A distributed acoustic wave sensing system data noise reduction method, device and storage medium

Abstract

The present invention relates to a method, device, and storage medium for data denoising in a distributed acoustic wave sensing system. The method comprises: step S1, acquiring data from the distributed acoustic wave sensing system; step S2, preprocessing the data acquired in step S1; step S3, constructing a fusion algorithm model comprising a fusion algorithm input layer, a fusion algorithm hidden layer, and a fusion algorithm output layer; and step S4, using the fusion algorithm model constructed in step S3 to perform denoising on the data preprocessed in step S2. Compared with existing technologies, the present invention offers advantages such as signal enhancement and noise suppression, thereby obtaining a detection signal with a high signal-to-noise ratio.

Description

Technical Field

[0001] The present invention relates to the field of distributed optical fiber sensing, and in particular to a data noise reduction method, device and storage medium for a distributed acoustic wave sensing system based on a fusion algorithm. Background Art

[0002] Over the past few decades, fiber optic sensors have garnered widespread attention due to their advantages, including resistance to electromagnetic interference, compact size, remote monitoring capabilities, and low cost. Furthermore, in some applications, optical fiber can be used for both communication and sensing, further demonstrating the technology's advantages. A variety of fiber optic sensing solutions exist, enabling the measurement of a variety of parameters along the fiber. Due to the high sensitivity of the φ-OTDR (phase-sensitive optical time-domain reflectometer) commonly used in DAS (distributed acoustic sensing) systems, which can monitor vibration signals caused by acoustic waves, it has broad application prospects in areas such as structural monitoring, oil exploration, seismic monitoring, and perimeter security. However, the high-precision detection and widespread adoption of DAS systems remain challenging. One reason for this is that DAS testing is susceptible to interference from noise, which significantly impacts test results and, in severe cases, can even cause signals to be buried in the noise, making them difficult to detect.

[0003] At present, there are mainly the following noise suppression methods for DAS systems:

[0004] The first is digital cumulative averaging, the most widely used data processing method in optical time-domain reflectometry systems. It achieves this by summing and averaging multiple measurement results to suppress noise and extract the signal. While digital cumulative averaging is easy to implement in DAS systems, it also has drawbacks such as reduced system test bandwidth, limited ability to suppress non-random noise, and the need for extended periods of time to collect sufficient data samples.

[0005] The second method is to use the wavelet transform method to denoise the data. The wavelet transform is a transform domain processing method. The basic principle of filtering out noise is that in the wavelet domain, as the decomposition scale changes, the wavelet coefficients of the useful signal and the noise have different change trends. The wavelet coefficients with larger values ​​are controlled by the signal, and the wavelet coefficients with smaller values ​​are controlled by the noise. In this way, it is necessary to select a suitable threshold to distinguish the useful signal from the noise; but the wavelet transform method is a type of linear model, and it is not adaptable in selecting basis functions, determining the threshold size and the degree of decomposition, which limits its application.

[0006] The third method is to use adaptive filtering. Adaptive filtering is a widely used signal processing technology for extracting detection information in harsh environments. It belongs to a processing method in the filtering domain. Changes in the external environment will cause changes in the noise in the DAS system. This will make it difficult for the fixed-order filter designed based on experience to function. However, the adaptive filtering system has a noise signal as a reference, so it can adjust the filtering parameters and frequency response according to the changes in noise, thereby ensuring a relatively good filtering effect.

[0007] Although these signal processing methods achieve good denoising effects, the three noise reduction algorithms mentioned above still have some problems that have not been completely solved in φ-OTDR applications with a wide variety of noise types. For example, it is difficult to determine the optimal parameters in the algorithm, the denoising of low signal-to-noise ratio signals is not ideal, a large number of frequency characteristics of the signal are lost while removing noise, and the amount of data that needs to be processed per unit time in practical applications increases by several orders of magnitude. This makes traditional signal processing solutions no longer able to meet actual needs.

[0008] In order to further promote the development and application of distributed fiber optic sensing technology in high-sensitivity and long-distance fields, it is urgent to further optimize the noise reduction algorithm to improve system performance. Summary of the Invention

[0009] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method, device and storage medium for data noise reduction in a distributed acoustic wave sensing system.

[0010] The purpose of the present invention can be achieved by the following technical solutions:

[0011] According to a first aspect of the present invention, a method for data denoising of a distributed acoustic wave sensing system is provided, the method comprising:

[0012] Step S1, data acquisition of a distributed acoustic wave sensing system;

[0013] Step S2, preprocessing the data obtained in step S1;

[0014] Step S3, constructing a fusion algorithm model, which includes a fusion algorithm input layer, a fusion algorithm hidden layer, and a fusion algorithm output layer;

[0015] In step S4, the fusion algorithm model constructed in step S3 is used to perform noise reduction on the data pre-processed in step S2.

[0016] As a preferred technical solution, the step S1 is specifically as follows:

[0017] Step S101: a modulated optical pulse signal emitted by a distributed acoustic wave sensing system is input into a sensing optical fiber;

[0018] Step S102, obtaining a phase information of a measurement along the optical fiber after processing the optical signal returned from the sensing optical fiber;

[0019] Step S103 : performing N measurements to form a two-dimensional curve consisting of N pieces of phase information distributed along the optical fiber.

[0020] As a preferred technical solution, the sensing optical fiber is an ordinary single-mode optical fiber, a bend-resistant optical fiber, a few-mode optical fiber, or a scattering-enhanced optical fiber with a grating engraved on the above optical fibers.

[0021] As a preferred technical solution, the step S2 is specifically as follows:

[0022] An N×N fiber coupler is used to output N signals, which are then passed through a DC isolation module and then filtered out using a bandpass filter with a center frequency of Δf. The signal is then subjected to noise reduction processing before demodulation.

[0023] As a preferred technical solution, N is 2 or 3. When N is 2, the two signals of the fiber coupler have a 90° phase difference. Therefore, when merging the two signals, one of the signals needs to be multiplied by -1 and then added. After noise reduction processing of the combined signal, I / Q (Inphase / Quadrature) phase demodulation or Hilbert transform (Hilbert) demodulation can be used. When N = 3, the phase difference between the three output signals of the 3×3 coupler is 120°. After entering the noise reduction processing, DCM (Differential-cross-multiplying, DCM) demodulation, inverse tangent demodulation, I / Q demodulation, and Hilbert demodulation methods can be selected.

[0024] As a preferred technical solution, the fusion algorithm input layer in step S3 is single input or dual input.

[0025] As a preferred technical solution, the single input is a single-frame signal or a multi-frame signal on a long time axis, wherein when a single-frame signal is used as the input of the neural network, only the signal of the independent frame is learned, and when a multi-frame signal is used as the input of the network, the association between the previous and next frames of the signal can be learned.

[0026] As a preferred technical solution, the number of frames of the multi-frame signal segments must be greater than or equal to the repetition frequency of the detection pulse or the minimum frequency of the target signal, that is, each signal segment contains at least a complete cycle of the vibration signal with the minimum frequency.

[0027] As a preferred technical solution, the dual input is the previous frame signal x t-1 and the current frame signal x tThe two-dimensional folding matrix of , or the two-dimensional matrix of the previous multi-frame signal segment X_{tw:t-1} and the next multi-frame signal segment X_{t:t+w-1} on the long time axis.

[0028] As a preferred technical solution, the hidden layer of the fusion algorithm in step S3 includes a custom layer, a neural network layer and a fully connected layer.

[0029] As a preferred technical solution, the neural network layer is a convolutional neural network layer, a recurrent neural network layer, or a combination of a recurrent neural network layer and a recurrent neural network layer.

[0030] As a preferred technical solution, when the custom layer is a wavelet transform method, its optimal decomposition scale l, wavelet coefficient threshold λ and filter coefficient w are parameters to be learned, and the optimal parameters are determined by learning these parameters during training.

[0031] As a preferred technical solution, when the custom layer is an adaptive filtering noise reduction method, the decomposition order l and the weight coefficient w of each order are used as parameters to be learned, so that the l-order decomposition signal is multiplied by the respective weight coefficients to synthesize the filtered signal, and the energy loss E of the filtered signal is evaluated. loss As the value approaches 0, the R coefficient approaches 1.

[0032] As a preferred technical solution, the forms of CNN used in the hidden layer of the fusion algorithm in step S3 include visual geometry group model and dense network model.

[0033] As a preferred technical solution, the output layer of the fusion algorithm in step S3 uses the Huber error loss function to calculate the regression problem in training.

[0034] According to a second aspect of the present invention, a data noise reduction device for a distributed acoustic wave sensing system is provided, comprising:

[0035] Data acquisition module, used for data acquisition of distributed acoustic wave sensing system;

[0036] A preprocessing module, used for preprocessing the data acquired by the data acquisition module;

[0037] A model building module is used to build a fusion algorithm model, which includes a fusion algorithm input layer, a fusion algorithm hidden layer and a fusion algorithm output layer;

[0038] The denoising module is used to perform denoising on the data preprocessed by the preprocessing module using the fusion algorithm model constructed by the model building module.

[0039] According to a third aspect of the present invention, a distributed acoustic wave sensing system data denoising device is provided, which is characterized in that it includes a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the distributed acoustic wave sensing system data denoising method.

[0040] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, characterized in that a program is stored thereon, and when the program is executed by a processor, the distributed acoustic wave sensing system data noise reduction method is implemented.

[0041] Compared with the prior art, the present invention has the following advantages:

[0042] 1) This invention integrates traditional noise reduction methods into deep learning algorithms in the form of custom layers, leveraging the advantages of both traditional noise reduction algorithms and deep learning to obtain a more efficient noise reduction algorithm, achieve signal enhancement and noise suppression, and thus obtain a detection signal with a high signal-to-noise ratio;

[0043] 2) The present invention optimizes the traditional multi-parameter denoising algorithm through learning and constructs a denoising algorithm based on deep learning. It can achieve fast and efficient denoising even when only a small amount of data is collected, thereby obtaining a high signal-to-noise ratio signal while reducing the time cost of data denoising. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 Schematic diagram of the distributed acoustic wave sensing system framework for signal acquisition;

[0045] Figure 2(a) is a schematic diagram of data preprocessing for two detectors; Figure 2(b) is a schematic diagram of data preprocessing for three detectors;

[0046] Figure 3 Schematic diagram of the single-input traditional denoising algorithm and the deep learning algorithm framework of recurrent neural network or convolutional neural network;

[0047] Figure 4 A schematic diagram of a deep learning algorithm framework that connects a single-input traditional denoising algorithm with a recurrent neural network and a convolutional neural network.

[0048] Figure 5 Schematic diagram of the dual-input traditional denoising algorithm and the deep learning algorithm framework of recurrent neural network or convolutional neural network;

[0049] Figure 6 A schematic diagram of a deep learning algorithm framework that connects a dual-input traditional denoising algorithm with a recurrent neural network and a convolutional neural network.

[0050] Figure 7Schematic diagram of the Rayleigh scattering signal with and without noise reflected from the same optical fiber;

[0051] Figure 8 Schematic diagram of the input-output model of a single-input deep neural network;

[0052] Figure 9 Schematic diagram of the dual-input deep neural network input and output model;

[0053] Figure 10 is a flow chart of the method of the present invention;

[0054] Figure 11 Schematic diagram of the functional modules of the device of the present invention;

[0055] Figure 12 Schematic diagram of the structure of the device of the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0057] To address the problems of traditional noise reduction methods, such as difficulty determining optimal parameters and suboptimal noise reduction of low signal-to-noise ratio signals, the present invention provides a fusion method for noise reduction of data obtained from a distributed acoustic wave sensing system. This method not only removes noise while retaining useful signals, but also enhances the signal of test data in complex noisy environments, improving the signal-to-noise ratio. The method of the present invention specifically includes the following steps:

[0058] (1) Data acquisition of distributed acoustic wave sensing system

[0059] The distributed fiber-optic sensing system includes a narrow-linewidth laser, an acousto-optic modulator, an optical amplifier, a filter, a fiber circulator, a sensing fiber, and a photodetector. A modulated optical pulse signal emitted by the sensing system is input into the sensing fiber. The optical signal returning from the sensing fiber interferes with the local oscillator optical signal through heterodyne detection and is converted into an electrical signal. After analog-to-digital conversion and digital signal processing, phase information distributed along the fiber is obtained. By performing N measurements, a two-dimensional curve consisting of N pieces of phase information distributed along the fiber can be generated.

[0060] More preferably, the sensing optical fiber can be a common single-mode optical fiber, a bend-resistant optical fiber, a few-mode optical fiber, or a scattering-enhanced optical fiber with a grating engraved on the above optical fibers.

[0061] (2) Data preprocessing

[0062] After obtaining the multi-path interference signal, the data is preprocessed. If a 2×2 coupler is used to output two signals, they are differentially combined into one signal and then filtered through a bandpass filter with a center frequency of Δf to remove the carrier signal. If a 3×3 coupler is used to output three signals, the three signals are individually filtered. Finally, the filtered data is used for supervised deep learning to reduce data noise.

[0063] (3) Construction of fusion algorithm

[0064] The fusion algorithm integrates traditional noise reduction methods into a deep learning algorithm via custom layers. A deep neural network can be simplified into a structure consisting of an input layer, hidden layers, and an output layer. The data entering the input layer is denoted as X, and the data output by the output layer is denoted as Y. X and Y are the bandpass filtered data collected by the distributed acoustic wave sensing system. X and Y can be two-dimensional matrices folded from each frame of the signal sequence, or they can be two-dimensional matrices stacked from multiple frames of signal sequence. The former performs noise reduction on each frame of the signal, while the latter performs noise reduction on multiple frames simultaneously. The noise in the distributed acoustic wave sensing system is represented as N. The input to the neural network can be the sum of the high signal-to-noise ratio signal Y and the system noise N, i.e., X = Y + N.

[0065] More preferably, the output label Y of the neural network is obtained by connecting a section of ultra-low reflectivity fiber grating (FBG) string less than 1000m as a sensing fiber to a distributed acoustic wave sensing system, collecting a high signal-to-noise ratio signal and using it as the output label of the neural network.

[0066] More preferably, the noise N of the distributed acoustic wave sensing system is obtained by reducing the intensity of the system's output optical pulses and removing the connection of the sensing fiber, ensuring only the minimum analog-to-digital conversion input. This allows the system noise floor to be obtained and superimposed on the high signal-to-noise ratio signal Y to obtain the neural network input data X.

[0067] (4) Fusion algorithm input layer

[0068] The input X of the deep neural network can be a single input or a dual input. The single input can be a single frame signal or multiple frame signals on a long time axis. When the single frame signal is used as the input of the neural network, only the signal of the independent frame is learned. It can learn to obtain the properties of independently distributed noise, but cannot obtain the previous and next correlations of the noise on the long time axis. The data volume is large and the training is time-consuming. The multi-frame signal as the input of the network can learn the correlation between the previous and next frames of the signal, but the denoising effect of the single frame will be reduced. The advantage is that the data processing time is short and the speed is fast. The number of frames of the multi-frame signal segment must be greater than or equal to the repetition frequency of the detection pulse / the minimum frequency of the target signal, that is, each signal segment contains at least a complete cycle of the vibration signal with the minimum frequency.

[0069] More preferably, when the network input is dual input, the deep neural network can not only learn the features of a single input, but also comprehensively learn the features of multiple inputs to extract more feature signals, thereby having stronger denoising capabilities and more complete signal retention. A single frame as input can be the previous frame signal x t-1 and the current frame signal x t The two-dimensional folding matrix of the deep neural network learns the temporal characteristics of the noise from the signal with the previous and next frames, and then filters out more noise to obtain a signal with a high signal-to-noise ratio y t The multi-frame input can be a two-dimensional matrix of the previous multi-frame signal segment X_{tw:t-1} and the next multi-frame signal segment X_{t:t+w-1} on the time axis.

[0070] (5) Fusion algorithm hidden layer

[0071] More preferably, the hidden layer of the deep neural network mainly includes a custom layer, a convolutional neural network layer, a recurrent neural network layer and a fully connected layer. The custom layer is composed of a traditional noise reduction algorithm, which can be the wavelet transform and adaptive filtering noise reduction algorithm mentioned above. When the custom layer is a wavelet transform method, its optimal decomposition scale l, wavelet coefficient threshold λ and filter coefficient w are parameters to be learned, and these parameters are learned during training to determine the optimal parameters. When the custom layer is an adaptive filtering noise reduction method, its decomposition order l and the weight coefficient w of each order are used as parameters to be learned, so that the l-order decomposition signal is multiplied by the respective weight coefficients to synthesize the filtered signal, and the energy loss E of the filtered signal is evaluated. loss As the value approaches 0, the R coefficient approaches 1.

[0072] More preferably, when the custom layer is a wavelet transform method, it can be divided into two steps, namely wavelet decomposition and wavelet reconstruction. The size of the input image I is 2z×2n.

[0073] Wavelet decomposition consists of the following steps:

[0074] 1) Initialize the wavelet transform order to l, the threshold λ and the weight coefficient to w;

[0075] 2) When the order l is not 0, a set of low-pass filters H(w) and high-pass filters G(w) are constructed by the weight coefficients w;

[0076] 3) Use this set of decomposition filters H and G to filter the two-dimensional signal I to obtain m1 and m2 respectively:

[0077]

[0078]

[0079] 4) Then the output result is downsampled, that is, every other row is sampled to achieve wavelet decomposition;

[0080]

[0081]

[0082] 5) The decomposition results in two halves of length. One is the smoothed portion of the original signal, rm1, produced by the low-pass filter, and the other is the detailed portion, rm2, produced by the high-pass filter. rm1 and rm2 are concatenated to produce rm, which has the same size as I.

[0083] rm=[rm1,rm2]

[0084] 6) Use the same set of decomposition filters H and G to filter the two-dimensional signal rm to obtain m1 and m2 respectively:

[0085]

[0086]

[0087] 7) Then the output result is downsampled, that is, every other column is sampled to achieve wavelet decomposition;

[0088]

[0089]

[0090] 8) The decomposition result produces two parts, cm1 and cm2, of half length. Concatenate cm1 and cm2 to obtain cm, which has the same size as the original image I.

[0091] cm=[cm1,cm2]

[0092] 9) Execute the order l minus 1. When step 1 is satisfied, repeat steps 2 to 8 until the order is 0, and obtain the final two-dimensional wavelet decomposition matrix I new =cm, which contains 4 subgraphs, namely the approximate matrix LL, the horizontal matrix LH, the vertical matrix HL and the diagonal matrix HH.

[0093] 10) Evaluate the noise level. The hard threshold and soft threshold are obtained by initializing the threshold. First, the image I new Split into 4 sub-images LL, LH, HL and HH;

[0094]

[0095]

[0096]

[0097]

[0098] 11) After splicing the horizontal matrix LH, vertical matrix HL and diagonal matrix HH, calculate the new threshold λ new .

[0099] NL=[HL,LH,HH]

[0100]

[0101] 12) Image after wavelet decomposition I new With the new threshold λ new The hard threshold λ is obtained by the following formula hard and soft threshold λ soft ;

[0102] λ hard =I new ×(|I new |>λ new )

[0103] λ soft =[sign(I new )×(|I new |-λ new )]×(|I new |>λ new )

[0104] 13) Set the hard threshold λ hard and soft threshold λ soft Perform wavelet reconstruction respectively to obtain the denoised image I denoise Due to the hard and λ soft The wavelet reconstruction process is consistent with that in the following. λ Represents a hard / soft thresholded image.

[0105] Wavelet reconstruction includes the following steps:

[0106] 14) As in step 1), a set of low-pass filters H(w) and high-pass filters G(w) are constructed by the weight coefficient w, and G is flipped from left to right to obtain a new high-pass filter G new ;

[0107] G new =flip(G)

[0108] 15) Image I λ Split into 4 sub-images LL, LH, HL and HH;

[0109]

[0110]

[0111]

[0112]

[0113] 16) When the order l is not 0, perform the following process.

[0114] 17) combining the approximation matrix LL and the horizontal matrix LH into a single matrix AH and performing upsampling by interpolating every other row, with the interpolation matrix being a zero matrix;

[0115] AH=[LL1,0,LL2,0,...,LL z ,0,LH1,0,LH2,0,...,LH z ]

[0116] 18) Flip the matrix AH from left to right and then pass it through a low-pass filter to obtain AH new ;

[0117]

[0118] 19) The vertical matrix HL and the diagonal matrix HH are combined into a single matrix VD and up-sampling is achieved by interpolating every other row, with the interpolation matrix being a zero matrix;

[0119] VD=[HL1,0,HL2,0,...,HL z ,0,HH1,0,HH2,0,...,HH z ]

[0120] 20) After flipping the matrix VD from left to right and passing it through high-pass filtering, VD is obtained new ;

[0121]

[0122] 21) Add the filtered images in steps 13 and 15 to obtain a new matrix HG;

[0123] HG=VD new +AH new

[0124] 22) Flip HG and split it into two equal-sized parts by column, namely CL and CH;

[0125] HG=flip(HG)

[0126] CL=HG(:,1:n)

[0127] CH=HG(:,1+n:2×n)

[0128] 23) Perform upsampling by inserting every other row into the column, and the insertion matrix is ​​a zero matrix;

[0129] CL=[CL1,0,CL2,0,...,CL n ,0]

[0130] CH=[CH1,0,CH2,0,...,CH n ,0]

[0131] 24) After being flipped, they are respectively subjected to low-pass and high-pass filtering.

[0132]

[0133]

[0134] 25) After adding, flip the matrix to get the denoised image I denoise .

[0135] I denoise =flip(CL new +CH new )

[0136] 26) When the order l is reduced by 1 and step 16 is satisfied, steps 16 to 25 are repeated until the order is 0, and the final denoised image I after two-dimensional wavelet reconstruction is obtained. denoise .

[0137] 27) The custom layer includes steps 1 to 26. The denoised image obtained by the custom layer is further subjected to the subsequent neural network layer for further denoising. The loss function is evaluated during the training process, and the optimal decomposition scale l, wavelet coefficient threshold λ and filter coefficient w are optimized during the continuous training process.

[0138] More preferably, when the custom layer is adaptive filtering denoising, the denoised image I is obtained after the denoising process is implemented. denoise . Add the denoised image I to the loss function of the neural network denoise The energy loss E calculated after comparing with the original image I loss And the correlation coefficient R. The training process makes the energy loss close to 0 and the R coefficient close to 1 by training the decomposition order l and weight coefficient w.

[0139] More preferably, the forms of CNN used in the hidden layers of deep neural networks include the Visual Geometry Group (VGG) model and the DenseNet model. VGGNet has three advantages: 1) stronger nonlinear expression capabilities; 2) a significant reduction in the number of parameters; and 3) strong transferability. The advantages of DenseNet are: 1) higher computational efficiency, that is, only a very small number of feature maps need to be learned, and some redundant feature maps do not need to be learned again. 2) Feature reuse or feature multiplexing, DenseNet's skip connection mode allows each layer to access the feature maps of all previous layers, and can also effectively alleviate the problems of gradient diffusion and model degradation. 3) Implicit deep supervision.

[0140] Furthermore, the hidden layer of the deep neural network also includes a pooling layer, an activation function layer, and a batch normalization layer.

[0141] (6) Fusion algorithm output layer

[0142] The further output layer uses the Huber error loss function to calculate the regression problem during training.

[0143] The following is through Figure 1 The following examples are used to describe the construction method and operation of the sensor system of the present invention. It should be understood that the examples described herein are merely for illustrative purposes and are not intended to limit the scope of the invention. Furthermore, the embodiments described in the following examples may be combined with one another as long as they do not technically conflict.

[0144] Figure 1 Figure 1 shows a phase-type optical time-domain reflectometry (φ-OTDR) system for detecting backward Rayleigh scattering signals. The system comprises a narrow-linewidth laser 1, a first fiber coupler 2, an acousto-optic modulator 3, an acousto-optic modulator driver 4, a first erbium-doped fiber amplifier 5, a first bandpass filter 6, and a fiber circulator 7. A light pulse is injected into the first port of the fiber circulator through this path and then injected into the sensing fiber 8 from the second port. The Rayleigh scattered signal reflected from the sensing fiber returns to the second port of the fiber circulator and is output from the third port of the fiber circulator. The signal then passes through a second erbium-doped fiber amplifier 9, a second bandpass filter 10, and then combines with the light emitted from the narrow-linewidth laser 1 and transmitted through the transmission local oscillator fiber 11. After passing through a polarization controller 12, the signal is coupled to a second n×n coupler 13 and output from the coupler's n output ports to n photodetectors 14. Finally, the n detection signals are recorded and stored by a data acquisition system 15 for subsequent data processing.

[0145] Figure 2 shows the data preprocessing process after acquisition. Two detectors detect two signals with a phase difference of 180°. Subtracting the two signals yields a signal equivalent to the average of the two signals. This signal is filtered and used for the subsequent signal noise reduction process, as shown in Figure 2(a). Three detectors detect three signals with a phase difference of 120°. These are filtered after the DC isolation module removes the DC and are used for the subsequent signal noise reduction process, as shown in Figure 2(b).

[0146] Figure 3 This is an example of the signal denoising algorithm flow. In the figure, a, c, d, f, and g represent the input layer, custom layer, recurrent neural network layer or convolutional neural network layer and their variants, fully connected layer, and output layer, respectively.

[0147] Figure 4 This is an example of the signal denoising algorithm flow. In the figure, a, c, g represent the input layer, custom layer, recurrent neural network layer and its variants, convolutional neural network layer and its variants, fully connected layer, and output layer, respectively. The recurrent neural network layer and convolutional neural network layer are interchangeable.

[0148] Figure 5 This is an example of the signal denoising algorithm flow. In the figure, a through d, and f through g represent input 1 and input 2 of the input layer, the custom layer, the recurrent neural network layer or convolutional neural network layer and their variants, the fully connected layer, and the output layer, respectively.

[0149] Figure 6 This is an example of the signal denoising algorithm flow. Figures a through g represent input 1 and input 2 of the input layer, the custom layer, the recurrent neural network layer and its variants, the convolutional neural network layer and its variants, the fully connected layer, and the output layer, respectively. The recurrent neural network layer and the convolutional neural network layer are interchangeable.

[0150] Figure 7 is the Rayleigh scattered signal reflected from the same optical fiber. The solid line is the low-noise signal, which can be considered as a noise-free signal due to its high signal-to-noise ratio, and the dotted line is the noisy signal.

[0151] Figure 8 It is a single-input denoising model. The data input to the denoising algorithm are all two-dimensional data, which can be a fold of a frame of data, such as Figure 8 As shown in (a), it can also be a stack of multiple frames of data on a long time axis, such as Figure 8 The output is the corresponding noise-reduced signal.

[0152] Figure 9 It is a dual-input denoising model. The data input to the denoising algorithm are all two-dimensional data, which can be a fold of a frame of data, such as Figure 9 As shown in (a), it can also be a stack of multiple frames of data on a long time axis, such as Figure 9 If the input is a single frame of data, the two inputs are the previous frame time signal x t-1 and the current time signal x t The output is the noise reduction signal y of the current frame t If the input is a two-dimensional graph of multiple frames of data, the two inputs are the signals of the adjacent time segments X_{tw:t-1} and X_{t:t+w-1} respectively. The output is the noise reduction signal Y_{t:t+w-1} of the current time segment.

[0153] The above is an introduction to a method embodiment. The following further illustrates the solution of the present invention through an apparatus embodiment.

[0154] like Figure 11 As shown, a distributed acoustic wave sensing system data noise reduction device includes:

[0155] Data acquisition module 100, used for data acquisition of distributed acoustic wave sensing system;

[0156] A preprocessing module 200 is used to preprocess the data acquired by the data acquisition module;

[0157] A model building module 300 is used to build a fusion algorithm model, which includes a fusion algorithm input layer, a fusion algorithm hidden layer and a fusion algorithm output layer;

[0158] The noise reduction module 400 is used to perform noise reduction on the data preprocessed by the preprocessing module using the fusion algorithm model constructed by the model construction module.

[0159] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0160] The present invention also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 10 A data noise reduction method for a distributed acoustic wave sensing system is provided.

[0161] The present invention also provides Figure 12 The one shown corresponds to Figure 10 Schematic diagram of the data noise reduction device of the distributed acoustic wave sensing system. Figure 12 As mentioned above, at the hardware level, the distributed acoustic wave sensing system data noise reduction device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 10Of course, in addition to software implementation, the present invention does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0162] Improvements to a technology can be clearly distinguished as either hardware improvements (for example, improvements to circuit structures such as diodes, transistors, and switches) or software improvements (improvements to process flows). However, with technological advancements, many process flow improvements today can now be considered direct improvements to hardware circuit structures. Designers almost always program the improved process flow into the hardware circuit to obtain the corresponding hardware circuit structure. Therefore, it cannot be said that a process flow improvement cannot be implemented using a hardware module. For example, a programmable logic device (PLD), such as a field programmable gate array (FPGA), is an integrated circuit whose logical function is determined by user programming. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to hire a chip manufacturer to design and manufacture a dedicated integrated circuit chip. Moreover, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly done using "logic compiler" software. This is similar to the software compiler used when developing programs. Before compilation, the original code must also be written in a specific programming language, called a hardware description language (HDL). There is not just one HDL, but many, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art will also understand that by simply programming the method flow in one of these hardware description languages ​​and then programming it into an integrated circuit, a hardware circuit that implements the logic method flow can be easily obtained.

[0163] The controller can be implemented in any suitable manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that in addition to implementing the controller in a purely computer-readable program code format, the controller can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules that implement the method and structures within the hardware component.

[0164] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0165] For the convenience of description, the above device is described as being divided into various units according to their functions. Of course, when implementing the present invention, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0166] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0167] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0168] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0169] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0170] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0171] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0172] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0173] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0174] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0175] The present invention may be described in the general context of computer-executable instructions, such as program modules, executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media, including storage devices.

[0176] The various embodiments of the present invention are described in a progressive manner. Similar portions between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiment is generally similar to the method embodiment, so its description is relatively simple. For relevant portions, refer to the description of the method embodiment.

[0177] The foregoing is merely an embodiment of the present invention and is not intended to limit the present invention. It will be apparent to those skilled in the art that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are intended to be included within the scope of the claims of the present invention.

Claims

1. A method for data noise reduction in a distributed acoustic wave sensing system, characterized in that: The method includes: Step S1, data acquisition of a distributed acoustic wave sensing system; Step S2, preprocessing the data obtained in step S1; Step S3, constructing a fusion algorithm model, which includes a fusion algorithm input layer, a fusion algorithm hidden layer, and a fusion algorithm output layer; Step S4, using the fusion algorithm model constructed in step S3 to perform noise reduction on the data pre-processed in step S2; The fusion algorithm input layer in step S3 is single input or dual input; the dual input is the previous frame signal x t-1 and the current frame signal x t The two-dimensional folding matrix of , or the two-dimensional matrix of the previous multi-frame signal segment X_{tw:t-1} and the next multi-frame signal segment X_{t:t+w-1} on the long time axis; The hidden layer of the fusion algorithm in step S3 includes a custom layer, a neural network layer and a fully connected layer; When the custom layer is a wavelet transform method, its optimal decomposition scale l, wavelet coefficient threshold λ and filter coefficient w are parameters to be learned, and these parameters are learned during training to determine the optimal parameters; When the custom layer is an adaptive filtering noise reduction method, the decomposition order l and the weight coefficient w of each order are used as parameters to be learned, so that the l-order decomposition signal is multiplied by the respective weight coefficients to synthesize the filtered signal, and the energy loss E of the filtered signal is evaluated. loss As the value approaches 0, the R coefficient approaches 1.

2. The method for data noise reduction of a distributed acoustic wave sensing system according to claim 1, characterized in that: The step S1 is specifically as follows: Step S101: a modulated optical pulse signal emitted by a distributed acoustic wave sensing system is input into a sensing optical fiber; Step S102, obtaining a phase information of a measurement along the optical fiber after processing the optical signal returned from the sensing optical fiber; Step S103 : performing N measurements to form a two-dimensional curve consisting of N pieces of phase information distributed along the optical fiber.

3. The method for data noise reduction of a distributed acoustic wave sensing system according to claim 2, characterized in that: The sensing optical fiber is an ordinary single-mode optical fiber, a bend-resistant optical fiber, a few-mode optical fiber, or a scattering-enhanced optical fiber with a grating engraved on the above optical fibers.

4. The method for data noise reduction of a distributed acoustic wave sensing system according to claim 1, wherein: The step S2 is specifically as follows: An N×N fiber coupler is used to output N signals, which are then passed through a DC isolation module and then filtered out using a bandpass filter with a center frequency of Δf.

5. The method for data noise reduction of a distributed acoustic wave sensing system according to claim 4, characterized in that: The N is 2 or 3.

6. The method for data noise reduction of a distributed acoustic wave sensing system according to claim 1, characterized in that: The single input is a single-frame signal or a multi-frame signal on a long time axis, wherein when a single-frame signal is used as the input of the neural network, only the signal of the independent frame is learned, and when a multi-frame signal is used as the input of the network, the association between the previous and next frames of the signal can be learned.

7. The method for data noise reduction of a distributed acoustic wave sensing system according to claim 6, characterized in that: The number of frames of the multi-frame signal segments must be greater than or equal to the repetition frequency of the detection pulse or the minimum frequency of the target signal, that is, each signal segment contains at least a complete cycle of the vibration signal with the minimum frequency.

8. The method for data noise reduction of a distributed acoustic wave sensing system according to claim 1, characterized in that: The neural network layer is a convolutional neural network layer, a recurrent neural network layer, or a combination of a recurrent neural network layer and a convolutional neural network layer.

9. The method for data noise reduction of a distributed acoustic wave sensing system according to claim 1, characterized in that: The forms of CNN used in the hidden layer of the fusion algorithm in step S3 include visual geometry group model and dense network model.

10. The method for data noise reduction of a distributed acoustic wave sensing system according to claim 1, characterized in that: The output layer of the fusion algorithm in step S3 uses the Huber error loss function to calculate the regression problem in training.

11. A distributed acoustic wave sensing system data noise reduction device, characterized in that: include: Data acquisition module, used for data acquisition of distributed acoustic wave sensing system; A preprocessing module, used for preprocessing the data acquired by the data acquisition module; A model building module is used to build a fusion algorithm model, which includes a fusion algorithm input layer, a fusion algorithm hidden layer and a fusion algorithm output layer; A denoising module is used to perform denoising on the data preprocessed by the preprocessing module using the fusion algorithm model constructed by the model building module; The fusion algorithm input layer is single input or dual input; the dual input is the previous frame signal x t-1 and the current frame signal x t The two-dimensional folding matrix of , or the two-dimensional matrix of the previous multi-frame signal segment X_{tw:t-1} and the next multi-frame signal segment X_{t:t+w-1} on the long time axis; The hidden layers of the fusion algorithm include custom layers, neural network layers, and fully connected layers; When the custom layer is a wavelet transform method, its optimal decomposition scale l, wavelet coefficient threshold λ and filter coefficient w are parameters to be learned, and these parameters are learned during training to determine the optimal parameters; When the custom layer is an adaptive filtering noise reduction method, the decomposition order l and the weight coefficient w of each order are used as parameters to be learned, so that the l-order decomposition signal is multiplied by the respective weight coefficients to synthesize the filtered signal, and the energy loss E of the filtered signal is evaluated. loss As the value approaches 0, the R coefficient approaches 1.

12. A distributed acoustic wave sensing system data noise reduction device, characterized in that: The method comprises a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, the method is used to implement the data denoising method of the distributed acoustic wave sensing system according to any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that A program is stored thereon, and when the program is executed by a processor, the data noise reduction method of a distributed acoustic wave sensing system according to any one of claims 1 to 10 is implemented.

Images