Overhead ground wire monitoring ultrasonic guided wave signal denoising method, system, equipment and medium

Abstract

The invention discloses an overhead ground wire monitoring ultrasonic guided wave signal denoising method, system equipment and a medium. The method comprises the following steps: acquiring and preprocessing an original signal of an overhead ground wire; adding noise into the preprocessed signal to obtain a mixed signal; constructing a de-noising auto-encoder model fused through a one-dimensional convolutional neural network and a full-connection layer; inputting the mixed signal into a trained denoising auto-encoder model; an input signal is mapped to a hidden layer of the model through a model encoder to obtain a signal feature; and performing data reconstruction on the signal features in the hidden layer through a model decoder to obtain a de-noised signal. According to the method, the interference noise is added into the input signal for training, so that the robustness of the implicit layer on the interference noise existing in the input signal is improved; through a one-dimensional convolutional neural network and full-connection fusion mode, feature information in the ultrasonic guided wave signals is captured more accurately, the problem of signal distortion is avoided, and therefore the recognition capacity of tiny defects in the overhead ground wire is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of signal denoising, and in particular to an overhead ground wire monitoring ultrasonic guided wave signal denoising method, system, device and medium. BACKGROUND

[0002] The safety monitoring of overhead ground wires has become a key to ensuring the stable operation of power transmission lines. Ultrasonic guided wave detection technology is widely used in the detection of defects in overhead ground wires due to its long-distance and high-sensitivity advantages. However, the actual collected ultrasonic guided wave signals are easily disturbed by environmental noise, resulting in low signal-to-noise ratio and blurred features, which seriously affects the accuracy of defect identification. Traditional denoising methods, such as wavelet threshold denoising and adaptive filtering, have certain effects, but they have limitations in non-stationary noise scenarios.

[0003] At present, denoising autoencoders (DAE) based on deep learning have gradually become a research trend in signal denoising due to their automatic feature extraction and nonlinear mapping capabilities. However, existing models still have some problems: 1. Most denoising autoencoders based on 1DCNN and FC fusion, although they use 1DCNN to enhance the ability to extract ultrasonic guided wave (UGW) signal features, this fusion method fails to fully exploit the advantages of both network structures, and does not achieve deep collaboration between the two in the feature extraction and denoising process, resulting in the inability to fully exploit useful information in the signal; 2. The training data expansion method has deficiencies, leading to a decline in denoising performance when the model faces actual noise; 3. In some overhead ground wire monitoring scenarios where signal quality is extremely high, existing methods cannot better preserve the phase information of the signal while ensuring denoising effect. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides an overhead ground wire monitoring ultrasonic guided wave signal denoising method and system to solve the problems that the current model cannot achieve deep collaboration in the feature extraction and denoising process, cannot fully exploit useful information in the signal, the noise adding method is single and the denoising performance declines, and cannot better preserve the phase information of the signal while ensuring denoising effect.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides an overhead ground wire monitoring ultrasonic guided wave signal denoising method, comprising:

[0008] Obtaining an original signal of an overhead ground wire and performing preprocessing;

[0009] Adding noise to the preprocessed signal to obtain a mixed signal;

[0010] constructing a denoising autoencoder model fused by a one-dimensional convolutional neural network and a full connection layer;

[0011] inputting the mixed signal into the trained denoising autoencoder model;

[0012] the input signal is mapped to the hidden layer of the model by the model encoder to obtain signal features;

[0013] the signal features in the hidden layer are reconstructed by the model decoder to obtain a denoised signal.

[0014] As a preferred scheme of the overhead ground wire monitoring ultrasonic guided wave signal denoising method, the preprocessed signal is added to the noise to obtain a mixed signal, which comprises:

[0015] The noise power of the original signal is calculated, and the noise signal is subjected to Gaussian distribution according to the size of the back propagation of the signal-to-noise ratio at the noise power;

[0016] The preprocessed signal and the noise signal are superimposed to generate a mixed signal data set with different signal-to-noise ratios.

[0017] As a preferred scheme of the overhead ground wire monitoring ultrasonic guided wave signal denoising method, the denoising autoencoder model fused by the one-dimensional convolutional neural network and the full connection layer comprises:

[0018] an encoder and a decoder, and a PReLU activation function is used between the one-dimensional convolutional neural network layers.

[0019] As a preferred scheme of the overhead ground wire monitoring ultrasonic guided wave signal denoising method, the denoising autoencoder model fused by the one-dimensional convolutional neural network and the full connection layer further comprises:

[0020] The first two layers of the one-dimensional convolutional neural network encoding layer are respectively convolved with a convolution kernel and a filter, the input signal is mapped to the hidden layer of the model through nonlinear transformation combined with the full connection layer to obtain signal features.

[0021] As a preferred scheme of the overhead ground wire monitoring ultrasonic guided wave signal denoising method, the denoising autoencoder model fused by the one-dimensional convolutional neural network and the full connection layer further comprises:

[0022] The filter two-layer convolution kernel of the decoder is used for inverse convolution operation combined with the full connection layer to reconstruct the signal features in the hidden layer through nonlinear transformation.

[0023] The beneficial effects of the preferred scheme are that the convolution layer and the inverse convolution layer can realize efficient feature extraction and signal reconstruction through the cooperative work of the convolution layer and the inverse convolution layer.

[0024] As a preferred scheme of the overhead ground wire monitoring ultrasonic guided wave signal denoising method, the method further comprises training a denoising autoencoder model, and specifically comprises:

[0025] The denoising autoencoder model is initialized, model training parameters are set, and mean square error is used as a loss function;

[0026] The training set signal in the mixed signal is input into the denoising autoencoder model;

[0027] The parameters in the network are dynamically updated according to back propagation, the parameters of the model are adjusted, the value of the loss function is gradually reduced, and the pre-set standard is reached until the trained denoising autoencoder model is obtained.

[0028] The loss function is mainly used to measure the error between the reconstructed signal of the decoder and the useful signal corresponding to the noise input signal.

[0029] As a preferred scheme of the overhead ground wire monitoring ultrasonic guided wave signal denoising method, the method further comprises quantitatively evaluating the performance of the denoising autoencoder model, and specifically comprises:

[0030] The improved signal-to-noise ratio and mean square error of the numerical simulation signal before and after noise reduction are calculated;

[0031] The sizes of the improved signal-to-noise ratio and the mean square error are judged to determine the performance of the model.

[0032] In a second aspect, the present application provides an overhead ground wire monitoring ultrasonic guided wave signal denoising system, comprising:

[0033] The acquisition module is configured to acquire an overhead ground wire original signal and perform preprocessing;

[0034] The mixing module is configured to add noise to the preprocessed signal to obtain a mixed signal;

[0035] The model construction module is configured to construct a denoising autoencoder model fused by a one-dimensional convolutional neural network and a fully connected layer;

[0036] The input module is configured to input the mixed signal into the trained denoising autoencoder model;

[0037] The mapping module is configured to map the input signal to the hidden layer of the model through the model encoder to obtain signal features;

[0038] The reconstruction module is configured to reconstruct the signal features in the hidden layer through the model decoder to obtain a denoised signal.

[0039] In a third aspect, the present application provides a computer device, comprising:

[0040] a memory and a processor;

[0041] The memory is used for storing computer executable instructions, and the processor is used for executing the computer executable instructions, which realize the steps of the overhead ground wire monitoring ultrasonic guided wave signal denoising method.

[0042] In a fourth aspect, the present application provides a computer readable storage medium storing computer executable instructions, which realize the steps of the overhead ground wire monitoring ultrasonic guided wave signal denoising method when executed by a processor.

[0043] Compared with the prior art, the present application has the following beneficial effects: the present application increases the robustness of the implicit layer to the interference noise in the input signal by adding interference noise in the input signal for training; by combining the one-dimensional convolutional neural network (1DCNN) with the full connection (FC) fusion mode, the feature extraction capability of the 1DCNN can be utilized to more accurately capture the feature information in the ultrasonic guided wave signal, and effectively remove the noise components in the signal; at the same time, in the denoising process, by reasonable network structure design and training strategy, the signal distortion problem that may occur in the traditional denoising method is avoided, and the original characteristics of the signal can be well preserved. The PReLU activation function is used between layers of the 1DCNN, which can better adapt to the nonlinear characteristics of the signal, so that the error between the useful signal reconstructed by the decoder and the original useful signal is smaller, and the accuracy and reliability of the denoised signal are guaranteed, providing a high-quality data basis for subsequent signal analysis and defect judgment. Overall, the signal-to-noise ratio (SNR) of the ultrasonic guided wave signal after denoising is significantly improved, so that the weak defect echo signal originally covered by noise can be highlighted, and the identification ability of the micro defect in the overhead ground wire is improved. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0045] Figure 1 The overall flowchart of the overhead ground wire monitoring ultrasonic guided wave signal denoising method described in an embodiment of the present application.

[0046] Figure 2 The architecture of the DAE and the principle of layer-by-layer training in the overhead ground wire monitoring ultrasonic guided wave signal denoising method described in an embodiment of the present application. DETAILED DESCRIPTION

[0047] In order to make the above objectives, characteristics and advantages of the present application more apparent, comprehensible and easier to be understood, specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the protection scope of the present application.

[0048] Embodiment 1, reference Figure 1 For an embodiment of the present application, an overhead ground wire monitoring ultrasonic guided wave signal denoising method is provided, comprising:

[0049] S100: obtaining an overhead ground wire original signal and performing pretreatment;

[0050] S200: adding noise to the pretreated signal to obtain a mixed signal;

[0051] S300: constructing a denoising autoencoder model fused by a one-dimensional convolutional neural network and a full connection layer;

[0052] S400: inputting the mixed signal into the trained denoising autoencoder model;

[0053] S500: the input signal is mapped to the hidden layer of the model by the model encoder to obtain signal features;

[0054] S600: the signal features in the hidden layer are reconstructed by the model decoder to obtain a denoised signal.

[0055] It should be noted that in the process of monitoring the overhead ground wire, the ultrasonic guided wave detection data often contains weak defect echo signals. These weak signals are easily covered by background noise, making it difficult to accurately identify. Since the overhead ground wire is exposed to complex environments for a long time, it is disturbed by various external factors, making the noise components in the detection signal more complex and strong. The weak defect echo signal itself has low intensity, and in a strong noise environment, its characteristics are difficult to highlight, which brings great difficulty to subsequent signal analysis and defect judgment. If these weak defect echo signals cannot be effectively identified, the existence of small defects in the overhead ground wire cannot be found in time, which may cause serious safety accidents.

[0056] Meanwhile, the current denoising methods for ultrasonic guided wave signals also have obvious shortcomings. Traditional denoising methods often have difficulty in removing noise while well preserving the original characteristics of the signal. For example, although some simple filtering methods can reduce noise to some extent, they will blur the details of the signal, causing signal distortion and making subsequent defect judgment deviate. Moreover, the existing denoising methods have poor adaptability to complex noise, and when the noise characteristics change, the denoising effect will decrease significantly. In overhead ground wire monitoring, due to the complexity and variability of the detection environment, the noise characteristics are often unstable, which further limits the application effect of the existing denoising methods.

[0057] Moreover, the denoising model has limitations in feature extraction capability and denoising effect. Some models can extract part of the characteristics of the signal, but have insufficient extraction capability for complex nonlinear characteristics in the ultrasonic guided wave signal. For example, some models may not be able to fully capture the weak characteristic changes in the signal caused by defects, resulting in the denoised signal still containing a lot of noise components. In addition, the existing models have low computational efficiency when processing large-scale data, which cannot meet the real-time monitoring needs of overhead ground wires.

[0058] Therefore, by steps S100-S600, a model capable of efficiently and accurately extracting signal characteristics and denoising is constructed. By adding interference noise to the input signal for training, the robustness of the hidden layer to interference noise in the input signal is increased. By utilizing the powerful feature extraction capability of 1DCNN, the characteristic information in the ultrasonic guided wave signal can be more accurately captured, and the noise components in the signal can be effectively removed. Through reasonable network structure design and training strategy, the signal distortion problem that may occur in traditional denoising methods is avoided, and the overall quality of the ultrasonic guided wave signal and the defect recognition capability are improved.

[0059] Embodiment 2, refer to Figures 1-2 For an embodiment of the present application, based on the above embodiment, an overhead ground wire monitoring ultrasonic guided wave signal denoising method is provided.

[0060] In the embodiments of the present application, in step S100, the original signal of the overhead ground wire is acquired and preprocessed;

[0061] Specifically, the signal of the overhead ground wire can be collected by a professional ultrasonic guided wave detection device; during the collection process, the device is ensured to be in a normal working state to ensure the quality of the collected signal.

[0062] After the collection is completed, the signal is preliminarily checked and preprocessed, and the preprocessing includes: rejecting obviously abnormal or invalid signals, and retaining valid ultrasonic guided wave signals as original data for subsequent processing; since the amplitudes of the collected signals are positive and negative, and the amplitude size is uncertain, in order to avoid the signal amplitude fluctuation range being too large to reduce the artificial unsupervised learning ability, the input signal is normalized, and linear transformation is adopted to make the data points fall within the [0, 1] interval. Wherein, the transformation function is represented as:

[0063]

[0064] Wherein, x max is the maximum value of the signal, x min is the minimum value, and x new is the normalized signal value.

[0065] Through the above normalization processing, the preprocessed signal is obtained, which has a unified scale and is convenient for subsequent model processing.

[0066] In an optional embodiment, the preprocessing in S100 can also include cutting the long signal into fixed length segments (such as 1024 sampling points), adapting the input size of the one-dimensional convolutional neural network, and ensuring the consistency of the model input.

[0067] In the embodiments of the present application, the preprocessed signal is added with noise to obtain a mixed signal in step S200, including:

[0068] The noise power of the original signal is calculated, and the noise signal is subjected to Gaussian distribution according to the size of the signal-to-noise ratio back propagation at the noise power;

[0069] The preprocessed signal and the noise signal are superimposed to generate a mixed signal data set with different signal-to-noise ratios.

[0070] Specifically, the calculation formula can be represented as:

[0071]

[0072] y(n)=x(n)+noise(n)

[0073] Wherein, y(n) is the signal with noise; x(n) is the original signal; noise(n) is the noise; that is, the signal with noise y(n) is equal to the original signal x(n) plus the noise noise(n).

[0074] It should be noted that the mixed signal data set with different signal-to-noise ratios is generated by adding noise, which provides diversified data for subsequent model training.

[0075] Further, in the field of overhead ground wire monitoring ultrasonic guided wave signal denoising, there is currently a denoising autoencoder (DAE) method based on one-dimensional convolutional neural network (1DCNN) and fully connected (FC) fusion, but the model fusion method has limitations. 1DCNN is good at extracting local features, while fully connected layers have advantages in processing global features and complex mapping relationships. The existing fusion method simply combines the two, without realizing the deep collaboration of the two in feature extraction and denoising process, resulting in the inability to fully exploit useful information in the signal when dealing with complex noise and weak defect echo signals. At the same time, the model lacks dynamic adaptability to different types and intensities of noise. In actual overhead ground wire monitoring, noise sources are complex and diverse, such as environmental noise, equipment interference, etc., and noise intensity will change over time. The existing fusion model cannot automatically adjust the feature extraction and denoising strategy according to the real-time changes of noise, thereby affecting the denoising effect. Therefore, the above problems are optimized by the model construction of the following step S300, and the model structure is as shown in Figure 2 .

[0076] In the embodiments of the present application, the denoising autoencoder model (DAE) of one-dimensional convolutional neural network (1DCNN) and fully connected layer (FC) fusion constructed in step S300 includes:

[0077] An encoder and a decoder, and using a PReLU activation function between one-dimensional convolutional neural network layers.

[0078] Specifically, the PReLU activation function is represented as:

[0079]

[0080] In the embodiments of the present application, the denoising autoencoder model of one-dimensional convolutional neural network and fully connected layer fusion in step S300 further includes step A1:

[0081] A1: Convolve the first two layers of the one-dimensional convolutional neural network encoding layer with convolution kernels and filters, respectively, combine the fully connected layer, map the input signal to the hidden layer of the model through nonlinear transformation, and obtain the signal features.

[0082] Specifically, through nonlinear transformation, the encoder maps the input data to a hidden layer, which is a low-dimensional dense potential space, and can be represented as:

[0083] r=f(y)=f(ω ij y+b1)

[0084] Where ω ij and b1 are the weight matrix and bias matrix of the encoder, respectively.

[0085] Exemplarily, in the 1DCNNFC DAE, the first two layers of the one-dimensional convolutional neural network coding layer are respectively convolved with a convolution kernel with a size of 1x1 and a filter with a step of 1, to obtain signal features of 200x1 and 40x1.

[0086] It should be noted that the convolution layer is composed of convolution units, and the low convolution layer can only extract bottom features, and the deep convolution layer can extract deep features.

[0087] In the embodiment of the application, the denoising self-encoder model of the one-dimensional convolutional neural network and the fully connected layer in step S300 further includes step A2:

[0088] A2: The signal features of the hidden layer are reconstructed by non-linear transformation through the two-layer convolution kernel of the filter of the decoder and the fully connected layer.

[0089] Specifically, the decoder maps the key information in the hidden space back to the reconstructed data through another non-linear transformation, which can be represented as:

[0090]

[0091] Where ω mn , b2 are the bias matrix and the bias matrix of the decoder respectively, and f(y), g(r) are non-linear activation functions.

[0092] Exemplarily, the two-layer convolution kernel with a filter size of 1x1 and a step of 1 is used to perform deconvolution operation, to obtain the reconstructed denoising signal 800x1. At the same time, the PReLU activation function is used between the 1DCNN layers to enhance the non-linear expression ability of the model.

[0093] It should be noted that the combination of deconvolution and fully connected layer, such as the expansion of the latent space dimension, forms a complete process of "compression-reconstruction", which is the key design of the denoising self-encoder to realize signal recovery.

[0094] Overall, in the signal processing process, the 1DCNNFC DAE model realizes efficient feature extraction and signal reconstruction through the cooperative work of the convolution layer and the deconvolution layer. The fully connected layer acts as a bridge between the convolution features and the latent space, which not only compresses the redundant information under the noise interference, but also reconstructs the global structure of the useful signal. In combination with the convolution layer, a complete process of "local feature extraction-global feature abstraction-signal reconstruction" is realized.

[0095] Meanwhile, the signal of a specific size is taken as input, and the convolution operation and the deconvolution operation of the convolution kernel are performed, so that the deep features of the signal can be quickly and accurately extracted and the useful signal can be reconstructed. This efficient processing manner enables the model to have high calculation efficiency when processing large-scale ultrasonic guided wave signals, and can meet the demand of real-time monitoring of overhead ground wires and timely discover potential safety hazards.

[0096] In the embodiments of the present application, step S300 further includes training the denoising autoencoder model, specifically including steps B1-B3:

[0097] B1: initialize the denoising autoencoder model, set the model training parameters, and take the mean square error as the loss function;

[0098] Specifically, the training parameters include, for example, the learning rate, the batch size, the number of training rounds, etc.

[0099] Specifically, the mean square error (MSE) is taken as the loss function, and the calculation formula is represented as:

[0100]

[0101] B2: input the training set signal in the mixed signal to the denoising autoencoder model;

[0102] B3: dynamically update the parameters in the network according to the back propagation, adjust the parameters of the model, so that the value of the loss function gradually decreases, until the preset standard is reached, and the trained denoising autoencoder model is obtained.

[0103] In an alternative embodiment, the loss function can also be the mean absolute error (MAE).

[0104] Specifically, different signal-to-noise ratio mixed signal data sets are generated by adding noise, and the data is proportionally divided into training set signals and prediction set signals. In the training process, the parameters of the model are constantly adjusted, so that the value of the loss function gradually decreases, until the preset number of training rounds or the loss function converges, and the optimal solution of the internal parameters of the model is obtained; finally, the trained optimal model is read to perform denoising processing on the signals in the prediction set.

[0105] It should be noted that for the DAE, the greater the signal feature extraction capability of the encoder, the smaller the error between the reconstructed signal of the decoder and the useful signal corresponding to the noise input signal. Therefore, by setting the activation function and the loss function for training, the loss function is mainly used to measure the error between the reconstructed signal of the decoder and the useful signal corresponding to the noise input signal. Compared with the MAE which applies linear penalty to all errors, it may ignore the accurate reconstruction of key transient features such as guided wave pulse peaks, resulting in blurring of signal details, so the MSE is preferred.

[0106] Meanwhile, by adding noise to generate mixed signal data sets with different signal-to-noise ratios, and dividing the data into training sets and prediction sets in proportion for training and testing. This training method enables the model to adapt to signal denoising tasks in different noise environments, improving the model's generalization ability and adaptability. In practical applications, the detection environment of overhead ground wires is complex and variable, and the noise characteristics are unstable. The model of the present application can better cope with these changes and ensure good denoising effect in various situations.

[0107] In the embodiments of the present application, the mixed signal is input into the trained denoising autoencoder model in step S400;

[0108] In the embodiments of the present application, the signal input in step S500 is mapped to the hidden layer of the model by the model encoder to obtain signal features;

[0109] In the embodiments of the present application, the signal features in the hidden layer are reconstructed by the model decoder in step S600 to obtain the denoised signal.

[0110] Specifically, steps S400-S600 refer to Figure 2 The prediction set signal is input into the model, and after processing by the encoder and the decoder, the denoised signal is obtained.

[0111] It should be noted that the denoising autoencoder (DAE) is based on AE, and interference (i.e. mixed signal) is added to the input signal. In the model training process, DAE first obtains the initialized weight, and then dynamically updates the parameters in the network according to the backpropagation algorithm; therefore, DAE can increase the robustness of the hidden layer to the interference noise in the input signal. The purpose of denoising is to remove noise to ensure that the useful signal extracted is as close to the initial signal as possible.

[0112] In the embodiments of the present application, it also includes step S700: quantitatively evaluating the performance of the denoising autoencoder model, specifically including:

[0113] Calculate the improved signal-to-noise ratio (SNRimp) and mean square error (MSE) of the numerical simulation signal before and after noise reduction;

[0114] Determine the model performance by judging the size of the improved signal-to-noise ratio and the mean square error.

[0115] Specifically, the relevant formula can be expressed as:

[0116] SNR imp = SNR out - SNR in

[0117]

[0118] wherein, ∑x(n) 2 is the power of the input signal, is the power of the signal after removing the noise, and n is the data length of the signal.

[0119] It should be noted that, under normal circumstances, the SNR out of the signal is greater, the MSE is smaller, the proportion of noise in the signal is smaller, and the signal is closer to x(n), and the denoising performance of the model is stronger.

[0120] In an optional embodiment, the step S700 of quantitatively evaluating the denoising autoencoder model performance can also be achieved by comparing the visual images of the signals before and after denoising. The denoising effect of the model can be directly observed by observing the images.

[0121] It should be noted that the SNRimp reflects the improvement degree of the signal-to-noise ratio of the signal after denoising, and the MSE measures the error size between the signal after denoising and the original useful signal. Through these quantitative indicators, the denoising effect of the model can be directly evaluated. The quantitative evaluation can provide a clear basis for the optimization and improvement of the model, and ensure that the model can achieve good denoising performance under different conditions.

[0122] In summary, through the above scheme process, effective denoising of the overhead ground wire monitoring ultrasonic guided wave signal can be achieved, and the signal quality is improved, thereby providing a reliable data basis for subsequent defect identification and analysis.

[0123] Embodiment 3, the above is a schematic scheme of an overhead ground wire monitoring ultrasonic guided wave signal denoising method. It should be noted that the technical scheme of the overhead ground wire monitoring ultrasonic guided wave signal denoising system belongs to the same concept as the technical scheme of the overhead ground wire monitoring ultrasonic guided wave signal denoising method described above. The technical scheme of the overhead ground wire monitoring ultrasonic guided wave signal denoising system in this embodiment is not described in detail, and can be referred to the description of the technical scheme of the overhead ground wire monitoring ultrasonic guided wave signal denoising method described above.

[0124] The embodiment also provides an overhead ground wire monitoring ultrasonic guided wave signal denoising system, comprising:

[0125] An acquisition module is configured to acquire an overhead ground wire original signal and perform preprocessing.

[0126] A mixing module is configured to add noise to the preprocessed signal to obtain a mixed signal.

[0127] A model construction module is configured to construct a denoising autoencoder model fused by a one-dimensional convolutional neural network and a fully connected layer.

[0128] An input module is configured to input the mixed signal to the trained denoising autoencoder model.

[0129] The mapping module is configured to map the input signal to a hidden layer of the model through a model encoder to obtain signal features.

[0130] The reconstruction module is configured to reconstruct data of the signal features in the hidden layer through a model decoder to obtain a denoised signal.

[0131] The embodiment further provides a computer device suitable for the case of denoising an ultrasonic guided wave signal for overhead ground wire monitoring, which comprises a memory and a processor; the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions to implement the method for denoising an ultrasonic guided wave signal for overhead ground wire monitoring proposed in the above embodiment.

[0132] The embodiment further provides a storage medium having a computer program stored thereon, and the program is executed by a processor to implement the method for denoising an ultrasonic guided wave signal for overhead ground wire monitoring proposed in the above embodiment.

[0133] The storage medium proposed in the embodiment belongs to the same inventive concept as the method for denoising an ultrasonic guided wave signal for overhead ground wire monitoring proposed in the above embodiment, and the technical details not described in the embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0134] From the above description about the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary universal hardware, and of course can also be realized by hardware. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product can be stored in a computer readable storage medium, such as a floppy disk, a ROM, a RAM, a FLASH, a hard disk, or an optical disc, and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods of various embodiments of the present application.

[0135] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and all modifications or replacements should be covered in the scope of the claims of the present application.

Claims

1. A method for denoising ultrasonic guided wave signals for overhead ground wire monitoring, characterized in that: include: Obtain the original signal of the overhead ground wire and perform preprocessing; Add noise to the preprocessed signal to obtain a mixed signal; Build a denoising autoencoder model by fusing a one-dimensional convolutional neural network and a fully connected layer; Input the mixed signal into the trained denoising autoencoder model; The input signal is mapped to the hidden layer of the model through the model encoder to obtain signal features; The signal features in the hidden layer are reconstructed by the model decoder to obtain a denoised signal.

2. The method for denoising an ultrasonic guided wave signal for monitoring an overhead ground wire according to claim 1, wherein: Add noise to the preprocessed signal to obtain a mixed signal, including: Calculating the noise power of the original signal, and making the noise signal obey Gaussian distribution according to the magnitude of the reverse propagation of the signal-to-noise ratio at the noise power; The preprocessed signal is superimposed with the noise signal to generate mixed signal data sets with different signal-to-noise ratios.

3. The method for denoising an ultrasonic guided wave signal for monitoring an overhead ground wire according to claim 2, wherein: The denoising autoencoder model integrating the one-dimensional convolutional neural network and the fully connected layer includes: The encoder and decoder use the PReLU activation function between the 1D convolutional neural network layers.

4. The method for denoising an ultrasonic guided wave signal for monitoring an overhead ground wire according to claim 3, wherein: The denoising autoencoder model integrating the one-dimensional convolutional neural network and the fully connected layer further includes: The first two layers of the one-dimensional convolutional neural network encoding layer are convolved with the convolution kernel and the filter respectively, and combined with the fully connected layer, the input signal is mapped to the hidden layer of the model through nonlinear transformation to obtain the signal features.

5. The method for denoising an ultrasonic guided wave signal for monitoring an overhead ground wire according to claim 4, wherein: The denoising autoencoder model integrating the one-dimensional convolutional neural network and the fully connected layer further includes: The decoder's two-layer convolution kernel filter is used to perform deconvolution operations combined with a fully connected layer to reconstruct the signal features of the hidden layer through nonlinear transformation.

6. The method for denoising an ultrasonic guided wave signal for monitoring an overhead ground wire according to claim 5, wherein: It also includes training the denoising autoencoder model, specifically including: Initialize the denoising autoencoder model, set the model training parameters, and use the mean square error as the loss function; Input the training set signal in the mixed signal into the denoising autoencoder model; The parameters in the network are dynamically updated according to back propagation, and the parameters of the model are adjusted to gradually reduce the value of the loss function until it reaches the preset standard, thereby obtaining the trained denoising autoencoder model.

7. The method for denoising an ultrasonic guided wave signal for monitoring an overhead ground wire according to claim 6, wherein: It also includes quantitative evaluation of the denoising autoencoder model performance, including: Calculate the improved signal-to-noise ratio and mean square error of the numerical simulation signal before and after noise reduction; Determine the size of the improved signal-to-noise ratio and mean square error to determine the model performance.

8. An ultrasonic guided wave signal denoising system for monitoring overhead ground wires, applying the method according to any one of claims 1 to 7, characterized in that: include: An acquisition module is used to obtain the original signal of the overhead ground wire and perform preprocessing; A mixing module, used to add noise to the preprocessed signal to obtain a mixed signal; The model building module is used to build a denoising autoencoder model by fusing a one-dimensional convolutional neural network and a fully connected layer; An input module, used to input the mixed signal into the trained denoising autoencoder model; A mapping module, configured to map an input signal to a hidden layer of the model through the model encoder to obtain signal features; The reconstruction module is used to reconstruct the signal features in the hidden layer through the model decoder to obtain a denoised signal.

9. A computer device comprising: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the overhead ground wire monitoring ultrasonic guided wave signal denoising method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the method for denoising ultrasonic guided wave signals for monitoring overhead ground wires according to any one of claims 1 to 7.

Images