Power equipment voiceprint sample weighting expansion method and fault detection method

Abstract

The invention relates to the technical field of sample expansion, and provides a power equipment voiceprint sample weighted expansion method and a fault detection method, and the method comprises the steps: carrying out the clustering of equipment voiceprint defect data in each defect sample group, obtaining a plurality of clusters, selecting a plurality of data points as the centroid based on the shortest distance between the data points in the clusters, and carrying out the clustering of the data points, taking all centroids selected from one defect sample group as a centroid set; for each centroid set, selecting voiceprint defect data, forming a voiceprint sample sequence, calculating an amplitude spectrum for each sample in the voiceprint sample sequence, extracting a peak sequence from the amplitude spectrum according to a peak condition, performing weighting processing on the peak sequence to obtain an enhanced peak sequence, and then performing data processing on the enhanced peak sequence of each sample to form a voiceprint sample sequence; summing according to the sample weight to obtain a mixed peak sequence, and calculating to obtain a new sample based on the mixed peak sequence. Data homogenization is avoided, large-scale computing power support is not needed, and the development cost is effectively reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of sample expansion, and in particular relates to a method for weighted expansion of voiceprint samples of electric power equipment and a fault detection method. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the development of technologies such as the Internet of Things, big data, and artificial intelligence, intelligent operation and inspection technologies have enabled real-time monitoring, status assessment, and fault diagnosis of power grid equipment, reducing the operating costs and fault handling time of power companies. However, the current monitoring technology for the internal operation sound and vibration of power grid equipment is not well developed and the training of the model requires a large number of samples to improve the recognition accuracy. Therefore, the corresponding sample expansion method also needs to be improved.

[0004] At present, abnormal defects in on-site power equipment are low-probability events, and the occurrence of equipment defects is somewhat random, resulting in insufficient data collection and uneven types of voiceprint samples. The difficulty of sample collection restricts the improvement of model algorithm capabilities. The commonly used expansion technology is to convert audio signals into two-dimensional image signals after feature analysis and extraction, and generate images that approximate real data through continuous model learning. Although it solves the problem of data expansion difficulty to a certain extent, it still has the following problems:

[0005] (1) The current data generation has an unfixed degree of freedom and a high degree of homogeneity. The generated defect features are unclear, which seriously affects the accuracy of the model. There are two difficulties: low fault detection rate and high fault false detection rate.

[0006] (2) Current technologies have high requirements for the quality of raw data and the generated models themselves. However, due to the complex and changeable equipment environment and differences in data distribution, existing models are difficult to migrate and reuse, resulting in the need to continuously develop similar models, which consumes a lot of manpower, computing power and other resources.

[0007] (3) Current technology uses deep learning algorithms, which make data collection difficult, development costs high, and model training and generation time-consuming. Summary of the invention

[0008] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a weighted expansion method for voiceprint samples of power equipment and a fault detection method. Through similar feature clustering and the shortest distance between data points, samples with a higher degree of differentiation are selected as expansion objects to avoid data homogeneity. Through multi-sample weighted expansion, the generation of transformer on-site voiceprint data in more reasonable scenarios can be achieved without the need for large-scale computing power support, effectively reducing development costs.

[0009] In order to achieve the above object, the present invention adopts the following technical solution:

[0010] A first aspect of the present invention provides a method for weighted expansion of voiceprint samples of electric power equipment, comprising:

[0011] Acquire voiceprint defect data of several devices, classify them by defect type, and obtain multiple defect sample groups;

[0012] For each defect sample group, cluster the device voiceprint defect data in the group to obtain multiple clusters, and for each cluster, select several data points as centroids based on the shortest distance between data points in the cluster, and then take all the centroids selected in a defect sample group as a centroid set;

[0013] For each centroid set, voiceprint defect data from different measuring points of the same device or voiceprint defect data from different devices are selected to form a voiceprint sample sequence, and the amplitude spectrum is calculated for each sample in the voiceprint sample sequence. The peak sequence is extracted from the amplitude spectrum according to the peak condition, and the peak sequence is weighted to obtain the enhanced peak sequence. The enhanced peak sequences of each sample are summed according to the sample weight to obtain a mixed peak sequence, and a new sample is calculated based on the mixed peak sequence.

[0014] Furthermore, the step of selecting the centroid includes:

[0015] (1) For each cluster, randomly select a data point as a centroid and include it in the centroid library;

[0016] (2) For each data point in the cluster that has not yet been selected, calculate the shortest distance to the most recently selected centroid;

[0017] (3) for each data point in the cluster that has not been selected, based on the shortest distance, select a data point as a centroid to be included in the centroid library;

[0018] (4) For each cluster, repeat steps (2) and (3) until the number of centroids in the centroid library reaches the target value;

[0019] (5) For each defect sample group, the centroid libraries of all corresponding clusters are merged to obtain the centroid set.

[0020] Furthermore, the probability of the data point being selected is proportional to the shortest distance.

[0021] Furthermore, the weighted processing of the peak sequence adopts random weighted processing;

[0022] Alternatively, the sample weight is determined according to the distance between the sample source sensor in the enhanced peak sequence and the sensor biased by the new sample simulation.

[0023] Furthermore, the amplitude spectrum is calculated using a fast Fourier transform algorithm;

[0024] The new samples are calculated using an inverse fast Fourier transform algorithm.

[0025] A second aspect of the present invention provides a fault detection method, comprising:

[0026] Obtain the voiceprint data of the device to be detected;

[0027] Based on the voiceprint data of the device to be detected, the trained fault detection model is used to perform fault detection on the device to be detected;

[0028] Among them, the training steps of the fault detection model include: using the weighted expansion method of the power equipment voiceprint samples described in the first aspect to expand the training set; using the expanded training set to train the fault detection model.

[0029] The third aspect of the present invention provides a power equipment voiceprint sample weighted expansion system, which includes:

[0030] A classification module is configured to: obtain a number of device voiceprint defect data, and classify them according to defect types to obtain multiple defect sample groups;

[0031] A clustering module is configured to: for each defect sample group, cluster the device voiceprint defect data in the group to obtain multiple clusters, and for each cluster, select a number of data points as centroids based on the shortest distance between data points in the cluster, and then use all the centroids selected in a defect sample group as a centroid set;

[0032] The expansion module is configured as follows: for each centroid set, voiceprint defect data from different measuring points of the same device or voiceprint defect data from different devices are selected to form a voiceprint sample sequence, and the amplitude spectrum is calculated for each sample in the voiceprint sample sequence, a peak sequence is extracted from the amplitude spectrum according to the peak condition, and the peak sequence is weighted to obtain an enhanced peak sequence, and the enhanced peak sequences of each sample are summed according to the sample weight to obtain a mixed peak sequence, and a new sample is calculated based on the mixed peak sequence.

[0033] A fourth aspect of the present invention provides a fault detection system, comprising:

[0034] A data acquisition module, which is configured to: acquire voiceprint data of a device to be detected;

[0035] The fault detection module is configured to: perform fault detection on the device to be detected based on the voiceprint data of the device to be detected and using a trained fault detection model;

[0036] Among them, the training steps of the fault detection model include: using the power equipment voiceprint sample weighted expansion system described in the fourth aspect to expand the training set; using the expanded training set to train the fault detection model.

[0037] A fifth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in a method for weighted expansion of voiceprint samples of electric power equipment or a method for fault detection as described above.

[0038] The sixth aspect of the present invention provides a computer device, including a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor, wherein when the processor executes the program, the steps in a method for weighted expansion of voiceprint samples of electric power equipment or a fault detection method as described above are implemented.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention selects samples with a higher degree of differentiation as expansion objects through similar feature clustering and the shortest distance between data points to avoid data homogeneity.

[0041] The present invention targets at the scene that the voiceprint has the characteristics of non-single sound source and non-point sound source, and the sound signal collected by the sensor is actually the superposition of different sound sources after direct radiation and reflection. Through weighted expansion of too many samples, the transformer on-site voiceprint data generation in more reasonable scenarios can be realized, and it can be used in various equipment defect scenarios.

[0042] The present invention adopts clustering algorithm and weighted expansion technology, so that the algorithm does not require professional developers and large-scale computing power support, effectively reducing development costs.

[0043] The present invention specifies a weight sequence through multi-sample weighted expansion, so that the generated sample data can be controlled. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0045] Figure 1 It is a flow chart of a method for weighted expansion of voiceprint samples of electric power equipment according to the first embodiment of the present invention;

[0046] Figure 2 is a flow chart of data fusion weighting according to the first embodiment of the present invention;

[0047] Figure 3is a schematic diagram of sample 1 according to embodiment 1 of the present invention;

[0048] Figure 4 is a schematic diagram of sample 2 of embodiment 1 of the present invention;

[0049] Figure 5 This is a diagram showing the effect of sample hybrid weighted directional enhancement expansion according to the first embodiment of the present invention;

[0050] Figure 6 It is a structural diagram of a computer device according to Embodiment 5 of the present invention. DETAILED DESCRIPTION

[0051] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0052] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0053] Terminology explanation:

[0054] Clustering: Clustering is to divide a data set into different classes or clusters according to a specific standard (such as distance), so that the similarity of data objects in the same cluster is as large as possible, and the difference of data objects in different clusters is as large as possible. In other words, after clustering, data of the same class is gathered together as much as possible, and data of different classes are separated as much as possible.

[0055] FFT algorithm: a general term for efficient and fast calculation methods for discrete Fourier transform using computers. It is obtained by improving the discrete Fourier transform algorithm based on the odd, even, imaginary, and real characteristics of discrete Fourier transform. It can greatly reduce the number of multiplications required for computer calculation of discrete Fourier transform, especially the more the number of sample points N to be transformed, the more significant the savings in FFT algorithm calculation. In addition, IFFT is the inverse transform of FFT.

[0056] Spectrogram: Spectrogram is a speech spectrum diagram, which is generally obtained by processing the received time domain signal. The horizontal axis is time, the vertical axis is frequency, and the coordinate point value is the energy of the speech data. Since it uses a two-dimensional plane to express three-dimensional information, the size of the energy value is represented by color. The darker the color, the stronger the speech energy at that point.

[0057] Embodiment 1

[0058] This embodiment provides a method for weighted expansion of voiceprint samples of electric power equipment.

[0059] The present embodiment provides a method for weighted expansion of voiceprint samples of power equipment, which can be applied to products such as remote intelligent patrol systems to solve the problems of small amount and insufficient diversity of voiceprint sample data, improve the accuracy of related defect identification of power equipment under different working conditions, and enhance the engineering practical capabilities of artificial intelligence.

[0060] The present embodiment provides a method for weighted expansion of voiceprint samples of electric power equipment. First, the voiceprint sample data of on-site equipment is fully quantified and collected, and the unordered samples are constructed into a sample library according to the defect type; then a clustering algorithm is applied to each sample, and a centroid selection algorithm is used to select the source samples after clustering; finally, the expanded sample data is obtained by summing and calculating according to certain sample weights through a multi-sample weighted expansion method.

[0061] This embodiment provides a method for weighted expansion of voiceprint samples of electric power equipment, such as Figure 1 As shown, the following steps are included:

[0062] Step 1: Multi-sample representative data selection method.

[0063] By clustering similar features of voiceprint samples of the same category, samples with high differentiation are selected as expansion objects, representative selection of multi-sample data is achieved, and the problem of high data homogeneity in sample expansion is solved.

[0064] Step 101: Sample classification.

[0065] The collected on-site equipment sample data (equipment voiceprint defect data) are classified according to the defect type, such as DC bias magnetism, heavy overload, partial discharge, cooling noise and other types of defects, and the disordered samples are processed into ordered samples. A total of Q sample data are classified into k types of defect sample groups, each containing j samples (regarded as j points).

[0066] Step 102: For each type of defect sample group, cluster the data in the group to obtain m clusters.

[0067] (1) Cluster the samples within each defect sample group according to the clustering algorithm and randomly set m points in the feature space as the initial cluster centers;

[0068] (2) For each other point, calculate the distance to the m cluster centers. For unknown points, select the nearest cluster center point as the label category;

[0069] (3) After marking the categories of all points, recalculate the new center point of each cluster;

[0070] (4) If the new center point is the same as the original center point, end clustering; otherwise, re-execute step (2).

[0071] Step 103: Sample selection.

[0072] (1) In a clustered defect sample group, taking DC bias magnetization as an example, first randomly select a point in a cluster as the first centroid and include it in the centroid library;

[0073] (2) For each point X in the cluster that has not been selected, calculate the shortest distance D(X) from it to the most recently selected centroid;

[0074] (3) Select a new data point as the next centroid and include it in the centroid library. The selection method is that the probability of the data point being selected is proportional to D(X), that is, probability P(x) = D(X) / ΣD(X), and select the point with the largest P(X) value;

[0075] (4) Repeat steps (2) and (3) until q centroids appear in the centroid library.

[0076] In summary, q centroids constitute the sample group of the cluster, and a defective sample group obtains m*q centroids through sample selection to form the centroid set corresponding to the defective sample group. This selection strategy ensures the dispersion between the centroids, improves the representativeness of the samples, and can better support training.

[0077] Step 2: Voiceprint multi-sample data fusion weighting technology.

[0078] Considering that the voiceprint of the transformer site has the characteristics of non-single sound source and non-point sound source, and the sound signal collected by the sensor is actually the superposition of different sound sources after direct and reflected, that is, under the same working conditions, the voiceprint data collected after moving the sensor position may appear as the weighted superposition of voiceprint data at other positions. In order to realize the generation of transformer site voiceprint data in more reasonable scenarios, reference is made to Figure 2 The specific steps of the flow chart based on multi-sample mixed weighted enhancement and expansion of voiceprint spectrogram are as follows:

[0079] Step 201: For each centroid set, select voiceprint defect data of different measuring points of the same transformer or different transformers to obtain n data points to form a voiceprint sample sequence (X1, X2, ..., Xn), and select a reference voiceprint sample in the voiceprint sample sequence based on expert experience. For example, the voiceprint sample sequence includes voiceprint signal 1 and voiceprint signal 2, and voiceprint signal 2 is selected as the reference voiceprint sample.

[0080] Step 202: Calculate the effective value based on the reference voiceprint sample: N is the number of data points in the reference voiceprint sample, x is the reference voiceprint sample, i is the i-th data in the benchmark voiceprint sample.

[0081] Step 203: for each sample X in the voiceprint sample sequence i , the amplitude spectrum and phase spectrum are calculated respectively through the FFT (Fast Fourier Transform) algorithm.

[0082] Step 204: For each sample X i , extract the peak amplitude sequence (peak sequence) according to the set peak conditions.

[0083] The specific peak condition can be adjusted according to actual conditions. For example, the peak value is defined as a value greater than the maximum value within a certain range in the amplitude spectrum to avoid random peaks in the background noise signal from being enhanced and causing signal pollution. Taking the amplitude sequence (amplitude spectrum) of X = [0 1 10 2 1 2 1 2 7 1] as an example, the peak sequence is peaks 10, 2, and 7 corresponding to indexes 2, 5, and 8. The peak value greater than 2 is set as a valid peak, that is, only peaks 10 and 7 at indexes 2 and 8 will be retained, and subsequent random weighted enhancement processing will be performed.

[0084] Step 205: Perform random weighting processing on each peak sequence to obtain an enhanced peak sequence.

[0085] Similarly, according to the example in step 204, the random weighting range is set to 5-0.2. For example, the final generated random sequence is w=[0.7 3]. That is, the peak values ​​10 and 7 at the above indexes 2 and 8 are weighted to 7 and 21 respectively, generating a randomly weighted amplitude spectrum.

[0086] Step 206: The enhanced peak sequence of each sample is summed up according to the sample weight to obtain a mixed peak sequence.

[0087] The sample weight is determined based on the distance between the sample source sensor and the sensor that the new sample is biased to simulate in the enhanced peak sequence. Taking the audio obtained by sensors at two positions of the same transformer as an example, the weights can be 0.5 and 0.5 respectively to simulate the voiceprint features in the middle of the two microphones; or based on 0.2 and 0.8, the voiceprint features of the sensor close to one side can be simulated and the spectrum can be fused. The specific weight can be determined according to the situation and can generally be set to the same weight.

[0088] (6) Perform IFFT (Inverse Fast Fourier Transform) calculation according to the mixed peak sequence, the effective value of the reference voiceprint sample and the phase spectrum of the reference voiceprint sample to obtain the multi-sample mixed enhanced sample (new sample).

[0089] like Figure 5 As shown, for Figure 3 and Figure 4 The effect diagram after multi-sample mixed weighted directional enhancement expansion. Since sample 2 is used as the reference signal and is given a higher weight, the overall expanded sample is Figure 4 The similarity is high, but it can also be seen that the expanded sample has a lot of detailed information of sample 1 within the low frequency band of 200Hz.

[0090] The present embodiment provides a method for weighted expansion of voiceprint samples of electric equipment, which performs weighted fusion of similar voiceprint samples by weighting sample features, realizes the expansion of the main device voiceprint in the case of superposition of non-point sound source reflections, and solves the problem of small amount and insufficient diversity of voiceprint sample data. The present embodiment provides a method for weighted expansion of voiceprint samples of electric equipment, which clusters similar features of voiceprint samples of the same category, uses a centroid selection algorithm to select samples with a high degree of differentiation as expansion objects, realizes representative selection of multi-sample data, and solves the problem of high degree of data homogeneity in sample expansion.

[0091] The method for weighted expansion of voiceprint samples of electric equipment provided in this embodiment has higher technical accuracy: through weighted expansion of multiple samples, a weight sequence is specified so that the generated sample data can be controlled; through clustering of similar features, samples with a higher degree of differentiation are selected as expansion objects to avoid data homogeneity.

[0092] The weighted expansion method for power equipment voiceprint samples provided in this embodiment is more technically versatile: the on-site voiceprint has the characteristics of non-single sound source and non-point sound source, and the sound signal collected by the sensor is actually the superposition of different sound sources after direct radiation and reflection. The weighted expansion technology is used to realize the generation of transformer on-site voiceprint data in more reasonable scenarios, and it can be used in various equipment defect scenarios.

[0093] The technical development cost of a weighted expansion method for voiceprint samples of electric power equipment provided in this embodiment is low: by adopting clustering algorithm and weighted expansion technology, the algorithm does not require professional developers or large-scale computing power support, effectively reducing development costs.

[0094] Embodiment 2

[0095] This embodiment provides a fault detection method, which is characterized by comprising:

[0096] Obtain the voiceprint data of the device to be detected;

[0097] Based on the voiceprint data of the device to be detected, the trained fault detection model is used to perform fault detection on the device to be detected;

[0098] The training step of the fault detection model includes: expanding the training set by using the weighted expansion method of the voiceprint samples of electric power equipment described in the first embodiment; and training the fault detection model by using the expanded training set.

[0099] The training set includes several device voiceprint defect data and several device voiceprint normal data.

[0100] Embodiment 3

[0101] This embodiment provides a power equipment voiceprint sample weighted expansion system, which specifically includes:

[0102] A classification module is configured to: obtain a number of device voiceprint defect data, and classify them according to defect types to obtain multiple defect sample groups;

[0103] A clustering module is configured to: for each defect sample group, cluster the device voiceprint defect data in the group to obtain multiple clusters, and for each cluster, select a number of data points as centroids based on the shortest distance between data points in the cluster, and then use all the centroids selected in a defect sample group as a centroid set;

[0104] The expansion module is configured as follows: for each centroid set, voiceprint defect data from different measuring points of the same device or voiceprint defect data from different devices are selected to form a voiceprint sample sequence, and the amplitude spectrum is calculated for each sample in the voiceprint sample sequence, a peak sequence is extracted from the amplitude spectrum according to the peak condition, and the peak sequence is weighted to obtain an enhanced peak sequence, and the enhanced peak sequences of each sample are summed according to the sample weight to obtain a mixed peak sequence, and a new sample is calculated based on the mixed peak sequence.

[0105] It should be noted here that each module in this embodiment corresponds to each step in Example 1 one by one, and the specific implementation process is the same, which will not be repeated here.

[0106] Embodiment 4

[0107] This embodiment provides a fault detection system, which specifically includes:

[0108] A data acquisition module, which is configured to: acquire voiceprint data of a device to be detected;

[0109] The fault detection module is configured to: perform fault detection on the device to be detected based on the voiceprint data of the device to be detected and using a trained fault detection model;

[0110] The training step of the fault detection model includes: using a power equipment voiceprint sample weighted expansion system described in Example 3 to expand the training set; and using the expanded training set to train the fault detection model.

[0111] It should be noted here that each module in this embodiment corresponds to each step in Example 2 one by one, and the specific implementation process is the same, which will not be repeated here.

[0112] Embodiment 4

[0113] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the method for weighted expansion of voiceprint samples of electric power equipment described in the first embodiment or the method for fault detection described in the third embodiment are implemented.

[0114] Embodiment 5

[0115] This embodiment provides a computer device, such as Figure 6 As shown, it includes a computer-readable storage medium 1003, a processor 1001, a communication interface 1002, and a computer program stored on the computer-readable storage medium 1003 and executable on the processor 1001, wherein the processor 1001, the communication interface 1002, and the computer-readable storage medium 1003 can be connected via a bus or other means. The communication interface 1002 is used to receive and send data, and when the processor 1001 executes the program, the steps in the method for weighted expansion of voiceprint samples of electric power equipment described in the first embodiment or the method for fault detection described in the third embodiment are implemented.

[0116] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0117] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0118] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1The steps for the functions specified in one or more boxes.

[0119] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for weighted expansion of voiceprint samples of electric power equipment, characterized in that: include: Acquire voiceprint defect data of several devices, classify them by defect type, and obtain multiple defect sample groups; For each defect sample group, cluster the device voiceprint defect data in the group to obtain multiple clusters, and for each cluster, select several data points as centroids based on the shortest distance between data points in the cluster, and then take all the centroids selected in a defect sample group as a centroid set; For each centroid set, voiceprint defect data from different measuring points of the same device or voiceprint defect data from different devices are selected to form a voiceprint sample sequence, and the amplitude spectrum is calculated for each sample in the voiceprint sample sequence. The peak sequence is extracted from the amplitude spectrum according to the peak condition, and the peak sequence is weighted to obtain the enhanced peak sequence. The enhanced peak sequences of each sample are summed according to the sample weight to obtain a mixed peak sequence, and a new sample is calculated based on the mixed peak sequence.

2. A method for weighted expansion of voiceprint samples of electric power equipment as claimed in claim 1, characterized in that: The step of selecting the centroid includes: (1) For each cluster, randomly select a data point as a centroid and include it in the centroid library; (2) For each data point in the cluster that has not yet been selected, calculate the shortest distance to the most recently selected centroid; (3) for each data point in the cluster that has not been selected, based on the shortest distance, select a data point as a centroid to be included in the centroid library; (4) For each cluster, repeat steps (2) and (3) until the number of centroids in the centroid library reaches the target value; (5) For each defect sample group, the centroid libraries of all corresponding clusters are merged to obtain the centroid set.

3. A method for weighted expansion of voiceprint samples of electric power equipment as claimed in claim 2, characterized in that: The probability of the data point being selected is proportional to the shortest distance.

4. A method for weighted expansion of voiceprint samples of electric power equipment as claimed in claim 1, characterized in that: The weighted processing of the peak sequence adopts random weighted processing; Alternatively, the sample weight is determined according to the distance between the sample source sensor in the enhanced peak sequence and the sensor biased by the new sample simulation.

5. A method for weighted expansion of voiceprint samples of electric power equipment as claimed in claim 1, characterized in that: The amplitude spectrum is calculated using a fast Fourier transform algorithm; The new samples are calculated using an inverse fast Fourier transform algorithm.

6. A fault detection method, characterized in that: include: Obtain the voiceprint data of the device to be detected; Based on the voiceprint data of the device to be detected, the trained fault detection model is used to perform fault detection on the device to be detected; The training step of the fault detection model includes: expanding the training set by using the weighted expansion method of voiceprint samples of electric power equipment described in any one of claims 1 to 5; and training the fault detection model by using the expanded training set.

7. A power equipment voiceprint sample weighted expansion system, characterized in that: include: A classification module is configured to: obtain a number of device voiceprint defect data, and classify them according to defect types to obtain multiple defect sample groups; A clustering module is configured to: for each defect sample group, cluster the device voiceprint defect data in the group to obtain multiple clusters, and for each cluster, select a number of data points as centroids based on the shortest distance between data points in the cluster, and then use all the centroids selected in a defect sample group as a centroid set; The expansion module is configured as follows: for each centroid set, voiceprint defect data from different measuring points of the same device or voiceprint defect data from different devices are selected to form a voiceprint sample sequence, and the amplitude spectrum is calculated for each sample in the voiceprint sample sequence, a peak sequence is extracted from the amplitude spectrum according to the peak condition, and the peak sequence is weighted to obtain an enhanced peak sequence, and the enhanced peak sequences of each sample are summed according to the sample weight to obtain a mixed peak sequence, and a new sample is calculated based on the mixed peak sequence.

8. A fault detection system, characterized in that: include: A data acquisition module, which is configured to: acquire voiceprint data of a device to be detected; The fault detection module is configured to: perform fault detection on the device to be detected based on the voiceprint data of the device to be detected and using a trained fault detection model; The training step of the fault detection model includes: expanding the training set using the power equipment voiceprint sample weighted expansion system described in claim 7; and training the fault detection model using the expanded training set.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the method for weighted expansion of voiceprint samples of electric power equipment described in any one of claims 1 to 5 or the method for fault detection described in claim 6 are implemented.

10. A computer device comprising a computer-readable storage medium, a processor, and a computer program stored in the computer-readable storage medium and executable on the processor, characterized in that: When the processor executes the program, the steps in the method for weighted expansion of voiceprint samples of electric power equipment described in any one of claims 1 to 5 or the method for fault detection described in claim 6 are implemented.

Images