Power transformer fault voiceprint detection method based on extreme learning machine
Through the power transformer fault soundprint detection method based on the limit learning machine, the improved Mel filter and limit learning machine model is used to solve the problem of insufficient accuracy of the power transformer fault soundprint information extraction, and achieve fast and accurate fault detection and safe equipment status monitoring.
Patent Information
- Application Number
- CN202410918958.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-10
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-07-10
AI Technical Summary
The prior art is difficult to quickly and accurately extract the soundprint information of power transformer failures, resulting in insufficient accuracy of equipment status monitoring and fault warning, and there is a personal safety threat.
The power transformer fault soundprint detection method based on the limit learning machine is adopted, and the soundprint features are extracted through audio signal preprocessing and improved Mel filters, and the limit learning machine model is constructed for fault soundprint detection.
It realizes fast and accurate fault soundprint detection, reduces system complexity, improves detection accuracy and speed, and ensures the safety of staff.
Smart Images

Figure CN119007748A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method for detecting a fault voiceprint of a power transformer, and in particular to a method for detecting a fault voiceprint of a power transformer based on an extreme learning machine. Background Art
[0002] As a key equipment in the power system, the trouble-free operation of power transformers plays a vital role in reliable power supply. When a transformer fails, it may cause power outages in some areas, or even affect the safe operation of the power system, causing serious economic losses. With the expansion of the scale of power grids and the development of intelligence, the status monitoring and fault warning of substation equipment have become important means to ensure the stable operation of power systems. Substation staff can judge the operating status of the transformer by listening to the operating sound of the transformer during on-site inspections, but approaching the transformer when the transformer makes abnormal sounds may threaten the personal safety of the staff. Using modern digital signal processing technology to diagnose transformer voiceprint faults can not only improve the accuracy of transformer voiceprint recognition, but also ensure the personal safety of staff.
[0003] During the operation of substation equipment, the voiceprint information will change with the change of the working status of the equipment. Therefore, by monitoring and analyzing the voiceprint of the transformer, it is possible to preliminarily determine whether the operating status of the transformer is normal, timely discover potential fault hazards, and provide a basis for the maintenance and overhaul of the transformer. Based on the voiceprint feature information of the transformer in different operating states, a mathematical model can be established to describe the normal operating state and fault state of the equipment. For example, BP neural network, naive Bayes, SVM, CNN and other algorithms can be used to realize transformer fault detection.
[0004] However, the audio signal of the substation has the characteristics of non-steady state and large fluctuation, because the operating status of the power equipment and environmental factors will affect the audio signal, resulting in signal instability and volatility. When the operating status of the transformer fluctuates, the extracted feature quantities overlap between normal and fault conditions. In a short period of time, these fluctuations can be regarded as relatively stable signals for processing. Appropriate signal processing technology and algorithms are needed to pre-process the audio signal, extract the feature information reflecting the operating status of the equipment, and realize real-time monitoring of the equipment working status and fault warning. Summary of the invention
[0005] The problem to be solved by the present invention is to provide a power transformer fault voiceprint detection method based on an extreme learning machine, which analyzes the voiceprint information of the substation equipment during operation, quickly and accurately extracts key feature information, and realizes real-time monitoring of the equipment working status and fault warning.
[0006] The present invention adopts the following technical solution: a method for detecting fault voiceprint of a power transformer based on an extreme learning machine, comprising the following steps:
[0007] S1. Audio data preprocessing: Collect the audio signal of the transformer fault, divide the audio signal into frames using the overlapping frame method, preprocess each frame of data after framing through fast Fourier transform, convert the time domain signal into the frequency domain, and obtain the spectrum data;
[0008] S2. Voiceprint feature extraction: The voiceprint feature parameters of the audio signal in the training set are extracted through the improved Mel filter to construct the training set;
[0009] S3. Build a fault voiceprint detection model based on extreme learning machine and perform sample training to obtain a trained fault voiceprint detection model;
[0010] S4, establishing a test set, selecting three consecutive frames from the middle position of each audio signal for voiceprint feature evaluation, obtaining real-time test samples, and forming a test set;
[0011] S5. Perform fault detection, input the real-time test sample into the trained fault voiceprint detection model, perform test recognition, and obtain the voiceprint recognition result of the transformer fault.
[0012] Specifically, in step S1, the audio signal of the transformer fault has a sampling frequency of 48kHz and a length of 10s;
[0013] The overlapping frame method is used to divide the audio signal into frames. The frame length of each frame is 8192 points, the frame shift is 4096 points, and each audio signal is divided into 116 frames. The data x of the i-th frame of the audio signal i Expressed as: x i =[x i1 ,x i2 ,…,x i8192 ], 1≤i≤116;
[0014] For each frame of data x i Perform fast Fourier transform, remove negative frequencies, retain positive frequencies, and obtain spectrum data y i , expressed as: y i =[y i1 ,y i2 ,…,y i4096 ].
[0015] Specifically, in step S2, the Mel filter is improved by replacing the triangular filter group in the traditional Mel filter with a rectangular filter group to reduce the impact of the amplitude modulation extracted by filtering on the voiceprint features. According to the principle of low-frequency dense and high-frequency sparse, the number of low-frequency filters is increased to improve the spectral resolution and reduce the complexity.
[0016] Further, in step S2, based on the spectrum data obtained in step S1, the energy spectrum of the audio signal in the training set with frequency domains of 0-1200 Hz and 19000-20400 Hz is extracted through the improved Mel filter, and the output of the i-th frame of the audio signal is z i , the transformation formula is as follows:
[0017]
[0018] Among them, the output of the improved Mel filter is a 1×M dimensional vector, z i =[z i1 ,z i2 ,…,z iM ].
[0019] Specifically, in step S3, the extreme learning machine is a training algorithm based on a single hidden layer feedforward neural network, including an input layer, a hidden layer and an output layer; the extreme learning machine determines the weight and bias of the input layer of the single hidden layer feedforward neural network by random selection and assigns values;
[0020] Construct a fault voiceprint detection model based on extreme learning machine. For N different audio signal samples (x j ,t j ), x j =[x j1 ,x j2 ,…,x jn ] T ,t j =[t j1 ,t j2 ,…t jm ] T , t j The label of the audio is represented by a single hidden layer feed-forward neural network with K hidden nodes, expressed as:
[0021]
[0022] Where g(x) is the activation function, j=1,…,N,w i is the weight from the input layer to the hidden layer, w i =[w i1 ,w i2 ,…,w in ] T represents the bias connecting the ith hidden node, w i ·x j w i and x j The inner product of i is the bias of the hidden layer, β j Represents the weight from the hidden layer to the output layer of the jth sample;
[0023] The single hidden layer feedforward neural network infinitely approximates N samples, that is, there is β i ,ω i 、b i So that:
[0024]
[0025] The K equations are expressed as:
[0026] Hβ=T
[0027] in:
[0028]
[0029] Where H is the output matrix of the hidden layer of the single hidden layer feedforward neural network, the i-th column of H is the output of the i-th node of the hidden layer of the neural network, and β is the weight from the hidden layer to the output layer.
[0030] Furthermore, a fault voiceprint detection model based on extreme learning machine is constructed. The input and output dimensions of the extreme learning machine are set according to the extracted feature dimension and label dimension. The label is encoded using one-hot encoding, and the normal sound is t normal =[1,0] T , the fault sound is t abnormal =[0,1] T , set the number of hidden nodes of the extreme learning machine to 1200, and select the activation function as Sigmoid, which is expressed as follows:
[0031]
[0032] Among them, e -x It represents the reciprocal of the exponential function with the natural base e as the base. The model is trained through the training set, and the trained fault voiceprint detection model is saved for real-time detection of power transformer fault voiceprints.
[0033] Specifically, in step S4, a test set is established, and three consecutive frames are selected from the middle position of each audio signal and numerically averaged using the following formula to obtain a real-time test sample P j :
[0034]
[0035] Among them, p j_58 ,p j_59 ,p j_60 It is the feature-extracted data of the 58th to 60th frames of the jth audio signal.
[0036] Specifically, in step S5, fault detection is performed, and the test sample P jInput the trained fault voiceprint detection model and get the j-th sample detection output t j :
[0037] t j =[t j1 ,t j2 ] T
[0038] If t j1 =[1,0] T , the test result is normal; t j2 =[0,1] T , the test result is abnormal.
[0039] On the other hand, the technical solution of the present invention further provides: an electronic device, comprising:
[0040] one or more processors;
[0041] a storage device having one or more programs stored thereon;
[0042] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the above-mentioned methods for detecting fault voiceprints of power transformers based on extreme learning machines.
[0043] The technical solution of the present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in any of the above-mentioned methods for detecting fault voiceprints of power transformers based on extreme learning machines are implemented.
[0044] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:
[0045] 1. The power transformer fault voiceprint detection method of the present invention analyzes the voiceprint information of the substation during operation, and uses the ELM algorithm to detect the transformer fault voiceprint. The ELM algorithm randomly selects the input weights of the SLFNs. During the execution of the ELM algorithm, no manual adjustment of parameters is required, which avoids the repeated iteration process of the traditional training algorithm, converges quickly, greatly reduces the training time, improves the detection accuracy and speed, and is easy to implement in engineering.
[0046] 2. The power transformer fault voiceprint detection method of the present invention can improve the spectral resolution of transformer voiceprint feature extraction through an improved Mel filter, reduce system complexity, and effectively improve the fault detection accuracy of the ELM algorithm. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 This is a flow chart of the steps of the power transformer fault voiceprint detection method of the present invention;
[0048] Figure 2 A curve diagram showing the corresponding relationship between the Mel frequency and the actual frequency in an embodiment of the present invention;
[0049] Figure 3 The Mel filter in the embodiment of the present invention is composed of;
[0050] Figure 4 The frequency domain diagram of the soundprint under four different operating conditions of the transformer in the embodiment of the present invention;
[0051] Figure 5 This is a time-frequency comparison diagram of voiceprint feature extraction before and after the improvement of the present invention;
[0052] Figure 6 This is a structural diagram of the SLFN of the present invention. DETAILED DESCRIPTION
[0053] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the application is further elaborated in detail below in conjunction with the accompanying drawings. The described embodiments are only a part of the embodiments involved in the present invention. All non-innovative embodiments of other researchers in the field on this embodiment belong to the protection scope of the present invention. At the same time, for the step numbering in the embodiment of the present invention, it is only set for the convenience of explanation, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.
[0054] The present invention proposes a fault voiceprint detection method for power transformer based on extreme learning machine (ELM). By analyzing the voiceprint information of substation equipment during operation, real-time monitoring of equipment working status and fault warning are achieved. Figure 1 As shown, the following steps are included:
[0055] S1. Audio data preprocessing: Collect the audio signal of the transformer fault, build a training set, use the overlapping frame method to frame the audio signal of the training set, preprocess each frame of data after framing through fast Fourier transform, convert the time domain signal to the frequency domain, and obtain the spectrum data;
[0056] S2. Voiceprint feature extraction: The voiceprint feature parameters of the audio signal in the training set are extracted through the improved Mel filter to construct the training set;
[0057] S3. Build a fault voiceprint detection model based on ELM and perform sample training to obtain a trained fault voiceprint detection model;
[0058] S4, establishing a test set, selecting three consecutive frames from the middle position of each audio signal for voiceprint feature evaluation, obtaining real-time test samples, and forming a test set;
[0059] S5. Perform fault detection, input the real-time test sample into the trained fault voiceprint detection model, perform test recognition, and obtain the voiceprint recognition result of the transformer fault.
[0060] In one embodiment of the present invention, an audio signal of a power transformer fault is collected, the sampling frequency is 48kHz, and the length is 10s. The power transformer fault voiceprint detection method based on the extreme learning machine is as follows:
[0061] S1. Audio data preprocessing
[0062] In signal processing methods, framing is a common processing strategy, which divides the speech signal into multiple short-term signal segments, each of which is called a frame. In order to maintain continuity and relevance, it is necessary to ensure that there are overlapping parts between adjacent frames. This method is called overlapping framing.
[0063] In this embodiment, the collected audio signal is divided into multiple short-time signal segments, each of which is called a frame. In order to maintain continuity and relevance, it is necessary to ensure that there is an overlap between adjacent frames. Assuming that the frame length is 8192 points and the frame shift is 4096 points, a certain audio segment can be divided into 116 frames.
[0064] The number of points of Fast Fourier Transform (FFT) is usually consistent with the frame length. When the sampling rate of the audio is constant, the longer the frame length, the higher the frequency resolution, but the amount of FFT calculation will also increase. Taking all factors into consideration, this embodiment selects a frame length of 8192 points and a frame shift of 4096 points. For each audio with a length of 10s and a sampling rate of 48kHz, it can be divided into 116 frames. This not only ensures the frequency resolution, but also controls the amount of calculation within an acceptable range.
[0065] Specifically, the audio frame data is represented as x i =[x i1 ,x i2 ,…,x i8192 ], 1≤i≤116, perform FFT on each frame, remove negative frequencies, retain positive frequencies to obtain spectrum data y i =[y i1 ,y i2 ,…,y i4096 ],1≤i≤116.
[0066] S2. Voiceprint feature extraction
[0067] Commonly used sampling rates of microphones include 44.1kHz, 48kHz, 96kHz, etc. If feature extraction is not performed, the amount of data will be too large, which will bring great inconvenience to the detection and identification work. Therefore, this embodiment uses FFT to convert the time domain signal to the frequency domain, and uses a dimensionality reduction filter to extract feature parameters, which can not only reduce the data dimension, but also improve the category representation ability, thereby improving the performance of the classification algorithm.
[0068] Studies have shown that the human ear is more sensitive to low-frequency signals than high-frequency signals. The design of the Mel filter bank is based on the human ear's perception of different frequencies. Therefore, in the Mel filter bank, the number of filters at low frequencies is denser, and the number of filters at high frequencies is sparser. The specific idea is to map the frequency domain signal to the Mel frequency domain and then divide it into Mel frequency bands of equal width. The Mel filter is equivalent to the human cochlea, simulating the sensitivity of the human ear to low and high frequencies, and reflecting the nonlinear response of the human ear to frequency, such as Figure 2 shown.
[0069] The traditional Mel filter is composed of several non-uniformly distributed triangular filters. The filters at low frequencies are densely distributed, with higher frequency resolution, making it easier to extract the features of the low-frequency part, such as Figure 3 shown.
[0070] The frequency response of the filter is:
[0071]
[0072] Among them, k is the frequency, f(m) is the center frequency of the mth triangular filter, and the energy output by each triangular filter is the extracted feature.
[0073] Mel filters have been widely used in voiceprint recognition and fault diagnosis, and have achieved good results. However, for transformer voiceprints that are stable in the time domain and only distributed in high and low frequencies in the frequency domain, Mel filters cannot accurately extract the required voiceprints.
[0074] The frequency domain diagram of the soundprint under four different operating conditions of the transformer, such as Figure 4 As shown in the figure, considering that the operating frequency of the transformer is mainly concentrated in the multiple of 50Hz and does not exceed 1kHz, the significant changes in these frequencies are usually related to abnormal faults of the transformer. Therefore, it is particularly important to maintain the resolution of this frequency band during the frequency extraction process. Although the Mel filter is approximately linear within 1kHz, according to calculations, it is difficult to align the peak frequency of the Mel filter group with the multiple of 50Hz. This means that the use of the Mel filter group may cause the loss of key frequency features below 1kHz. In addition, the fault sound of partial discharge has a higher power output at 19000-20400Hz.
[0075] In view of these potential problems, the application of Mel filter banks in transformer voiceprint feature extraction needs to be carefully considered. After in-depth analysis of a large amount of audio data time-frequency diagrams in the State Grid database, it was found that the mid-frequency band of transformer fault sounds contains almost no effective feature information. However, for certain types of fault sounds, such as partial discharge, there are frequency components with high power in the high frequency band (around 20kHz). Therefore, in the feature extraction process, we should focus on the feature information of the low-frequency band and the high-frequency band. In addition, the frequency response of the Mel filter bank is complex, which will affect the running speed when the number of dimensions increases.
[0076] Therefore, in order to solve the above problems, this embodiment improves the transformer voiceprint feature and optimizes the Mel filter in feature extraction. The time-frequency comparison of voiceprint feature extraction before and after the improvement is as follows: Figure 5 As shown, it can be seen that the method of this embodiment can effectively reduce the complexity of the algorithm.
[0077] Specifically, in this embodiment, the Mel filter is improved to extract the energy spectrum of 0-1200Hz and 19000-20400Hz in the frequency domain, and an efficient rectangular filter group is used instead of a triangular filter group to reduce the impact of the amplitude modulation extracted by filtering on the voiceprint features; according to the principle of low-frequency dense and high-frequency sparse, the number of low-frequency filters is increased, the spectral resolution is improved, and the algorithm complexity is effectively reduced.
[0078] The output of the improved Mel filter can be represented by a 1×M dimensional vector. Preferably, in this embodiment, M is 61.
[0079] After processing by the improved Mel filter, the output of the i-th frame is z i =[z i1 ,z i2 ,…,z i61 ],1≤i≤116, the transformation formula is as follows:
[0080]
[0081] S3. Build a fault voiceprint detection model based on ELM
[0082] Extreme Learning Machine is a training algorithm for a single hidden layer feedforward neural network (SLFN). It has an extremely fast training speed and can achieve high generalization performance, making it suitable for situations with a large number of training samples.
[0083] Different from the traditional training algorithm, the ELM algorithm of this embodiment randomly assigns the weights and biases of the input layer, and then directly solves the weights from the hidden layer to the output layer by using the method of finding the Moore-Penrose generalized inverse matrix.
[0084] A typical SLFN consists of an input layer, a hidden layer, and an output layer, such as Figure 6 As shown in the figure, its main feature is that the number of neurons in its hidden layer is much larger than the number of neurons in the input layer. As a result, SLFN can quickly generate random connection weights during training, thereby accelerating the training process. In SLFN, the connection weights between the input layer and the super hidden layer are usually randomly initialized, while the connection weights of the output layer can be directly calculated by analytical solution or pseudo-inverse method without the need for iterative back-propagation algorithm.
[0085] In this embodiment, ELM randomly selects the input weights of SLFN and determines the output weights through analysis and judgment. Compared with traditional SLFN, ELM has significant advantages in practical application: fast training speed, no need to adjust parameters repeatedly, excellent generalization performance, and easy access to the global optimal solution.
[0086] The learning mode of the fault voiceprint detection model in this embodiment is: for any N different samples (x j ,t j ), where x j =[x j1 ,x j2 ,…,x jn ] T ,t j =[t j1 ,t j2 ,…t jm ] T , t j The label that represents the audio.
[0087] In particular, in this embodiment, m=2, the normal sound is [1,0], and the fault sound is [0,1].
[0088] The standard SLFN with K hidden nodes is expressed as:
[0089]
[0090] j=1,…,N
[0091] Among them, g(x) is the activation function, w i =[w i1 ,w i2 ,...,w in ] T is the bias connecting the ith hidden node, w i ·x j w i and x j The inner product of .
[0092] Such a SLFN can infinitely approximate these N samples: That is, there is β i ,ω i , b i So that:
[0093]
[0094] These K equations can be written as:
[0095] Hβ=T
[0096] in:
[0097]
[0098] Where H is the output matrix of the hidden layer of the neural network, and the i-th column of H is the output of the i-th node of the hidden layer of the neural network.
[0099] It should be noted that if the activation function g(x) is infinitely differentiable, then the weight w from the input layer to the hidden layer can be randomly selected. i and the bias b of the hidden layer i , SLFN can be regarded as a linear system, and the entire network can be determined by simply analytically solving the weight β from the hidden layer to the output layer.
[0100] Hβ=T, H is an N×K matrix. If N=K, then H is reversible; if N≠K, then find the Moore-Penrose generalized inverse matrix of H. The network obtained in this way not only has the minimum training error, but also the minimum norm of the weights.
[0101] In this embodiment, the input and output dimensions of ELM are set to 61 and 2 according to the extracted feature dimension and label dimension. The label is encoded using one-hot encoding, and the normal sound is t normal =[1,0] T , the fault sound is t abnormal =[0,1] T , set the number of hidden nodes of ELM to 1200, and select the activation function as Sigmoid:
[0102]
[0103] The trained hidden layer model is saved for subsequent fault voiceprint detection.
[0104] In this embodiment, the ELM algorithm does not require manual parameter adjustment during execution, which avoids the repeated iteration of the traditional training algorithm, converges quickly, and greatly reduces the training time. With its excellent generalization ability and extremely fast training speed, ELM is superior to SVM in detection accuracy and speed. The time required for voiceprint detection is greatly shortened, and the accuracy is further improved to 98.6%.
[0105] S4. Test set establishment
[0106] The feature extraction of the test set is the same as that of the training set. The difference is that the training set needs to be divided into frames, while the test set only needs to take the 3 frames in the middle for feature averaging.
[0107] In this embodiment, the feature-extracted data p of frames 58, 59, and 60 are j_58 ,p j_59 ,p j_60 Perform the following processing:
[0108]
[0109] Get the real-time test sample P j .
[0110] S5. Fault detection
[0111] Input the real-time recorded audio data into the trained model and output the label t i =[t i1 ,t i2 ] T As the test result, if it is normal, it is [1,0] T , and the exception is [0,1] T .
[0112] In summary, the power transformer fault voiceprint detection method of the present invention analyzes the voiceprint information of the substation during operation, and uses the ELM algorithm to detect the transformer fault voiceprint. ELM randomly selects the input weights of SLFNs. During the execution of the ELM algorithm, no manual adjustment of parameters is required, which avoids the repeated iteration process of the traditional training algorithm, converges quickly, greatly reduces the training time, improves the detection accuracy and speed, and is easy to implement in engineering. At the same time, the improved Mel filter can improve the spectral resolution of transformer voiceprint feature extraction, reduce the system complexity, and effectively improve the fault detection accuracy of the ELM algorithm.
[0113] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.
Claims
1. A method for detecting fault voiceprint of power transformer based on extreme learning machine, characterized in that: The steps include: S1. Audio data preprocessing: Collect the audio signal of the transformer fault, divide the audio signal into frames using the overlapping frame method, preprocess each frame of data after framing through fast Fourier transform, convert the time domain signal into the frequency domain, and obtain the spectrum data; S2, voiceprint feature extraction: through the improved Mel filter, extract the voiceprint feature parameters and construct the training set; S3. Build a fault voiceprint detection model based on extreme learning machine and perform sample training to obtain a trained fault voiceprint detection model; S4, establishing a test set, selecting three consecutive frames from the middle position of each audio signal to evaluate the voiceprint features, obtaining real-time test samples, and forming a test set; S5. Perform fault detection, input the real-time test sample into the trained fault voiceprint detection model, perform test recognition, and obtain the voiceprint recognition result of the transformer fault.
2. The power transformer fault voiceprint detection method according to claim 1 is characterized in that: In step S1, the audio signal of the transformer fault has a sampling frequency of 48kHz and a length of 10s; The overlapping frame method is used to divide the audio signal into frames. The frame length of each frame is 8192 points, the frame shift is 4096 points, and each audio signal is divided into 116 frames. The data x of the i-th frame of the audio signal i Expressed as: x i =[x i1 ,x i2 ,…,x i8192 ], 1≤i≤116; For each frame of data x i Perform fast Fourier transform, remove negative frequencies, retain positive frequencies, and obtain spectrum data y i , expressed as: y i =[y i1 ,y i2 ,…,y i4096 ].
3. The power transformer fault voiceprint detection method according to claim 2 is characterized in that: In step S2, the Mel filter is improved by replacing the triangular filter group in the traditional Mel filter with a rectangular filter group to reduce the impact of the amplitude modulation extracted by filtering on the voiceprint features. According to the principle of low-frequency density and high-frequency sparseness, the number of low-frequency filters is increased to improve the spectral resolution and reduce the complexity.
4. The power transformer fault voiceprint detection method according to claim 3 is characterized in that: In step S2, based on the spectrum data obtained in step S1, the energy spectrum of the audio signal in the training set with frequency domains of 0-1200 Hz and 19000-20400 Hz is extracted through the improved Mel filter. The output of the i-th frame of the audio signal is z i , the transformation formula is as follows: Among them, the output of the improved Mel filter is a 1×M dimensional vector, z i =[z i1 ,z i2 ,…,z iM ].
5. The power transformer fault voiceprint detection method according to claim 3 is characterized in that: In step S3, the extreme learning machine is a training algorithm based on a single hidden layer feedforward neural network, including an input layer, a hidden layer and an output layer; the extreme learning machine determines the weight and bias of the input layer of the single hidden layer feedforward neural network by random selection and assigns values; Construct a fault voiceprint detection model based on extreme learning machine. For N different audio signal samples (x j ,t j ), x j =[x j1 ,x j2 ,…,x jn ] T ,t j =[t j1 ,t j2 ,…t jm ] T , t j Representing the audio label, there is a single hidden layer forward neural network with K hidden nodes, expressed as: Where g(x) is the activation function, j=1,…,N,w i is the weight from the input layer to the hidden layer, w i =[w i1 ,w i2 ,…,w in ] T represents the bias connecting the ith hidden node, w i ·x j w i and x j The inner product of i is the bias of the hidden layer, β j Represents the weight from the hidden layer to the output layer of the jth sample; The single hidden layer feedforward neural network infinitely approximates N samples, that is, there is β i ,ω i 、b i So that: The K equations are expressed as: Hβ=T in: Where H is the output matrix of the hidden layer of the single hidden layer feedforward neural network, the i-th column of H is the output of the i-th node of the hidden layer of the neural network, and β is the weight from the hidden layer to the output layer.
6. The power transformer fault voiceprint detection method according to claim 5, characterized in that: In step S3, a fault voiceprint detection model based on an extreme learning machine is constructed. The input and output dimensions of the extreme learning machine are set according to the extracted feature dimension and label dimension. The label is encoded using one-hot encoding. The normal sound is t normal =[1,0] T , the fault sound is t abnormal =[0,1] T , set the number of hidden nodes of the extreme learning machine to 1200, and select the activation function as Sigmoid, which is expressed as follows: Among them, e -x It represents the reciprocal of the exponential function with the natural base e as the base. The model is trained through the training set, and the trained fault voiceprint detection model is saved for real-time detection of power transformer fault voiceprints.
7. The power transformer fault voiceprint detection method according to claim 6, characterized in that: In step S4, a test set is established, and three consecutive frames are selected from the middle position of each audio signal and numerically averaged using the following formula to obtain a real-time test sample P: j : Among them, p j_58 ,p j_59 ,p j_60 It is the feature-extracted data of the 58th to 60th frames of the jth audio signal.
8. The power transformer fault voiceprint detection method according to claim 7, characterized in that: In step S5, fault detection is performed and the test sample P j Input the trained fault voiceprint detection model and get the j-th sample detection output t j : t j =[t j1 ,t j2 ] T If t j1 =[1,0] T , the test result is normal; t j2 =[0,1] T , the test result is abnormal.
9. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the power transformer fault voiceprint detection method as described in any one of claims 1 to 8.
10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the steps in the power transformer fault voiceprint detection method described in any one of claims 1 to 8 are implemented.
Citation Information
Patent Citations
Fault diagnosis method of power transformer based on extreme learning machine
CN102944769A
Acoustic feature extraction method, device, storage medium and terminal device
CN109147818A
Transformer fault diagnosis method
CN110969262A
Transformer fault identification method based on kernel function extreme learning machine
CN112766140A
Construction method of sound classification model, and sound classification method and system
CN114255783A
Cited By
New energy station transformer voiceprint fault diagnosis method based on wave search optimization
CN119479700A
