CN117390413A - Recognition method for distributed power optical fiber vibration signal noise reduction and time sequence feature extraction - Google Patents

Recognition method for distributed power optical fiber vibration signal noise reduction and time sequence feature extraction Download PDF

Info

Publication number
CN117390413A
CN117390413A CN202311355783.0A CN202311355783A CN117390413A CN 117390413 A CN117390413 A CN 117390413A CN 202311355783 A CN202311355783 A CN 202311355783A CN 117390413 A CN117390413 A CN 117390413A
Authority
CN
China
Prior art keywords
optical fiber
time
noise reduction
fiber vibration
sequence
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
Application number
CN202311355783.0A
Other languages
Chinese (zh)
Other versions
CN117390413B (en
Inventor
姜万昌
姜雨馨
苏畅
于皓宇
李雪
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.)
Northeast Electric Power University
Original Assignee
Northeast Dianli 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 Northeast Dianli University filed Critical Northeast Dianli University
Priority to CN202311355783.0A priority Critical patent/CN117390413B/en
Publication of CN117390413A publication Critical patent/CN117390413A/en
Application granted granted Critical
Publication of CN117390413B publication Critical patent/CN117390413B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • 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
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

The invention discloses a recognition method for noise reduction and time sequence feature extraction of a distributed power optical fiber vibration signal, relates to the field of power optical fiber signal processing, and solves the problems that the prior art cannot simultaneously perform signal noise reduction and extract the time sequence feature of the vibration signal, and the recognition accuracy rate of vibration events is low and the false alarm rate is high; then, marking event type labels on the data of the sampling points to construct an optical fiber vibration event data set; constructing an ASVD model according to the optical fiber vibration event data set, training, and testing the trained ASVD model to obtain a classification result of the optical fiber vibration event; the method improves generalization and robustness of the model in a new application scene; the mining of the time sequence state characteristics of the vibration signals is increased, and the distinguishing capability of the model on multiple events is further improved.

Description

Recognition method for distributed power optical fiber vibration signal noise reduction and time sequence feature extraction
Technical Field
The invention relates to the field of power optical fiber signal processing, in particular to a recognition method for distributed power optical fiber vibration signal noise reduction and time sequence feature extraction.
Background
The phase-sensitive light-sensitive time domain reflection technology provides a vibration sensing mode, has the advantages of distributed measurement, wide monitoring range, high sensitivity and the like, and is widely applied to the safety monitoring fields of seismic exploration, perimeter security protection, natural gas pipelines and the like. In practical monitoring application, the environment where the sensing optical fiber is located is complex and changeable, and the acquired signal contains a large amount of background noise, so that the signal-to-noise ratio of the signal is low, and the difficulty of interference identification is increased. At present, intelligent identification and differentiation of fiber along-line vibration sources remains a challenging task.
With the application of artificial intelligence in the fields of image, voice recognition and the like, more and more researches are conducted on recognizing vibration sources of a distributed optical fiber vibration sensing system by using a machine learning method.
Currently, the prior art mainly has the following defects:
1. today, there have been many studies to apply convolutional neural networks to extract and classify the characteristics of different vibration signals. However, due to the complex nature of vibration events, the use of convolutional neural networks alone for vibration signal classification may result in poor recognition on datasets constructed in certain complex scenarios. Furthermore, vibration events often imply significant temporal characteristics, whereas convolutional neural networks are mainly used to extract local structural features of the vibration signal, rather than mining for temporal features, resulting in some similar events that cannot be classified.
2. In the task of vibration signal identification, there has not been enough intensive research to mine the evolution of the sequence state of vibration signals in the time dimension. Some scholars have tried. For example, in the "ieee access" in 2020, "a novel DAS signal recognition method based on 1DCNNs-BiLSTM network space-time information extraction," a one-dimensional CNN is used to extract the time structural feature of a signal, but this method also has obvious drawbacks that a convolution kernel can only capture the time dynamics in a kernel window, and the time sequence dependency of the signal cannot be represented. Attention-based spatiotemporal convolution network for Φ -OTDR event classification "in 2021," 19th International Conference on Optical Communications andNetworks (ICOCN) "utilized a spatiotemporal convolution network (ATCN) of channel attention to capture temporal features while considering causal relationships of time domain signals and emphasizing key channel features by attention. The main disadvantages are that when processing long time sequence signals, the convolution kernel is limited in size, and long time signal state dependence cannot be captured, namely, the characterization capability is limited and the time sequence mining is insufficient.
3. In the actual monitoring process, due to the complexity of the optical fiber laying environment, the signals acquired from the sensors are inevitably mixed with a large amount of background noise, so that the characteristics are submerged by the noise, and the characteristics related to vibration cannot be accurately represented by the model, thereby influencing the subsequent recognition effect. Early signal optimization algorithms had short time fourier transforms, filters, time domain averages, etc. For example, publication No.: CN114510960a, patent name: a method for identifying the mode of distributed optical fiber sensor system features that the filter and wavelet transform are used to make noise reduction on signal data. The main disadvantage is that the noise reduction of the signal is stripped from the recognition process as part of the preprocessing, ignoring the inherent relationship between noise reduction and recognition, and possibly causing information loss and degradation of recognition performance. With the development of deep learning, some intelligent signal denoising algorithms have been proposed. For example, publication No.: CN109726642a, patent name: a noise reduction method of distributed optical fiber vibration signals based on variation modal decomposition is provided, and the noise reduction algorithm based on variation modal decomposition is provided, on the basis of carrying out variation modal decomposition on the vibration signals to filter high-frequency noise, wavelet packet threshold operation is carried out on each modal component so as to eliminate shot noise and low-frequency noise, and the signal to noise ratio of the optical fiber vibration signals can be effectively improved. Similar to the technology related to this patent, most of the research is focused on improving the signal-to-noise ratio of the vibration signal to improve the signal quality, and then identifying the noise-reduced signal input model, however, to date, no document has integrated the signal noise reduction into the identification model.
Therefore, in order to solve the above-mentioned problems, an identification method based on adaptive shrinkage noise reduction and sequence state extraction is proposed. According to the method, noise reduction is integrated into the recognition model, so that more compact coupling can be realized, the noise reduction and the recognition process are mutually influenced and mutually optimized, and the overall recognition performance and the robustness of the model are improved.
Disclosure of Invention
The invention aims at: the distributed power optical fiber vibration signal noise reduction and feature extraction recognition method solves the problems that in the prior art, signal noise reduction cannot be carried out simultaneously, time sequence features of vibration signals are extracted, and vibration event recognition accuracy is low and false alarm rate is high.
The identification method for the distributed power optical fiber vibration signal noise reduction and time sequence feature extraction comprises the following steps:
step one, acquiring a space-time matrix signal of an optical fiber vibration event, cutting the matrix signal, and retaining data of an effective space sampling point; then, marking event type labels for the data of the sampling points, and constructing an optical fiber vibration event data set;
step two, constructing an ASVD model by adopting the optical fiber vibration event data set in the step one, training, and testing the trained ASVD model to obtain a classification result of the optical fiber vibration event;
the specific training process is as follows:
setting a depth residual error network in an ASVD model as a trunk model, and carrying out noise reduction treatment on a feature map obtained by a convolution layer of each layer after convolution operation according to channels to obtain a noise-reduced feature sequence X;
and (3) carrying out time sequence state feature extraction on the feature sequence X after noise reduction by adopting a long-short-time memory network to obtain time domain features of the vibration signals, and completing training of an ASVD model.
The invention has the beneficial effects that:
1. the identification method can simultaneously carry out noise reduction and feature extraction on the vibration signal, avoids the trouble of manually carrying out data noise reduction pretreatment, and improves the generalization and robustness of the model in a new application scene; compared with the existing deep learning framework, the method has the advantages that the excavation of the time sequence state characteristics of the vibration signals is increased, the distinguishing capability of the model on multiple events is further improved, and the problem of vibration time identification and classification can be effectively solved.
2. The method of the invention introduces a channel shrinkage denoising module (CSDU, channel-wise shrinkagedenoising unit) which can learn to generate different soft thresholds according to different vibration events; in addition, different adaptive soft thresholds can be generated according to different channel characteristic differences, so that more accurate and reliable vibration signal noise reduction is realized.
3. According to the invention, the time domain feature extraction of the vibration signal is realized by using a long-short-time memory network. A three-layer cyclic neural network is constructed to capture the state evolution of the time domain sequence, and comprises two bidirectional LSTM (long-short-time memory network) and a unidirectional LSTM, wherein the bidirectional LSTM can capture wider dependency and understand the sequence more comprehensively; since vibration signal acquisition and characterization is time-sequential from left to right, the time-sequential state is modeled with unidirectional LSTM at the last layer. The deep dynamic time sequence modeling mode can fully mine time sequence information, and further improves the characterization capability.
Drawings
FIG. 1 is a flow chart of a method for identifying noise reduction and time sequence feature extraction of a distributed power optical fiber vibration signal according to the present invention;
FIG. 2 is a schematic block diagram of a shrink noise reduction module;
FIG. 3 is a schematic diagram of a confusion matrix for an ASVD model under a test set of six classes of events.
Detailed Description
Referring to fig. 1 to 3 for describing the method for identifying noise reduction and time sequence feature extraction of a distributed power optical fiber vibration signal according to the present embodiment, the method uses vibration source type identification application of a power communication optical fiber as an example, and provides a method for identifying and classifying vibration signal features of a distributed optical fiber, as shown in fig. 1, including the following steps:
step 1, acquiring space-time matrix signals of a vibration source by using phase sensitive optical time domain reflectometer equipment, marking various events with type labels, and constructing a vibration event data set;
and 1-1, taking one fiber core of the existing power communication optical fiber as a detection optical fiber, wherein the total length is 10.1km. In order to facilitate subsequent data processing, all data are uniformly and fixedly cut, only 12 spatial sampling point signals near the vibration point are reserved as signal samples of the vibration event, the spatial sampling interval is 10 meters, and the spatial sampling interval corresponds to 12 spatial sampling points of the tail part of the optical fiber (corresponds to the length of the optical fiber of 120 meters of the tail part of the optical fiber). The distributed optical fiber vibration sensing system is used for collecting vibration signals of effective space sampling points when vibration events occur, the sampling frequency is set to 8000Hz, 1.25s are accumulated to form a space-time matrix, each row of signals of the matrix are represented by the light signal intensities of different positions along the optical fiber collected at a certain moment, the column signals are represented by the signal intensities of a certain space position point in sample collection time, and the signal samples are a two-dimensional space-time matrix:
X i ＝[X i (1)，X i (2)，...，X i (M)]
the signal acquired at a certain moment i is represented by M, which is the number of effective space sampling points, in this embodiment m=12;
X＝[X 1 ，X 2 ，...，X N ]
wherein, X represents a space-time matrix of a vibration signal sample, N represents the time point of the signal, and is determined by the signal duration and the sampling frequency, in this embodiment, n=10000;
step 1-2: the power optical fiber vibration event data set comprises six types of events including background noise, excavation, knocking, watering, shaking and walking, and the label is set according to the six types of event types in the embodiment: 0.1, 2, 3, 4, 5, according to step 1-1, constructing a fiber vibration event data set for six types of events, as shown in table 1:
TABLE 1
Step 2: an ASVD model is constructed and trained based on the fiber vibration event data set until the model converges. And the test set is used to evaluate model classification effects.
Step 2-1, dividing the optical fiber vibration event data set into a training set and a test set according to the proportion of 8:2, as shown in table 1:
step 2-2, constructing an ASVD model on the data set, wherein the structure of the ASVD model is shown in figure 1;
step 2-2-1, setting a depth residual network ResNet as a main model in an ASVD model, and carrying out noise reduction treatment on a characteristic diagram obtained by convolution of each layer according to channels after the convolution operation of each layer, wherein the specific operation is shown as a channel shrinkage denoising module (CSDU, channel-wise shrinkage denoising unit) in FIG. 2, and comprises the following steps:
setting the output of a certain convolution layer as Z;
Z＝[z 1 ,…,z c ]
wherein z is k Taking absolute values of characteristic diagram elements of all output channels for the kth characteristic diagram after convolution, performing Global Average Pooling (GAP) operation on the characteristic diagrams of all output channels to obtain element average values of all the characteristic diagrams, and performing affine transformation on the average values by using MLP (multi-layer perceptron) to obtain a scaling parameter mu in the (0, 1) range k ：
Wherein, gamma k Is the kth feature mapIs a characteristic value of mu k Scaling parameter μ= [ μ ] for the scaling parameter of the kth feature map 1 ,...,μ k ,...,μ c ]The soft threshold is calculated as:
wherein c, w, h are the number of channels, width and height of the feature map, x c,w,h Is the characteristic diagram element of the c-th channel, theta k Is the soft threshold for the kth feature map. The output of the noise reduction of the feature map elements according to the obtained soft threshold is as follows:
z′ k ＝sign(x k )·mar((|x k |-θ k ,0)
where sign (-) is a sign function, x k All feature map elements for the kth channel. Each channel corresponds to a soft threshold value, and soft threshold value operation is carried out on each channel according to the formula to obtain a noise-reduced characteristic diagram Z '= [ Z' 1 ,…,z' c ]And takes it as input to the next convolutional layer.
At the last layer, the feature map is reassigned in time to obtain a sequence X, and the time length of the sequence is T. Specifically, for any time step t, X t Is the element of all feature maps of the last layer at time t.
And 2-2-2, performing time sequence state mining on the feature sequence obtained after noise reduction, wherein in order to better mine the evolution of the time sequence state, the structure of two bidirectional long-short time memory network blocks and a unidirectional long-short time memory network is adopted in the embodiment.
Wherein, a bidirectional long-short-time memory network is used for mining the sequence state, and for any time step t, an input X is given t In the bidirectional architecture, the forward and reverse hidden states of the time step are respectivelyAnd->The forward and reverse hidden state updates are:
where LSTM is a recursive layer.
Forward hidden stateAnd reverse hidden state->And connecting the two layers to obtain the hidden state required to be fed into the output layer. Finally, let(s)>Output O calculated by output layer t The method comprises the following steps:
O t ＝H t W hq +b q
wherein the weight matrixAnd bias->Is a model parameter of the output layer, h is the number of hidden layer units, and q is the number of output layer units.
In the real mode, two serial bidirectional circulating networks are adopted, and the output O= [ O ] of the first bidirectional long-short-time memory network 1 ，O 2 ，...，O T ]As the input of the next bidirectional long and short time memory network, repeating the steps described in step 2.2.2 to obtain the final output of the bidirectional long and short time memory network
In this embodiment, a sequential mode of unidirectional long-short-term memory network capturing sequence states is adopted, and an output sequence is obtained as follows:
s＝h T
wherein LSTM is a recursive layer,is input at t time, o t Is output at t time, h t ，c t Is the hidden state at the t moment, h t ，c t Is the hidden state at the last instant, i.e. at the time T-1, T represents the sequence length, and s therefore represents the hidden state of the last time step.
Step 2-2-3: given a state vector s, it is placed in a two-layer multi-layer perceptron for classification.
g＝σ(sw 1 +b 1 )
P＝φ(σ(gw 2 +b 2 ))
In the method, in the process of the invention,are all learnable variables, wherein w 1 、b 1 Weights and biases for hidden layers; w (w) 2 、b 2 For output layer weights and biases. u is the number of hidden layer units, g is the hidden layer output, σ is the activation function Relu, and φ is the activation function Softmax. Finally, a probability vector of 1 XK is obtained to represent the estimated probability value P= [ P ] of the corresponding sample under K classes 1 ，p 2 ，...，p K ]In the present embodiment, k=6.
Step 2-3: and performing identification test on the test set of six types of events by adopting an ASVD model to obtain a confusion matrix shown in figure 3. The abscissa of the confusion matrix map represents the predicted category of the model, the ordinate represents the true category of the data, and the value on the diagonal line is represented as the number of correct classifications, so that the larger the value on the diagonal line is, the smaller the value of the rest positions is, and the better the classification effect of the model is represented. The average accuracy of the model provided by the invention is 96.4% according to the confusion matrix, the average false alarm rate of six types of events is 0.7%, and the average false alarm rate is 3.53%.
The technical features of the above-described embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above-described embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples illustrate only a few embodiments of the invention, which are described in detail and are not to be construed as limiting the scope of the invention. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the invention, which are all within the scope of the invention. Accordingly, the scope of protection of the present invention is to be determined by the appended claims.

Claims (6)

1. The identification method for the distributed power optical fiber vibration signal noise reduction and time sequence feature extraction is characterized by comprising the following steps of: the method is realized by the following steps:
step one, acquiring a space-time matrix signal of an optical fiber vibration event, cutting the matrix signal, and retaining data of an effective space sampling point; then, marking event type labels for the data of the sampling points, and constructing an optical fiber vibration event data set;
step two, constructing an ASVD model by adopting the optical fiber vibration event data set in the step one, training, and testing the trained ASVD model to obtain a classification result of the optical fiber vibration event;
the specific training process is as follows:
setting a depth residual error network in an ASVD model as a trunk model, and carrying out noise reduction treatment on a feature map obtained by a convolution layer of each layer after convolution operation according to channels to obtain a noise-reduced feature sequence X;
and (3) carrying out time sequence state feature extraction on the feature sequence X after noise reduction by adopting a long-short-time memory network to obtain time domain features of the vibration signals, and completing training of an ASVD model.
2. The method for identifying the noise reduction and the time sequence feature extraction of the distributed power optical fiber vibration signal according to claim 1, wherein the method comprises the following steps:
step one, acquiring signals of all space points when an optical fiber vibration event occurs by using a distributed optical fiber vibration sensing system, and accumulating the signals of the space points to form a space-time matrix signal; cutting the signals, reserving 12 space sampling points at the tail of the optical fiber, and obtaining a two-dimensional space-time matrix of effective space sampling points;
after the labels are set according to the event types, a fiber vibration event data set comprising six different types is constructed according to the two-dimensional space-time matrix.
3. The method for identifying the noise reduction and the time sequence feature extraction of the distributed power optical fiber vibration signal according to claim 1, wherein the method comprises the following steps: in the second step, a channel shrinkage denoising module is adopted to perform denoising treatment on the feature map according to channels, and the specific process is as follows:
firstly, taking absolute values of feature map elements output by each channel, then performing GAP operation on the feature maps output by each channel to obtain element average values of each feature map, and performing affine transformation on the element average values by using MLP to obtain a scaling parameter in the (0, 1) range;
obtaining soft threshold values of all channels according to the scaling parameters; each channel corresponds to a soft threshold value, and noise reduction is carried out on the feature map elements according to the soft threshold value of each channel, so that a noise-reduced feature map is obtained and is used as the input of the next convolution layer;
and in the final layer of convolution layer, the denoised feature map is redistributed according to time to obtain a denoised feature sequence X.
4. The distributed power fiber vibration signal noise reduction and timing feature of claim 3The identification method for extracting the symptoms is characterized by comprising the following steps: scaling parameter μ of kth feature map k Expressed by the following formula:
wherein, gamma k Obtaining soft threshold value theta of the kth channel according to the scaling parameter as the characteristic value of the kth characteristic diagram k Expressed by the following formula:
wherein c, w, h are the number of channels, width and height of the feature map, x c,w,h A feature map element of the c-th channel;
soft threshold θ employing the kth channel k All feature map elements x for the kth channel k Noise reduction is carried out, and a feature map z 'after noise reduction is output' k Expressed by the following formula:
z′ k ＝sign(x k )·max(|x k |-θ k ，0)
soft threshold operation is carried out on each channel to obtain a feature map Z ' = [ Z ' after noise reduction ' 1 ,…,z′ c ]And takes it as input to the next convolutional layer.
5. The method for identifying the noise reduction and the time sequence feature extraction of the distributed power optical fiber vibration signal according to claim 1, wherein the method comprises the following steps: in the second step, the long-short-time memory network adopts the structure of two bidirectional long-short-time memory networks and one unidirectional long-short-time memory network;
at time t, a feature sequence X is input t Forward hidden stateAnd reverse hidden state->The updating is as follows:
wherein LSTM is a recursive layer; connecting the forward hidden state and the reverse hidden state, and memorizing the output value O of the network at the moment t in a bidirectional long-short time t Expressed by the following formula:
O t ＝H t W hq +b q
wherein H is t For the hidden state of the output layer, q is the number of output layer units, W hq B is a weight parameter q Is a bias parameter;
taking the output value O of the first bidirectional long-short time memory network as the input of the next bidirectional long-short time memory network, repeating the operation to obtain the output O of the second bidirectional long-short time memory network final 。
6. The method for identifying the noise reduction and the time sequence feature extraction of the distributed power optical fiber vibration signal according to claim 5, wherein the method comprises the following steps: the method adopts a sequence mode of capturing the state of the characteristic sequence by a unidirectional long-short time memory network, and obtains an output sequence as follows:
S＝h T
wherein LSTM is a recursive layer, o t For output at time t, h t ，c t In order to be in a hidden state at the time t,for input at time t, h t ，c t The hidden state at the moment T-1, T is the sequence length, s is the hidden state of the last time step;
according to the state vector s, the classification output of the multi-layer perceptron is as follows:
g＝σ(sw 1 +b 1 )
P＝φ(σ(gw 2 +b 2 ))
wherein w is 1 、b 1 To conceal layer weights and offsets, w 2 、b 2 For output layer weights and offsets, g is hidden layer output, σ is the activation function Relu, φ is the activation function Softmax; and finally obtaining the estimated probability value P of the corresponding sample under the K classes.
CN202311355783.0A 2023-10-19 2023-10-19 Recognition method for distributed power optical fiber vibration signal noise reduction and time sequence feature extraction Active CN117390413B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311355783.0A CN117390413B (en) 2023-10-19 2023-10-19 Recognition method for distributed power optical fiber vibration signal noise reduction and time sequence feature extraction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311355783.0A CN117390413B (en) 2023-10-19 2023-10-19 Recognition method for distributed power optical fiber vibration signal noise reduction and time sequence feature extraction

Publications (2)

Publication Number Publication Date
CN117390413A true CN117390413A (en) 2024-01-12
CN117390413B CN117390413B (en) 2024-06-14

Family

ID=89438588

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311355783.0A Active CN117390413B (en) 2023-10-19 2023-10-19 Recognition method for distributed power optical fiber vibration signal noise reduction and time sequence feature extraction

Country Status (1)

Country Link
CN (1) CN117390413B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118445718A (en) * 2024-05-08 2024-08-06 东北电力大学 Electric power communication optical cable DAS signal event identification method based on space-time feature extraction

Citations (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180080812A1 (en) * 2017-07-25 2018-03-22 University Of Electronic Science And Technology Of China Distributed optical fiber sensing signal processing method for safety monitoring of underground pipe network
US20180357542A1 (en) * 2018-06-08 2018-12-13 University Of Electronic Science And Technology Of China 1D-CNN-Based Distributed Optical Fiber Sensing Signal Feature Learning and Classification Method
CN110995339A (en) * 2019-11-26 2020-04-10 电子科技大学 Method for extracting and identifying time-space information of distributed optical fiber sensing signal
CN111242021A (en) * 2020-01-10 2020-06-05 电子科技大学 Distributed optical fiber vibration signal feature extraction and identification method
CN112418014A (en) * 2020-11-09 2021-02-26 南京信息工程大学滨江学院 Modulation signal identification method based on wavelet transformation and convolution long-short term memory neural network
CN113049084A (en) * 2021-03-16 2021-06-29 电子科技大学 Attention mechanism-based Resnet distributed optical fiber sensing signal identification method
WO2022027423A1 (en) * 2020-08-06 2022-02-10 大象声科（深圳）科技有限公司 Deep learning noise reduction method and system fusing signal of bone vibration sensor with signals of two microphones
CN114199362A (en) * 2021-12-17 2022-03-18 齐鲁工业大学 Distributed optical fiber vibration sensor mode identification method
CN114239749A (en) * 2021-12-23 2022-03-25 桂林电子科技大学 Modulation identification method based on residual shrinkage and bidirectional long-short term memory network
CN114510960A (en) * 2021-12-28 2022-05-17 齐鲁工业大学 Method for recognizing distributed optical fiber sensor system mode
CN114692681A (en) * 2022-03-18 2022-07-01 电子科技大学 Distributed optical fiber vibration and sound wave sensing signal identification method based on SCNN
CN116842379A (en) * 2023-06-06 2023-10-03 山东省计算中心（国家超级计算济南中心） Mechanical bearing residual service life prediction method based on DRSN-CS and BiGRU+MLP models
CN116838955A (en) * 2023-06-30 2023-10-03 浙江大学 Two-stage oil and gas pipeline line interference identification method

Patent Citations (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180080812A1 (en) * 2017-07-25 2018-03-22 University Of Electronic Science And Technology Of China Distributed optical fiber sensing signal processing method for safety monitoring of underground pipe network
US20180357542A1 (en) * 2018-06-08 2018-12-13 University Of Electronic Science And Technology Of China 1D-CNN-Based Distributed Optical Fiber Sensing Signal Feature Learning and Classification Method
CN110995339A (en) * 2019-11-26 2020-04-10 电子科技大学 Method for extracting and identifying time-space information of distributed optical fiber sensing signal
CN111242021A (en) * 2020-01-10 2020-06-05 电子科技大学 Distributed optical fiber vibration signal feature extraction and identification method
WO2022027423A1 (en) * 2020-08-06 2022-02-10 大象声科（深圳）科技有限公司 Deep learning noise reduction method and system fusing signal of bone vibration sensor with signals of two microphones
CN112418014A (en) * 2020-11-09 2021-02-26 南京信息工程大学滨江学院 Modulation signal identification method based on wavelet transformation and convolution long-short term memory neural network
CN113049084A (en) * 2021-03-16 2021-06-29 电子科技大学 Attention mechanism-based Resnet distributed optical fiber sensing signal identification method
CN114199362A (en) * 2021-12-17 2022-03-18 齐鲁工业大学 Distributed optical fiber vibration sensor mode identification method
CN114239749A (en) * 2021-12-23 2022-03-25 桂林电子科技大学 Modulation identification method based on residual shrinkage and bidirectional long-short term memory network
CN114510960A (en) * 2021-12-28 2022-05-17 齐鲁工业大学 Method for recognizing distributed optical fiber sensor system mode
CN114692681A (en) * 2022-03-18 2022-07-01 电子科技大学 Distributed optical fiber vibration and sound wave sensing signal identification method based on SCNN
CN116842379A (en) * 2023-06-06 2023-10-03 山东省计算中心（国家超级计算济南中心） Mechanical bearing residual service life prediction method based on DRSN-CS and BiGRU+MLP models
CN116838955A (en) * 2023-06-30 2023-10-03 浙江大学 Two-stage oil and gas pipeline line interference identification method

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
HUANG JC等: "A Fuber Vibration Signal Recognition Method Based on CNN-CBAM-LSTM", 《APPLIED SCIENCES-BASEL》, vol. 12, no. 17, 17 September 2022 (2022-09-17) *
卢娜;安博文;李玉涟;卢学佳;: "基于时域特征的光纤安防系统信号识别算法", 传感器与微系统, no. 04, 20 April 2017 (2017-04-20) *
郝尚青;刘建;崔振伟;杨传江;: "基于光纤传感与智能识别的管道安全预警技术", 石油规划设计, no. 02, 20 March 2017 (2017-03-20) *
陈勐勐;王峰;张旭苹;卫开夏;顾书龙;: "POTDR振动传感系统的数据处理与分析方法", 电子测量与仪器学报, no. 11, 15 November 2016 (2016-11-15) *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118445718A (en) * 2024-05-08 2024-08-06 东北电力大学 Electric power communication optical cable DAS signal event identification method based on space-time feature extraction

Also Published As

Publication number Publication date
CN117390413B (en) 2024-06-14

Similar Documents

Publication Publication Date Title
CN110570613A (en) Fence vibration intrusion positioning and mode identification method based on distributed optical fiber system
CN114239749B (en) Modulation identification method based on residual shrinkage and two-way long-short-term memory network
CN117390413B (en) Recognition method for distributed power optical fiber vibration signal noise reduction and time sequence feature extraction
CN116584951B (en) Electrocardiosignal detection and positioning method based on weak supervision learning
CN108171119B (en) SAR image change detection method based on residual error network
CN111091101A (en) High-precision pedestrian detection method, system and device based on one-step method
CN111695413A (en) Signal first arrival pickup method and device combining U-Net and Temporal encoding
CN114998837B (en) Real-time identification method, device, system and medium for intrusion event
CN113435276A (en) Underwater sound target identification method based on antagonistic residual error network
CN110348434A (en) Camera source discrimination method, system, storage medium and calculating equipment
CN116956222A (en) Multi-complexity behavior recognition system and method based on self-adaptive feature extraction
Duan et al. Multi-scale convolutional neural network for SAR image semantic segmentation
CN116388933A (en) Communication signal blind identification system based on deep learning
CN110555125A (en) Vehicle retrieval method based on local features
CN115060184A (en) Optical fiber perimeter intrusion detection method and system based on recursive graph
CN111242028A (en) Remote sensing image ground object segmentation method based on U-Net
CN111797690B (en) Fiber perimeter intrusion recognition method and device based on wavelet neural network grating array
CN112330562B (en) Heterogeneous remote sensing image transformation method and system
CN113990303A (en) Environmental sound identification method based on multi-resolution cavity depth separable convolution network
CN115100733B (en) RFID gesture recognition method, computer device, product and storage medium
CN112883905B (en) Human behavior recognition method based on coarse-grained time-frequency features and multi-layer fusion learning
CN116504253A (en) Bird voice recognition method and system based on frequency dynamic convolution model
CN112906679B (en) Pedestrian re-identification method, system and related equipment based on human shape semantic segmentation
CN114925722A (en) Perimeter security intrusion signal detection method based on generalized S transformation and transfer learning
CN114626412A (en) Multi-class target identification method and system for unattended sensor system

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