Power equipment fault detection method and system based on sound feature analysis
By calculating the deviation coefficient and temporal degradation coefficient of the acoustic fingerprint sequence of power equipment, and combining the conditional GAN model and convolutional neural network, the problem of sample imbalance in power equipment fault detection is solved, and more accurate fault detection is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 国网黑龙江省电力有限公司大兴安岭供电公司
- Filing Date
- 2025-10-15
- Publication Date
- 2026-07-03
AI Technical Summary
In existing technologies, deep learning-based fault detection methods for power equipment suffer from a lack of fault data, leading to imbalanced sample categories, overfitting, and inaccurate detection results.
By acquiring multiple voiceprint sequences from power equipment, dividing them into voiceprint segments, calculating the deviation coefficient and temporal degradation coefficient, generating new voiceprint sequences using a conditional GAN model, and training them through a convolutional neural network, the sample quality and detection accuracy are improved.
Dynamically measuring the similarity between voiceprint segments improves the accuracy of fault detection, generates sample data that more closely resembles real-world situations, and enhances detection performance.
Smart Images
Figure CN121354589B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of fault detection technology, and in particular to a method and system for detecting faults in power equipment based on sound feature analysis. Background Technology
[0002] Power equipment plays a crucial role in power grid operation, performing vital transmission and distribution functions. Faults in power equipment can not only damage the equipment itself but also trigger widespread power outages, economic losses, and even safety accidents. By conducting power equipment fault detection, real-time monitoring and early warning of equipment status can be achieved, potential anomalies can be identified, and maintenance personnel can be guided to take targeted measures. This ensures the safe and stable operation of the power grid, extends equipment lifespan, and reduces maintenance costs and accident risks.
[0003] Electrical equipment generates specific acoustic signals during operation. Different operating states and fault types often correspond to different sound characteristics, such as discharge sounds, vibration sounds, or mechanical friction sounds. Sound acquisition has the advantages of being non-contact, low-cost, and real-time. Data can be obtained simply by deploying sensors around the equipment without affecting its normal operation. By combining modern signal processing and artificial intelligence algorithms, key features can be extracted from these sound signals and abnormal patterns can be identified, enabling accurate diagnosis of early equipment faults.
[0004] In existing technologies, deep learning methods are often used to detect faults in power equipment based on sound feature analysis. However, deep learning requires a sufficient amount of fault data for training. Power equipment is usually designed to be highly reliable, and the probability of serious faults is extremely low. This directly leads to an extreme scarcity of fault audio samples, resulting in very few fault samples that can be used for training, while the number of normal samples is extremely large. This leads to an extreme imbalance in sample categories, which can easily cause the model to overfit. That is, the model tends to judge all inputs as normal, resulting in inaccurate power equipment fault detection results obtained by deep learning. Summary of the Invention
[0005] To address the aforementioned technical problems, the purpose of this application is to provide a method and system for detecting power equipment faults based on sound feature analysis. The specific technical solution adopted is as follows:
[0006] Firstly, a method for detecting faults in power equipment based on sound feature analysis is provided, the method comprising:
[0007] Obtain multiple acoustic fingerprint sequences of faulty power equipment and divide each acoustic fingerprint sequence into multiple acoustic fingerprint segments;
[0008] The deviation coefficient of the voiceprint segment is determined based on the difference between the amplitude of multiple sampling points in each voiceprint segment and the preset amplitude.
[0009] Based on the distance between the frequency value corresponding to each maximum value in each frequency curve and other frequency values, and the amplitude corresponding to other frequency values, the clustering coefficient of each maximum value is determined. Based on the difference between the clustering coefficient of each maximum value in each frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve, the similarity between each frequency curve and each other frequency curve is determined. Each frequency curve is obtained by performing a Fourier transform on each voiceprint segment.
[0010] The temporal degradation coefficient of a voiceprint segment is determined based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence.
[0011] The fault weight of a voiceprint segment is determined based on the deviation coefficient of each voiceprint segment, the temporal degradation coefficient of that voiceprint segment, and the multiple deviation coefficients of multiple target voiceprint segments; the target voiceprint segment indicates the voiceprint segment with the largest deviation coefficient among the multiple voiceprint segments of the corresponding voiceprint sequence.
[0012] The fault weights of multiple voiceprint segments in each voiceprint sequence and multiple voiceprint segments with fault type labels are input into a pre-trained conditional GAN model to generate a new voiceprint sequence; the new voiceprint sequence is used to train a convolutional neural network.
[0013] Optionally, determining the deviation coefficient of a voiceprint segment based on the difference between the amplitude of multiple sampling points in each voiceprint segment and a preset amplitude includes:
[0014] The pulse weight of each voiceprint segment is determined based on the difference between the amplitude of multiple sampling points in each voiceprint segment and the preset amplitude, the number of multiple sampling points, and the amplitude of multiple sampling points.
[0015] The deviation coefficient of a voiceprint segment is determined based on the pulse weight of each voiceprint segment, the amplitude of multiple sampling points in that voiceprint segment, and the preset amplitude.
[0016] Optionally, determining the deviation coefficient of a voiceprint segment based on the pulse weight of each voiceprint segment, the amplitude of multiple sampling points in the voiceprint segment, and a preset amplitude includes:
[0017] Calculate the absolute difference between the mean amplitude of multiple sampling points in each voiceprint segment and the preset amplitude to obtain the amplitude difference of each voiceprint segment;
[0018] The deviation coefficient of a voiceprint segment is determined based on the amplitude difference of each voiceprint segment and the pulse weight of that voiceprint segment.
[0019] Optionally, determining the clustering coefficient of each maximum value based on the distance between the frequency value corresponding to each maximum value and other frequency values, and the amplitude corresponding to the other frequency values, includes:
[0020] Perform a Fourier transform on each voiceprint segment to obtain the frequency curve corresponding to each voiceprint segment; where the horizontal axis of the frequency curve is the frequency value and the vertical axis is the amplitude.
[0021] The aggregation coefficient of each maximum value is determined based on the number of multiple frequency values in each frequency curve, the distance between the frequency value corresponding to each maximum value and each other frequency value, and the amplitude of the other frequency value.
[0022] Optionally, determining the similarity between each frequency curve and each other frequency curve based on the difference between the clustering coefficient of each maximum value in each frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve includes:
[0023] The similarity between a frequency curve and other frequency curves is determined by the larger of the difference between the clustering coefficient of each maximum value in each frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve, and the larger of the clustering coefficient of each maximum value in the frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve.
[0024] Optionally, determining the temporal degradation coefficient of a voiceprint segment based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence, includes:
[0025] Calculate the similarity between the frequency curve corresponding to each voiceprint segment and each other frequency curve in the same voiceprint sequence to obtain the first similarity score;
[0026] Calculate the similarity between the other frequency curve and the frequency curve corresponding to the voiceprint segment to obtain the second similarity.
[0027] The smaller value between the first similarity and the second similarity is determined as the comprehensive similarity between the frequency curve corresponding to the voiceprint segment and the other frequency curves;
[0028] The temporal degradation coefficient of a voiceprint segment is determined based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the comprehensive similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence.
[0029] Optionally, determining the temporal degradation coefficient of a voiceprint segment based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the comprehensive similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence, includes:
[0030] The stability coefficient of the voiceprint sequence to which the voiceprint segment belongs is determined based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the comprehensive similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence.
[0031] The deviation value between the voiceprint segment and the preceding voiceprint segment is determined based on the deviation coefficient of each voiceprint segment and the deviation coefficient of each preceding voiceprint segment in the same voiceprint sequence.
[0032] The degradation coefficient of a voiceprint segment is determined based on the stability coefficient of the voiceprint sequence to which each voiceprint segment belongs and the deviation value between the voiceprint segment and each preceding voiceprint segment.
[0033] The temporal degradation coefficient of a voiceprint segment is determined based on the degradation coefficient of each preceding voiceprint segment in the voiceprint sequence to which the voiceprint segment belongs, and the stability coefficient of the voiceprint sequence to which the voiceprint segment belongs.
[0034] Optionally, determining the fault weight of a voiceprint segment based on the deviation coefficient of each voiceprint segment, the temporal degradation coefficient of the voiceprint segment, and multiple deviation coefficients of multiple target voiceprint segments includes:
[0035] The fault difference coefficient of a voiceprint segment is determined based on the difference between the deviation coefficient of each voiceprint segment and the deviation coefficient of each target voiceprint segment, and the difference between the timing degradation coefficient of the voiceprint segment and the timing degradation coefficient of the target voiceprint segment.
[0036] The fault difference coefficient of each voiceprint segment is normalized, and the difference between the preset value and the normalized fault difference coefficient is determined as the fault weight of that voiceprint segment.
[0037] Optionally, after inputting the fault weights of multiple voiceprint segments in each voiceprint sequence and the multiple voiceprint segments labeled with fault types into a pre-trained conditional GAN model to generate a new voiceprint sequence, the method further includes:
[0038] The new voiceprint sequence is converted into a time-spectrum image and input into a convolutional neural network. The convolutional neural network is then trained in a supervised manner to obtain a trained convolutional neural network.
[0039] Secondly, a power equipment fault detection system based on sound feature analysis is provided, the system comprising:
[0040] The acquisition module is used to acquire multiple acoustic fingerprint sequences of faulty power equipment and divide each acoustic fingerprint sequence into multiple acoustic fingerprint segments.
[0041] The first determining module is used to determine the deviation coefficient of the voiceprint segment based on the difference between the amplitude of multiple sampling points in each voiceprint segment and the preset amplitude.
[0042] The second determining module is used to determine the clustering coefficient of each maximum value based on the distance between the frequency value corresponding to each maximum value in each frequency curve and other frequency values, and the amplitude corresponding to other frequency values. It also determines the similarity between each frequency curve and each other frequency curve based on the difference between the clustering coefficient of each maximum value in each frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve. Each frequency curve is obtained by performing a Fourier transform on each voiceprint segment.
[0043] The third determining module is used to determine the temporal degradation coefficient of a voiceprint segment based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs and the similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence.
[0044] The fourth determination module is used to determine the fault weight of a voiceprint segment based on the deviation coefficient of each voiceprint segment, the temporal degradation coefficient of the voiceprint segment, and multiple deviation coefficients of multiple target voiceprint segments; the target voiceprint segment indicates the voiceprint segment with the largest deviation coefficient among multiple voiceprint segments in the corresponding voiceprint sequence.
[0045] The generation module is used to input the fault weights of multiple voiceprint segments in each voiceprint sequence and multiple voiceprint segments with fault type labels into a pre-trained conditional GAN model to generate new voiceprint sequences; the new voiceprint sequences are used to train convolutional neural networks.
[0046] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined arbitrarily to obtain various preferred embodiments of this application.
[0047] This application offers the following advantages: It calculates the deviation coefficient based on the difference between the amplitude of sampling points in each voiceprint segment and a preset amplitude to assess the anomaly level of the signal segment. By transforming each voiceprint segment into a frequency curve using Fourier transform, the aggregation coefficient of each maximum value is determined based on the distance between the frequency value corresponding to each maximum and other frequency values, as well as the amplitude corresponding to other frequency values. Furthermore, the similarity between frequency curves is calculated based on the difference in aggregation coefficients of corresponding maximum values between different frequency curves. On this basis, the temporal degradation coefficient is determined by combining the deviation coefficients of each voiceprint segment in the same sequence and the similarity between the frequency curve corresponding to that voiceprint segment and other frequency curves in the sequence, reflecting the temporal evolution of fault characteristics. Finally, by combining the deviation coefficients of the voiceprint segment, the temporal degradation coefficient, and the deviation coefficient of the target voiceprint segment in the sequence, the fault weight of that voiceprint segment is determined. The fault weights of multiple voiceprint segments in each voiceprint sequence, along with multiple voiceprint segments labeled with fault types, are input into a pre-trained conditional GAN model to generate a new voiceprint sequence. This new voiceprint sequence will be used for subsequent training of the convolutional neural network. When comparing different samples, the similarity between voiceprint segments can be dynamically measured according to the fault development stage, thereby more accurately assessing the fault weight of each voiceprint segment, improving the quality and usability of generated samples, and enhancing the accuracy of fault detection. Attached Figure Description
[0048] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0049] Figure 1 This is a flowchart of a power equipment fault detection method based on sound feature analysis in one embodiment;
[0050] Figure 2 This is a schematic diagram of the structure of a power equipment fault detection system based on sound feature analysis in one embodiment;
[0051] Figure 3 This is a schematic diagram of the structure of an electronic device in one embodiment. Detailed Implementation
[0052] To further illustrate the technical means and effects adopted by this application to achieve the intended inventive objective, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a power equipment fault detection method and system based on sound feature analysis proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.
[0054] The following description, in conjunction with the accompanying drawings, details a specific scheme for a power equipment fault detection method based on sound feature analysis provided in this application. For example... Figure 1 As shown, the method includes:
[0055] S11. Obtain multiple voiceprint sequences of the faulty power equipment and divide each voiceprint sequence into multiple voiceprint segments.
[0056] This application addresses power equipment prone to partial discharge, insulation aging, or mechanical failure, such as transformers, switchgear, circuit breakers, ring main units, and busbar connection points. It involves deploying broadband acoustic sensors near the weakest areas of the equipment's casing and continuously collecting data at a frequency of 20kHz to obtain multiple acoustic fingerprint sequences (sample data). Each acoustic fingerprint sequence is then manually categorized according to the fault type: normal, discharge fault, insulation aging, mechanical friction, and loosening fault.
[0057] Since deep learning requires sufficient and diverse fault data for training, new sample data needs to be generated based on existing real sample data. In power equipment fault detection, acoustic fingerprint sequences often contain signals from different time periods, some of which contain key fault features, while others may mainly consist of background noise or normal operating sounds. This application mainly analyzes acoustic fingerprint sequences characterizing faults (discharge faults, insulation aging, mechanical friction, and loosening faults).
[0058] To preserve the fault characteristics of real samples in the generated sample data and thus improve the authenticity of the generated samples, it is necessary to divide each voiceprint sequence into time periods and assign different weights based on the salience of fault characteristics in each time period. This allows for the selective retention and enhancement of key features during sample expansion, thereby generating data that is closer to the real situation and more valuable for model training.
[0059] For any given voiceprint sequence, divide the voiceprint sequence into windows of 1 second each to obtain several voiceprint feature segments (voiceprint segments). The voiceprint segments in the voiceprint sequence are ordered in chronological order.
[0060] S12. Determine the deviation coefficient of the voiceprint segment based on the difference between the amplitude of multiple sampling points in each voiceprint segment and the preset amplitude.
[0061] The deviation coefficient indicates the degree of abnormality of the voiceprint segment signal relative to a preset amplitude, i.e., a normal reference. The preset amplitude is determined based on the average amplitude of multiple voiceprint sequences that characterize normality.
[0062] In one embodiment, the deviation coefficient of the voiceprint segment is determined based on the difference between the amplitude of multiple sampling points in each voiceprint segment and a preset amplitude, including:
[0063] The pulse weight of each voiceprint segment is determined based on the difference between the amplitude of multiple sampling points in each voiceprint segment and the preset amplitude, the number of multiple sampling points, and the amplitude of multiple sampling points.
[0064] The deviation coefficient of a voiceprint segment is determined based on the pulse weight of each voiceprint segment, the amplitude of multiple sampling points in that voiceprint segment, and the preset amplitude.
[0065] When electrical equipment malfunctions, its abnormal operating state is typically manifested in the acoustic signature sequence as an overall increase in amplitude. This is due to increased energy release, elevated background noise or vibration levels, and an increase in the number and amplitude of pulses. For example, partial discharge, insulation breakdown, or mechanical impact can generate more frequent and stronger transient impact signals. Therefore, the focus should first be on the pulse characteristics of the acoustic signature segment, quantifying the abnormal behavior of that segment by analyzing its amplitude and number of pulses. Pulse weighting of each voiceprint segment The calculation formula is:
[0066] ;
[0067] in, For the first Pulse weighting of each voiceprint segment For the first The number of multiple sampling points in each voiceprint segment, where b is the b-th sampling point, and norm() is the normalization function. For the first The first voiceprint segment The amplitude of each sampling point The preset amplitude is the average amplitude of multiple normal voiceprint sequences.
[0068] The larger the value, the more likely it is to be the first. The first voiceprint segment The more a sampling point deviates from the normal amplitude, the greater its weight is when analyzing the pulse; that is, more attention is paid to the amplitude of that sampling point.
[0069] In one embodiment, the deviation coefficient of the voiceprint segment is determined based on the pulse weight of each voiceprint segment, the amplitude of multiple sampling points in the voiceprint segment, and a preset amplitude, including:
[0070] Calculate the absolute difference between the mean amplitude of multiple sampling points in each voiceprint segment and the preset amplitude to obtain the amplitude difference of each voiceprint segment;
[0071] The deviation coefficient of a voiceprint segment is determined based on the amplitude difference of each voiceprint segment and the pulse weight of that voiceprint segment.
[0072] The deviation of the voiceprint segment from the normal situation is mainly reflected in the difference in the overall amplitude. Therefore, the deviation of the voiceprint segment from the normal situation is used as the basis for constructing the anomaly. Then, the deviation degree of the voiceprint is enhanced by the pulse degree, so as to determine the deviation coefficient of each voiceprint segment.
[0073] No. Deviation coefficient of each voiceprint segment The calculation formula is:
[0074] ;
[0075] in, For the first Deviation coefficient of each voiceprint segment For the first The amplitude difference of each voiceprint segment It can be done through the first The absolute difference between the average amplitude of each voiceprint segment and the preset amplitude is obtained. For the first Pulse weights for each voiceprint segment.
[0076] S13. Based on the distance between the frequency value corresponding to each maximum value in each frequency curve and other frequency values, and the amplitude corresponding to other frequency values, determine the clustering coefficient of each maximum value, and based on the difference between the clustering coefficient of each maximum value in each frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve, determine the similarity between each frequency curve and each other frequency curve.
[0077] Each frequency curve is obtained by performing a Fourier transform on each voiceprint segment. The horizontal axis of the frequency curve represents the frequency value, and the vertical axis represents the amplitude.
[0078] The concentration factor indicates the degree of energy concentration of each maximum value in the entire frequency curve. The larger the value, the higher the amplitude near the maximum value and the rapid decay of the amplitude at adjacent frequency points, indicating highly concentrated energy. The smaller the value, the more dispersed the amplitude, the absence of obvious main peaks, and the flatter the energy, which is mostly background noise or normal operating components.
[0079] The development of a fault is sequential, and the acoustic characteristics of different periods (such as the initial and final stages) exhibit systematic differences in energy, frequency, and other aspects, which directly affects the value of the deviation coefficient. Therefore, direct similarity comparisons cannot be performed; instead, it is necessary to determine the fault period in which each acoustic segment is located to correct or compensate for the interference of the "period" variable on the similarity measurement.
[0080] In one embodiment, the clustering coefficient of each maximum value is determined based on the distance between the frequency value corresponding to each maximum value and other frequency values in each frequency curve, and the amplitude corresponding to the other frequency values, including:
[0081] Perform a Fourier transform on each voiceprint segment to obtain the frequency curve corresponding to each voiceprint segment; where the horizontal axis of the frequency curve is the frequency value and the vertical axis is the amplitude.
[0082] The aggregation coefficient of each maximum value is determined based on the number of multiple frequency values in each frequency curve, the distance between the frequency value corresponding to each maximum value and each other frequency value, and the amplitude of the other frequency value.
[0083] For a single voiceprint sequence, since different voiceprint segments exhibit different fault characteristics within the entire sequence, it is necessary to transform the sequence to the frequency domain for comparison. For any given voiceprint segment, a Fourier transform is used to convert it to the frequency domain, yielding its frequency curve. Each voiceprint segment corresponds to one frequency curve, with the horizontal axis representing frequency and the vertical axis representing amplitude.
[0084] In spectrum analysis, defects in electrical equipment often alter the energy distribution of acoustic signals, causing abnormal clustering of certain frequency components. For example, partial discharge or arcing faults can lead to a significant enhancement of high-frequency energy and the formation of spikes in specific frequency bands, while mechanical friction or loosening can concentrate energy and periodically enhance it in low-frequency or harmonic bands. Therefore, it is necessary to quantify the degree of clustering at each location in the spectrum curve.
[0085] Since each voiceprint segment corresponds to a frequency curve, therefore the first The voiceprint segment corresponds to the first The frequency curve. The first frequency curve Clustering coefficient of each maxima The calculation formula is:
[0086] ;
[0087] in, For the first The first frequency curve The clustering coefficient with a maximum value, For the first The number of multiple frequency values in a frequency curve For the first One frequency value, It is a natural exponential function. For the first The first frequency curve The frequency value corresponding to the i-th maximum value is related to the i-th The first frequency curve The distance between frequency values For the first The first frequency curve The amplitude of each frequency value. If the two frequency curves of the same voiceprint sequence have a stronger clustering at the same frequency, it indicates that the frequency distributions of the two frequency curves have commonalities.
[0088] In one embodiment, determining the similarity between each frequency curve and each other frequency curve based on the difference between the clustering coefficient of each maximum value in each frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve includes:
[0089] The similarity between a frequency curve and other frequency curves is determined by the larger of the difference between the clustering coefficient of each maximum value in each frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve, and the larger of the clustering coefficient of each maximum value in the frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve.
[0090] Record any maximum value in any frequency curve as the target maximum value, and record the maximum value in another frequency curve of the same voiceprint sequence that is closest to the target maximum value as the corresponding maximum value of the target maximum value.
[0091] No. The frequency curve and the first Similarity of frequency curves The calculation formula is:
[0092] ;
[0093] in, For the first The frequency curve and the first The similarity of the frequency curves For the first The number of maxima of a frequency curve. For the first A maximum value, For the first The first frequency curve The clustering coefficient with a maximum value, For the first The first frequency curve The maximum value at the th ... The clustering coefficients of the corresponding maxima of the frequency curves, where MAX() is the maximum value function and norm() is the normalization function.
[0094] The larger the value, the more it indicates and If the values are relatively large, the weight is higher, meaning the focus is mainly on the similarity of clustering near frequencies with high clustering levels.
[0095] S14. Determine the temporal degradation coefficient of the voiceprint segment based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence.
[0096] The timing degradation coefficient indicates the stage or severity of the fault in the voiceprint segment. The larger the value, the more severe the fault is and the closer it is to the end stage.
[0097] In one embodiment, the temporal degradation coefficient of a voiceprint segment is determined based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence, including:
[0098] Calculate the similarity between the frequency curve corresponding to each voiceprint segment and each other frequency curve in the same voiceprint sequence to obtain the first similarity score;
[0099] Calculate the similarity between the other frequency curve and the frequency curve corresponding to the voiceprint segment to obtain the second similarity.
[0100] The smaller value between the first similarity and the second similarity is determined as the comprehensive similarity between the frequency curve corresponding to the voiceprint segment and the other frequency curves;
[0101] The temporal degradation coefficient of a voiceprint segment is determined based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the comprehensive similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence.
[0102] Calculate the first The frequency curve and the first frequency curve in the same voiceprint sequence The similarity between the frequency curves is used as the first similarity, and then the similarity is calculated for the second frequency curve in the same voiceprint sequence. The frequency curve and the first The similarity between the frequency curves is used as the second similarity, and the smaller value between the first and second similarities is used as the third similarity. The frequency curve and the first The overall similarity of the frequency curves is calculated as follows: since the correspondences obtained when the two curves obtain their corresponding maxima may differ, the calculation results will vary. Therefore, the minimum value of the two curves is taken as the first value. The frequency curve and the first Overall similarity of frequency curves .
[0103] In one embodiment, the temporal degradation coefficient of a voiceprint segment is determined based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the comprehensive similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence, including:
[0104] The stability coefficient of the voiceprint sequence to which the voiceprint segment belongs is determined based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the comprehensive similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence.
[0105] The deviation value between the voiceprint segment and the preceding voiceprint segment is determined based on the deviation coefficient of each voiceprint segment and the deviation coefficient of each preceding voiceprint segment in the same voiceprint sequence.
[0106] The degradation coefficient of a voiceprint segment is determined based on the stability coefficient of the voiceprint sequence to which each voiceprint segment belongs and the deviation value between the voiceprint segment and each preceding voiceprint segment.
[0107] The temporal degradation coefficient of a voiceprint segment is determined based on the degradation coefficient of each preceding voiceprint segment in the voiceprint sequence to which the voiceprint segment belongs, and the stability coefficient of the voiceprint sequence to which the voiceprint segment belongs.
[0108] The stability coefficient indicates the internal consistency and stability of the fault characteristics exhibited throughout the entire acoustic signature sequence.
[0109] The degradation coefficient indicates the degradation trend of the voiceprint segment relative to the earlier stable state in the sequence.
[0110] When determining the current stage of a fault, the judgment is mainly based on the characteristic that the degree of fault deviation gradually increases over time. However, the stability of the fault varies at different stages, so it is necessary to select the fault development trend that shows relatively stable fault characteristics over time for judgment.
[0111] Therefore Stability coefficient of the voiceprint sequence for each voiceprint segment The calculation formula is:
[0112] ;
[0113] in, For the first The stability coefficient of the voiceprint sequence described in each voiceprint segment. For the first The number of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs. For the first The frequency curves corresponding to other voiceprint segments. For the first Deviation coefficients for other voiceprint segments For the first The frequency curve and the first The overall similarity of the frequency curves corresponding to other voiceprint segments. It is a normalized exponential function. The larger the value, the more likely it is to be the first. The larger the deviation coefficient of the voiceprint segment, the more significant the deviation. This is mainly due to the first segment. The frequency curve of the first voiceprint segment and the first The overall similarity of the frequency curves of the first voiceprint segment to the second... The stability of the characteristics of each voiceprint segment is measured.
[0114] Then, based on the temporal development of the voiceprint segments with larger characteristic stability coefficients, the degree of degradation corresponding to each voiceprint segment is judged, that is, the degradation period of each voiceprint segment is judged. Degradation coefficient of each voiceprint segment The calculation formula is:
[0115] ;
[0116] in, For the first The degradation coefficient of each voiceprint segment For the first Within the same voiceprint sequence of the voiceprint segment, located at the... The number of preceding voiceprint segments before each voiceprint segment. For the first One front voiceprint segment For the first The first voiceprint segment before The stability coefficient of the voiceprint sequence to which each voiceprint segment belongs. For the first Within the same voiceprint sequence of the voiceprint segment, located at the... The first voiceprint segment before The deviation value of the first pre-voiceprint segment, if the first The deviation coefficient of the first voiceprint segment is greater than that of the second. If the voiceprint segment is positive, the deviation value is 1; otherwise, the deviation value is 0.
[0117] The larger the value, the more likely it is to be the first. If the deviation of the first voiceprint segment from the previous voiceprint segments with higher stability coefficients is significantly greater, then it indicates that the second... Each voiceprint segment exhibits characteristics of the fault deterioration timing.
[0118] Then the first Timing degradation coefficient of each voiceprint segment The calculation formula is:
[0119] ;
[0120] in, For the first The temporal degradation coefficient of each voiceprint segment For the first One front voiceprint segment For the first Within the same voiceprint sequence of the voiceprint segment, located at the... The number of preceding voiceprint segments before each voiceprint segment. For the first The first voiceprint segment before The stability coefficient of the voiceprint sequence to which each voiceprint segment belongs. For the first in the same voiceprint sequence The first voiceprint segment before The degradation coefficient of each front voiceprint segment.
[0121] Since the judgment is mainly based on the temporal degradation properties of the voiceprint segments with stable characteristics, therefore... As a weight, when the degradation coefficients of most voiceprint segments with high feature stability are relatively large, it indicates that the overall degradation coefficient is relatively large, indicating a clear fault development trend. At the same time, an accumulation method is used here to describe the accumulation of degradation degree, thereby quantifying the degree of fault development.
[0122] S15. Determine the fault weight of the voiceprint segment based on the deviation coefficient of each voiceprint segment, the temporal degradation coefficient of the voiceprint segment, and the multiple deviation coefficients of multiple target voiceprint segments.
[0123] The target voiceprint segment indicates the voiceprint segment with the largest deviation coefficient among multiple voiceprint segments in the corresponding voiceprint sequence.
[0124] Fault weights indicate the importance or representativeness of each voiceprint segment in characterizing its fault type.
[0125] In one embodiment, the fault weight of a voiceprint segment is determined based on the deviation coefficient of each voiceprint segment, the temporal degradation coefficient of the voiceprint segment, and multiple deviation coefficients of multiple target voiceprint segments, including:
[0126] The fault difference coefficient of a voiceprint segment is determined based on the difference between the deviation coefficient of each voiceprint segment and the deviation coefficient of each target voiceprint segment, and the difference between the timing degradation coefficient of the voiceprint segment and the timing degradation coefficient of the target voiceprint segment.
[0127] The fault difference coefficient of each voiceprint segment is normalized, and the difference between the preset value and the normalized fault difference coefficient is determined as the fault weight of that voiceprint segment.
[0128] The preset value can be set according to the actual situation, for example, 1.
[0129] For any given voiceprint sequence, the voiceprint segment with the largest deviation coefficient is denoted as the target voiceprint segment of that voiceprint sequence.
[0130] When measuring the difference between any voiceprint segment and the target voiceprint segment based on the deviation coefficient, since there may be significant differences due to different stages of fault development, the similarity of the fault is compensated by the difference in the timing degradation coefficient (i.e., the difference in the stage of fault development).
[0131] No. Fault difference coefficient of each voiceprint segment The calculation formula is:
[0132] ;
[0133] in, For the first Fault difference coefficient of each voiceprint segment The number of multiple voiceprint sequences. For the first A voiceprint sequence For the first Deviation coefficient of each voiceprint segment Let be the deviation coefficient of the target voiceprint segment in the i-th voiceprint sequence. For the first The temporal degradation coefficient of each voiceprint segment is the temporal degradation coefficient of the target voiceprint segment in the i-th voiceprint sequence.
[0134] This reflects the degree of difference between the deviation coefficient of the voiceprint segment and the representative voiceprint segment. When it is large, then The difference is smaller, thus reducing the degree of difference in the deviation coefficient, that is, compensating for the similarity of the faults (i.e., the subsequent weight of the fault performance).
[0135] No. Fault weight of each voiceprint segment The calculation formula is:
[0136] ;
[0137] in, For the first Fault weights for each voiceprint segment For the first The fault difference coefficient of each voiceprint segment, norm() is the normalization function.
[0138] S16. Input the fault weights of multiple voiceprint segments in each voiceprint sequence and the multiple voiceprint segments with fault type labels into the pre-trained conditional GAN model to generate new voiceprint sequences.
[0139] Among them, the new voiceprint sequence is used to train the convolutional neural network.
[0140] Multiple voiceprint segments labeled with fault type and their corresponding fault weights from each voiceprint sequence are input into the generator to output a new voiceprint sequence. Simultaneously, a discriminator performs adversarial loss to distinguish between real and generated voiceprints, and a weighted reconstruction error is calculated between the generator output and the corresponding real voiceprint segment features. This results in a larger reconstruction loss for high-weight segments and a smaller loss for low-weight segments, guiding the model to prioritize preserving the voiceprint features of key fault segments while allowing for diverse variations in non-key segments. Through iterative optimization of the adversarial loss and weighted reconstruction loss, a synthetic voiceprint sample that accurately preserves fault features while possessing diversity is ultimately obtained.
[0141] In one embodiment, after inputting the fault weights of multiple voiceprint segments in each voiceprint sequence and multiple voiceprint segments labeled with fault types into a pre-trained conditional GAN (Generative Adversarial Network) model to generate a new voiceprint sequence, the method further includes:
[0142] The new voiceprint sequence is converted into a time-spectrum image and input into a convolutional neural network. The convolutional neural network is then trained in a supervised manner to obtain a trained convolutional neural network.
[0143] The acquired acoustic signature sequences are converted into time-spectrum maps and used as input to the model. Supervised training is then performed on the convolutional neural network (CNN) to allow it to learn feature patterns of normal and various fault states, resulting in a well-trained CNN. During actual operation, the real-time acquired acoustic signature sequences are converted back into time-spectrum maps and input into the trained CNN. The output determines whether a fault has occurred and its type or severity, enabling automated, real-time detection and early warning of power equipment faults.
[0144] This application calculates the deviation coefficient based on the difference between the amplitude of the sampling points in each voiceprint segment and the preset amplitude to assess the abnormality of the signal segment. Each voiceprint segment is transformed into a frequency curve using Fourier transform. The clustering coefficient of each maximum is determined based on the distance between the frequency value corresponding to each maximum and other frequency values, as well as the amplitude corresponding to other frequency values. The similarity between frequency curves is calculated based on the difference in the clustering coefficients of corresponding maximums between different frequency curves. Furthermore, by combining the deviation coefficients of each voiceprint segment in the same sequence and the similarity between the frequency curve corresponding to that voiceprint segment and other frequency curves in the sequence, the temporal degradation coefficient is determined to reflect the temporal evolution of fault characteristics. Finally, by combining the deviation coefficients of the voiceprint segment, the temporal degradation coefficient, and the deviation coefficient of the target voiceprint segment in the sequence, the fault weight of that voiceprint segment is determined. The fault weights of multiple voiceprint segments in each voiceprint sequence, along with multiple voiceprint segments labeled with fault types, are input into a pre-trained conditional GAN model to generate a new voiceprint sequence. The generated new voiceprint sequence will be used for subsequent training of the convolutional neural network. When comparing different samples, the similarity between voiceprint segments can be dynamically measured according to the fault development stage, thereby more accurately assessing the fault weight of each voiceprint segment and improving the quality and usability of the generated samples.
[0145] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.
[0146] This application also provides a power equipment fault detection system based on sound feature analysis, such as... Figure 2 As shown, the system includes:
[0147] The acquisition module 21 is used to acquire multiple acoustic fingerprint sequences of faulty power equipment and divide each acoustic fingerprint sequence into multiple acoustic fingerprint segments;
[0148] The first determining module 22 is used to determine the deviation coefficient of the voiceprint segment based on the difference between the amplitude of multiple sampling points in each voiceprint segment and the preset amplitude.
[0149] The second determining module 23 is used to determine the clustering coefficient of each maximum value based on the distance between the frequency value corresponding to each maximum value in each frequency curve and other frequency values, and the amplitude corresponding to other frequency values, and to determine the similarity between each frequency curve and each other frequency curve based on the difference between the clustering coefficient of each maximum value in each frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve; each frequency curve is obtained by performing a Fourier transform on each voiceprint segment;
[0150] The third determining module 24 is used to determine the temporal degradation coefficient of the voiceprint segment based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs and the similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence.
[0151] The fourth determining module 25 is used to determine the fault weight of a voiceprint segment based on the deviation coefficient of each voiceprint segment, the temporal degradation coefficient of the voiceprint segment, and multiple deviation coefficients of multiple target voiceprint segments; the target voiceprint segment indicates the voiceprint segment with the largest deviation coefficient among multiple voiceprint segments in the corresponding voiceprint sequence.
[0152] The generation module 26 is used to input the fault weights of multiple voiceprint segments in each voiceprint sequence and multiple voiceprint segments with fault type labels into a pre-trained conditional GAN model to generate a new voiceprint sequence; the new voiceprint sequence is used to train a convolutional neural network.
[0153] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs.
[0154] Figure 3 This is a schematic diagram of the structure of an electronic device according to an example embodiment of this application. The electronic device includes a memory, a processor, and a computer program stored in the memory and used to run on the processor. When the processor executes the computer program, it implements the method described in any of the above embodiments. Figure 3The electronic device 30 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.
[0155] like Figure 3 As shown, the electronic device 30 can be manifested as a general-purpose computing device, such as a server device. The components of the electronic device 30 may include, but are not limited to: at least one processor 31, at least one memory 32, and a bus 33 connecting different system components (including memory 32 and processor 31).
[0156] Bus 33 includes a data bus, an address bus, and a control bus.
[0157] The memory 32 may include volatile memory, such as random access memory (RAM) 321 and / or cache memory 322, and may further include read-only memory (ROM) 323.
[0158] The memory 32 may also include a program tool 325 (or utility) having a set (at least one) program module 324, such program module 324 including but not limited to: an operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.
[0159] The processor 31 executes various functional applications and data processing, such as the methods provided in any of the above embodiments, by running computer programs stored in the memory 32.
[0160] Electronic device 30 can also communicate with one or more external devices 34 (e.g., keyboard, pointing device, etc.). This communication can be performed via input / output (I / O) interface 35. Furthermore, electronic device 30 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public network, such as the Internet) via network adapter 36. As shown, network adapter 36 communicates with other modules of electronic device 30 via bus 33. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with electronic device 30, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.
[0161] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.
[0162] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method provided in any of the above embodiments.
[0163] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.
[0164] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0165] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the method described in any of the above embodiments.
[0166] The program code for executing the computer program product of this application can be written in any combination of one or more programming languages. The program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.
[0167] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0168] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application.
[0169] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
Claims
1. A power equipment failure detection method based on sound feature analysis, characterized by, The method includes: Acquire multiple acoustic fingerprint sequences of faulty power equipment and divide each acoustic fingerprint sequence into multiple acoustic fingerprint segments; The deviation coefficient of the voiceprint segment is determined based on the difference between the amplitude of multiple sampling points in each voiceprint segment and the preset amplitude. Based on the distance between the frequency value corresponding to each maximum value and other frequency values in each frequency curve, and the amplitude corresponding to other frequency values, the clustering coefficient of each maximum value is determined. Based on the difference between the clustering coefficient of each maximum value in each frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve, the similarity between each frequency curve and each other frequency curve is determined. Each frequency curve is obtained by Fourier transforming each voiceprint segment. Based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence, the temporal degradation coefficient of the voiceprint segment is determined, including: Calculate the similarity between the frequency curve corresponding to each voiceprint segment and each other frequency curve in the same voiceprint sequence to obtain the first similarity score; Calculate the similarity between the other frequency curve and the frequency curve corresponding to the voiceprint segment to obtain the second similarity. The smaller value between the first similarity and the second similarity is taken as the comprehensive similarity between the frequency curve corresponding to the voiceprint segment and the other frequency curves. Based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the comprehensive similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence, the temporal degradation coefficient of the voiceprint segment is determined, including: The stability coefficient of the voiceprint sequence to which the voiceprint segment belongs is determined based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the comprehensive similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence. The deviation value between the voiceprint segment and the preceding voiceprint segment is determined based on the deviation coefficient of each voiceprint segment and the deviation coefficient of each preceding voiceprint segment in the same voiceprint sequence. The degradation coefficient of a voiceprint segment is determined based on the stability coefficient of the voiceprint sequence to which each voiceprint segment belongs and the deviation value between the voiceprint segment and each preceding voiceprint segment. The temporal degradation coefficient of a voiceprint segment is determined based on the degradation coefficient of each preceding voiceprint segment in the voiceprint sequence to which the voiceprint segment belongs and the stability coefficient of the voiceprint sequence to which the voiceprint segment belongs. Based on the deviation coefficient of each voiceprint segment, the temporal degradation coefficient of that voiceprint segment, and multiple deviation coefficients of multiple target voiceprint segments, the fault weight of that voiceprint segment is determined, including: The fault difference coefficient of a voiceprint segment is determined based on the difference between the deviation coefficient of each voiceprint segment and the deviation coefficient of each target voiceprint segment, and the difference between the timing degradation coefficient of the voiceprint segment and the timing degradation coefficient of the target voiceprint segment. The fault difference coefficient of each voiceprint segment is normalized, and the difference between the preset value and the normalized fault difference coefficient is used as the fault weight of the voiceprint segment; the target voiceprint segment is the voiceprint segment with the largest deviation coefficient among multiple voiceprint segments in the corresponding voiceprint sequence. The fault weights of multiple voiceprint segments in each voiceprint sequence and multiple voiceprint segments with fault type labels are input into the trained conditional GAN model to generate new voiceprint sequences for training convolutional neural networks.
2. The power equipment failure detection method based on sound feature analysis as claimed in claim 1, wherein, The step of determining the deviation coefficient of a voiceprint segment based on the difference between the amplitude of multiple sampling points in each voiceprint segment and a preset amplitude includes: The pulse weight of each voiceprint segment is determined based on the difference between the amplitude of multiple sampling points in each voiceprint segment and the preset amplitude, the number of multiple sampling points, and the amplitude of multiple sampling points. The deviation coefficient of a voiceprint segment is determined based on the pulse weight of each voiceprint segment, the amplitude of multiple sampling points in that voiceprint segment, and the preset amplitude.
3. The power equipment failure detection method based on sound feature analysis as claimed in claim 2, wherein, The step of determining the deviation coefficient of a voiceprint segment based on the pulse weight of each voiceprint segment, the amplitude of multiple sampling points in the voiceprint segment, and a preset amplitude includes: Calculate the absolute difference between the mean amplitude of multiple sampling points in each voiceprint segment and the preset amplitude to obtain the amplitude difference of each voiceprint segment; The deviation coefficient of a voiceprint segment is determined based on the amplitude difference of each voiceprint segment and the pulse weight of that voiceprint segment.
4. The power equipment fault detection method based on sound feature analysis as described in claim 1, characterized in that, The determination of the clustering coefficient for each maximum value based on the distance between the frequency value corresponding to each maximum value and other frequency values, and the amplitude corresponding to the other frequency values, includes: Perform a Fourier transform on each voiceprint segment to obtain the frequency curve corresponding to each voiceprint segment; where the horizontal axis of the frequency curve is the frequency value and the vertical axis is the amplitude. The aggregation coefficient of each maximum value is determined based on the number of multiple frequency values in each frequency curve, the distance between the frequency value corresponding to each maximum value and each other frequency value, and the amplitude of the other frequency value.
5. The power equipment fault detection method based on sound feature analysis as described in claim 4, characterized in that, The step of determining the similarity between each frequency curve and each other frequency curve based on the difference between the clustering coefficient of each maximum value in each frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve includes: The similarity between a frequency curve and other frequency curves is determined by the larger of the difference between the clustering coefficient of each maximum value in each frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve, and the larger of the clustering coefficient of each maximum value in the frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve.
6. The power equipment fault detection method based on sound feature analysis as described in claim 1, characterized in that, After inputting the fault weights of multiple voiceprint segments in each voiceprint sequence and the multiple voiceprint segments labeled with fault types into the trained conditional GAN model to generate a new voiceprint sequence, the process further includes: The new voiceprint sequence is converted into a time-spectrum image and input into a convolutional neural network. The convolutional neural network is then trained in a supervised manner to obtain a trained convolutional neural network.
7. A power equipment fault detection system based on sound feature analysis, characterized in that, The system includes: The acquisition module is used to acquire multiple acoustic fingerprint sequences of faulty power equipment and divide each acoustic fingerprint sequence into multiple acoustic fingerprint segments. The first determining module is used to determine the deviation coefficient of the voiceprint segment based on the difference between the amplitude of multiple sampling points in each voiceprint segment and the preset amplitude. The second determining module is used to determine the clustering coefficient of each maximum value based on the distance between the frequency value corresponding to each maximum value in each frequency curve and other frequency values, and the amplitude corresponding to other frequency values. Based on the difference between the clustering coefficient of each maximum value in each frequency curve and the clustering coefficient of the corresponding maximum value in each other frequency curve, the similarity between each frequency curve and each other frequency curve is determined. Each frequency curve is obtained by Fourier transforming each voiceprint segment. The third determining module is used to determine the temporal degradation coefficient of a voiceprint segment based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence, including: Calculate the similarity between the frequency curve corresponding to each voiceprint segment and each other frequency curve in the same voiceprint sequence to obtain the first similarity score; Calculate the similarity between the other frequency curve and the frequency curve corresponding to the voiceprint segment to obtain the second similarity. The smaller value between the first similarity and the second similarity is taken as the comprehensive similarity between the frequency curve corresponding to the voiceprint segment and the other frequency curves. Based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the comprehensive similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence, the temporal degradation coefficient of the voiceprint segment is determined, including: The stability coefficient of the voiceprint sequence to which the voiceprint segment belongs is determined based on the deviation coefficients of multiple voiceprint segments in the voiceprint sequence to which each voiceprint segment belongs, and the comprehensive similarity between the frequency curve corresponding to the voiceprint segment and each other frequency curve in the same voiceprint sequence. The deviation value between the voiceprint segment and the preceding voiceprint segment is determined based on the deviation coefficient of each voiceprint segment and the deviation coefficient of each preceding voiceprint segment in the same voiceprint sequence. The degradation coefficient of a voiceprint segment is determined based on the stability coefficient of the voiceprint sequence to which each voiceprint segment belongs and the deviation value between the voiceprint segment and each preceding voiceprint segment. The temporal degradation coefficient of a voiceprint segment is determined based on the degradation coefficient of each preceding voiceprint segment in the voiceprint sequence to which the voiceprint segment belongs and the stability coefficient of the voiceprint sequence to which the voiceprint segment belongs. The fourth determination module is used to determine the fault weight of a voiceprint segment based on the deviation coefficient of each voiceprint segment, the timing degradation coefficient of the voiceprint segment, and multiple deviation coefficients of multiple target voiceprint segments, including: The fault difference coefficient of a voiceprint segment is determined based on the difference between the deviation coefficient of each voiceprint segment and the deviation coefficient of each target voiceprint segment, and the difference between the timing degradation coefficient of the voiceprint segment and the timing degradation coefficient of the target voiceprint segment. The fault difference coefficient of each voiceprint segment is normalized, and the difference between the preset value and the normalized fault difference coefficient is used as the fault weight of the voiceprint segment; the target voiceprint segment is the voiceprint segment with the largest deviation coefficient among multiple voiceprint segments in the corresponding voiceprint sequence. The generation module is used to input the fault weights of multiple voiceprint segments in each voiceprint sequence and multiple voiceprint segments with fault type labels into the trained conditional GAN model to generate new voiceprint sequences for training convolutional neural networks.