Optical fiber vibration signal denoising reconstruction method and device, equipment and storage medium
Patent Information
- Application Number
- CN202510946080.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-09
- Publication Date
- 2025-10-28
Smart Images
Figure CN120850152A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of signal denoising technology, and in particular to a method, apparatus, device and storage medium for denoising and reconstructing optical fiber vibration signals. Background Art
[0002] With the widespread application of Distributed Acoustic Sensing (DAS) technology in fields such as oil and gas pipeline monitoring, perimeter security, and geological exploration, accurately extracting abnormal events or valuable signals from large amounts of high-frequency sampled fiber vibration data has become a key issue in research and engineering practice. Although DAS systems possess high sensitivity, high spatial resolution, and wide coverage, they still face many challenges in practical applications, especially in complex environments where the acquired vibration signals are often accompanied by various types of environmental noise, such as mechanical vibration, geological disturbances, and background interference caused by climate change, severely obscuring the effective signals. Traditionally, signal processing methods such as filtering, wavelet analysis, and empirical mode decomposition have been used to preprocess the raw data. While these methods have achieved some success in certain scenarios, they rely on manual parameter setting and lack adaptive modeling capabilities for complex spatiotemporal characteristics, making them particularly difficult to handle non-stationary and nonlinear noise.
[0003] In recent years, deep learning methods have been widely adopted in fiber optic vibration signal recognition, especially models such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), which have demonstrated powerful feature extraction and classification capabilities in ideal or semi-controlled environments. However, these models generally rely on large-scale labeled data and are highly sensitive to the training environment, making them difficult to generalize to real-world scenarios. To improve the adaptability of models to real-world scenarios, some studies have introduced data augmentation and transfer learning strategies, expanding the sample distribution by perturbing the data or transferring model parameters trained under ideal conditions to real-world tasks. However, these methods are prone to performance degradation when the source and target domains are inconsistent, and may even lead to negative transfer, failing to effectively address the core issue of noise interference.
[0004] In summary, existing fiber optic vibration signal processing methods still suffer from several bottlenecks in dealing with complex environmental interference, including weak generalization ability, reliance on large amounts of manually tuned or labeled data, and insufficient modeling of multi-source noise. Therefore, a novel training framework is urgently needed that can guide the model to learn essential features from real-world data while using an ideal signal as a reference, and simultaneously construct a robust discrimination mechanism to suppress noise effects. The virtual-real coupled learning-driven fiber optic vibration signal denoising and reconstruction technology proposed in this application is designed to address these issues. It integrates a dual-structure design of ideal and real data encoders, adversarial noise separation capabilities, and adaptive feedback from multi-scale evaluation metrics, thus providing a more solid theoretical foundation and promising application prospects for intelligent signal processing in complex environments using distributed fiber optic sensing systems. Summary of the Invention
[0005] This application provides a method, apparatus, device, and storage medium for denoising and reconstructing optical fiber vibration signals, aiming to solve the problem that existing optical fiber vibration signal training methods are difficult to combat complex environmental noise. It fully captures effective abnormal signal feature information through a method of co-training with ideal data and real data.
[0006] In a first aspect, this application provides a method for denoising and reconstructing optical fiber vibration signals, comprising:
[0007] Abnormal data in the fiber optic vibration signal is filtered out to obtain preprocessed signal data;
[0008] An ideal data encoder is constructed using ideal signal data, and key features of the signal are extracted into a latent feature space using the ideal data encoder.
[0009] A real-world signal encoder is constructed using real-world signal data to map complex signals in real-world scenarios to a latent feature space.
[0010] The ideal data encoder and the actual signal encoder are jointly trained to decode the features in the latent space into reconstructed signals, respectively.
[0011] During joint training, a comprehensive loss function that includes environmental noise correction terms and signal reconstruction error terms is adopted;
[0012] Based on the ideal data encoder and the actual signal encoder after joint training, an adversarial learning mechanism is introduced for training to construct a fiber optic vibration signal denoising and reconstruction model.
[0013] The preprocessed signal data is denoised and reconstructed based on the fiber optic vibration signal denoising and reconstruction model.
[0014] The robustness and adaptability of the fiber optic vibration signal denoising and reconstruction model were verified by introducing various types of noise interference or applying the model to different practical scenarios.
[0015] In one possible design, anomalous data in the fiber optic vibration signal is filtered out to obtain preprocessed signal data, including:
[0016] Abnormal data in the fiber optic vibration signal dataset is filtered out based on a noise reduction autoencoder. The filtering process is represented as follows:
[0017]
[0018] ∈=μ E +kσ E (4)
[0019] like:
[0020] E(x i )>∈ (5)
[0021] Then determine x i This is an outlier;
[0022] In the formula, z i The encoder converts the noise signal The representation mapped to the latent space, It is a noise signal, f θ θ is the encoder's mapping function, and θ is the encoder's parameter. It uses latent space to represent z i The i-th element of the reconstructed signal, g φ φ is the mapping function of the decoder, φ is the parameter of the decoder, ∈ is the threshold set according to the mean and standard deviation, and μ is the value of μ. E It is the mean of the reconstruction error of all normal data, σ E The standard deviation of the reconstruction error of all normal data, where k is a constant, E(x) i ) is signal x i With reconstructed signal Reconstruction error between them.
[0023] In one possible design, the key features of the signal are extracted using an ideal data encoder as follows:
[0024]
[0025] y i =max(x i ,x i+1 ,…,x i+j (7)
[0026] In the formula, yi Let W represent the value of the feature map of the i-th output in the convolution operation, f() represent the activation function, and W j It is the j-th convolutional kernel used for feature extraction, x i+j Let represent the (i+j)th element of the input signal, b be the bias term used to adjust the output value, g be the parameter that determines the size of the convolution kernel within the input signal window, j be the parameter that determines the size of the pooling window within the pooling range, and x be the parameter that determines the pooling range. i It is the i-th element of the input signal, x i+1 It is the (i+1)th element of the input signal.
[0027] In one possible design, the actual signal encoder maps complex signals from a real-world scenario to a latent feature space using the following formula:
[0028]
[0029] z = μ(x) i )+σ(x i )·m (9)
[0030] In the formula, μ and σ 2 Let m be the mean and variance of the latent variable z, respectively, and m be the value from the standard normal distribution. The noise in the sampling, σ(x) i ) is signal x i The variance.
[0031] In one possible design, the ideal data encoder and the actual signal encoder are jointly trained to decode features in the latent space into reconstructed signals, including:
[0032] The ideal data encoder and the actual signal encoder are jointly trained, wherein the decoding process of the ideal data encoder is as follows:
[0033]
[0034] In the formula, b' is the bias term used to adjust the output value, and W' j These are the weights of the deconvolution kernel, y i+j This represents the (i+j)th element of the latent representation output by the encoder. It is the signal reconstructed by an ideal data encoder;
[0035] The decoding process of the actual signal encoder is as follows:
[0036] z = μ(x) i )+σ(x i )·m (11)
[0037]
[0038] In the formula, μ(x) i ) is the mean of the input data, σ(x) i ) is the standard deviation of the input data, and m is the standard normal distribution. The noise in the mid-sample, f D () is the data space mapped by the latent space variable z. It is the signal reconstructed by the actual data encoder. It is the final reconstructed signal, and α is a weighting coefficient used to adjust the contribution of the ideal signal and the actual signal in the final reconstructed signal.
[0039] In one possible design, the comprehensive loss function is expressed as:
[0040]
[0041] Where, L reconstruction It is the reconstruction error, L KL It is the corrected KL divergence, L VAE It is the total loss function of the variational autoencoder, λ noise It is a weight that controls the effect of noise on the KL divergence, used to adjust the influence of noise. It is the expectation of the noise term. It is a noise signal, and N is the number of signal samples used in the calculation.
[0042] In one possible design, the training process, based on the jointly trained ideal data encoder and the actual signal encoder, is represented by introducing an adversarial learning mechanism as follows:
[0043] The adversarial network is generated using the following formula:
[0044]
[0045] In the formula, G() is the generator network, and D() is the discriminator's judgment result on the data. It is a discriminator neural network with parameters as follows: ε() is the sigmoid activation function, which outputs a probability value representing the true probability of the data, where x is the input signal. It is the output signal after introducing the adversarial learning mechanism;
[0046] The generative adversarial network is trained based on a loss function, which is expressed as follows:
[0047]
[0048] L VAE+GAN =L VAE +λ D LD +λ G L G (twenty two)
[0049] Where, L D It is the loss of the discriminator, p data (x) is the distribution of the real data, p G (x) is the distribution of the generated data by the generator, L G It is the loss of the generator, L VAE+GAN λ is the total loss function of the fiber optic vibration signal denoising and reconstruction model. D and λ G These are hyperparameters in adversarial learning. It is a loss of real data. It is the loss of data generated by the generator.
[0050] Secondly, this application provides an optical fiber vibration signal denoising and reconstruction device, the device comprising:
[0051] The data preprocessing module is configured to filter out abnormal data in the fiber optic vibration signal to obtain preprocessed signal data.
[0052] The first encoder construction module is configured to construct an ideal data encoder using ideal signal data, and extract key features of the signal into a latent feature space through the ideal data encoder.
[0053] The second encoder construction module is configured to construct an actual signal encoder using actual signal data, mapping complex signals in the actual scene to the latent feature space.
[0054] The joint training module is configured to jointly train the ideal data encoder and the actual signal encoder to decode the features in the latent space into reconstructed signals, respectively.
[0055] The loss optimization module is configured to use a comprehensive loss function that includes an environmental noise correction term and a signal reconstruction error term during joint training.
[0056] The adversarial learning module is configured to be trained based on the ideal data encoder and the actual signal encoder after joint training, and to construct a fiber optic vibration signal denoising and reconstruction model.
[0057] The denoising and reconstruction module is configured to perform denoising and reconstruction processing on the preprocessed signal data based on the fiber vibration signal denoising and reconstruction model.
[0058] The model verification module is configured to verify the robustness and adaptability of the fiber optic vibration signal denoising and reconstruction model by introducing various types of noise interference or applying the fiber optic vibration signal denoising and reconstruction model to different practical scenarios.
[0059] Thirdly, embodiments of this application provide an electronic device, including: at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, causing the at least one processor to perform the fiber optic vibration signal denoising and reconstruction method as described in the first aspect and various possible designs of the first aspect.
[0060] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions. When a processor executes the computer-executable instructions, it implements the fiber optic vibration signal denoising and reconstruction method described in the first aspect and various possible designs of the first aspect.
[0061] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the fiber optic vibration signal denoising and reconstruction method described in the first aspect and various possible designs of the first aspect.
[0062] The fiber optic vibration signal denoising and reconstruction method, apparatus, equipment, and storage medium provided in this application have at least the following beneficial effects:
[0063] This application utilizes the complementary nature of ideal and real-world data training to effectively capture abnormal vibration signals in complex noisy environments, achieving more accurate signal feature extraction. It is particularly suitable for fiber optic vibration monitoring systems in noisy environments. This invention combines an autoencoder and adversarial learning mechanisms to effectively reduce noise interference in signal feature extraction, thereby achieving more accurate vibration identification. This application not only significantly improves the effectiveness of vibration signal feature extraction but also overcomes the limitations of traditional methods in complex, noisy environments. Attached Figure Description
[0064] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0065] Figure 1 A flowchart of a fiber optic vibration signal denoising and reconstruction method provided in this application embodiment;
[0066] Figure 2 This is a schematic diagram illustrating the abnormal signal values identified by the noise-reducing self-encoder method on a certain optical fiber vibration signal curve, provided in an embodiment of this application.
[0067] Figure 3 A schematic diagram illustrating the reconstruction effect of an ideal optical fiber vibration signal data using a convolutional autoencoder method, provided in an embodiment of this application.
[0068] Figure 4 A schematic diagram illustrating the reconstruction effect of a specific optical fiber vibration signal data using the variational autoencoder method provided in this embodiment of the application.
[0069] Figure 5 The following are signal reconstruction effect diagrams provided in the embodiments of this application for four types of vibration events;
[0070] Figure 6 This is a structural diagram of the fiber optic vibration signal denoising and reconstruction device provided in an embodiment of this application.
[0071] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION
[0072] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.
[0073] The collection, storage, use, processing, transmission, provision, and disclosure of financial data or user data involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.
[0074] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.
[0075] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.
[0076] This application provides a method for denoising and reconstructing fiber optic vibration signals, such as... Figure 1 The diagram shown is a detailed flowchart of the fiber optic vibration signal denoising and reconstruction method provided in this application embodiment. The fiber optic vibration signal denoising and reconstruction method includes the following steps S100-S500.
[0077] S100: Filter out abnormal data in the fiber optic vibration signal to obtain preprocessed signal data.
[0078] In some embodiments, the method for filtering out abnormal data in an optical fiber vibration signal dataset is as follows:
[0079] Anomalies in the fiber optic vibration signal dataset are filtered out using a noise-reducing autoencoder-based method.
[0080]
[0081] ∈=μ E +kσ E (4)
[0082] like:
[0083] E(x i )>∈ (5)
[0084] Then determine x i This is an outlier. Figure 2 An example graph of abnormal data values identified from a certain fiber optic vibration signal curve is given.
[0085] Among them, z i It is encoder f θ noise signal The representation mapped to the latent space, It is a noise signal, f θ θ is the encoder's mapping function, and θ is the encoder's parameter. It is g φ z is represented using latent space. i The i-th element of the reconstructed signal, g φ φ is the mapping function of the decoder, φ is the parameter of the decoder, ∈ is the threshold set according to the mean and standard deviation, and μ is the value of μ. E It is the mean of the reconstruction error of all normal data, σ E The standard deviation of the reconstruction error of all normal data is given by k, which is a constant and typically takes a value between 2 and 3.
[0086] S200: Construct an ideal data encoder using ideal signal data, and extract key features of the signal into the latent feature space through the ideal data encoder.
[0087] In some embodiments, an ideal data encoder is constructed using ideal fiber vibration signal data and interference from minute signals is suppressed by employing a convolutional autoencoder method:
[0088] The input signal is weighted and summed using convolution operations, then nonlinearly transformed using an activation function, and finally dimensionality reduction is achieved by using a sliding window with a max pooling formula.
[0089]
[0090] y i =max(x i ,x i+1 ,…,x i+j (7)
[0091] Among them, y i Let W represent the value of the feature map of the i-th output in the convolution operation, f() represent the activation function, and W j It is the j-th convolutional kernel used for feature extraction, x i+j denoted as the (i+j)th element of the input signal, b is the bias term used to adjust the output value, k is the size of the convolution kernel that determines the range of the input signal window, and j is the size of the pooling window that determines the pooling range.
[0092] When a signal contains obvious local patterns (such as periodic vibrations), convolution operations can effectively extract local features of the signal without being easily affected by global noise.
[0093] Figure 3 A schematic diagram comparing the ideal data of a certain original optical fiber vibration signal and the ideal data of the reconstructed optical fiber vibration signal is given.
[0094] S300: Constructs a real signal encoder using actual signal data to map complex signals in real-world scenarios to a latent feature space.
[0095] In some embodiments, complex signals in real-world scenarios are mapped into a latent spatial representation to cope with environmental noise interference, using a variational autoencoder method:
[0096]
[0097] z = μ(x) i )+σ(x i )·m (9)
[0098] Where μ and σ 2 Let m be the mean and variance of the latent variable z, respectively, and m be the value from the standard normal distribution. The noise in the sampled data.
[0099] Variational autoencoders (VAEs) can generate new samples and model the underlying distribution of data. If there are many different operating modes in an optical fiber vibration signal, VAEs can be useful, not only for denoising but also for understanding the signal's underlying structure. The generative capabilities of VAEs facilitate effective signal modeling even in the absence of labeled data.
[0100] Figure 4 A schematic diagram comparing the actual data of a certain original optical fiber vibration signal and the actual data of the reconstructed optical fiber vibration signal is given.
[0101] S400: Jointly trains the ideal data encoder and the actual signal encoder to decode the features in the latent space into reconstructed signals.
[0102] In some embodiments, the two encoders in steps S200 and S300 are trained and decoded to obtain the reconstructed signal: the decoding method of the ideal data encoder is deconvolution.
[0103]
[0104] Where b' is the bias term used to adjust the output value, and W' j These are the weights of the deconvolution kernel, y i+j This represents the (i+j)th element of the latent representation output by the encoder.
[0105] The low-dimensional signal is restored back to a high-dimensional space similar to the original signal through deconvolution, resulting in a reconstructed signal of ideal data.
[0106] The actual data encoder uses a reparameterization technique for decoding:
[0107] z = μ(x) i )+σ(x i )·m (11)
[0108]
[0109] Where, μ(x) i ) is the mean of the input data, σ(x) i ) is the standard deviation of the input data, and m is the standard normal distribution. The noise in the mid-sample. f D () is the data space mapped by the latent space variable z.
[0110] By using reparameterization techniques, the originally non-differentiable random sampling operation can be transformed into a differentiable form, thereby enabling the model to be optimized through gradient descent and effectively train the generative decoder.
[0111] The signal reconstruction capability is improved by continuously training and optimizing the two encoders.
[0112] S500: During joint training, a comprehensive loss function is used that includes an environmental noise correction term and a signal reconstruction error term.
[0113] In some embodiments, the error loss function designed for effectively repairing signal loss caused by noise and interference, including an environmental noise correction term and signal reconstruction error, is as follows:
[0114]
[0115] L reconstruction It is the reconstruction error, L KL It is the corrected KL divergence, L VAE λ is the total loss function of the variational autoencoder. noise It is a weight that controls the effect of noise on the KL divergence, used to adjust the influence of noise. It is the expectation of the noise term.
[0116] Among them, the total loss function L of the variational autoencoder VAE The smaller the value, the stronger the model's ability to reconstruct the signal.
[0117] Figure 5 A schematic diagram of the loss function for reconstructing the signal from the model is given.
[0118] S600: Based on the ideal data encoder and the actual signal encoder after joint training, an adversarial learning mechanism is introduced for training to construct a fiber optic vibration signal denoising and reconstruction model.
[0119] In some embodiments, introducing an adversarial learning mechanism for training is used to correct noise in real-world scene data. The implementation method is as follows:
[0120] Generative Adversarial Networks:
[0121]
[0122] Where G() is the generator network, and D() is the discriminator's judgment result on the data. It is a discriminator neural network with parameters as follows: ε() is the Sigmoid activation function, which outputs a probability value representing the true probability of the data.
[0123]
[0124] L VAE+GAN =L VAE +λ D L D +λ G L G (twenty one)
[0125] Among them, L D It is the loss of the discriminator, p data (x) is the distribution of the real data, p G (x) is the distribution of the generated data by the generator, L G It is the loss of the generator, L VAE+GAN λ is the total loss function of the VAE+GAN model. D and λ G It is a hyperparameter in adversarial learning.
[0126] Among them, the total loss function L of VAE and GAN models VAE+GAN The smaller the value, the better the environmental noise correction effect.
[0127] S700: Based on the fiber optic vibration signal denoising and reconstruction model, the preprocessed signal data is denoised and reconstructed.
[0128] Specifically, based on the fiber optic vibration signal denoising and reconstruction model, the model removes noise from the environment and reconstructs useful signal information by modeling and extracting features from complex noise. Specifically, the input preprocessed signal data is fed into the model, and the latent features are decoded by the decoder part, and finally the denoised signal is output.
[0129] S800: By introducing various types of noise interference or applying the fiber optic vibration signal denoising and reconstruction model to different practical scenarios, the robustness and adaptability of the fiber optic vibration signal denoising and reconstruction model are verified.
[0130] In some embodiments, the fiber optic vibration signal denoising and reconstruction model based on ideal-real reference learning established in the above steps is used to process a certain fiber optic vibration signal. Several types of noise are added first; in this embodiment, Gaussian noise, impulse noise, and random distortion noise are introduced. As an example, the calculation formula is as follows:
[0131]
[0132] n(t)=α·x(t)+β (24)
[0133] Where, σ n Let be the standard deviation of Gaussian noise, A be the pulse amplitude of the pulse signal, p be the probability of pulse noise, α be the signal distortion coefficient of random noise, β be the random noise shift factor, and x(t) be the original signal.
[0134] After introducing noise, multiple quantitative evaluation metrics are used to assess the model performance and noise reduction effect. To ensure the model's generalization ability on unknown types of fiber optic vibration signals, a comprehensive evaluation is conducted using the following metrics: Signal-to-Noise Ratio (SNR), Mean Squared Error (MSE), and Structural Similarity Index (SSIM). SNR reflects the ratio between signal strength and noise strength. A higher SNR indicates a larger proportion of useful components in the signal and a smaller impact from noise. MSE reflects the squared mean of the differences between the model's predicted and actual values; a lower MSE indicates a smaller difference between the predicted and actual values. SSIM reflects the similarity of the signal structure, with values ranging from 0 to 1; values closer to 1 indicate a stronger ability to recover the signal. Its calculation formula is as follows:
[0135]
[0136]
[0137] Where, E[x(t)| 2 ] is the mean square value of the signal, E[n(t)]| 2 [] is the mean square value of the noise. It is the mean of the predicted signal. It is the variance of the predicted signal. C1 and C2 are the covariances of the signal and the predicted signal, respectively. C1 and C2 are constants, which are usually chosen to be very small to stabilize the formula.
[0138] To verify the effectiveness of the method proposed in this application, fiber optic vibration monitoring data from a Linyuan oil refinery was used as the research object, and the proposed fiber optic vibration signal denoising and reconstruction method was experimentally verified. Experimental results show that, compared with other single methods, the method of this invention significantly improves the signal-to-noise ratio of the fiber optic vibration signal. The reconstructed fiber optic vibration signal curve can basically capture the intrusion situation of the monitored pipeline, which is of great significance for subsequent classification of fiber optic vibration types.
[0139] In terms of fiber optic vibration data selection, four types of vibration events—digging, striking, shaking, and walking—were chosen for signal reconstruction experiments. Figure 5 The reconstruction results of four types of vibration event signals are presented.
[0140] This application addresses the shortcomings of existing training methods in effectively handling fiber optic vibration signal noise and capturing useful fiber optic vibration signal features. It proposes a virtual-real coupled learning-driven method for denoising and reconstructing fiber optic vibration signals, effectively suppressing noise by combining ideal and real-world data during training. First, a denoising autoencoder is used to filter out abnormal data in the fiber optic vibration signal. Then, a convolutional autoencoder and a variational autoencoder are jointly trained. Finally, deconvolution and reparameterization are used to reconstruct the signal. This method of combining ideal and real-world data effectively suppresses noise and captures useful signal features.
[0141] Compared to traditional methods that use CNN or RNN alone, the method of this invention effectively integrates the advantages of both ideal and real data, further improving the robustness and stability of signal and noise processing, and is particularly superior in capturing useful signal features.
[0142] This application also provides an optical fiber vibration signal denoising and reconstruction device, such as... Figure 6 As shown, the fiber optic vibration signal denoising and reconstruction device includes:
[0143] The data preprocessing module 601 is configured to filter out abnormal data in the fiber optic vibration signal to obtain preprocessed signal data.
[0144] The first encoder construction module 602 is configured to construct an ideal data encoder using ideal signal data, and extract key features of the signal into a latent feature space through the ideal data encoder.
[0145] The second encoder construction module 603 is configured to construct an actual signal encoder using actual signal data, and map complex signals in the actual scene to the latent feature space.
[0146] The joint training module 604 is configured to jointly train the ideal data encoder and the actual signal encoder to decode the features in the latent space into reconstructed signals, respectively.
[0147] The loss optimization module 605 is configured to use a comprehensive loss function that includes an environmental noise correction term and a signal reconstruction error term during joint training.
[0148] The adversarial learning module 606 is configured to introduce an adversarial learning mechanism for training based on the jointly trained ideal data encoder and the actual signal encoder, and to construct a fiber optic vibration signal denoising and reconstruction model.
[0149] The denoising and reconstruction module 607 is configured to perform denoising and reconstruction processing on the preprocessed signal data based on the fiber vibration signal denoising and reconstruction model.
[0150] The model verification module 608 is configured to verify the robustness and adaptability of the fiber optic vibration signal denoising and reconstruction model by introducing various types of noise interference or applying the fiber optic vibration signal denoising and reconstruction model to different practical scenarios.
[0151] This application provides an electronic device. The electronic device may include a processor and a memory, wherein the processor and the memory can communicate; exemplarily, the processor and the memory communicate via a communication bus.
[0152] The processor executes computer execution instructions stored in memory, causing the processor to perform the scheme in the above embodiments. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0153] The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. Transceivers are used to enable communication between database access devices and other computers (e.g., clients, read-write libraries, and read-only libraries). Memory may include random access memory (RAM) and may also include non-volatile memory.
[0154] The electronic device provided in this application embodiment can be the terminal device described in the above embodiments.
[0155] This application also provides a computer-readable storage medium storing computer instructions. When the computer instructions are executed on a computer, the computer performs the technical solution of the fiber optic vibration signal denoising and reconstruction method described in the above embodiment.
[0156] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When the at least one processor executes the computer program, it can implement the technical solution of the fiber optic vibration signal denoising and reconstruction method in the above embodiments.
[0157] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.
[0158] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.
[0159] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.
[0160] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.
[0161] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.
[0162] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.
[0163] Buses can be Industry Standard Architecture (ISA) buses, Peripheral Component Interconnect (PCI) buses, or Extended Industry Standard Architecture (EISA) buses, etc. Buses can be categorized into address buses, data buses, control buses, etc.
[0164] The aforementioned storage medium can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium accessible to general-purpose or special-purpose computers.
[0165] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor. The processor and storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and storage medium can exist as discrete components in an electronic control unit or main control device.
[0166] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.
[0167] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.
Claims
1. A method for denoising and reconstructing fiber optic vibration signals, characterized in that, The method comprises: Abnormal data in the fiber optic vibration signal is filtered out to obtain preprocessed signal data; An ideal data encoder is constructed using ideal signal data, and key features of the signal are extracted into a latent feature space using the ideal data encoder. A real-world signal encoder is constructed using real-world signal data to map complex signals in real-world scenarios to a latent feature space. The ideal data encoder and the actual signal encoder are jointly trained to decode the features in the latent space into reconstructed signals, respectively. During joint training, a comprehensive loss function that includes environmental noise correction terms and signal reconstruction error terms is adopted; Based on the ideal data encoder and the actual signal encoder after joint training, an adversarial learning mechanism is introduced for training to construct a fiber optic vibration signal denoising and reconstruction model. The preprocessed signal data is denoised and reconstructed based on the fiber optic vibration signal denoising and reconstruction model. The robustness and adaptability of the fiber optic vibration signal denoising and reconstruction model were verified by introducing various types of noise interference or applying the model to different practical scenarios.
2. The fiber optic vibration signal denoising and reconstruction method according to claim 1, characterized in that, Abnormal data in the fiber optic vibration signal is filtered out to obtain preprocessed signal data, including: Abnormal data in the fiber optic vibration signal dataset is filtered out based on a noise reduction autoencoder. The filtering process is represented as follows: ∈=μ E +kσ E (4) like: Former i )>∈ (5) Then determine x i This is an outlier; In the formula, z i The encoder converts the noise signal The representation mapped to the latent space, It is a noise signal, f θ θ is the encoder's mapping function, and θ is the encoder's parameter. It uses latent space to represent z i The i-th element of the reconstructed signal, g φ φ is the mapping function of the decoder, φ is the parameter of the decoder, ∈ is the threshold set according to the mean and standard deviation, and μ is the value of μ. E It is the mean of the reconstruction error of all normal data, σ E The standard deviation of the reconstruction error of all normal data, where k is a constant, E(x) i ) is signal x i With reconstructed signal Reconstruction error between them.
3. The fiber optic vibration signal denoising and reconstruction method according to claim 1, characterized in that, The method for extracting key features of a signal using an ideal data encoder is as follows: y i =max(x i ,x i+1 ,…,x i+j ) (7) In the formula, y i Let W represent the value of the feature map of the i-th output in the convolution operation, f() represent the activation function, and W j It is the j-th convolutional kernel used for feature extraction, x i+j Let represent the (i+j)th element of the input signal, b be the bias term used to adjust the output value, g be the parameter that determines the size of the convolution kernel within the input signal window, j be the parameter that determines the size of the pooling window within the pooling range, and x be the parameter that determines the pooling range. i It is the i-th element of the input signal, x i+1 It is the (i+1)th element of the input signal.
4. The fiber optic vibration signal denoising and reconstruction method according to claim 1, characterized in that, The actual signal encoder maps complex signals in a real-world scenario to a latent feature space using the following formula: z=μ(x i )+σ(x i )·m (9) In the formula, μ and σ 2 Let m be the mean and variance of the latent variable z, respectively, and m be the value from the standard normal distribution. The noise in the sampling, σ(x) i ) is signal x i The variance.
5. The fiber optic vibration signal denoising and reconstruction method according to claim 1, characterized in that, The ideal data encoder and the actual signal encoder are jointly trained to decode features in the latent space into reconstructed signals, including: The ideal data encoder and the actual signal encoder are jointly trained, wherein the decoding process of the ideal data encoder is as follows: In the formula, b' is the bias term used to adjust the output value, and W' j These are the weights of the deconvolution kernel, y i+j This represents the (i+j)th element of the latent representation output by the encoder. It is the signal reconstructed by an ideal data encoder; The decoding process of the actual signal encoder is as follows: z=μ(x i )+σ(x i )·m (11) In the formula, μ(x) i ) is the mean of the input data, σ(x) i ) is the standard deviation of the input data, and m is the standard normal distribution. The noise in the mid-sample, f D () is the data space mapped by the latent space variable z. It is the signal reconstructed by the actual data encoder. It is the final reconstructed signal, and α is a weighting coefficient used to adjust the contribution of the ideal signal and the actual signal in the final reconstructed signal.
6. The fiber optic vibration signal denoising and reconstruction method according to claim 1, characterized in that, The comprehensive loss function is expressed as follows: In the formula, L reconstruction It is the reconstruction error, L KL It is the corrected KL divergence, L VAE It is the total loss function of the variational autoencoder, λ noise It is a weight that controls the effect of noise on the KL divergence, used to adjust the influence of noise. It is the expectation of the noise term. It is a noise signal, and N is the number of signal samples used in the calculation.
7. The fiber optic vibration signal denoising and reconstruction method according to claim 1, characterized in that, The training process based on the jointly trained ideal data encoder and the actual signal encoder, by introducing an adversarial learning mechanism, is represented as follows: The adversarial network is generated using the following formula: In the formula, G() is the generator network, and D() is the discriminator's judgment result on the data. It is a discriminator neural network with parameters as follows: ε() is the sigmoid activation function, which outputs a probability value representing the true probability of the data, where x is the input signal. It is the output signal after introducing the adversarial learning mechanism; The generative adversarial network is trained based on a loss function, which is expressed as follows: L VAE+GAN =L VAE +λ D L D +λ G L G (22) In the formula, L D It is the loss of the discriminator, p data (x) is the distribution of the real data, p G (x) is the distribution of the generated data by the generator, L G It is the loss of the generator, L VAE+GAN λ is the total loss function of the fiber optic vibration signal denoising and reconstruction model. D and λ G These are hyperparameters in adversarial learning. It is a loss of real data. It is the loss of data generated by the generator.
8. A fiber optic vibration signal denoising and reconstruction device, characterized in that, The device comprises: The data preprocessing module is configured to filter out abnormal data in the fiber optic vibration signal to obtain preprocessed signal data. The first encoder construction module is configured to construct an ideal data encoder using ideal signal data, and extract key features of the signal into a latent feature space through the ideal data encoder. The second encoder construction module is configured to construct an actual signal encoder using actual signal data, mapping complex signals in the actual scene to the latent feature space. The joint training module is configured to jointly train the ideal data encoder and the actual signal encoder to decode the features in the latent space into reconstructed signals, respectively. The loss optimization module is configured to use a comprehensive loss function that includes an environmental noise correction term and a signal reconstruction error term during joint training. The adversarial learning module is configured to be trained based on the ideal data encoder and the actual signal encoder after joint training, and to construct a fiber optic vibration signal denoising and reconstruction model. The denoising and reconstruction module is configured to perform denoising and reconstruction processing on the preprocessed signal data based on the fiber vibration signal denoising and reconstruction model. The model verification module is configured to verify the robustness and adaptability of the fiber optic vibration signal denoising and reconstruction model by introducing various types of noise interference or applying the fiber optic vibration signal denoising and reconstruction model to different practical scenarios.
9. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes the computer execution instructions stored in the memory to implement the fiber optic vibration signal denoising and reconstruction method as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the fiber optic vibration signal denoising and reconstruction method as described in any one of claims 1-7.