样本卷积神经网络学习与预测", 《地球物理学报》 * |
| 韩友德: "局部二值模式和卷积神经网络在多姿态人脸识别中的应用研究", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111983718A (en) * | 2020-07-30 | 2020-11-24 | 中煤科工集团西安研究院有限公司 | Remote advanced detection method for directional drilling and tunneling working face |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110598796B (en) | 2022-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zuo et al. | Compositional data analysis in the study of integrated geochemical anomalies associated with mineralization | |
| Xiao et al. | GIS-based 3D prospectivity mapping: A case study of Jiama copper-polymetallic deposit in Tibet, China | |
| Ghasemzadeh et al. | Stream sediment geochemical data analysis for district-scale mineral exploration targeting: Measuring the performance of the spatial U-statistic and CA fractal modeling | |
| Yousefi et al. | Weighted drainage catchment basin mapping of geochemical anomalies using stream sediment data for mineral potential modeling | |
| Agterberg | Geomathematics: theoretical foundations, applications and future developments | |
| Yu et al. | Application of hierarchical clustering, singularity mapping, and Kohonen neural network to identify Ag-Au-Pb-Zn polymetallic mineralization associated geochemical anomaly in Pangxidong district | |
| Holden et al. | Automatic identification of responses from porphyry intrusive systems within magnetic data using image analysis | |
| CN113488117B (en) | Deep gold ore deposit ore-forming and prospecting method with deep learning capability | |
| CN111221048B (en) | Boulder boundary identification and imaging method based on cross-hole resistivity CT multi-scale inversion | |
| Kim et al. | Application of fuzzy logic and geometric average: A Cu sulfide deposits potential mapping case study from Kapsan Basin, DPR Korea | |
| Ghezelbash et al. | Incorporation of principal component analysis, geostatistical interpolation approaches and frequency-space-based models for portraying the Cu-Au geochemical prospects in the Feizabad district, NW Iran | |
| Sadeghi et al. | Concentration-distance from centroids (C-DC) multifractal modeling: A novel approach to characterizing geochemical patterns based on sample distance from mineralization | |
| Zaitouny et al. | Fast automatic detection of geological boundaries from multivariate log data using recurrence | |
| Serrano et al. | Impact of two geostatistical hydro-facies simulation strategies on head statistics under non-uniform groundwater flow | |
| Shi et al. | Training image selection for development of subsurface geological cross-section by conditional simulations | |
| Zhang et al. | Comparison of 3D prospectivity modeling methods for Fe-Cu skarn deposits: A case study of the Zhuchong Fe-Cu deposit in the Yueshan orefield (Anhui), eastern China | |
| Liu et al. | Developments in quantitative assessment and modeling of mineral resource potential: an overview | |
| Yang et al. | Deep learning inversion of gravity data for detection of CO2 plumes in overlying aquifers | |
| Başokur et al. | Object-based model verification by a genetic algorithm approach: Application in archeological targets | |
| CN110598796B (en) | Machine learning-based CSAMT (continuous source mining) electrical characteristic enhancement and classification method for deep goaf | |
| Soltani-Mohammadi et al. | Grade estimation using a hybrid method of back-propagation artificial neural network and particle swarm optimization with integrated samples coordinate and local variability | |
| Tobely et al. | Position detection of unexploded ordnance from airborne magnetic anomaly data using 3-D self organized feature map | |
| Zhang et al. | Data-processing and recognition of seepage and microseepage anomalies of acid-extractable hydrocarbons in the south slope of the Dongying depression, eastern China | |
| Sadeghi | SPS Model: a significant algorithm to reduce the time and computer memory required in geostatistical simulations | |
| Xie et al. | GIS prospectivity mapping and 3D modeling validation for potential uranium deposit targets in Shangnan district, China |
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 |