Distribution network fault characteristic quantity extraction method, system and device and storage medium

Abstract

The invention relates to the technical field of power faults, and discloses a distribution network fault characteristic quantity extraction method, system and device and a storage medium, and the method comprises the steps: obtaining a sound wave electric signal, processing the sound wave electric signal, and obtaining a screened sound wave electric signal; inputting the screened sound wave electric signals into a neural network model, and obtaining a first fault type and a second fault type through a convolutional layer; the fault type is obtained by judging the first fault type and the second fault type, the fault recognition effect is further improved, and missed recognition is avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power failure, and in particular to a distribution network fault feature extraction method, system, device and storage medium. BACKGROUND

[0002] Power equipment failure not only causes power outages and economic losses, but also can cause more serious safety accidents. Therefore, it is of great theoretical significance and practical value to study the diagnosis and prediction methods of equipment failure. Through effective fault diagnosis, the reliability and stability of the power system can be improved, and the downtime and maintenance cost can be reduced. In addition, accurate fault prediction can help operation and maintenance personnel take preventive measures in advance to avoid accidents. On the other hand, with the continuous advancement of the digital transformation of the power grid, various sensor technologies are widely used in the state detection of power equipment, and the power Internet of Things is built based on this, which can improve the ability of power grid equipment fault prediction and is crucial to the safety of the power grid system.

[0003] In the vast structure of the power system, electrical primary equipment is the core component of energy conversion and transmission. However, electrical primary equipment has been running under high load for a long time, and its internal components gradually age and wear out, leading to frequent failures. These failures not only cause equipment damage and power outages, but also can cause fires, explosions and other safety accidents, posing a great threat to social economy and people's life and property safety. Therefore, it is of great significance to carry out research on electrical equipment fault diagnosis to ensure the safe and stable operation of the power system and improve the reliability of power supply. In existing technologies, light-weight deep neural networks are used to identify motor bearing signals, which can accurately detect equipment failure. However, this method usually lacks explainability and it is difficult to guarantee the fault diagnosis effect in practical applications. Some methods use adaptive neural fuzzy networks to identify equipment vibration signals, which can effectively and accurately diagnose equipment failure. However, this method has high complexity and it is difficult to guarantee the fault diagnosis efficiency in practical applications. Moreover, when collecting multi-feature voiceprint signals of electrical primary equipment, the original signals contain many sharply fluctuating signals due to the interference of the microphone's own sensitivity characteristics and external environment, which affects the subsequent fault diagnosis effect. SUMMARY

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

[0005] Therefore, the present application provides a distribution network fault feature extraction method, system, device and storage medium to solve the problem that the sound wave signal collection is easily disturbed and the feature extraction is inaccurate in existing distribution network fault diagnosis, resulting in poor fault recognition 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 a distribution network fault feature extraction method, comprising:

[0008] Obtaining an acoustic electric signal, processing the acoustic electric signal to obtain a screened acoustic electric signal;

[0009] Inputting the screened acoustic electric signal into a neural network model to obtain a first fault type and a second fault type through a convolution layer;

[0010] Obtaining a fault type by judging the first fault type and the second fault type.

[0011] As a preferred scheme of the distribution network fault feature extraction method, the method comprises the following steps:

[0012] Adjusting a first position of a microphone within a first preset time to obtain a first acoustic electric signal and a second acoustic electric signal;

[0013] Adjusting a second position of the microphone within a second preset time to obtain a third acoustic electric signal and a fourth acoustic electric signal.

[0014] As a preferred scheme of the distribution network fault feature extraction method, the method comprises the following steps:

[0015] Obtaining a first mean acoustic wave graph of a voiceprint signal of the second acoustic electric signal, and obtaining a first average energy value according to the first mean acoustic wave graph;

[0016] Filtering an acoustic wave graph of the first acoustic electric signal according to the first average energy value to obtain a first filtered acoustic wave graph;

[0017] Obtaining a second mean acoustic wave graph of a voiceprint signal of the fourth acoustic electric signal, and obtaining a second average energy value according to the second mean acoustic wave graph;

[0018] Filtering an acoustic wave graph of the third acoustic electric signal according to the second average energy value to obtain a second filtered acoustic wave graph.

[0019] The preferred technical scheme has the beneficial effect that the energy threshold is used to filter the acoustic signal, effectively removes noise interference, and improves the accuracy and reliability of fault feature extraction.

[0020] As a preferred scheme of the distribution network fault feature extraction method, the method comprises the following steps:

[0021] Inputting the first filtered acoustic wave graph into the neural network model to obtain the first fault type through two convolution layers;

[0022] The second filtered acoustic wave image is input into the neural network model, first layer features are extracted through a first layer convolutional layer, the first layer features and the first filtered acoustic wave image are output second layer features through a second layer convolutional layer, and a second fault type is obtained according to the second layer features.

[0023] The preferred technical scheme has the beneficial effects that different levels of feature information are extracted through the multi-layer convolutional neural network, accurate identification of the fault type is realized, and the accuracy and efficiency of fault diagnosis are improved.

[0024] As a preferred scheme of the power distribution network fault feature quantity extraction method, the judgment on the first fault type and the second fault type comprises:

[0025] If the first fault type and the second fault type are the same, the second fault type is output.

[0026] If the first fault type and the second fault type are different, electrical quantity abnormal data is output.

[0027] Third layer features are extracted from the second layer features and the electrical quantity abnormal data through a third layer convolutional layer, and a fault type is output through the third layer features.

[0028] The preferred technical scheme has the beneficial effects that the fault type is comprehensively judged through the multi-layer convolutional network combined with the electrical quantity abnormal data, and the diagnostic accuracy and reliability are further improved.

[0029] As a preferred scheme of the power distribution network fault feature quantity extraction method, the first average energy value comprises:

[0030] The acoustic wave image of the first acoustic electric signal is segmented, and the energy of each segment is compared with the first average energy value.

[0031] The part lower than the first average energy value is set to 0, and the corresponding electric signal sampling point in the acoustic wave image of the first acoustic electric signal is modified.

[0032] As a preferred scheme of the power distribution network fault feature quantity extraction method, the method further comprises:

[0033] The frequency band lower than the first average energy value is attenuated and removed.

[0034] If the energy of the high-frequency frequency band is lower than the first average energy value based on the speech signal, a frequency band is filtered in the frequency domain.

[0035] In a second aspect, the present application provides a power distribution network fault feature quantity extraction system, comprising:

[0036] An acoustic wave data processing module is configured to acquire an acoustic wave electrical signal, process the acoustic wave electrical signal, and obtain a screened acoustic wave electrical signal.

[0037] A model identification module is configured to input the screened acoustic wave electrical signal into a neural network model, and obtain a first fault type and a second fault type through a convolution layer.

[0038] A fault judgment module is configured to judge the first fault type and the second fault type, and obtain a fault type.

[0039] In a third aspect, the present application provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the power distribution network fault feature extraction method when executing the computer program.

[0040] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the power distribution network fault feature extraction method when executed by a processor.

[0041] Compared with the prior art, the present application has the following advantages: the present application collects acoustic wave signals through multiple microphones, screens the acoustic wave signals, combines a neural network model to judge faults, effectively removes noise interference, and improves the accuracy and reliability of fault feature extraction; the present application extracts multi-level features through a multi-layer convolutional neural network, enhances the ability to capture overall information, and improves the precision of fault diagnosis; the present application comprehensively considers abnormal electrical quantity data, further verifies the fault type, avoids misjudgment, and ensures the accuracy of the diagnosis result. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0043] Figure 1 A whole flow logic diagram of a power distribution network fault feature extraction method provided by an embodiment of the present application. DETAILED DESCRIPTION

[0044] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are 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.

[0045] Embodiment 1, refer to Figure 1 For an embodiment of the present application, a network fault feature extraction method is provided, comprising:

[0046] S100: Obtain an acoustic electric signal, process the acoustic electric signal to obtain a screened acoustic electric signal;

[0047] S200: Input the screened acoustic electric signal into a neural network model, obtain a first fault type and a second fault type through a convolution layer;

[0048] S300: Obtain a fault type by judging the first fault type and the second fault type.

[0049] It should be noted that the acoustic electric signal is screened to remove noise interference and ensure signal quality; the convolution layer of the neural network is used to extract features and accurately identify the fault type; the fault type is comprehensively judged, combined with abnormal electrical quantity data for further verification to avoid misjudgment.

[0050] In the embodiments of the present application, the above step S100 comprises the following sub-steps A1-A2;

[0051] In A1: In a first preset time, adjust the first position of the microphone to obtain a first acoustic electric signal and a second acoustic electric signal;

[0052] In A2: In a second preset time, adjust the second position of the microphone to obtain a third acoustic electric signal and a fourth acoustic electric signal.

[0053] In an optional embodiment, the position of the microphone can be mechanically driven to move the microphone, and the microphone is fixed by using a mechanical arm (or a multi-degree-of-freedom mechanical structure) to control the position of the microphone through the movement of the mechanical arm; in the first preset time, the mechanical arm moves the microphone to the first position to obtain the first acoustic electric signal and the second acoustic electric signal; in the second preset time, the mechanical arm moves the microphone to the second position to obtain the third acoustic electric signal and the fourth acoustic electric signal

[0054] In an alternative embodiment, adjusting the position of the microphone can also fix the layout of the multi-microphone array, and a plurality of fixed-position microphones are arranged around the device to form a microphone array. In the first preset time, the microphone at the first position is activated to obtain the first and second sound wave electrical signals. In the second preset time, the microphone at the second position is activated to obtain the third and fourth sound wave electrical signals.

[0055] In the embodiment of the application, adjusting the position of the microphone includes rotating the rotating disc, moving the four connecting rods away from the rotating disc along the direction of the sliding groove, and in the first preset time, the first microphone obtains the first sound wave electrical signal of the device, and the four second microphones respectively obtain the four second sound wave electrical signals.

[0056] After obtaining, the rotating disc is reversely rotated, the four connecting rods are moved to approach the rotating disc along the direction of the sliding groove, and in the second preset time, the first microphone obtains the third sound wave electrical signal of the device, and the four second microphones respectively obtain the four fourth sound wave electrical signals.

[0057] Specifically, in the running process of the electrical primary equipment, the multi-feature vibration voiceprint signal is collected. The electrical primary equipment generates mechanical vibration in the running state. These vibrations are transmitted through the device structure and the air medium around the device, forming a detectable sound wave signal. The microphone, as the receiving end of the sound wave, vibrates with the fluctuation of the air pressure, and then converts the mechanical vibration into an initial electrical signal. The electrical signal is gain-processed by the preamplifier to improve the signal strength. The data acquisition card converts the amplified analog electrical signal into a digital signal at a high sampling rate and resolution.

[0058] When collecting the multi-feature voiceprint signal of the electrical primary equipment, the original signal has many dramatic jumps due to the interference of the microphone's own sensitive characteristics and external environment, which affects the subsequent fault diagnosis effect. The specific values of the first preset time and the second preset time need to be set according to the actual application scene and the characteristics of the device.

[0059] The original signal is screened. When the first microphone obtains the first sound wave electrical signal of the device, the four second microphones are located on a circle with the first microphone as the center and a radius of r1. When the first microphone obtains the third sound wave electrical signal of the device, the four second microphones are located on a circle with the first microphone as the center and a radius of r2. The value of r1:r2 is not less than 2.

[0060] The second sound wave electrical signals obtained by the four second microphones are away from the device and are used to obtain blank sound waves other than the device.

[0061] The fourth sound wave electrical signal acquired by the four second microphones is slightly close to the device, and the fourth sound wave electrical signal includes part of the sound wave information of the device, and also includes the sound wave of the environment and the device itself.

[0062] It should be noted that in the existing screening method, the threshold method is usually used for screening the original signal, but the existing threshold setting needs to be set according to the characteristics of the preset fault waveform library of the device. However, due to the interference of the sensitive characteristics of the microphone itself and the external environment and other factors, the existing threshold setting is not accurate. The present application effectively screens noise by adjusting the microphone position and collecting sound wave signals at different distances, thereby improving the accuracy of fault diagnosis.

[0063] In the embodiment of the present application, after the A1-A2 steps are completed in the above step S100, the following steps A3-A10 are further included;

[0064] In A3, the first average sound wave graph of the voiceprint signal of the second sound wave electrical signal is acquired, and the first average energy value is obtained according to the first average sound wave graph;

[0065] In A4, the sound wave graph of the first sound wave electrical signal is filtered according to the first average energy value, and the first filtered sound wave graph is obtained;

[0066] In A5, the second average sound wave graph of the voiceprint signal of the fourth sound wave electrical signal is acquired, and the second average energy value is obtained according to the second average sound wave graph;

[0067] In A6, the sound wave graph of the third sound wave electrical signal is filtered according to the second average energy value, and the second filtered sound wave graph is obtained;

[0068] In A7, the sound wave graph of the first sound wave electrical signal is segmented, and the energy of each segment is compared with the first average energy value;

[0069] In A8, the part lower than the first average energy value is set to 0, and the corresponding electrical signal sampling points in the sound wave graph of the first sound wave electrical signal are modified;

[0070] In A9, the frequency band corresponding to the first average energy value is attenuated and removed;

[0071] In A10, if the energy of the high frequency band is lower than the first average energy value obtained based on the voice signal, a filtering operation is performed on the frequency band in the frequency domain;

[0072] In an optional embodiment, the processing and screening of the sound wave electrical signal can be signal denoising based on wavelet transform. The sound wave electrical signal is decomposed by wavelet to obtain coefficients of different frequency bands. The high frequency band coefficients are threshold processed to remove noise components. The denoised sound wave electrical signal is obtained through wavelet reconstruction.

[0073] In an alternative embodiment, the processing and screening of the acoustic electrical signal can be based on adaptive filter signal processing, initializing the weights of the adaptive filter, filtering the input acoustic electrical signal, calculating the filtered output signal, adjusting the weights of the filter according to the error signal (difference between the expected output and the actual output), repeating the above steps until the performance of the filter converges.

[0074] In the embodiment of the present application, the first preset time is divided into M time periods, and the first average energy value E1 is expressed as:

[0075]

[0076] wherein Ei is the energy of each time period, M is the M Ei, and E1 is the first average energy value;

[0077] The acoustic image of the first acoustic electrical signal is divided into M time periods, the energy of each time period is compared with the first average energy value, and the part lower than the first average energy value is set to 0, and the corresponding electrical signal sampling point in the acoustic image of the first acoustic electrical signal is modified.

[0078] Similar operations are performed on the frequency components, and the frequency band lower than the first average energy value is attenuated or even removed. For example, if the energy of the high-frequency band containing a large amount of noise is lower than the first average energy value based on the speech signal, the frequency band is filtered in the frequency domain to reduce the influence of the frequency band in the reconstructed acoustic image.

[0079] In the second preset time, the time period of the second preset time is divided into M time periods, and the second average energy value E2 is expressed as:

[0080]

[0081] wherein Ei is the energy of each time period, M is the M Ei, and E2 is the second average energy value.

[0082] The acoustic image of the third acoustic electrical signal is divided into M time periods, Ei is the energy of each time period, and the weight wi of each time period of the acoustic image of the third acoustic electrical signal is expressed as:

[0083]

[0084] The signal of each time period of the acoustic image of the third acoustic electrical signal is multiplied by the corresponding weight wi, and then the weighted signal of each time period is summed and averaged to obtain the filtered signal y, which is expressed as:

[0085]

[0086] The screening acoustic electric signal includes a first filtered acoustic image and a second filtered acoustic image.

[0087] It should be noted that by calculating the average energy value by segmentation, the energy threshold screening and weighting processing are performed on the acoustic electric signal, the low-energy noise and high-frequency interference are effectively removed, the useful signal with high energy is retained, meanwhile, the signal weight is further optimized by dynamic weighting adjustment, the signal-to-noise ratio of the acoustic signal is significantly improved, more accurate and reliable data basis is provided for subsequent fault feature extraction, and thus the accuracy and reliability of fault diagnosis are improved.

[0088] In the embodiment of the present application, the step S200 includes the following sub-steps B1-B2.

[0089] In B1, the first filtered acoustic image is input into the neural network model, and the first fault type is obtained through two convolution layers.

[0090] In B2, the second filtered acoustic image is input into the neural network model, the first layer feature is extracted through the first layer convolution layer, the first layer feature and the first filtered acoustic image are output through the second layer convolution layer to obtain the second layer feature, and the second fault type is obtained according to the second layer feature.

[0091] In an optional embodiment, the identification of the fault type can be converting the first filtered acoustic image and the second filtered acoustic image into one-dimensional feature vectors respectively; inputting the feature vectors into a multi-layer perception model; performing feature extraction and nonlinear transformation through a hidden layer of the multi-layer perception; using a Softmax activation function in an output layer to map the feature vectors to different fault types to obtain the fault type.

[0092] In an optional embodiment, the identification of the fault type can be converting the first filtered acoustic image and the second filtered acoustic image into time series data respectively; inputting the time series data into an LSTM network; extracting long-term dependency and short-term features in the time series through a hidden layer of the LSTM; using a Softmax activation function in an output layer to map the extracted features to different fault types to obtain the fault type.

[0093] In the embodiment of the present application, the identification of the fault type includes inputting the first filtered acoustic image into a neural network model, obtaining the first fault type through two convolution layers; inputting the second filtered acoustic image into the neural network model, extracting the first layer feature through the first layer convolution layer, outputting the second layer feature through the second layer convolution layer from the first layer feature and the first filtered acoustic image, and obtaining the second fault type according to the second layer feature.

[0094] Specifically, when the first filtered acoustic wave image is input into the first convolutional layer, the convolution kernel slides over the input data to extract local features. The first convolutional layer is configured with 32 convolution kernels of size 3*3*1. After filtering by the first layer, an output feature map of size 26*26*32 is formed. The 26*26*32 feature map obtained by the first convolutional layer is subjected to a maximum pooling operation, and the number of features after pooling is reduced to 13*13*32. The second convolutional layer has a kernel depth of 32, i.e., 3*3*32. The second convolutional layer outputs a feature map of size 11*1*64, and the number of features after pooling is reduced to 5*5*64.

[0095] In the second convolutional layer, the input is no longer the original acoustic wave image, but the feature map output by the first convolutional layer. These feature maps already contain some basic features of the input data. The second convolutional layer further extracts and combines these features to form higher-level feature representations. These high-level features can better reflect the essential properties and fault types of the input data.

[0096] The second filtered acoustic wave image is input into the neural network model, and the first layer features are extracted by the first convolutional layer. The first layer features and the first filtered acoustic wave image are output by the second convolutional layer to output the second fault type.

[0097] The input layer receives the second filtered acoustic wave image, which is converted into a matrix of size 64*64*1 (height* width*single channel) after quantization and enters the input layer.

[0098] The first convolutional layer uses multiple 3*3 convolution kernels (e.g., 16), with a step size of 1 and a ReLU activation function. Through the working principle of the above convolutional layer, this layer can extract preliminary local features such as sound frequency and amplitude, and output 16 feature maps of size 62*62 (since the convolution kernel is 3*3 and the step size is 1, the calculation is 64-3+1=62).

[0099] The second convolutional layer: then the feature map of the first layer (i.e., the first layer feature) and the first layer feature and the first filtered acoustic wave Figure 1 are input into the second convolutional layer. The second convolutional layer uses 5*5 convolution kernels (e.g., 32), with a step size of 2 and a ReLU activation function. This layer further abstracts and combines features based on the first layer features, and outputs 32 feature maps of size 30*30 (the calculation method is similar to that of the first layer output).

[0100] Pooling layer: After the second convolutional layer, a max-pooling layer is added, for example, with a 2x2 pooling window and a stride of 2; the role of the pooling layer is to downsample the feature maps output by the second convolutional layer to reduce the number of features and data volume while preserving important feature information to some extent; After the pooling layer, 32 feature maps of size 15x15 are output.

[0101] Fully connected layer: The feature maps output from the pooling layer are stretched and tiled into a single long vector (for example, a 15x15x32 = 7200-element vector), which is used as input to the fully connected layer. The fully connected layer can contain multiple neurons (e.g., 128) and use a Sigmoid activation function for non-linear transformation to further process the feature vectors and integrate features at different levels.

[0102] Output layer: Finally, the results output by the fully connected layer enter the output layer, which has a number of neurons equal to the number of fault types. For example, if there are three different fault types (normal and two fault conditions) to be diagnosed, the output layer has three neurons, which can use a Softmax activation function to convert the output vector into a distribution representing the probability of each fault type, and the output layer outputs the class with the highest probability as the predicted second fault type.

[0103] It should be noted that the multi-layer convolutional neural network is used to extract features from the sound wave signal and identify the fault type. First, the convolutional layer can automatically extract local features such as sound frequency and amplitude from the signal, effectively capturing fault features. Second, the pooling layer reduces the data volume while preserving important features, improving computational efficiency. Finally, the fully connected layer and the output layer process the features and output the probability distribution of the fault type, achieving accurate classification. Overall, the accuracy and efficiency of fault diagnosis are significantly improved, making it suitable for complex sound wave signal fault recognition.

[0104] In the embodiment of the present application, the above step S300 includes the following sub-steps C1-C3;

[0105] In C1: if the first fault type and the second fault type are the same, output the second fault type;

[0106] In C2: if the first fault type and the second fault type are different, output the electrical quantity abnormal data;

[0107] In C3: according to the second layer feature and the electrical quantity abnormal data, a third layer feature is extracted through a third layer convolutional layer, and the fault type is output through the third layer feature.

[0108] Specifically, if the first fault type and the second fault type are the same, the second fault type is output; if the first fault type and the second fault type are different, the electrical quantity abnormality data of the equipment is obtained; the second-layer features and the electrical quantity abnormality data are extracted through the third-layer convolution layer, and the fault type is output through the third-layer features. The electrical quantity abnormality data further verifies the fault type of the equipment from another angle, further improving the accuracy.

[0109] It should be noted that if the first fault type and the second fault type are the same, the second fault type is output; if the first fault type and the second fault type are different, the electrical quantity abnormal data of the equipment is obtained; the second layer features and the electrical quantity abnormal data are extracted through the third layer convolution layer to extract the third layer features, and the fault type is output through the third layer features; the neural network model of the present invention can use the pooling mechanism to capture multi-level feature information, optimize the feature extraction process, and realize the effective complementarity of features of different scales; this structural design not only enhances the ability to capture overall information, but also improves the feature expression ability of the network. At the same time, the present invention can more accurately extract the distribution network fault feature quantity.

[0110] The above is a schematic scheme of a distribution network fault feature extraction method according to this embodiment. It should be noted that the technical scheme of this distribution network fault feature extraction system and the technical scheme of the distribution network fault feature extraction method described above are based on the same concept. For details not described in detail in the technical scheme of the distribution network fault feature extraction system according to this embodiment, please refer to the description of the technical scheme of the distribution network fault feature extraction method described above.

[0111] The distribution network fault feature extraction system in this embodiment includes:

[0112] an acoustic wave data processing module, configured to obtain acoustic wave electrical signals, and process the acoustic wave electrical signals to obtain screening acoustic wave electrical signals;

[0113] a model identification module, configured to input the screening acoustic wave electrical signal into a neural network model and obtain a first fault type and a second fault type through a convolutional layer;

[0114] The fault judgment module is used to obtain a fault type by judging the first fault type and the second fault type.

[0115] This embodiment further provides a computer device suitable for extracting distribution network fault feature values, including:

[0116] Memory and processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a distribution network fault feature extraction method proposed in the above embodiment.

[0117] The embodiment further provides a storage medium, which stores a computer program, and the computer program is executed by a processor to implement the network distribution fault feature quantity extraction method proposed in the above embodiment.

[0118] The storage medium proposed in the embodiment and the network distribution fault feature quantity extraction method proposed in the above embodiment belong to the same inventive concept, 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.

[0119] Through the above description of 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 or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH memory, a hard disk, or an optical disc, and includes a number of instructions to make a computing device (which can be a personal computer, a server, or a network device, etc.) execute the methods of various embodiments of the present application.

[0120] 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. 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 of them should be covered in the scope of the claims of the present application.

Claims

1. A method for extracting distribution network fault characteristics, characterized in that: include: Acquiring an acoustic wave electrical signal, and processing the acoustic wave electrical signal to obtain a screening acoustic wave electrical signal; Inputting the screening acoustic wave electric signal into a neural network model, and obtaining a first fault type and a second fault type through a convolution layer; The fault type is obtained by judging the first fault type and the second fault type.

2. A method for extracting distribution network fault characteristics according to claim 1, characterized in that: Acquiring acoustic and electrical signals includes: Adjusting a first position of the microphone within a first preset time to obtain a first acoustic wave electrical signal and a second acoustic wave electrical signal; Within the second preset time, the second position of the microphone is adjusted to obtain a third sound wave electrical signal and a fourth sound wave electrical signal.

3. A method for extracting distribution network fault characteristics according to claim 2, characterized in that: Processing the acoustic wave electrical signal includes: Obtaining a first mean acoustic wave graph of the voiceprint signal of the second acoustic wave electrical signal, and obtaining a first average energy value according to the first mean acoustic wave graph; filtering the acoustic wave graph of the first acoustic wave electrical signal according to the first average energy value to obtain a first filtered acoustic wave graph; Obtaining a second mean acoustic wave graph of the voiceprint signal of the fourth acoustic wave electrical signal, and obtaining a second average energy value according to the second mean acoustic wave graph; The acoustic wave graph of the third acoustic wave electrical signal is filtered according to the second average energy value to obtain a second filtered acoustic wave graph.

4. A method for extracting distribution network fault characteristics according to claim 3, characterized in that: The first fault type and the second fault type obtained by the convolutional layer include: The first filtered acoustic wave image is input into the neural network model, and the first fault type is obtained through two convolutional layers; The second filtered acoustic wave graph is input into the neural network model, the first layer features are extracted through the first convolution layer, the first layer features and the first filtered acoustic wave graph are output through the second convolution layer to output the second layer features, and the second fault type is obtained according to the second layer features.

5. A method for extracting distribution network fault characteristics according to claim 4, characterized in that: Determining the first fault type and the second fault type includes: If the first fault type and the second fault type are the same, output the second fault type; If the first fault type and the second fault type are different, outputting electrical quantity abnormality data; The third-layer features are extracted through the third convolution layer based on the second-layer features and electrical quantity abnormal data, and the fault type is output through the third-layer features.

6. A method for extracting distribution network fault characteristics according to claim 3, characterized in that: The first average energy value includes: Segmenting the acoustic wave graph of the first acoustic wave electrical signal, and comparing the energy of each segment with the first average energy value; The portion lower than the first average energy value is set to 0, and the corresponding electrical signal sampling points in the acoustic wave graph of the first acoustic wave electrical signal are modified.

7. A method for extracting distribution network fault characteristics according to claim 3 or 6, characterized in that: Also includes: Attenuate and remove the frequency band corresponding to the frequency lower than the first average energy value; If the energy of the high frequency band is lower than a first average energy value obtained based on the speech signal, a filtering operation is performed on the frequency band in the frequency domain.

8. A distribution network fault feature extraction system, applying a distribution network fault feature extraction method according to any one of claims 1 to 7, characterized in that: include: an acoustic wave data processing module, configured to obtain acoustic wave electrical signals, and process the acoustic wave electrical signals to obtain screening acoustic wave electrical signals; a model identification module, configured to input the screening acoustic wave electrical signal into a neural network model and obtain a first fault type and a second fault type through a convolutional layer; The fault judgment module is used to obtain a fault type by judging the first fault type and the second fault type.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for extracting distribution network fault characteristics according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor, the steps of a distribution network fault feature extraction method according to any one of claims 1 to 7 are implemented.

Images