CN111046737B - Efficient intelligent sensing acquisition method for microseism signal detection - Google Patents

Efficient intelligent sensing acquisition method for microseism signal detection Download PDF

Info

Publication number
CN111046737B
CN111046737B CN201911112716.XA CN201911112716A CN111046737B CN 111046737 B CN111046737 B CN 111046737B CN 201911112716 A CN201911112716 A CN 201911112716A CN 111046737 B CN111046737 B CN 111046737B
Authority
CN
China
Prior art keywords
data
micro
seismic
microseismic
acquisition
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201911112716.XA
Other languages
Chinese (zh)
Other versions
CN111046737A (en
Inventor
佟训乾
宾康成
张晓普
林君
孙锋
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.)
Jilin University
Original Assignee
Jilin University
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 Jilin University filed Critical Jilin University
Priority to CN201911112716.XA priority Critical patent/CN111046737B/en
Publication of CN111046737A publication Critical patent/CN111046737A/en
Application granted granted Critical
Publication of CN111046737B publication Critical patent/CN111046737B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Evolutionary Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Signal Processing (AREA)
  • Medical Informatics (AREA)
  • Geophysics And Detection Of Objects (AREA)

Abstract

The invention discloses a high-efficiency intelligent acquisition method for micro-seismic signals, which comprises the steps of compressed sensing sampling of micro-seismic data, extraction of effective micro-seismic events in the compressed data and reconstruction of the micro-seismic event compressed data. Specifically, all distributed acquisition nodes firstly perform compressed sensing on the micro-seismic data so as to obtain a measurement value of the micro-seismic signal. The compressed sensing removes redundant data of the micro-seismic signals in the time domain, and avoids subsequent unnecessary calculation overhead. The measurement of the microseismic signals is then input into a pre-trained deep neural network, thereby completing the identification of valid microseismic events. Upon identifying a valid event, the acquisition node transmits the compressed sampled data containing the microseismic event to a data center. Finally, the data center performs approximate optimization through a singular value decomposition and clustering method to continuously update sparse basis of the seismic data, and uses l1The norm spectral projection gradient (SPGL1) algorithm reconstructs the original microseismic event data. In conclusion, the novel data acquisition method can improve the data acquisition efficiency of the whole system in the aspects of data recording and data transmission.

Description

Efficient intelligent sensing acquisition method for microseism signal detection
Technical Field
The invention relates to the field of design of a microseism signal detection method, in particular to a high-efficiency intelligent sensing acquisition method for microseism signal detection.
Background
With the rapid development of artificial intelligence algorithms, academic circles pay more and more attention to the application of the algorithms in the field of resource exploration. For example, in the global engineering frontier published in 2018 of the Chinese institute of engineering, the intelligentization, high-efficiency exploration and exploitation of oil and gas resources are listed as the frontier of engineering research, and meanwhile, the intelligent acquisition, high-efficiency transmission and intelligent analysis of data are indicated as the trend of future development. Therefore, the research on the acquisition method for improving the use efficiency of the hardware resources of the seismic acquisition equipment is carried out by combining the relevant theory of artificial intelligence, and the research has great significance.
The seismic data efficient acquisition method based on the compressed sensing principle fully utilizes the constraint condition of sparsity of the seismic data, and reduces the acquisition amount and the transmission amount of the required seismic data no matter aiming at the compressed acquisition in nodes or the sparse acquisition in a spatial domain, so that the method has higher acquisition efficiency compared with the traditional data compression and encoding method. However, due to the diversity of seismic data features, most methods for efficiently acquiring seismic data by applying compressed sensing in acquisition nodes use a fixed transform domain as a sparse domain of the seismic data (such as Curvelet, Seislet and the like) or perform dictionary learning (such as principal component analysis, K-SVD and the like) by using conditions of linearity, orthogonality and the like as constraints, so that high-quality characterization of different types of seismic data is difficult to achieve. Moreover, these methods do not fully consider the high-order combination of sparse domain features during seismic data reconstruction, and it is difficult to recover details in seismic data with high precision on the premise of using fewer feature components, so that the reconstruction capability for seismic data with high compression ratio is not strong.
Disclosure of Invention
The invention aims to provide an efficient intelligent sensing acquisition method for microseismic signal detection,
the classification method based on machine learning is combined with compression sampling, and signal detection is carried out on the basis of generalized random undersampling, so that only data related to microseism signals are transmitted. Aiming at the problem of sparse representation of the micro-seismic data, the sparse representation is combined with machine learning, a micro-seismic signal sparse domain is rapidly solved in a self-adaptive mode, and a proper sparse basis is provided for an SPGL1 algorithm, so that the reconstruction of the micro-seismic signals is completed, the energy consumption of data recording is reduced, and the acquisition scale of the system is greatly improved.
In order to achieve the purpose, the invention provides the following scheme:
an efficient intelligent perception acquisition method for microseismic signal detection, comprising:
1) carrying out compression sampling on the microseism data in a time domain according to the measurement matrix to obtain a compression sampling vector value b;
2) constructing a basic structure of a microseism signal detection model by analyzing the relationship among local feature extraction, time sequence state memory and smooth distribution of the microseism signal, and optimizing model parameters in an iterative mode;
3) the micro-seismic signal detection method based on machine learning obtains a trained detection model, sends parameters of the model to each acquisition node to distinguish micro-seismic signals for each element in the vector value b, and transmits data containing the micro-seismic signals to a data center;
4) and extracting the micro-seismic signal characteristics contained in each acquisition node through singular value decomposition to be used as an initial dictionary. Updating a position limiting and clustering method to perform approximate optimization so as to ensure sparse representation capability and calculation speed in an updating process and provide sparse basis for a self-adaptive microseism data reconstruction method;
5) and performing microseismic data reconstruction on the sparse basis by adopting an SPGL1 algorithm.
Further, the air conditioner is provided with a fan,
step 1) selecting a microseismic signal y belonging to RTObtaining compressed data b ∈ R by jitter samplingMExpressed as:
b＝yΦ
wherein phi ∈ RT×MTo observe
Figure BDA0002273197460000021
A matrix; phi epsilon to Rr×MξTo observe the non-zero element model of the matrix, the elements in Φ are only two:0 and 1, with and only one non-zero element in each column of Φ; each row has at most one non-zero element, and the row number of the non-zero element in the previous column in phi is necessarily smaller than that of the non-zero element in the next column.
Further, step 2) includes dividing the sampled data into two categories in a point-by-point manner: data related to micro-seismic signals and data unrelated to micro-seismic signals, comprising the steps of:
the local feature extraction adopts three layers of one-dimensional convolution neural networks, wherein the first layer of network consists of 4 one-dimensional convolution kernels with the size of 3, 4 one-dimensional convolution kernels with the size of 5 and 2 one-dimensional convolution kernels with the size of 7; the second layer network is composed of 12 one-dimensional convolution kernels with the size of 3 and 8 with the size of 5; the third layer network is composed of 32 one-dimensional convolution kernels with the size of 3, the input is formed by splicing the output result of the first layer network and the output result of the second layer network, the number of input channels is 30, and the local feature extraction output is the feature extraction result of 72 channels; in the process of calculating convolution, all vectors of multiple channels are aligned at the back end, and then the unaligned part at the front end is filled with 0.
Furthermore, the time sequence state memory adopts a 2-layer GRU network to extract the time sequence characteristics of each local morphological characteristic in the microseism signal, and the input end is 72-channel characteristic mapping in the local characteristic extraction; the first layer of network comprises 8 GRUs, the second layer of network only comprises 1 GRU, and a Sigmoid classifier is arranged on the second layer of GRU network and is responsible for mapping the output value of the second layer of GRU network into a (0,1) interval and outputting the value as a time sequence state memory module.
Further, the air conditioner is provided with a fan,
adding a smoothing term containing e and negative exponential order decay into the smooth distribution, and outputting Q at the ith pointiIs represented as follows:
Qi＝σ(WsPi+VsIi+bs)
wherein, WS，VS，bSParameters needing to be learned in the smooth distribution module; piFor smooth distribution of the input value of the module at the ith point, the time sequence state memory module at the ith pointOutputting the value; i isiCorrection factor for point I, IiThe solution is as follows:
Figure BDA0002273197460000031
Figure BDA0002273197460000032
wherein, when i + j exceeds the sampling point range, Pi+j＝0；CHIs a normalization factor; h is a smooth scale, and the value of H is 2;
smoothly distributed according to output QiIf Q, classifying the collected dataiIf the sampling point contains the microseism signal, the microseism signal is transmitted to a data center; if QiIf the number is less than 0.5, the acquisition node is judged to be irrelevant to the microseism signal, and the data is not sent.
Further, the step 3) fully excavates the common characteristics of the micro seismic signals in each acquisition node, and acquires sparse bases of the micro seismic signals in a self-adaptive mode; the data center forms a measurement matrix used by each acquisition node according to acquisition time stamp information of data finally sent by the acquisition node, and uses l1And restoring the compressed and sampled data into original seismic data by a norm spectrum projection gradient method.
Further, the air conditioner is provided with a fan,
the clustering dictionary learning method based on singular value decomposition reduces the error expressed by a sample set T on the premise of not increasing the sparsity of a coefficient matrix A by adjusting each atom in a sparse basis Ψ of micro-seismic data, realizes the optimization of the sparse basis Ψ, and solves the following problems in the training optimization of an over-complete dictionary Ψ:
Figure BDA0002273197460000041
wherein A isTΨ at the basis for the sample set TTLower seriesA number matrix; a isiIs ATThe vector of the ith row is psiTThe coefficient corresponding to the vector of the ith column; u is microseismic data at the base psiTLower sparsity at psiTIn the updating process of each atom in the coefficient matrix A, only the elements corresponding to the non-zero items in the coefficient matrix A are updated when the atoms in the sparse basis Ψ are updated each timeTAtom Ψ in the ith columniThe update process of (a) is specifically expressed in the form:
Figure BDA0002273197460000042
E′i＝EiΩi
Figure BDA0002273197460000043
a′i＝aiΩi
ωl＝{l|1≤l≤D，ai(l)≠0}
wherein, EiIs due to deletion of the atom ΨiResulting in errors in the set of microseismic signal samples T; omegaiIs ΨTThe selection matrix updated by the ith column of atoms has the size of R x omegalAnd wherein the position is (ω)l(l) L) is 1 and the other elements are 0; in each iteration process, updating all atoms and corresponding representation coefficients thereof by the clustering dictionary learning method based on singular value decomposition, and when the finally learned sparse basis reaches a stable state, finishing the iteration by the algorithm and outputting the final sparse basis psi.
Further, the micro-seismic data reconstruction of the sparse basis using the SPGL1 algorithm is represented as follows:
Figure BDA0002273197460000044
where σ is the reconstructed noise level, let τ | | | s | | y1Of sparsityThe measurement index, in order to balance tau and sigma to achieve the optimal effect, a function gamma needs to be constructed to describe the relationship between tau and sigma,
Γ(τ)＝σ
through Newton iterative format, approximate solution is obtained for tau:
Figure BDA0002273197460000051
where k is the number of iterations, in terms of τk+1Using a spectral projection gradient method to solve the following equation:
Figure BDA0002273197460000052
in finding sk+1Then, the gamma (tau) is alignedk+1) And Γ' (τ)k+1) And (3) solving an approximate solution:
Γ(τk+1)＝||b-sk+1ΨΦ||2
Figure BDA0002273197460000053
by iterating over and over so that τk+1Sparse constraint τ towards gradual optimizationσFurther, the corresponding s is obtainedσTo optimize the solution, the recovered raw seismic data is finally
Figure BDA0002273197460000054
According to the specific embodiment provided by the invention, the invention has the following technical effects: the problems of low signal-to-noise ratio and overhigh data flow load of the conventional seismic acquisition system are solved. Adopting a generalized jitter random sampling method to carry out undersampling on the seismic data; identifying the micro-seismic signals through a machine learning algorithm which integrates a convolutional neural network, a recurrent neural network and a probability map model according to the characteristics of the micro-seismic signals in the aspects of local morphology, time sequence relation and probability distribution; the method has the advantages that the characteristics of signals are extracted through singular value decomposition and serve as initial sparse bases, approximate value optimization is carried out through a position limiting and clustering method so as to continuously update the sparse bases, micro seismic data reconstruction is carried out on the sparse bases through the SPGL1 algorithm, the energy consumption of data recording is reduced, meanwhile, the collection scale supported by the system is greatly improved, and the working efficiency of field seismic exploration is effectively improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a schematic overall flow diagram of a method for efficient intelligent acquisition of microseismic signals in accordance with the present invention;
FIG. 2 is a schematic flow chart of a time domain random sampling method according to an embodiment of the present invention for an efficient intelligent micro-seismic signal acquisition method;
FIG. 3 is a schematic flow diagram of a microseismic signal detection method of an embodiment of the present invention for efficient intelligent collection of microseismic signals;
FIG. 4 is a schematic diagram of a network structure of a local feature extraction module according to an embodiment of the method for efficient intelligent acquisition of microseismic signals of the present invention;
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
As shown in fig. 1, an efficient intelligent sensing acquisition method for micro-seismic signal detection includes a micro-seismic data compression sampling method, a micro-seismic signal detection method, and a data reconstruction method:
firstly, the micro-seismic data is compressed and sampled in a time domain in a jitter random sampling mode. The processors of the acquisition nodes generate jittering random factors in a uniformly distributed mode, and the sampling time of the analog-to-digital converter is controlled according to the random factors, so that time domain compression sampling of the micro-seismic data is realized;
secondly, designing a machine learning algorithm fusing a convolutional neural network, a recurrent neural network and a probability map model according to the characteristics of the seismic signals in three aspects of local form, time sequence relation and probability distribution, and realizing that the sampled data is divided into two types in a point-by-point mode as shown in fig. 3: and the data related to the micro-seismic signals and the data unrelated to the micro-seismic signals are used for solving the problem of detecting the micro-seismic signals based on undersampled data.
And finally, extracting the micro-seismic signal characteristics contained in each acquisition node through singular value decomposition to serve as an initial dictionary. And updating the position limitation and clustering method to perform approximate optimization so as to ensure the sparse representation capability and the calculation speed of the updating process and provide a proper sparse basis for the self-adaptive micro seismic data reconstruction method. By using a1Norm spectral projection gradient method (spectral projected gradient for)1minimization, SPGL1) algorithm to reconstruct the microseismic data of the sparse basis, so that the main information in the microseismic data can be recovered.
With reference to fig. 2 and 4, performing compression sampling on the microseism data in a time domain according to the measurement matrix to obtain a compression sampling vector value b; the method comprises the following steps:
by microseismic signals y ∈ RTObtaining compressed data b ∈ R by jitter samplingMExpressed as:
b＝yΦ
Figure BDA0002273197460000071
wherein phi ∈ RT×MIs an observation matrix; phi epsilon to Rr×MξTo observe the non-zero element model of the matrix, the elements in Φ are only two: 0 and 1, with and only one non-zero element in each column of Φ; each row has at most one non-zero element, and the row number of the non-zero element in the previous column in phi is necessarily smaller than that of the non-zero element in the next column.
Constructing a basic structure of a microseism signal detection model by analyzing the relationship among local feature extraction, time sequence state memory and smooth distribution of the microseism signal, and optimizing model parameters in an iterative mode; the phi generation algorithm used is shown in fig. 2; the method comprises the following steps:
the local feature extraction module adopts three layers of one-dimensional convolution neural networks, the network structure is shown in figure 4, and the first layer of network consists of 4 one-dimensional convolution kernels with the size of 3, 4 one-dimensional convolution kernels with the size of 5 and 2 one-dimensional convolution kernels with the size of 7; the second layer network is composed of 12 one-dimensional convolution kernels with the size of 3 and 8 with the size of 5; the third network is composed of 32 one-dimensional convolution kernels with the size of 3, and the input of the third network is formed by splicing the output result of the first network and the output result of the second network, so that the number of input channels is 30. Finally, the local feature extraction module outputs the feature extraction result of 72 channels. In the process of calculating convolution, aligning all vectors of multiple channels at the rear end, and then completing unaligned parts at the front end by 0;
the time sequence state memory module adopts 2 layers of GRU networks to extract the time sequence characteristics of each local morphological characteristic in the microseismic signals, and the input end of the time sequence state memory module is 72-channel characteristic mapping in the local characteristic extraction module. The first layer network contains 8 GRUs, and the second layer network has only 1 GRU. A Sigmoid classifier is arranged on the second layer GRU network and is responsible for mapping the output value of the second layer GRU network into a (0,1) interval and outputting the value as a time sequence state memory module;
adding a smoothing term containing e and negative exponential order attenuation into a smooth distribution module, wherein the output Q of the ith pointiCan be expressed as follows:
Qi＝σ(WsPi+VsIi+bs)
wherein, WS，VS，bSParameters needing to be learned in the smooth distribution module; p isiThe input value of the smooth distribution module at the ith point is also the output value of the time sequence state memory module at the ith point; i isiIs the correction factor for the ith point. I isiThe solution is as follows:
Figure BDA0002273197460000081
Figure BDA0002273197460000082
wherein, when i + j exceeds the sampling point range, Pi+j＝0；CHIs a normalization factor; h is the smoothing scale. The size of the wavelength of the seismic signal in the time domain is more than 3. In order to ensure that complete smoothness can be obtained under the minimum seismic wave size and the judgment value of the seismic signal is not greatly weakened due to an overlarge smooth window, the value of H is set to be 2.
Finally, the smooth distribution module is according to QiThe values of (a) classify the collected data. If QiIf the sampling point contains the microseism signal, the microseism signal is transmitted to a data center; if QiIf the number is less than 0.5, the acquisition node is judged to be irrelevant to the microseism signal, and the data is not sent.
The clustering dictionary learning method based on singular value decomposition reduces errors of expressing T on the premise of not increasing sparsity of a coefficient matrix A by adjusting each atom in a sparse basis Ψ, and accordingly Ψ optimization is achieved. The training optimization for Ψ is actually solving the following problem:
Figure BDA0002273197460000083
wherein A isTIs T at the base ΨTA coefficient matrix of; a isiIs ATThe vector of line i, i.e. ΨTThe coefficient corresponding to the vector of the ith column (i.e. the ith atom in the dictionary); u is microseismic data at the base ΨTThe lower sparsity. At the point of pair psiTSince only the elements corresponding to the non-zero entries in the system participate in the representation of the signal during the update of each atom in (b), only the elements corresponding to the non-zero entries in the coefficient matrix a are updated each time the atom in the sparse basis Ψ is updated. ΨTAtom Ψ in the ith columniThe update process of (a) can be specifically expressed in the following form:
Figure BDA0002273197460000084
E′i＝EiΩi
Figure BDA0002273197460000085
a′i＝aiΩi
ωl＝{l|1≤l≤D，ai(l)≠0}
wherein E isiIs due to deletion of the atom ΨiResulting in errors in the set of microseismic signal samples T; omegaiIs ΨTThe selection matrix updated by the ith column of atoms has the size of R x omegalAnd wherein the position is (ω)l(l) L) is 1 and the remaining elements are 0. In each iteration process, the clustering dictionary learning method based on singular value decomposition updates all atoms and corresponding representation coefficients thereof. When the finally learned sparse basis reaches a stable state, the algorithm ends iteration and outputs the final sparse basis Ψ.
The solution to the microseismic signal reconstruction problem using the SPGL1 method is as follows:
Figure BDA0002273197460000091
where σ is the reconstructed noise level. Let τ | | s | | non-phosphor1And the index is a measurement index of sparsity. In order to balance τ and σ to achieve the optimal effect, a function Γ needs to be constructed to describe the relationship between τ and σ, that is:
Γ(τ)＝σ
through Newton iterative format, approximate solution is obtained for tau:
Figure BDA0002273197460000092
where k is the number of iterations. According to τk+1Using a spectral projected gradient method (projected gradient), the following equation is solved:
Figure BDA0002273197460000093
in finding sk+1Then, the gamma (tau) is alignedk+1) And r' (τ)k+1) And (3) solving an approximate solution:
Γ(τk+1)＝||b-sk+1ΨΦ||2
Figure BDA0002273197460000094
by iterative iteration of the above formula, let τk+1Sparse constraint τ towards gradual optimizationσFurther, the corresponding s is obtainedσIs an optimized solution. Finally, the recovered original seismic data is
Figure BDA0002273197460000095
In summary, the data center reconstructs the measurement matrix according to the acquisition time stamp of each acquisition node, and then uses the SPGL1 algorithm to solve S under the condition that b, Φ, and Ψ are known, so as to obtain the recovered seismic data
Figure BDA0002273197460000096
The principle and the implementation mode of the invention are explained by applying a specific example, and the description of the embodiment is only used for helping to understand the method and the core idea of the invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.

Claims (1)

1. An efficient intelligent sensing acquisition method for microseismic signal detection, comprising:
1) carrying out compression sampling on the microseism data in a time domain according to the measurement matrix to obtain a compression sampling vector value b;
2) constructing a basic structure of a microseismic signal detection model by analyzing the relationship among local feature extraction, time sequence state memory and smooth distribution of the microseismic signal, and optimizing model parameters in an iterative manner;
3) the micro-seismic signal detection method based on machine learning obtains a trained detection model, sends parameters of the model to each element in the vector value b of each acquisition node to judge a micro-seismic signal, and transmits data containing the micro-seismic signals to a data center;
4) extracting the micro-seismic signal characteristics contained in each acquisition node through singular value decomposition as an initial dictionary, updating a position limitation and clustering method to perform approximate value optimization so as to ensure sparse representation capacity and calculation speed in the updating process and provide sparse basis for a self-adaptive micro-seismic data reconstruction method;
5) performing microseismic data reconstruction on the sparse basis by adopting an SPGL1 algorithm;
step 1) selecting a microseismic signal y belonging to RTObtaining compressed data b ∈ R by jitter samplingMExpressed as:
b＝yΦ
wherein phi ∈ RT×MIs composed of
Figure FDA0003656279930000011
Observing a matrix; phi epsilon to Rr×MξTo observe the non-zero element model of the matrix, the elements in Φ are only two: 0 and 1, with and only one non-zero element in each column of Φ; each row has at most one non-zero element, and the row number of the non-zero element in the previous column in phi is certainly less than that of the non-zero element in the next column;
step 2) comprises dividing the sampled data into two categories in a point-by-point manner: data related to micro-seismic signals and data unrelated to micro-seismic signals, comprising the steps of:
the local feature extraction adopts three layers of one-dimensional convolution neural networks, and the first layer of network consists of 4 one-dimensional convolution kernels with the size of 3, 4 one-dimensional convolution kernels with the size of 5 and 2 one-dimensional convolution kernels with the size of 7; the second layer network is composed of 12 one-dimensional convolution kernels with the size of 3 and 8 with the size of 5; the third layer network is composed of 32 one-dimensional convolution kernels with the size of 3, the input is formed by splicing the output result of the first layer network and the output result of the second layer network, the number of input channels is 30, and the local feature extraction output is the feature extraction result of 72 channels; in the process of calculating convolution, aligning all vectors of multiple channels at the rear end, and then completing unaligned parts at the front end by 0;
the time sequence state memory adopts a 2-layer GRU network to extract the time sequence characteristics of each local morphological characteristic in the microseism signal, and the input end is 72-channel characteristic mapping in the local characteristic extraction; the first layer network comprises 8 GRUs, the second layer network only comprises 1 GRU, and a Sigmoid classifier is arranged on the second layer GRU network and is responsible for mapping the output value of the second layer GRU network into a (0,1) interval and outputting the value as a time sequence state memory module;
adding a smoothing term containing e and negative exponential order decay into the smooth distribution, and outputting Q at the ith pointiIs represented as follows:
Qi＝σ(WsPi+VsIi+bs)
wherein, WS，VS，bSParameters needing to be learned in the smooth distribution module are set; piFor smoothingThe input value of the distribution module at the ith point is also the output value of the time sequence state memory module at the ith point; I.C. AiCorrection factor for point I, IiThe solution is as follows:
Figure FDA0003656279930000021
Figure FDA0003656279930000022
wherein, when i + j exceeds the sampling point range, Pi+j＝0；CHIs a normalization factor; h is a smooth scale, and the value of H is 2;
smoothly distributed according to output QiIf Q, classifying the collected dataiIf the sampling point contains the microseism signal, the microseism signal is transmitted to a data center; if QiIf the number is less than 0.5, judging that the acquisition node is irrelevant to the microseism signal, and not sending the data;
the step 3) fully excavates the common characteristics of the micro seismic signals in each acquisition node, and acquires sparse bases of the micro seismic signals in a self-adaptive mode; the data center forms a measurement matrix used by each acquisition node according to acquisition time stamp information of data finally sent by the acquisition node, and uses l1Restoring the compressed and sampled data into original seismic data by a norm spectrum projection gradient method;
the clustering dictionary learning method based on singular value decomposition reduces the error expressed by a sample set T on the premise of not increasing the sparsity of a coefficient matrix A by adjusting each atom in a sparse basis Ψ of micro-seismic data, realizes the optimization of the sparse basis Ψ, and solves the following problems in the training optimization of an over-complete dictionary Ψ:
Figure FDA0003656279930000031
wherein A isTΨ at the basis for the sample set TTA coefficient matrix of; a isiIs ATVector of line i of center, ΨTThe coefficient corresponding to the vector of the ith column; u is microseismic data at the base ΨTLower sparsity in the pair psiTIn the updating process of each atom in the coefficient matrix A, only the elements corresponding to the non-zero items in the coefficient matrix A are updated when the atoms in the sparse basis Ψ are updated each timeTAtom Ψ in the ith columniThe update process of (a) is specifically expressed in the form:
Figure FDA0003656279930000032
E′i＝EiΩi
Figure FDA0003656279930000033
a′i＝aiΩi
ωl＝{l|1≤l≤D,ai(l)≠0}
wherein, EiIs due to deletion of the atom ΨiResulting in errors in the set of microseismic signal samples T; omegaiIs ΨTThe selection matrix updated by the ith column of atoms has the size of R x omegalAnd wherein the position is (ω)l(l) L) is 1 and the other elements are 0; in each iteration process, updating all atoms and corresponding representation coefficients thereof by a clustering dictionary learning method based on singular value decomposition, and when the finally learned sparse basis reaches a stable state, finishing the iteration by the algorithm and outputting a final sparse basis psi;
the reconstruction of the microseismic data of the sparse basis using the SPGL1 algorithm is represented as follows:
Figure FDA0003656279930000034
where σ is the reconstructed noise level, let τ | | | s | | y1For the sparsity measure, in order to balance τ and σ to achieve the optimal effect, a function Γ needs to be constructed to describe the relationship between τ and σ,
Γ(τ)＝σ
through Newton iterative format, approximate solution is obtained for tau:
Figure FDA0003656279930000035
where k is the number of iterations, according to τk+1Using a spectral projection gradient method to solve the following equation:
Figure FDA0003656279930000041
in finding sk+1Then, the gamma (tau) is alignedk+1) And Γ' (τ)k+1) And (3) solving an approximate solution:
Γ(τk+1)＝||b-sk+1ΨΦ||2
Figure FDA0003656279930000042
by iterating over and over so that τk+1Sparse constraint τ towards gradual optimizationσFurther, the corresponding s is obtainedσTo optimize the solution, the recovered raw seismic data is finally
Figure FDA0003656279930000043
CN201911112716.XA 2019-11-14 2019-11-14 Efficient intelligent sensing acquisition method for microseism signal detection Active CN111046737B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911112716.XA CN111046737B (en) 2019-11-14 2019-11-14 Efficient intelligent sensing acquisition method for microseism signal detection

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911112716.XA CN111046737B (en) 2019-11-14 2019-11-14 Efficient intelligent sensing acquisition method for microseism signal detection

Publications (2)

Publication Number Publication Date
CN111046737A CN111046737A (en) 2020-04-21
CN111046737B true CN111046737B (en) 2022-07-08

Family

ID=70232907

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911112716.XA Active CN111046737B (en) 2019-11-14 2019-11-14 Efficient intelligent sensing acquisition method for microseism signal detection

Country Status (1)

Country Link
CN (1) CN111046737B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111929724B (en) * 2020-07-28 2022-05-31 中国地震局地震研究所 Earthquake signal compressed sensing acquisition transmission method facing early warning
CN112257560B (en) * 2020-10-20 2022-01-07 华北电力大学 Microseismic event detection method and system by utilizing cumulative similarity
CN112731518B (en) 2020-12-09 2022-06-07 中国科学院地质与地球物理研究所 Artificial intelligence real-time micro-seismic monitoring node
CN113609939A (en) * 2021-07-26 2021-11-05 吉林大学 Moving target efficient identification method based on compressed observation
CN115996144A (en) * 2022-12-08 2023-04-21 西南石油大学 Fracturing data transmission method and system based on compressed sensing and RSA encryption algorithm

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5402520A (en) * 1992-03-06 1995-03-28 Schnitta; Bonnie S. Neural network method and apparatus for retrieving signals embedded in noise and analyzing the retrieved signals
CN107666322A (en) * 2017-09-08 2018-02-06 山东科技大学 A kind of adaptive microseism data compression sensing method based on dictionary learning
CN108897045A (en) * 2018-08-28 2018-11-27 中国石油天然气股份有限公司 Deep learning model training method and seismic data noise attenuation method, device and equipment
CN109063687A (en) * 2018-08-29 2018-12-21 长江大学 A kind of microseism P wave recognition methods and system based on depth convolutional neural networks
CN109490957A (en) * 2018-11-28 2019-03-19 华北电力大学 A kind of Reconstruction of seismic data method based on space constraint compressed sensing
CN110082821A (en) * 2019-03-26 2019-08-02 长江大学 A kind of no label frame microseism signal detecting method and device

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5402520A (en) * 1992-03-06 1995-03-28 Schnitta; Bonnie S. Neural network method and apparatus for retrieving signals embedded in noise and analyzing the retrieved signals
CN107666322A (en) * 2017-09-08 2018-02-06 山东科技大学 A kind of adaptive microseism data compression sensing method based on dictionary learning
CN108897045A (en) * 2018-08-28 2018-11-27 中国石油天然气股份有限公司 Deep learning model training method and seismic data noise attenuation method, device and equipment
CN109063687A (en) * 2018-08-29 2018-12-21 长江大学 A kind of microseism P wave recognition methods and system based on depth convolutional neural networks
CN109490957A (en) * 2018-11-28 2019-03-19 华北电力大学 A kind of Reconstruction of seismic data method based on space constraint compressed sensing
CN110082821A (en) * 2019-03-26 2019-08-02 长江大学 A kind of no label frame microseism signal detecting method and device

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
An Efficient Neural-Network-Based Microseismic Monitoring Platform for Hydraulic Fracture on an Edge Computing Architecture;Xiaopu Zhang等;《sensors》;20180605;第18卷;第1-19页 *
An Efficient Seismic Data Acquisition Based on Compressed Sensing Architecture With Generative Adversarial Networks;XIAOPU ZHANG等;《IEEE access》;20190815;第7卷;第105948-105961页 *
Joint seismic data denoising and interpolation with double-sparsity dictionary learning;lingchen zhu等;《journal of geophysics and engineering》;20170523;第802-810页 *

Also Published As

Publication number Publication date
CN111046737A (en) 2020-04-21

Similar Documents

Publication Publication Date Title
CN111046737B (en) Efficient intelligent sensing acquisition method for microseism signal detection
CN109241491A (en) The structural missing fill method of tensor based on joint low-rank and rarefaction representation
CN112508110A (en) Deep learning-based electrocardiosignal graph classification method
CN110796625A (en) Image compressed sensing reconstruction method based on group sparse representation and weighted total variation
CN111723701B (en) Underwater target identification method
CN112541572A (en) Residual oil distribution prediction method based on convolutional encoder-decoder network
Alqadi et al. Proposed Implementation Method to Improve LSB Efficiency
CN112991483B (en) Non-local low-rank constraint self-calibration parallel magnetic resonance imaging reconstruction method
CN105989336A (en) Scene identification method based on deconvolution deep network learning with weight
CN113113030A (en) High-dimensional damaged data wireless transmission method based on noise reduction self-encoder
CN113865859A (en) Multi-scale multi-source heterogeneous information fusion gearbox state fault diagnosis method
CN108765313A (en) High spectrum image denoising method based on low-rank representation in class
Qian et al. DTAE: Deep tensor autoencoder for 3-D seismic data interpolation
CN114548532A (en) VMD-based TGCN-GRU ultra-short-term load prediction method and device and electronic equipment
CN114202459A (en) Blind image super-resolution method based on depth prior
CN107622476B (en) Image Super-resolution processing method based on generative probabilistic model
CN111714124B (en) Magnetic resonance film imaging method, device, imaging equipment and storage medium
CN116310851B (en) Remote sensing image change detection method
CN116306780B (en) Dynamic graph link generation method
CN110288002B (en) Image classification method based on sparse orthogonal neural network
CN115330901B (en) Image reconstruction method and device based on compressed sensing network
CN111368680A (en) Wave atom transformation-based deep learning anti-aliasing seismic data regularization method
CN110956221A (en) Small sample polarization synthetic aperture radar image classification method based on deep recursive network
CN105931184A (en) SAR image super-resolution method based on combined optimization
CN110838096B (en) Seismic image completion method based on information entropy norm

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