A method, device and storage medium for intelligently locating abnormal discharge of a converter valve

Abstract

The present invention relates to an intelligent positioning method, device and storage medium for abnormal discharge of a converter valve. The method uses a distributed microphone array to collect the on-site acoustic signal of the converter valve and the simulated acoustic signal of the converter valve, and assigns labels according to different abnormal discharge positions; performs preliminary audio processing on the acoustic signal of the converter valve, and obtains the spectrogram and sound image of the processed on-site acoustic signal of the converter valve and the simulated acoustic signal of the converter valve; uses a trained deep convolutional neural network containing a multi-attention mechanism to extract the on-site acoustic signal features of the converter valve from the spectrogram and the sound image; outputs the abnormal discharge positioning result when the on-site acoustic signal features match the existing features in the current abnormal discharge feature library, otherwise, assigns a new label to the on-site acoustic signal of the converter valve, and uses the signal features and their new labels to update the deep convolutional neural network and the abnormal discharge feature library. Compared with the prior art, the present invention has the advantages of low environmental sensitivity, intelligent automatic positioning, etc.

Description

Technical Field

[0001] The present invention relates to the technical field of high-voltage direct current transmission, and in particular to a method, a device and a storage medium for intelligently locating abnormal discharge of a converter valve. Background Art

[0002] The converter valve is a key core equipment of high-voltage direct current transmission, used to achieve AC-DC conversion of electric energy, and is known as the "heart" of the converter station. In actual operation, the converter valve will be subjected to abnormal operating conditions such as load, overvoltage, and harsh electromagnetic environment. These abnormal operating conditions and natural aging of components can easily lead to abnormal discharge of the converter valve components.

[0003] The existing online monitoring systems and manual inspection methods have blind spots, which makes it difficult to detect some abnormal discharge faults in time. The acoustic signals generated by the converter valve during operation contain a large amount of status information and have unique characteristics like human fingerprints. When the valve group generates abnormal discharge of devices in different positions, its voiceprint will also change accordingly. These subtle changes are difficult to accurately capture by the human ear. The voiceprint detection method based on acoustic signals using deep learning does not require human participation and can automatically perform intelligent analysis and diagnosis of abnormal discharge of the converter valve. For example, Chinese patent CN113791323A discloses a portable abnormal discharge ultrasonic detection device based on intelligent voiceprint recognition technology. The device collects ultrasonic signals generated by local discharge through a microphone array unit, and has a deep learning model in its voiceprint recognition and diagnosis unit. The built-in GPU chip extracts the spectral characteristics of the acoustic signal based on artificial intelligence algorithms and deep convolutional neural networks, and automatically diagnoses and classifies different discharge types according to the spectral characteristics. However, there are many components such as thyristors, capacitors, resistors, and trigger plates in the converter valve assembly, and the converter valve tower is located in a closed space, with severe reverberation. At the same time, the sound field between the valve towers will also cause mutual interference. The existing technologies including the above-mentioned solutions rarely involve consideration of environmental factors when solving the problem of accurate positioning of abnormal discharge of converter valves. Therefore, how to reduce the adverse effects of environmental factors such as sound field interference on the positioning of abnormal discharge of converter valves and further improve the accuracy of intelligent positioning have become problems that need to be solved in this field. Summary of the invention

[0004] The purpose of the present invention is to provide a method, device and storage medium for intelligently locating abnormal discharge of a converter valve in order to overcome the defect of the above-mentioned prior art that lacks consideration of environmental factors such as acoustic field interference.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] According to a first aspect of the present invention, a method for intelligently locating abnormal discharge of a converter valve is provided, comprising the following steps:

[0007] S1, using a distributed microphone array to collect converter valve acoustic signals, and assigning corresponding labels according to different abnormal discharge positions, wherein the converter valve acoustic signals include converter valve on-site acoustic signals and converter valve simulated acoustic signals;

[0008] S2, performing preliminary audio processing on the converter valve acoustic signal, and obtaining a first spectrogram and an acoustic image of the processed converter valve on-site acoustic signal and a second spectrogram and an acoustic image of the processed converter valve simulated acoustic signal, wherein the preliminary audio processing process includes sequentially performing echo elimination, noise suppression, and amplitude amplification;

[0009] S3, using a trained deep convolutional neural network including a multi-attention mechanism to locate abnormal discharge of a converter valve, and extracting acoustic signal features of the converter valve site from the first spectrogram and the sonogram, wherein the deep convolutional neural network is trained using the second spectrogram and the sonogram, and the abnormal discharge signal features of the converter valve extracted by the deep convolutional neural network during the training process and their labels are used to construct an initial abnormal discharge feature library;

[0010] S4, judging whether the acoustic signal feature of the current converter valve site matches the existing feature in the current abnormal discharge feature library: if not, executing S5; if yes, outputting the abnormal discharge positioning result;

[0011] S5, assign a new label to the current converter valve on-site acoustic signal, use the current converter valve on-site acoustic signal features and their new labels to update the current deep convolutional neural network and the current abnormal discharge feature library, and return to S4.

[0012] As a preferred technical solution, the distribution of the distributed microphone array includes arranging a microphone array between every two converter valve towers, and symmetrically arranging the microphone arrays at both ends of the entire converter valve group.

[0013] As a preferred technical solution, the echo cancellation result is achieved by using a time domain adaptive filter.

[0014] As a preferred technical solution, the noise suppression result is achieved through noise estimation, a posteriori / a priori signal-to-noise ratio calculation and Wiener filtering process.

[0015] As a preferred technical solution, the amplitude amplification process includes using an acoustic signal enhancement algorithm to amplify the amplitude of the noise-reduced converter valve acoustic signal.

[0016] As a preferred technical solution, the process of acquiring the spectrogram and the sonogram includes obtaining the corresponding spectrogram through framing, pre-emphasis, windowing and wavelet transform operations on the acoustic signal of the converter valve after preliminary audio processing, and obtaining the corresponding sonogram through acoustic imaging technology.

[0017] As an optimal technical solution, the deep convolutional neural network containing a multi-attention mechanism includes multiple efficient channel attention mechanism-convolutional neural network modules, multi-head self-attention modules, global average pooling modules, feature fusion modules and fully connected modules connected in series in sequence. The first spectrogram and sonogram, the second spectrogram and sonogram are processed by the efficient channel attention mechanism-convolutional neural network module, the multi-head self-attention module and the global average pooling module in sequence to obtain corresponding feature vectors. The feature vectors are processed by the feature fusion module to obtain fusion features. The fusion features are processed by the fully connected module to obtain the probabilities of different abnormal discharge categories, thereby realizing the abnormal discharge positioning of the converter valve.

[0018] As a preferred technical solution, the efficient channel attention mechanism-convolutional neural network module includes a convolutional neural network unit and an efficient channel attention mechanism unit, and the efficient channel attention mechanism unit is connected in series after the convolutional neural network unit.

[0019] According to a second aspect of the present invention, there is provided an intelligent positioning device for abnormal discharge of a converter valve, comprising a memory, a processor, and a program stored in the memory, wherein the method described is implemented when the processor executes the program.

[0020] According to a third aspect of the present invention, there is provided a storage medium having a program stored thereon, wherein the program implements the method described above when executed.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1. The present invention processes the acoustic signals of the converter valve collected from the distributed microphone array through echo elimination, noise suppression and amplitude amplification operations, which can effectively separate the echo and noise signals, and use a deep neural network to locate abnormal discharge, fully considering the adverse effects of environmental factors such as sound field interference on the abnormal discharge positioning of the converter valve, and has the advantages of low environmental sensitivity and intelligent automatic positioning, which can effectively reduce the influence of the measurement environment on the positioning results and improve the accuracy of the positioning of the abnormal discharge signal of the converter valve;

[0023] 2. After the collected converter valve acoustic signal is preliminarily processed, the present invention obtains the first spectrogram and sonogram of the processed converter valve on-site acoustic signal and the second spectrogram and sonogram of the processed converter valve simulated acoustic signal, and inputs the spectrograms and sonograms of the two signals into a deep convolutional neural network with multiple attention mechanism (MA-DCNN) for abnormal discharge positioning, which enriches the input information of MA-DCNN, improves the feature extraction capability of MA-DCNN, and helps to improve the accuracy of positioning;

[0024] 3. The present invention can ensure the positioning capability of the proposed method for unknown abnormal discharges when applied on site by updating the MA-DCNN and the converter valve abnormal discharge feature library, and further reduce the impact of environmental changes on abnormal discharge positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a schematic diagram of the process of the method provided in Example 1 of the present invention;

[0026] Figure 2 This is a schematic diagram of the arrangement of the distributed microphone array provided in Embodiment 1 of the present invention;

[0027] Figure 3 Schematic diagram of the ECA-CNN module provided in Example 1 of the present invention;

[0028] Figure 4 Schematic diagram of the MSA module provided in Example 1 of the present invention;

[0029] Figure 5 This is a schematic diagram of the MA-DCNN provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0030] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0031] Example 1

[0032] like Figure 1 As shown, this embodiment provides an intelligent positioning method for abnormal discharge of a converter valve, which is based on a microphone array and includes the following steps:

[0033] Step S1, using a distributed microphone array to collect converter valve acoustic signals, and assigning corresponding labels according to different abnormal discharge locations, wherein the converter valve acoustic signals include converter valve on-site acoustic signals and converter valve simulated acoustic signals.

[0034] Specifically, a distributed microphone array is used to collect the on-site acoustic signals of the converter valve and the simulated acoustic signals of the converter valve abnormal discharge fault simulation platform built in the laboratory, and different labels are assigned according to the different abnormal discharge locations. The distributed microphone array is arranged as follows: a microphone array is arranged between every two converter valve towers, and two microphone arrays are arranged at both ends of the entire converter valve group, and the microphone arrays are arranged symmetrically at both ends of the converter valve group. The arrangement structure of each microphone array is a 7×1 linear array. Figure 2 As shown, for the case of 6 converter valves, 7 linear microphone arrays are arranged on both sides of the converter valve group, for a total of 14 microphone arrays, and each microphone array is arranged in a 7×1 linear array structure.

[0035] Step S2, perform preliminary audio processing on the converter valve acoustic signal, and obtain the first spectrogram and acoustic image of the processed converter valve on-site acoustic signal and the second spectrogram and acoustic image of the processed converter valve simulated acoustic signal. The preliminary audio processing process includes echo elimination, noise suppression and amplitude amplification in sequence. The specific steps are as follows:

[0036] Step S21, i.e. the echo cancellation step, uses a time domain adaptive filter to eliminate the echo signal. The input signal is divided into two paths, one of which is transmitted to the near end and reflected by the ambient medium to form an echo signal which is then collected by the microphone array, while the other is transmitted as a reference signal to the time domain adaptive filter to output an analog echo signal. The microphone array received signal is subtracted from the analog echo signal to eliminate the echo, and finally an error signal is obtained. One path outputs the error signal, and the other path is transmitted to the time domain adaptive filter. The filter parameters are adjusted using an adaptive algorithm to minimize the error signal, thereby achieving the best echo cancellation result.

[0037] Step S22, i.e. the noise suppression step, estimates the noise signal based on the echo cancellation and removes it from the mixed signal collected by the microphone array to obtain a noise-reduced acoustic signal. First, the quantile noise estimation method is used to obtain an initial estimate of the noise, and the posterior / prior signal-to-noise ratio is calculated based on the estimate; then, the probability of acoustics / noise in each frame is analyzed by using a transformation algorithm based on the likelihood ratio function and a model that integrates the acoustic / noise classification features, and the probability is used to judge the noise and further update a more accurate noise estimate; based on the more accurate noise estimate, the posterior / prior signal-to-noise ratio is recalculated and the Wiener filter expression is derived; finally, the Wiener filter is used to complete the noise reduction and obtain the noise-reduced converter valve acoustic signal;

[0038] Step S23, i.e., the amplitude amplification step, uses an acoustic signal enhancement algorithm to amplify the amplitude of the acoustic signal of the converter valve after noise reduction. Specifically, the volume is adjusted by adjusting the sampling point amplitude of the discrete digital signal using a digital gain method to achieve acoustic signal amplitude amplification.

[0039] Step S24, namely the spectrogram and acoustic image acquisition step, specifically includes acquiring the first spectrogram and acoustic image of the processed on-site acoustic signal of the converter valve, and acquiring the second spectrogram and acoustic image of the processed simulated acoustic signal of the converter valve. The spectrograms of the processed on-site acoustic signal of the converter valve and the simulated acoustic signal can be obtained through operations such as framing, pre-emphasis, windowing, and wavelet transformation, and the acoustic images of the processed on-site acoustic signal of the converter valve and the simulated acoustic signal can be obtained through acoustic imaging technology.

[0040] Step S3, using the trained deep convolutional neural network with multiple attention mechanism (MA-DCNN) to locate the abnormal discharge of the converter valve, MA-DCNN uses the first spectrogram and sonogram of the processed converter valve on-site acoustic signal as a test set, and extracts the converter valve on-site acoustic signal features from them, and uses the test set to test the effectiveness of MA-DCNN's abnormal discharge positioning, wherein MA-DCNN uses the second spectrogram and sonogram of the processed converter valve simulated acoustic signal as a training set, and uses the training set to train MA-DCNN's converter valve abnormal discharge feature extraction and positioning capabilities, and the converter valve abnormal discharge signal features and labels extracted by MA-DCNN during the training process are used to construct an initial abnormal discharge feature library.

[0041] MA-DCNN is composed of multiple layers of convolutional neural networks (CNN). The convolutional layer in the convolutional neural network automatically extracts features from the input data. It contains multiple convolution kernels. Each element of the convolution kernel corresponds to a weight coefficient and a bias. When the convolution kernel is working, it will regularly scan the input features, perform matrix element multiplication and summation of the input features within the convolution kernel receptive field, and superimpose the bias, as shown in the following formula:

[0042]

[0043] Among them, Z l represents the feature map output of the lth layer, that is, the feature map input of the l+1th layer, Z l+1represents the feature map output of the l+1th layer after convolution calculation, b is the bias, and Z(i,j) corresponds to the pixel value of the feature map, where (i,j)∈{0,1,...,L l+1}, L l+1 =(L l +2p-f) / s0+1 is the characteristic graph Z l+1 The size of the feature map, Ch is the number of channels of the feature map, f, s0 and p correspond to the convolution kernel size, convolution step size and number of padding layers respectively. By setting different convolution layer parameters, effective discriminant information of the input image can be extracted and learned layer by layer. A CNN module consists of a convolution layer, a batch normalization layer, a maximum pooling layer and a ReLU activation layer.

[0044] After the CNN module, the Efficient Channel Attention (ECA) mechanism is connected in series to adaptively score and weight the features learned by different convolution kernels, enhance the features related to abnormal discharge information, suppress the features with interference, and improve the network performance. Assume that the feature map output by the convolution layer is Where W, H and Ch are the width, height and number of channels of the feature map respectively. First, Z obtains the aggregated features g(Z) of each channel through the global average pooling layer, and then uses the number of channels Ch to adaptively calculate the number of cross-channels k. Finally, the aggregated features g(Z) are subjected to a one-dimensional convolution operation with a convolution kernel size of k and an activation function σ(·) to calculate the weight ω of each channel and assign it to the original feature Z. As shown below:

[0045]

[0046]

[0047] ω=σ(C1D k (g(Z))) (4)

[0048] In the formula, γ and b ECA are constants, and γ=2, b ECA =1, σ(·) is the Sigmoid function.

[0049] A CNN module and an ECA module are connected in series to form an ECA-CNN module, whose structure is as follows: Figure 3 shown.

[0050] like Figure 4 As shown, Multi-head Self Attention (MSA) is then used to further process the feature map to improve the global information extraction capability of the neural network, as shown below:

[0051] MultiHead(Z)=Concat(head1,...,head h )W O (5)

[0052] head i =Attention(Q,K,V) (6)

[0053]

[0054] Among them, W O is the parameter matrix, i is the index of the number of multi-head attention heads (0≤i≤h), and the number of multi-head attention heads is h=8. They are convolution functions that replace the original linear projection to project the feature map Z into the query matrix Q, the key matrix K, and the value matrix V. The number of heads in the multi-head attention is different, and the size of the convolution kernel is also set differently, achieving the effect of multi-scale convolution. The attention function is implemented by scaling the dot product operation:

[0055]

[0056] in, is the scaling factor, d k is the dimension of the key matrix K.

[0057] Afterwards, the global average pooling layer (GAP) will be used instead of the fully connected layer to process the feature maps output by the multi-head self-attention.

[0058] like Figure 5 As shown in FIG. 1 , the spectrogram and sonogram obtained in step S2 are processed in parallel using the same structure of three ECA-CNN modules, a multi-head self-attention layer, and a global average pooling layer in series to obtain two feature vectors. The two feature vectors are then fused through a one-dimensional convolutional layer to obtain the final fused feature x fusion .

[0059] After obtaining the fusion features, the Softmax function is used to perform the fully connected layer operation to realize the abnormal discharge location of the converter valve. The Softmax function can convert the output feature vector of the deep convolutional neural network into the probability size belonging to different abnormal discharge categories. The conversion formula is as follows:

[0060]

[0061] Among them, x represents the feature vector output by the fully connected layer, y represents the different location categories of abnormal discharge, C represents the number of all abnormal discharge locations of the converter valve, and p(y|x) is the normalized conditional probability value. By optimizing the cross entropy loss function:

[0062] L = -log[p(Y|X)] (10)

[0063] The optimal classification probability value can be learned, and then the back propagation learning algorithm can be used to feedback optimize the weight coefficients in the entire neural network layer by layer. Finally, an optimal neural network structure and positioning result can be automatically learned at the same time.

[0064] The final MA-DCNN complete structure is as follows Figure 5 As shown. MA-DCNN includes multiple efficient channel attention mechanism-convolutional neural network modules (i.e., ECA-CNN modules), multi-head self-attention modules (i.e., MSA modules), global average pooling modules, feature fusion modules, and fully connected modules connected in series. The first spectrogram and sonogram, and the second spectrogram and sonogram are processed by the efficient channel attention mechanism-convolutional neural network module, the multi-head self-attention module, and the global average pooling module in sequence to obtain the corresponding feature vectors. The feature vectors are processed by the feature fusion module to obtain fusion features. The fusion features are processed by the fully connected module to obtain the probabilities of different abnormal discharge categories, thereby realizing the abnormal discharge positioning of the converter valve.

[0065] Step S4, judging whether the acoustic signal feature of the current converter valve site matches the existing feature in the current abnormal discharge feature library: if not, executing step S5; if yes, outputting the abnormal discharge positioning result;

[0066] Step S5, assigning a new label to the current converter valve on-site acoustic signal, using the current converter valve on-site acoustic signal features and their new labels to update the current deep convolutional neural network and the current abnormal discharge feature library, and returning to step S4.

[0067] In steps S4 to S5, when monitoring the abnormal discharge of the converter valve on site, if the on-site acoustic signal features extracted by MA-DCNN do not match the existing features in the current abnormal discharge feature library, a new label is assigned to the signal, and the MA-DCNN is updated using the signal and its label, and the features and labels of the signal extracted by the updated MA-DCNN are added to the converter valve abnormal discharge feature library.

[0068] In summary, this embodiment provides an intelligent positioning method for abnormal discharge of converter valves based on distributed microphone arrays. The acoustic signal of the converter valve is collected from the distributed microphone array through echo elimination, noise suppression and acoustic signal enhancement algorithm processing, which can effectively separate the echo and noise signals and reduce the influence of the measurement environment on the positioning result; the spectrogram and the sound image are obtained according to the processed acoustic signal, and are input into MA-DCNN for abnormal discharge positioning, which enriches the input information of MA-DCNN and improves the feature extraction ability of MA-DCNN; through the update of MA-DCNN and the abnormal discharge feature library of converter valves, the positioning ability of the proposed method for unknown abnormal discharges can be guaranteed when applied on site. The intelligent positioning method for abnormal discharge of converter valves based on distributed microphone arrays provided in this embodiment has the characteristics of low environmental sensitivity and intelligent automatic positioning.

[0069] Example 2

[0070] This embodiment provides an intelligent positioning device for abnormal discharge of a converter valve, including a memory, a processor, and a program stored in the memory, and the method in embodiment 1 is implemented when the processor executes the program. The device processor includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or computer program instructions loaded from a storage unit to a random access memory (RAM). In the RAM, various programs and data required for the operation of the device can also be stored. The CPU, ROM, and RAM are connected to each other through a bus. The input / output (I / O) interface is also connected to the bus. Multiple components in the device are connected to the I / O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunication networks. The processing unit performs the various methods and processes described in embodiment 1, such as steps S1 to S5 in the aforementioned embodiments. For example, in some embodiments, steps S1 to S5 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more of the steps S1 to S5 described above may be executed. Alternatively, in other embodiments, the CPU may be configured to execute steps S1 to S5 in embodiment 1 in any other appropriate manner (e.g., by means of firmware). The functions described above may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0071] Further, the present embodiment also provides a storage medium, on which a program is stored, and the method in embodiment 1 is implemented when the program is executed. The program code for implementing the method can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as an independent software package and partially on a remote machine or completely on a remote machine or server. In the context of the present invention, a computer-readable medium can be a tangible medium that can contain or store a program for use by an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of machine-readable storage media would include electrical connections based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fibers, a portable compact disk-read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0072] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for intelligently locating abnormal discharge of a converter valve, characterized in that: The following steps are involved: S1, using a distributed microphone array to collect converter valve acoustic signals, and assigning corresponding labels according to different abnormal discharge positions, wherein the converter valve acoustic signals include converter valve on-site acoustic signals and converter valve simulated acoustic signals; S2, performing preliminary audio processing on the converter valve acoustic signal, and obtaining a first spectrogram and an acoustic image of the processed converter valve on-site acoustic signal and a second spectrogram and an acoustic image of the processed converter valve simulated acoustic signal, wherein the preliminary audio processing process includes sequentially performing echo elimination, noise suppression, and amplitude amplification; S3, using a trained deep convolutional neural network including a multi-attention mechanism to locate abnormal discharge of a converter valve, and extracting acoustic signal features of the converter valve site from the first spectrogram and the sonogram, wherein the deep convolutional neural network is trained using the second spectrogram and the sonogram, and the abnormal discharge signal features of the converter valve extracted by the deep convolutional neural network during the training process and their labels are used to construct an initial abnormal discharge feature library; S4, judging whether the acoustic signal feature of the current converter valve site matches the existing feature in the current abnormal discharge feature library: if not, executing S5; if yes, outputting the abnormal discharge positioning result; S5, assign a new label to the current converter valve on-site acoustic signal, use the current converter valve on-site acoustic signal features and their new labels to update the current deep convolutional neural network and the current abnormal discharge feature library, and return to S4.

2. The intelligent positioning method for abnormal discharge of a converter valve according to claim 1 is characterized in that: The distributed microphone array is distributed in the following manner: a microphone array is arranged between every two converter valve towers, and microphone arrays are symmetrically arranged at both ends of the entire converter valve group.

3. The intelligent positioning method for abnormal discharge of converter valve according to claim 1, characterized in that: The echo cancellation result is achieved by using a time domain adaptive filter.

4. The intelligent positioning method for abnormal discharge of a converter valve according to claim 1, characterized in that: The noise suppression result is achieved through noise estimation, a posteriori / a priori signal-to-noise ratio calculation and Wiener filtering process.

5. The intelligent positioning method for abnormal discharge of a converter valve according to claim 1, characterized in that: The amplitude amplification process includes using an acoustic signal enhancement algorithm to amplify the amplitude of the noise-reduced converter valve acoustic signal.

6. The intelligent positioning method for abnormal discharge of a converter valve according to claim 1, characterized in that: The acquisition process of the spectrogram and the sonogram includes obtaining the corresponding spectrogram through framing, pre-emphasis, windowing and wavelet transform operations for the acoustic signal of the converter valve after preliminary audio processing, and obtaining the corresponding sonogram through acoustic imaging technology.

7. The intelligent positioning method for abnormal discharge of a converter valve according to claim 1, characterized in that: The deep convolutional neural network including a multi-attention mechanism includes a plurality of efficient channel attention mechanism-convolutional neural network modules, a multi-head self-attention module, a global average pooling module, a feature fusion module and a fully connected module connected in series in sequence. The first spectrogram and sonogram, the second spectrogram and sonogram are processed by the efficient channel attention mechanism-convolutional neural network module, the multi-head self-attention module and the global average pooling module in sequence to obtain corresponding feature vectors. The feature vectors are processed by the feature fusion module to obtain fusion features. The fusion features are processed by the fully connected module to obtain the probabilities of different abnormal discharge categories, thereby realizing the abnormal discharge positioning of the converter valve.

8. The intelligent positioning method for abnormal discharge of a converter valve according to claim 7, characterized in that: The efficient channel attention mechanism-convolutional neural network module includes a convolutional neural network unit and an efficient channel attention mechanism unit, and the efficient channel attention mechanism unit is connected in series after the convolutional neural network unit.

9. An intelligent positioning device for abnormal discharge of a converter valve, comprising a memory, a processor, and a program stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 8 is implemented.

10. A storage medium having a program stored thereon, characterized in that: When the program is executed, the method according to any one of claims 1 to 8 is implemented.

Images