Partial discharge detection method and device

By extracting and processing the characteristic information of spectral and acoustic data, using the Hidden Markov model analysis, the problems of low detection accuracy and utilization in the local discharge detection method are solved, and higher detection accuracy and reliability are achieved.

CN118980891AInactive Publication Date: 2024-11-19MAOMING POWER SUPPLY BUREAU GUANGDONG POWER GRID CORP
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411059066.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-02
Publication Date
2024-11-19
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing partial discharge detection methods have low detection accuracy and sensitivity in complex electromagnetic environments, and have poor acousto-optical signal utilization, making it difficult to accurately reflect the discharge status of distribution network equipment.

Method used

By obtaining the spectral and sound spectrum data of the target device, the spectral characteristics such as the proportion of pulse repetition rate, the proportion of maximum pulse amplitude, the proportion of average pulse amplitude, and the proportion of Mel frequency cepspectral coefficients, the gammapass frequency cepspectral coefficients, etc. The sound spectrum features are vector quantized, and the pre-trained hidden Markov model is analyzed to accurately identify the local discharge state.

Benefits of technology

The deep fusion of spectral and sound spectrum information is achieved, the accuracy and reliability of local discharge detection is improved, the limitations of a single data source is overcome, and the accuracy of detection results is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118980891A_ABST
    Figure CN118980891A_ABST
Patent Text Reader

Abstract

The invention discloses a partial discharge detection method and device. The method comprises the following steps: acquiring spectral data and acoustic spectrum data of target equipment collected in a target sampling period; extracting spectral feature information in the spectral data, wherein the spectral feature information comprises at least one of the following features: a pulse repetition rate proportion, a maximum pulse amplitude proportion and an average pulse amplitude proportion; extracting sound spectrum characteristic information in the sound spectrum data, wherein the sound spectrum characteristic information comprises at least one of the following characteristics: a Mel frequency cepstrum coefficient and a Gammatone frequency cepstrum coefficient; performing vector quantization on each feature in the spectral feature information and the acoustic spectrum feature information, and taking all the obtained features as target feature information; and analyzing the target feature information by using a pre-trained hidden Markov model to obtain the partial discharge state of the target equipment. The technical problems of poor acousto-optic information utilization rate and low detection result accuracy in a related partial discharge detection scheme are solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of power grid fault detection, and in particular to a partial discharge detection method and device. Background Art

[0002] The distribution network is the main artery of regional power supply, and ensuring its safe and reliable operation is crucial for local transmission and distribution networks. During long-term operation, distribution network equipment is often affected by harsh environments and is prone to high-frequency abnormalities. The traditional manual inspection of distribution equipment one by one is not only labor-intensive, inefficient, and costly, but may also threaten the lives of operators in complex geographical environments. In addition, the electromagnetic environment of the distribution network is complex and changeable, and the types and numbers of interference signals are numerous, which seriously affects the accuracy and positioning accuracy of partial discharge detection. The partial discharge signal itself is highly complex and changeable. In order to avoid missed detection, partial discharge detection requires extremely high sensitivity, but this also makes it extremely susceptible to external electromagnetic interference.

[0003] Although the partial discharge detection method based on the fusion of acoustic and optical spectrum has made certain progress in improving the detection accuracy, it still faces many challenges. First, the acquisition of acoustic and optical signals is mostly limited to the surface of power equipment. This limitation makes the detection process susceptible to complex electromagnetic environment and geographical conditions, resulting in reduced detection accuracy. Especially in a strong magnetic field environment, the optical signal may be severely interfered, which in turn affects the effect of fusion detection. Secondly, the information utilization rate of the optical part is low, and the construction of the spectral feature spectrum is often monotonous, lacking rich feature information, and it is difficult to fully reflect the complex characteristics of partial discharge. This not only limits the sensitivity and accuracy of the detection method, but also affects its breadth and effectiveness in practical applications. Therefore, how to overcome the above-mentioned defects and improve the utilization rate and accuracy of the fusion of acoustic and optical spectrum in partial discharge detection has become a problem that needs to be solved urgently.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present application provide a partial discharge detection method and device to at least solve the technical problems of poor utilization of acoustic and optical information and low accuracy of detection results in related partial discharge detection schemes.

[0006] According to one aspect of an embodiment of the present application, a partial discharge detection method is provided, comprising: acquiring first spectrum data and first sound spectrum data of a target device collected within a target sampling period; extracting first spectrum feature information from the first spectrum data, wherein the first spectrum feature information includes at least one of the following features: pulse repetition rate ratio, maximum pulse amplitude ratio, and average pulse amplitude ratio; extracting first sound spectrum feature information from the first sound spectrum data, wherein the first sound spectrum feature information includes at least one of the following features: Mel frequency cepstrum coefficients and gammatone frequency cepstrum coefficients; vector quantizing each feature in the first spectrum feature information and the first sound spectrum feature information, respectively, and using all the obtained features as first target feature information; and analyzing the first target feature information using a pre-trained hidden Markov model to obtain a first partial discharge state of the target device.

[0007] Optionally, the first spectral data includes spectral data corresponding to multiple bands, and the first spectral characteristic information in the first spectral data is extracted, including: for each band, determining the number of discharge pulses in the band per unit time as the pulse repetition rate of the band based on the first spectral data, and determining a first sum of the pulse repetition rates of multiple bands, and determining the ratio of the pulse repetition rate of the band to the first sum as the pulse repetition rate proportion of the band; and / or, for each band, determining the maximum amplitude of the discharge pulse sequence of the band within the target sampling period as the maximum pulse amplitude of the band based on the first spectral data, and determining a second sum of the maximum pulse amplitudes of multiple bands, and determining the ratio of the maximum pulse amplitude of the band to the second sum as the maximum pulse amplitude proportion of the band; and / or, for each band, determining the average amplitude of the discharge pulse sequence of the band within the target sampling period as the average pulse amplitude of the band based on the first spectral data, and determining a third sum of the average pulse amplitudes of multiple bands, and determining the ratio of the average pulse amplitude of the band to the third sum as the average pulse amplitude proportion of the band.

[0008] Optionally, extracting the first sound spectrum feature information in the first sound spectrum data includes: performing frame processing on the first sound spectrum data to obtain multiple frames of first sub-sound spectrum data; performing windowing processing on each frame of the first sub-sound spectrum data using a Hamming window to obtain multiple frames of second sub-sound spectrum data; for each frame of the second sub-sound spectrum data, performing fast Fourier transform on the second sub-sound spectrum data to obtain spectrum data; performing filtering processing on the spectrum data using a Mel triangular filter group, and performing logarithmic operation and discrete cosine transform on the filtering result to obtain Mel-frequency cepstrum coefficients corresponding to the second sub-sound spectrum data; and / or performing filtering processing on the spectrum data using a gammatone filter group, and performing logarithmic operation and discrete cosine transform on the filtering result to obtain gammatone frequency cepstrum coefficients corresponding to the second sub-sound spectrum data.

[0009] Optionally, vector quantization is performed on each feature in the first spectral feature information and the first spectral feature information, respectively, including: using a splitting method to respectively determine a first initial code book corresponding to the first spectral feature information and a second initial code book corresponding to the first spectral feature information; using a nearest neighbor method to iteratively update the first initial code book to obtain a first target code book, and iteratively update the second initial code book to obtain a second target code book; using the first target code book to vector quantize each feature in the first spectral feature information; using the second target code book to vector quantize each feature in the first spectral feature information.

[0010] Optionally, a splitting method is used to respectively determine a first initial codebook corresponding to the first spectral feature information and a second initial codebook corresponding to the first spectral feature information, including: for the first spectral feature information, the following splitting process is cyclically executed: for each codeword in this cycle, the codeword is split into two codewords according to two disturbance coefficients, wherein the codeword of the first cycle is the average value of each feature in the first spectral feature information, and the two disturbance coefficients are 1+preset disturbance value and 1-preset disturbance value respectively; if the total number of codewords obtained by splitting in this cycle does not reach the preset number, all the codewords obtained by splitting are used as the codewords for the next cycle; if the total number of codewords obtained by splitting in this cycle does not reach the preset number, all the codewords obtained by splitting are used as the codewords for the next cycle; When the total number of codewords obtained by splitting reaches a preset number, the loop is stopped, and all the codewords obtained by splitting are used as the first initial codebook; for the first sound spectrum feature information, the following splitting process is executed cyclically: for each codeword in this loop, the codeword is split into two codewords according to two perturbation coefficients, wherein the codeword of the first loop is the average value of each feature in the first sound spectrum feature information; if the total number of codewords obtained by splitting in this loop does not reach the preset number, all the codewords obtained by splitting are used as the codewords of the next loop; if the total number of codewords obtained by splitting in this loop reaches the preset number, the loop is stopped, and all the codewords obtained by splitting are used as the second initial codebook.

[0011] Optionally, the first initial code book is iteratively updated using the nearest neighbor method to obtain a first target code book, and the second initial code book is iteratively updated to obtain a second target code book, including: for the first initial code book, the following update process is cyclically executed: determine the code book of this cycle, and cluster each feature in the first spectral feature information according to the nearest neighbor principle and the code book to obtain multiple clustering results, wherein the code book of the first cycle is the first initial code book; determine the first distance between each feature and the corresponding codeword in each clustering result, and determine the sum of all first distances as the total distance of this cycle; determine the difference between the total distance of this cycle and the total distance of the previous cycle, and determine the ratio of the difference to the total distance of this cycle as the distance correction amount, wherein the total distance of the previous cycle corresponding to the first cycle is the preset initial distance; if the distance correction amount is not less than the preset distance threshold, and the current number of cycles does not reach the preset number of cycles, determine the center of the multiple clustering results as the code book of the next cycle; if the distance correction amount is less than the preset distance threshold, or the current number of cycles reaches the preset number of cycles The number of loops is reached, the loop is stopped, and the codebook of this loop is used as the first target codebook; for the second initial codebook, the following update process is executed cyclically: the codebook of this loop is determined, and each feature in the first sound spectrum feature information is clustered according to the nearest neighbor principle and the codebook to obtain multiple clustering results, wherein the codebook of the first loop is the second initial codebook; the second distance between each feature and the corresponding codeword in each clustering result is determined, and the sum of all second distances is determined as the total distance of this loop; the difference between the total distance of this loop and the total distance of the previous loop is determined, and the ratio of the difference to the total distance of this loop is determined as the distance correction amount, wherein the total distance of the previous loop corresponding to the first loop is the preset initial distance; if the distance correction amount is not less than the preset distance threshold, and the current number of loops does not reach the preset number of loops, the center of the multiple clustering results is determined as the codebook of the next loop; if the distance correction amount is less than the preset distance threshold, or the current number of loops reaches the preset number of loops, the loop is stopped, and the codebook of this loop is used as the second target codebook.

[0012] Optionally, the training process of the hidden Markov model includes: obtaining multiple groups of second spectral data and second spectral data of the target device collected in multiple historical collection cycles; for each group of second spectral data and second spectral data, extracting second spectral feature information in the second spectral data, and extracting second spectral feature information in the second spectral data, respectively, vector quantizing each feature of the second spectral feature information and the second spectral feature information, and taking all the obtained features as a group of observation data; organizing all the observation data into an observation data sequence in chronological order; for each group of observation data, obtaining the local discharge state of the target device in the historical collection cycle corresponding to the observation data as the state data corresponding to the observation data, wherein the type of the local discharge state includes at least one of the following: slight discharge state, moderate discharge state, severe discharge state; organizing all the state data into a state data sequence in chronological order; iteratively training the hidden Markov model based on the observation data sequence and the state data sequence, and updating the model parameters, wherein the model parameters include: a state transfer matrix, an observation value probability matrix, and an initial probability matrix.

[0013] According to another aspect of an embodiment of the present application, a partial discharge detection device is also provided, including: an acquisition module, used to acquire first spectrum data and first sound spectrum data of a target device collected within a target sampling period; a first extraction module, used to extract first spectrum feature information in the first spectrum data, wherein the first spectrum feature information includes at least one of the following features: pulse repetition rate ratio, maximum pulse amplitude ratio, average pulse amplitude ratio; a second extraction module, used to extract first sound spectrum feature information in the first sound spectrum data, wherein the first sound spectrum feature information includes at least one of the following features: Mel frequency cepstral coefficients, gammatone frequency cepstral coefficients; a vector quantization module, used to respectively perform vector quantization on each feature in the first spectrum feature information and the first sound spectrum feature information, and use all the obtained features as first target feature information; a fault analysis module, used to analyze the first target feature information using a pre-trained hidden Markov model to obtain a first partial discharge state of the target device.

[0014] According to another aspect of an embodiment of the present application, a computer program product is further provided. The computer program product includes: a computer program, wherein when the computer program is executed by a processor, the above-mentioned partial discharge detection method is implemented.

[0015] According to another aspect of an embodiment of the present application, an electronic device is provided, which includes: a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to execute the above-mentioned partial discharge detection method through the computer program.

[0016] In an embodiment of the present application, by integrating multi-source detection technology, the first spectral data and the first acoustic spectrum data of the target device are collected simultaneously within the target sampling period. Subsequently, the spectral features such as the pulse repetition rate ratio, the maximum pulse amplitude ratio, the average pulse amplitude ratio in the spectral data, and the acoustic spectrum features such as the Mel frequency cepstral coefficient and the gammatone frequency cepstral coefficient in the acoustic spectrum data are extracted respectively. These features are efficiently processed using a vector quantization method to form the first target feature information. Finally, the feature information is analyzed by a pre-trained hidden Markov model to accurately identify the first partial discharge state of the target device. This technology realizes the deep fusion of spectral and acoustic spectrum information, effectively overcomes the limitations of a single data source, improves the accuracy and reliability of partial discharge detection, and thus solves the technical problems of poor utilization of acoustic and optical information and low accuracy of detection results in related partial discharge detection schemes. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 is a flow chart of an optional partial discharge detection method according to an embodiment of the present application;

[0019] Figure 2 is a schematic structural diagram of an optional partial discharge detection device according to an embodiment of the present application;

[0020] Figure 3 is a schematic diagram of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0022] It should be noted that the terms "first", "second", etc. in the specification, claims and drawings of the present application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0023] In order to better understand the embodiments of the present application, some nouns or terms that appear in the description of the embodiments of the present application are first translated and explained as follows:

[0024] Mel-Frequency Cepstral Coefficients (MFCC): A parameter widely used in speech signal processing and audio analysis. They are cepstral coefficients calculated from the Mel frequency scale and are used to represent the characteristics of speech signals. The Mel frequency scale is a nonlinear frequency scale that simulates the perceptual characteristics of the human auditory system. On this scale, the frequency intervals of the low-frequency part are larger, while the frequency intervals of the high-frequency part are smaller, which matches the sensitivity of the human auditory system to different frequencies.

[0025] Gammatone Filter Cepstral Coefficient (GFCC): It is a characteristic parameter based on the human cochlear auditory model, mainly used for audio data feature extraction and speech recognition. GFCC uses a set of overlapping bandpass filters (gammatone filters) to simulate the characteristics of the human cochlea. The impulse responses of these filters can describe the joint distribution characteristics of the sound signal in the time domain and frequency domain.

[0026] Vector Quantization (VQ) is a data compression technique mainly used in the fields of signal processing and image compression. The basic idea of ​​vector quantization is to represent the continuous values ​​of the input signal or image as a set of discrete vectors (called codewords or codebooks), which are selected from a predefined finite set. In this way, the original data can be compressed into a shorter representation while keeping the characteristics of the original data as much as possible.

[0027] Hidden Markov Model (HMM): is a statistical model used to describe a Markov process with hidden unknown parameters. Its main feature is that the system has two levels of states: observed state and hidden state. The observed state is what we can observe directly, while the hidden state cannot be observed directly, but can be inferred from the observed state. The core idea of ​​HMM is to infer the sequence of hidden states through the observed sequence data.

[0028] Example 1

[0029] According to an embodiment of the present application, a partial discharge detection method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0030] Figure 1 is a flow chart of a partial discharge detection method provided according to an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0031] Step S102, obtaining first spectrum data and first sound spectrum data of a target device collected within a target sampling period;

[0032] Step S104, extracting first spectrum characteristic information from the first spectrum data, wherein the first spectrum characteristic information includes at least one of the following characteristics: pulse repetition rate proportion, maximum pulse amplitude proportion, and average pulse amplitude proportion;

[0033] Step S106, extracting first sound spectrum feature information from the first sound spectrum data, wherein the first sound spectrum feature information includes at least one of the following features: Mel frequency cepstral coefficients and gammatone frequency cepstral coefficients;

[0034] Step S108, respectively performing vector quantization on each feature in the first spectrum feature information and the first sound spectrum feature information, and using all obtained features as first target feature information;

[0035] Step S110 , analyzing the first target feature information using a pre-trained hidden Markov model to obtain a first partial discharge state of the target device.

[0036] The following describes each step of the partial discharge detection method in conjunction with a specific implementation process.

[0037] First, first spectrum data and first sound spectrum data of a target device collected within a target sampling period are acquired.

[0038] Among them, the above-mentioned target device can be a distribution network equipment; the first spectral data usually covers spectral data corresponding to multiple bands. When analyzing it, the following features in the first spectral data can be extracted as the first spectral feature information, including: the pulse repetition rate proportion of each band, the maximum pulse amplitude proportion, and the average pulse amplitude proportion. It should be noted that these features can be used alone, or multiple features can be flexibly combined to optimize the analysis effect, thereby more accurately reflecting the complex characteristics of partial discharge.

[0039] Optionally, the extraction process of the pulse repetition rate ratio of each band can be performed by the following steps: for each band, the number of discharge pulses in the band per unit time is determined as the pulse repetition rate of the band according to the first spectrum data, and the first sum of the pulse repetition rates of multiple bands is determined, and the ratio of the pulse repetition rate of the band to the first sum is determined as the pulse repetition rate ratio of the band. The specific calculation formula is as follows:

[0040]

[0041] Where: n k is the pulse repetition rate ratio in the kth band, in %; N k is the pulse repetition rate of the kth band; h is the total number of bands.

[0042] Optionally, the extraction process of the maximum pulse amplitude ratio of each band can be performed by the following steps: for each band, the maximum amplitude of the discharge pulse sequence of the band within the target sampling period is determined as the maximum pulse amplitude of the band according to the first spectrum data, and the second sum of the maximum pulse amplitudes of multiple bands is determined, and the ratio of the maximum pulse amplitude of the band to the second sum is determined as the maximum pulse amplitude ratio of the band. The specific calculation formula is as follows:

[0043]

[0044] Where: a k is the proportion of the maximum pulse amplitude in the kth band, in %; A k is the maximum pulse amplitude of the kth band, in V; h is the total number of bands.

[0045] Optionally, the extraction process of the average pulse amplitude ratio of each band can be performed by the following steps: for each band, the average amplitude of the discharge pulse sequence of the band within the target sampling period is determined as the average pulse amplitude of the band according to the first spectrum data, and the third sum of the average pulse amplitudes of multiple bands is determined, and the ratio of the average pulse amplitude of the band to the third sum is determined as the average pulse amplitude ratio of the band. The specific calculation formula is as follows:

[0046]

[0047] Where: l k is the average pulse amplitude ratio of the kth band, in %; L k is the average pulse amplitude of the kth band, in V; h is the total number of bands.

[0048] Optionally, after acquiring the first sound spectrum data, first sound spectrum feature information in the first sound spectrum data is extracted, wherein the first sound spectrum feature information may include at least one of the following features: Mel frequency cepstral coefficients and gammatone frequency cepstral coefficients.

[0049] Specifically, the above-mentioned Mel-frequency cepstrum coefficients and gammatone frequency cepstrum coefficients can be obtained in the following manner: first, the first sound spectrum data is framed to obtain multiple frames of first sub-sound spectrum data; secondly, each frame of the first sub-sound spectrum data is windowed using a Hamming window to obtain multiple frames of second sub-sound spectrum data; then, each frame of the second sub-sound spectrum data is subjected to a fast Fourier transform (Fast Fourier Transform, FFT) to obtain spectrum data; then, the spectrum data is filtered using a Mel triangular filter group, and the filtering result is subjected to logarithmic operation and discrete cosine transform to obtain the Mel-frequency cepstrum coefficients corresponding to the second sub-sound spectrum data; the spectrum data is filtered using a gammatone filter group, and the filtering result is subjected to logarithmic operation and discrete cosine transform to obtain the gammatone frequency cepstrum coefficients corresponding to the second sub-sound spectrum data.

[0050] It should be noted that in order to reduce the impact of the truncation effect in the framing process and reduce the slope at both ends of the frame, it is necessary to perform a windowing operation on the speech frame. Considering that the spectral smoothness of the rectangular window is better, but the waveform details are easily lost and leakage will occur, and the main lobe width of the Hamming window is twice that of the rectangular window, and the attenuation is also faster than the rectangular window. Therefore, the embodiment of the present application selects the Hamming window to perform windowing processing on the first sub-spectrum data of each frame.

[0051] Optionally, after obtaining the above-mentioned first spectral feature information and first spectral feature information, vector quantization is performed on each feature therein respectively. This process can be carried out in the following manner: first, a first initial code book corresponding to the first spectral feature information and a second initial code book corresponding to the first spectral feature information are determined respectively by using a splitting method; then, the first initial code book is iteratively updated using a nearest neighbor method to obtain a first target code book, and the second initial code book is iteratively updated to obtain a second target code book; subsequently, each feature in the first spectral feature information is vector quantized using the first target code book, and each feature in the first spectral feature information is vector quantized using the second target code book.

[0052] Among them, as an optional implementation method, when using the splitting method to determine the first initial code book corresponding to the first spectral feature information, for the first spectral feature information, the following splitting process is cyclically executed: for each code word in this cycle, the code word is split into two code words according to two disturbance coefficients, wherein the code word of the first cycle is the average value of each feature in the first spectral feature information, and the two disturbance coefficients are 1+preset disturbance value and 1-preset disturbance value respectively; if the total number of code words obtained by splitting in this cycle does not reach the preset number, all the code words obtained by splitting are used as the code words for the next cycle; if the total number of code words obtained by splitting in this cycle reaches the preset number, the cycle is stopped, and all the code words obtained by splitting are used as the first initial code book.

[0053] For example, each feature in the first spectral feature information is recorded as y i , (i=1,2,…,T), T is the total number of spectral features, set the preset perturbation value to ε, the preset number to M, and perform the following steps:

[0054] S1, calculate the average value of all spectral features The average value is taken as the initial codeword, denoted as C1;

[0055] S2, split each existing codeword into two codewords using the following formula:

[0056] C′=(1+ε)C

[0057] C″=(1-ε)C

[0058] Where C is the codeword before splitting, C′ and C″ are the codewords after splitting respectively;

[0059] S3, determine whether the total number of codewords obtained by splitting reaches the preset number M, if not, return to step S2, if reached, determine the first initial codebook C j (j=1,2,…,M).

[0060] Similarly, when the splitting method is used to determine the second initial code book corresponding to the first sound spectrum feature information, a similar splitting process can be executed cyclically: for each code word in this cycle, the code word is split into two code words according to two perturbation coefficients, wherein the code word of the first cycle is the average value of each feature in the first sound spectrum feature information; if the total number of code words obtained by splitting in this cycle does not reach the preset number, all the code words obtained by splitting are used as the code words for the next cycle; if the total number of code words obtained by splitting in this cycle reaches the preset number, the cycle is stopped, and all the code words obtained by splitting are used as the second initial code book.

[0061] Optionally, after obtaining the first initial codebook and the second initial codebook, the first initial codebook may be iteratively updated using the nearest neighbor method to obtain a first target codebook, and the second initial codebook may be iteratively updated to obtain a second target codebook, and the specific method is as follows.

[0062] Optionally, for the first initial code book, the following update process is executed in a loop: determine the code book of this cycle, and cluster each feature in the first spectral feature information according to the nearest neighbor principle and the code book to obtain multiple clustering results, wherein the code book of the first cycle is the first initial code book; determine the first distance between each feature and the corresponding codeword in each clustering result, and determine the sum of all first distances as the total distance of this cycle; determine the difference between the total distance of this cycle and the total distance of the previous cycle, and determine the ratio of the difference to the total distance of this cycle as the distance correction amount, wherein the total distance of the previous cycle corresponding to the first cycle is the preset initial distance; if the distance correction amount is not less than the preset distance threshold, and the current number of cycles has not reached the preset number of cycles, determine the center of the multiple clustering results as the code book of the next cycle; if the distance correction amount is less than the preset distance threshold, or the current number of cycles reaches the preset number of cycles, stop the cycle, and use the code book of this cycle as the first target code book.

[0063] For example, obtain the first initial code book C j (j=1,2,…,M), set the preset distance threshold δ, initial distance d (0) , preset the number of cycles K, and then perform the following steps:

[0064] S1: Determine the codebook of the first cycle as the first initial codebook, and divide the codebook of this cycle into M spaces according to the nearest neighbor principle. like Then y i Classification and Among them, m is the number of cycles, y i represents the i-th feature in the first spectral feature information, d i Represents the distance between the i-th feature and the corresponding codeword;

[0065] S2, determine the total distance of this cycle: Where T is the total number of features, d (m) is the total distance of the mth cycle;

[0066] S3, determine the distance correction value of this cycle: Where, Δd (m) is the difference between the total distance of this cycle and the total distance of the previous cycle, δ (m) The distance correction for this cycle;

[0067] S4, determine whether the distance correction value is less than the preset distance threshold, i.e., δ (m) <δ, or whether the current loop number reaches the preset loop number, that is, m = K. If any of the conditions is met, the loop is stopped and the codebook of this loop is determined to be the final first target codebook; if both conditions are not met, return to step S1, at this time m = m + 1, and calculate the new cluster center as Among them, N j is the number of features in clustering result j.

[0068] Similarly, referring to the above step of determining the first target codebook, for the second initial codebook, a similar updating process is cyclically executed to obtain the second target codebook: determine the codebook of this cycle, and cluster each feature in the first sound spectrum feature information according to the nearest neighbor principle and the codebook to obtain multiple clustering results, wherein the codebook of the first cycle is the second initial codebook; determine the second distance between each feature and the corresponding codeword in each clustering result, and determine the sum of all second distances as the total distance of this cycle; determine the difference between the total distance of this cycle and the total distance of the previous cycle, and determine the ratio of the difference to the total distance of this cycle as the distance correction amount, wherein the total distance of the previous cycle corresponding to the first cycle is the preset initial distance; if the distance correction amount is not less than the preset distance threshold, and the current cycle number does not reach the preset cycle number, determine the center of the multiple clustering results as the codebook of the next cycle; if the distance correction amount is less than the preset distance threshold, or the current cycle number reaches the preset cycle number, stop the cycle, and use the codebook of this cycle as the second target codebook.

[0069] Optionally, after obtaining the first target codebook and the second target codebook, each feature in the first spectrum feature information is vector quantized using the first target codebook, and each feature in the first sound spectrum feature information is vector quantized using the second target codebook, and all the obtained features are used as the first target feature information.

[0070] Furthermore, the pre-trained hidden Markov model can be used to analyze the first target feature information to obtain the first partial discharge state of the target device, wherein the type of the partial discharge state may include at least one of the following: slight discharge state, moderate discharge state, and severe discharge state.

[0071] As an optional implementation, the hidden Markov model can be trained through the following steps:

[0072] S1, acquiring multiple groups of second spectrum data and second sound spectrum data of target devices collected in multiple historical collection cycles.

[0073] S2, for each set of second spectrum data and second sound spectrum data, extract the second spectrum feature information in the second spectrum data, and extract the second sound spectrum feature information in the second sound spectrum data, respectively perform vector quantization on each feature in the second spectrum feature information and the second sound spectrum feature information, and use all the obtained features as a set of observation data. It should be noted that the process of extracting feature information and vector quantization here can refer to the process of determining the first target feature information, which will not be repeated here. All observation data are organized into an observation data sequence in chronological order.

[0074] S3, for each group of observation data, obtain the partial discharge state of the target device in the historical acquisition period corresponding to the observation data as the state data corresponding to the observation data, wherein the type of partial discharge state includes at least one of the following: slight discharge state, moderate discharge state, severe discharge state; all state data are organized into a state data sequence in chronological order.

[0075] S4, iteratively training the hidden Markov model according to the observation data sequence and the state data sequence, and updating the model parameters, wherein the model parameters include: a state transfer matrix, an observation value probability matrix and an initial probability matrix.

[0076] Specifically, assuming that there are three types of state data, a total of Q groups of observation data are obtained, and the corresponding observation data sequence is recorded as [V1, V2, …, V Q ], and its corresponding state data sequence is recorded as [θ1,θ2…,θ Q ], the main parameters of the hidden Markov model are determined by training:

[0077] 1) State transfer matrix:

[0078] A={α ij =P(q t+1 =θ j ∣q t =θ i )}

[0079] Among them, q t is the state at time t, α ij It means that given the current state θ i In the case of j probability.

[0080] 2) Observation probability matrix:

[0081] B = {b jk =P(o t =V k ∣∣q t =θ j )}

[0082] Among them, t is the observed value at time t, b jk Indicates that under a given state θj, V is observed k The probability of , Q is the total number of observations.

[0083] 3) Initial probability matrix:

[0084] π={π i =P(q1=θ i )}

[0085] Among them, π i Represents the probability of being in state θi at the initial moment.

[0086] In practical applications, it is only necessary to input the current first target feature information into the trained hidden Markov model, and the model can analyze and predict the current accurate partial discharge state of the target device.

[0087] In an embodiment of the present application, by integrating multi-source detection technology, the first spectral data and the first acoustic spectrum data of the target device are collected simultaneously within the target sampling period. Subsequently, the spectral features such as the pulse repetition rate ratio, the maximum pulse amplitude ratio, the average pulse amplitude ratio in the spectral data, and the acoustic spectrum features such as the Mel frequency cepstral coefficient and the gammatone frequency cepstral coefficient in the acoustic spectrum data are extracted respectively. These features are efficiently processed using a vector quantization method to form the first target feature information. Finally, the feature information is analyzed by a pre-trained hidden Markov model to accurately identify the first partial discharge state of the target device. This technology realizes the deep fusion of spectral and acoustic spectrum information, effectively overcomes the limitations of a single data source, improves the accuracy and reliability of partial discharge detection, and thus solves the technical problems of poor utilization of acoustic and optical information and low accuracy of detection results in related partial discharge detection schemes.

[0088] Example 2

[0089] According to an embodiment of the present application, a partial discharge detection device for implementing the partial discharge detection method in Embodiment 1 is also provided. Figure 2 As shown, the partial discharge detection device at least includes: an acquisition module 21, a first extraction module 22, a second extraction module 23, a vector quantization module 24, and a fault analysis module 25, wherein:

[0090] An acquisition module 21 is used to acquire first spectrum data and first sound spectrum data of a target device collected within a target sampling period;

[0091] A first extraction module 22 is used to extract first spectrum characteristic information from the first spectrum data, wherein the first spectrum characteristic information includes at least one of the following characteristics: pulse repetition rate proportion, maximum pulse amplitude proportion, and average pulse amplitude proportion;

[0092] The second extraction module 23 is used to extract the first sound spectrum feature information in the first sound spectrum data, wherein the first sound spectrum feature information includes at least one of the following features: Mel frequency cepstral coefficients and gammatone frequency cepstral coefficients;

[0093] A vector quantization module 24 is used to perform vector quantization on each feature in the first spectrum feature information and the first sound spectrum feature information, and use all the obtained features as first target feature information;

[0094] The fault analysis module 25 is used to analyze the first target feature information by using a pre-trained hidden Markov model to obtain a first partial discharge state of the target device.

[0095] The functions of each module of the partial discharge detection device are described below in conjunction with a specific implementation process.

[0096] First, the acquisition module acquires first spectrum data and first sound spectrum data of the target device collected within a target sampling period, wherein the target device may be a power distribution network device, and the first spectrum data generally covers spectrum data corresponding to multiple bands.

[0097] When analyzing it, the first extraction module can extract the following features from the first spectral data as the first spectral feature information, including: the pulse repetition rate proportion of each band, the maximum pulse amplitude proportion, and the average pulse amplitude proportion. It should be noted that these features can be used alone or flexibly combined to optimize the analysis effect, thereby more accurately reflecting the complex characteristics of partial discharge.

[0098] Optionally, the process of extracting the pulse repetition rate proportion of each band can be carried out by the following steps: for each band, the number of discharge pulses in the band per unit time is determined as the pulse repetition rate of the band based on the first spectral data, and the first sum of the pulse repetition rates of multiple bands is determined, and the ratio of the pulse repetition rate of the band to the first sum is determined as the pulse repetition rate proportion of the band.

[0099] Optionally, the process of extracting the maximum pulse amplitude ratio of each band can be carried out by the following steps: for each band, the maximum amplitude of the discharge pulse sequence of the band within the target sampling period is determined as the maximum pulse amplitude of the band based on the first spectral data, and the second sum of the maximum pulse amplitudes of multiple bands is determined, and the ratio of the maximum pulse amplitude of the band to the second sum is determined as the maximum pulse amplitude ratio of the band.

[0100] Optionally, the process of extracting the average pulse amplitude ratio of each band can be carried out by the following steps: for each band, the average amplitude of the discharge pulse sequence of the band within the target sampling period is determined as the average pulse amplitude of the band based on the first spectral data, and the third sum of the average pulse amplitudes of multiple bands is determined, and the ratio of the average pulse amplitude of the band to the third sum is determined as the average pulse amplitude ratio of the band.

[0101] After the first sound spectrum data is acquired, the second extraction module extracts first sound spectrum feature information in the first sound spectrum data, wherein the first sound spectrum feature information may include at least one of the following features: Mel frequency cepstral coefficients and gammatone frequency cepstral coefficients.

[0102] Specifically, the second extraction module can obtain the above-mentioned Mel-frequency cepstrum coefficients and gammatone frequency cepstrum coefficients in the following manner: first, the first spectrum data is framed to obtain multiple frames of first sub-spectrum data; secondly, each frame of the first sub-spectrum data is windowed using a Hamming window to obtain multiple frames of second sub-spectrum data; then, each frame of the second sub-spectrum data is fast Fourier transformed to obtain spectrum data; then, the spectrum data is filtered using a Mel triangular filter group, and the filtering result is logarithmic and discrete cosine transformed to obtain the Mel-frequency cepstrum coefficients corresponding to the second sub-spectrum data; the spectrum data is filtered using a gammatone filter group, and the filtering result is logarithmic and discrete cosine transformed to obtain the gammatone frequency cepstrum coefficients corresponding to the second sub-spectrum data.

[0103] It should be noted that in order to reduce the impact of the truncation effect in the framing process and reduce the slope at both ends of the frame, it is necessary to perform a windowing operation on the speech frame. Considering that the spectral smoothness of the rectangular window is better, but the waveform details are easily lost and leakage will occur, and the main lobe width of the Hamming window is twice that of the rectangular window, and the attenuation is also faster than the rectangular window. Therefore, the embodiment of the present application selects the Hamming window to perform windowing processing on the first sub-spectrum data of each frame.

[0104] Optionally, after obtaining the above-mentioned first spectral feature information and first spectral feature information, the vector quantization module can respectively perform vector quantization on each feature therein, and this process can be carried out in the following manner: first, use the splitting method to respectively determine the first initial code book corresponding to the first spectral feature information and the second initial code book corresponding to the first spectral feature information; then use the nearest neighbor method to iteratively update the first initial code book to obtain the first target code book, and iteratively update the second initial code book to obtain the second target code book; then, use the first target code book to vector quantize each feature in the first spectral feature information, and use the second target code book to vector quantize each feature in the first spectral feature information.

[0105] Among them, as an optional implementation method, when using the splitting method to determine the first initial code book corresponding to the first spectral feature information, for the first spectral feature information, the following splitting process is cyclically executed: for each code word in this cycle, the code word is split into two code words according to two disturbance coefficients, wherein the code word of the first cycle is the average value of each feature in the first spectral feature information, and the two disturbance coefficients are 1+preset disturbance value and 1-preset disturbance value respectively; if the total number of code words obtained by splitting in this cycle does not reach the preset number, all the code words obtained by splitting will be used as the code words for the next cycle; if the total number of code words obtained by splitting in this cycle reaches the preset number, the cycle is stopped, and all the code words obtained by splitting will be used as the first initial code book.

[0106] Similarly, when the splitting method is used to determine the second initial code book corresponding to the first sound spectrum feature information, a similar splitting process can be executed cyclically: for each code word in this cycle, the code word is split into two code words according to two perturbation coefficients, wherein the code word of the first cycle is the average value of each feature in the first sound spectrum feature information; if the total number of code words obtained by splitting in this cycle does not reach the preset number, all the code words obtained by splitting are used as the code words for the next cycle; if the total number of code words obtained by splitting in this cycle reaches the preset number, the cycle is stopped, and all the code words obtained by splitting are used as the second initial code book.

[0107] Optionally, after obtaining the first initial codebook and the second initial codebook, the vector quantization module may iteratively update the first initial codebook using a nearest neighbor method to obtain a first target codebook, and iteratively update the second initial codebook to obtain a second target codebook, specifically in the following manner.

[0108] Optionally, for the first initial code book, the following update process is executed in a loop: determine the code book of this cycle, and cluster each feature in the first spectral feature information according to the nearest neighbor principle and the code book to obtain multiple clustering results, wherein the code book of the first cycle is the first initial code book; determine the first distance between each feature and the corresponding codeword in each clustering result, and determine the sum of all first distances as the total distance of this cycle; determine the difference between the total distance of this cycle and the total distance of the previous cycle, and determine the ratio of the difference to the total distance of this cycle as the distance correction amount, wherein the total distance of the previous cycle corresponding to the first cycle is the preset initial distance; if the distance correction amount is not less than the preset distance threshold, and the current number of cycles has not reached the preset number of cycles, determine the center of the multiple clustering results as the code book of the next cycle; if the distance correction amount is less than the preset distance threshold, or the current number of cycles reaches the preset number of cycles, stop the cycle, and use the code book of this cycle as the first target code book.

[0109] Similarly, referring to the above step of determining the first target codebook, for the second initial codebook, a similar updating process is cyclically executed to obtain the second target codebook: determine the codebook of this cycle, and cluster each feature in the first sound spectrum feature information according to the nearest neighbor principle and the codebook to obtain multiple clustering results, wherein the codebook of the first cycle is the second initial codebook; determine the second distance between each feature and the corresponding codeword in each clustering result, and determine the sum of all second distances as the total distance of this cycle; determine the difference between the total distance of this cycle and the total distance of the previous cycle, and determine the ratio of the difference to the total distance of this cycle as the distance correction amount, wherein the total distance of the previous cycle corresponding to the first cycle is the preset initial distance; if the distance correction amount is not less than the preset distance threshold, and the current cycle number does not reach the preset cycle number, determine the center of the multiple clustering results as the codebook of the next cycle; if the distance correction amount is less than the preset distance threshold, or the current cycle number reaches the preset cycle number, stop the cycle, and use the codebook of this cycle as the second target codebook.

[0110] Optionally, after obtaining the first target codebook and the second target codebook, the vector quantization module uses the first target codebook to vector quantize each feature in the first spectrum feature information, and uses the second target codebook to vector quantize each feature in the first spectrum feature information, and uses all the obtained features as the first target feature information.

[0111] Furthermore, the fault analysis module can use a pre-trained hidden Markov model to analyze the above-mentioned first target feature information to obtain a first partial discharge state of the target device, wherein the type of the partial discharge state may include at least one of the following: a slight discharge state, a moderate discharge state, and a severe discharge state.

[0112] As an optional implementation, the above hidden Markov model can be trained by the following steps: obtaining multiple groups of second spectrum data and second sound spectrum data of target devices collected in multiple historical collection cycles; for each group of second spectrum data and second sound spectrum data, extracting the second spectrum feature information in the second spectrum data, and extracting the second sound spectrum feature information in the second sound spectrum data, respectively performing vector quantization on each feature in the second spectrum feature information and the second sound spectrum feature information, and taking all the obtained features as a group of observation data. It should be noted that the process of extracting feature information and vector quantization here can refer to the process of determining the first target feature information, which will not be repeated here. All observation data are organized into an observation data sequence in chronological order; for each group of observation data, the partial discharge state of the target device in the historical collection cycle corresponding to the observation data is obtained as the state data corresponding to the observation data, wherein the type of partial discharge state includes at least one of the following: slight discharge state, moderate discharge state, severe discharge state; all state data are organized into a state data sequence in chronological order; the hidden Markov model is iteratively trained according to the observation data sequence and the state data sequence, and the model parameters are updated, wherein the model parameters include: state transition matrix, observation value probability matrix and initial probability matrix.

[0113] It should be noted that each module in the partial discharge detection device in the embodiment of the present application corresponds one-to-one to each implementation step of the partial discharge detection method in Example 1. Since a detailed description has been given in Example 1, some details not reflected in this embodiment can be referred to Example 1 and will not be repeated here.

[0114] Example 3

[0115] According to an embodiment of the present application, a computer program product is further provided. The computer program product includes a computer program. When the computer program is executed by a processor, the partial discharge detection method in Embodiment 1 is implemented.

[0116] According to an embodiment of the present application, a non-volatile storage medium is also provided, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the partial discharge detection method in Example 1 by running the computer program.

[0117] According to an embodiment of the present application, a processor is further provided, which is used to run a computer program, wherein the partial discharge detection method in Embodiment 1 is executed when the computer program is run.

[0118] According to an embodiment of the present application, an electronic device is further provided, comprising: a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to execute the partial discharge detection method in Embodiment 1 through the computer program.

[0119] Specifically, when the computer program is running, the following steps are executed: obtaining the first spectrum data and the first sound spectrum data of the target device collected within the target sampling period; extracting the first spectrum feature information in the first spectrum data, wherein the first spectrum feature information includes at least one of the following features: pulse repetition rate ratio, maximum pulse amplitude ratio, average pulse amplitude ratio; extracting the first sound spectrum feature information in the first sound spectrum data, wherein the first sound spectrum feature information includes at least one of the following features: Mel frequency cepstral coefficients, gammatone frequency cepstral coefficients; vector quantizing each feature in the first spectrum feature information and the first sound spectrum feature information, and using all the obtained features as the first target feature information; analyzing the first target feature information using a pre-trained hidden Markov model to obtain a first partial discharge state of the target device.

[0120] As an optional implementation, the electronic device may be in the form of a mobile terminal, a computer terminal or a similar computing device. Figure 3 FIG. 1 shows a hardware structure block diagram of an electronic device for implementing a partial discharge detection method. Figure 3 As shown, the electronic device 30 may include one or more (302a, 302b, ..., 302n are used to illustrate) processors 302 (the processor 302 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 304 for storing data, and a transmission device 306 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 3 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 3 More or fewer components as shown, or with Figure 3 Different configurations shown.

[0121] It should be noted that the one or more processors 302 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the electronic device 30. As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0122] The memory 304 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the partial discharge detection method in the embodiment of the present application. The processor 302 executes various functional applications and data processing by running the software programs and modules stored in the memory 304, that is, to implement the vulnerability detection method of the above-mentioned application. The memory 304 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 304 may further include a memory remotely arranged relative to the processor 302, and these remote memories may be connected to the electronic device 30 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0123] The transmission device 306 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the electronic device 30. In one example, the transmission device 306 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 306 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0124] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the electronic device 30 .

[0125] The serial numbers of the above embodiments are only for description and do not represent the advantages or disadvantages of the embodiments.

[0126] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0127] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0128] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed over multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0129] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0130] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of each embodiment method of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0131] The above are only preferred implementations of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A partial discharge detection method, characterized in that: include: Acquire first spectrum data and first sound spectrum data of a target device collected within a target sampling period; Extracting first spectrum characteristic information from the first spectrum data, wherein the first spectrum characteristic information includes at least one of the following characteristics: pulse repetition rate proportion, maximum pulse amplitude proportion, and average pulse amplitude proportion; Extracting first sound spectrum feature information from the first sound spectrum data, wherein the first sound spectrum feature information includes at least one of the following features: Mel frequency cepstrum coefficients and gammatone frequency cepstrum coefficients; Performing vector quantization on each feature in the first spectrum feature information and the first sound spectrum feature information respectively, and using all obtained features as first target feature information; The first target feature information is analyzed using a pre-trained hidden Markov model to obtain a first partial discharge state of the target device.

2. The method according to claim 1, characterized in that The first spectral data includes spectral data corresponding to a plurality of bands, and extracting first spectral feature information from the first spectral data includes: For each band, the number of discharge pulses in the band per unit time is determined as the pulse repetition rate of the band according to the first spectral data, and a first sum of the pulse repetition rates of multiple bands is determined, and the ratio of the pulse repetition rate of the band to the first sum is determined as the pulse repetition rate ratio of the band; and / or, For each band, determining the maximum amplitude of the discharge pulse sequence of the band within the target sampling period according to the first spectrum data as the maximum pulse amplitude of the band, determining the second sum of the maximum pulse amplitudes of multiple bands, and determining the ratio of the maximum pulse amplitude of the band to the second sum as the maximum pulse amplitude ratio of the band; and / or, For each band, the average amplitude of the discharge pulse sequence of the band within the target sampling period is determined according to the first spectral data as the average pulse amplitude of the band, and the third sum of the average pulse amplitudes of multiple bands is determined, and the ratio of the average pulse amplitude of the band to the third sum is determined as the average pulse amplitude ratio of the band.

3. The method according to claim 1, characterized in that Extracting first sound spectrum feature information from the first sound spectrum data includes: Performing frame processing on the first sound spectrum data to obtain multiple frames of first sub-sound spectrum data; Using a Hamming window to perform windowing processing on each frame of the first sub-sound spectrum data to obtain multiple frames of second sub-sound spectrum data; For each frame of the second sub-sound spectrum data, the second sub-sound spectrum data is subjected to a fast Fourier transform to obtain spectrum data; the spectrum data is filtered using a Mel triangular filter group, and a logarithmic operation and a discrete cosine transform are performed on the filtering result to obtain a Mel-frequency cepstrum coefficient corresponding to the second sub-sound spectrum data; and / or, the spectrum data is filtered using a gammatone filter group, and a logarithmic operation and a discrete cosine transform are performed on the filtering result to obtain a gammatone frequency cepstrum coefficient corresponding to the second sub-sound spectrum data.

4. The method according to claim 1, characterized in that: Performing vector quantization on each feature in the first spectrum feature information and the first sound spectrum feature information respectively includes: Determine a first initial codebook corresponding to the first spectrum feature information and a second initial codebook corresponding to the first sound spectrum feature information respectively by using a splitting method; Iteratively updating the first initial codebook using a nearest neighbor method to obtain a first target codebook, and iteratively updating the second initial codebook to obtain a second target codebook; Using the first target codebook to perform vector quantization on each feature in the first spectral feature information; The second target codebook is used to perform vector quantization on each feature in the first spectrum feature information.

5. The method according to claim 4, characterized in that Determining a first initial codebook corresponding to the first spectrum feature information and a second initial codebook corresponding to the first sound spectrum feature information respectively by using a splitting method includes: For the first spectral feature information, the following splitting process is executed cyclically: for each codeword in this cycle, the codeword is split into two codewords according to two disturbance coefficients, wherein the codeword of the first cycle is the average value of each feature in the first spectral feature information, and the two disturbance coefficients are 1+preset disturbance value and 1-preset disturbance value respectively; if the total number of codewords obtained by splitting in this cycle does not reach the preset number, all the codewords obtained by splitting are used as codewords for the next cycle; if the total number of codewords obtained by splitting in this cycle reaches the preset number, the cycle is stopped, and all the codewords obtained by splitting are used as the first initial codebook; For the first sound spectrum feature information, the following splitting process is executed cyclically: for each codeword in this cycle, the codeword is split into two codewords according to the two perturbation coefficients, wherein the codeword of the first cycle is the average value of each feature in the first sound spectrum feature information; if the total number of codewords obtained by splitting in this cycle does not reach a preset number, all the codewords obtained by splitting are used as codewords for the next cycle; if the total number of codewords obtained by splitting in this cycle reaches a preset number, the cycle is stopped, and all the codewords obtained by splitting are used as the second initial codebook.

6. The method according to claim 4, characterized in that Iteratively updating the first initial codebook using a nearest neighbor method to obtain a first target codebook, and iteratively updating the second initial codebook to obtain a second target codebook, comprising: For the first initial code book, the following update process is executed cyclically: determine the code book of this cycle, and cluster each feature in the first spectral feature information according to the nearest neighbor principle and the code book to obtain multiple clustering results, wherein the code book of the first cycle is the first initial code book; determine the first distance between each feature and the corresponding code word in each clustering result, and determine the sum of all the first distances as the total distance of this cycle; determine the difference between the total distance of this cycle and the total distance of the previous cycle, and determine the ratio of the difference to the total distance of this cycle as the distance correction amount, wherein the total distance of the previous cycle corresponding to the first cycle is the preset initial distance; if the distance correction amount is not less than the preset distance threshold, and the current number of cycles does not reach the preset number of cycles, determine the center of the multiple clustering results as the code book of the next cycle; if the distance correction amount is less than the preset distance threshold, or the current number of cycles reaches the preset number of cycles, stop the cycle, and use the code book of this cycle as the first target code book; For the second initial codebook, the following update process is executed cyclically: determining the codebook of this cycle, and clustering each feature in the first sound spectrum feature information according to the nearest neighbor principle and the codebook to obtain multiple clustering results, wherein the codebook of the first cycle is the second initial codebook; determining the second distance between each feature and the corresponding codeword in each clustering result, and determining the sum of all the second distances as the total distance of this cycle; determining the difference between the total distance of this cycle and the total distance of the previous cycle, and determining the ratio of the difference to the total distance of this cycle as the distance correction amount, wherein the total distance of the previous cycle corresponding to the first cycle is the preset initial distance; if the distance correction amount is not less than the preset distance threshold, and the current cycle number does not reach the preset cycle number, determining the center of the multiple clustering results as the codebook of the next cycle; if the distance correction amount is less than the preset distance threshold, or the current cycle number reaches the preset cycle number, stopping the cycle, and using the codebook of this cycle as the second target codebook.

7. The method according to claim 1, characterized in that The training process of the hidden Markov model includes: Acquire multiple groups of second spectrum data and second sound spectrum data of the target device collected in multiple historical collection cycles; For each set of the second spectrum data and the second sound spectrum data, extract the second spectrum feature information in the second spectrum data, and extract the second sound spectrum feature information in the second sound spectrum data, respectively perform vector quantization on each feature of the second spectrum feature information and the second sound spectrum feature information, and use all the obtained features as a set of observation data; and organize all the observation data into an observation data sequence in chronological order; For each set of observation data, the partial discharge state of the target device in the historical acquisition period corresponding to the observation data is obtained as the state data corresponding to the observation data, wherein the type of the partial discharge state includes at least one of the following: slight discharge state, moderate discharge state, and severe discharge state; all the state data are combined into a state data sequence in chronological order; The hidden Markov model is iteratively trained according to the observation data sequence and the state data sequence to update model parameters, wherein the model parameters include: a state transfer matrix, an observation value probability matrix and an initial probability matrix.

8. A partial discharge detection device, characterized in that: include: An acquisition module, used to acquire first spectrum data and first sound spectrum data of a target device collected within a target sampling period; A first extraction module is used to extract first spectrum characteristic information from the first spectrum data, wherein the first spectrum characteristic information includes at least one of the following characteristics: pulse repetition rate proportion, maximum pulse amplitude proportion, and average pulse amplitude proportion; A second extraction module is used to extract first sound spectrum feature information from the first sound spectrum data, wherein the first sound spectrum feature information includes at least one of the following features: Mel frequency cepstrum coefficients and gammatone frequency cepstrum coefficients; A vector quantization module, used to perform vector quantization on each feature in the first spectrum feature information and the first sound spectrum feature information, and use all obtained features as first target feature information; The fault analysis module is used to analyze the first target feature information by using a pre-trained hidden Markov model to obtain a first partial discharge state of the target device.

9. A computer program product, characterized in that include: A computer program, wherein when the computer program is executed by a processor, the partial discharge detection method according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: include: A memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the partial discharge detection method according to any one of claims 1 to 7 through the computer program.

Citation Information

Patent Citations

  • Speaker confirmation method based on VQ and SVM algorithms, and system thereof

    CN106373576A

  • Acousto-optic combined detection method and system for partial discharging of a high-voltage cable

    CN110275094A

  • Power equipment partial discharge detection method and equipment

    CN114755543A