CN116092525A - Electrical equipment state voice recognition method considering time-frequency domain feature fusion - Google Patents
Electrical equipment state voice recognition method considering time-frequency domain feature fusion Download PDFInfo
- Publication number
- CN116092525A CN116092525A CN202211686454.XA CN202211686454A CN116092525A CN 116092525 A CN116092525 A CN 116092525A CN 202211686454 A CN202211686454 A CN 202211686454A CN 116092525 A CN116092525 A CN 116092525A
- Authority
- CN
- China
- Prior art keywords
- frequency domain
- time
- sound
- feature
- features
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 83
- 230000004927 fusion Effects 0.000 title claims abstract description 34
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 35
- 238000013528 artificial neural network Methods 0.000 claims abstract description 22
- 238000012549 training Methods 0.000 claims abstract description 22
- 238000004806 packaging method and process Methods 0.000 claims abstract description 12
- 230000005236 sound signal Effects 0.000 claims abstract description 11
- 238000001914 filtration Methods 0.000 claims abstract description 7
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 230000009466 transformation Effects 0.000 claims abstract description 7
- 239000010410 layer Substances 0.000 claims description 34
- 238000007637 random forest analysis Methods 0.000 claims description 18
- 230000000694 effects Effects 0.000 claims description 13
- 238000011156 evaluation Methods 0.000 claims description 10
- 238000009826 distribution Methods 0.000 claims description 9
- 210000002569 neuron Anatomy 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 6
- 230000014759 maintenance of location Effects 0.000 claims description 5
- 230000008030 elimination Effects 0.000 claims description 3
- 238000003379 elimination reaction Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 239000011229 interlayer Substances 0.000 claims description 3
- 230000002159 abnormal effect Effects 0.000 abstract description 8
- 238000010801 machine learning Methods 0.000 abstract description 2
- 230000007547 defect Effects 0.000 description 6
- 238000012544 monitoring process Methods 0.000 description 6
- 238000003745 diagnosis Methods 0.000 description 5
- 238000012216 screening Methods 0.000 description 4
- 238000007789 sealing Methods 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 2
- 238000005096 rolling process Methods 0.000 description 2
- 101100136092 Drosophila melanogaster peng gene Proteins 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012856 packing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
- G10L25/51—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
- G10L25/18—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters the extracted parameters being spectral information of each sub-band
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
- G10L25/30—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Computational Linguistics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
- Examining Or Testing Airtightness (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a method for recognizing state sound of electrical equipment by considering time-frequency domain feature fusion. The method comprises the following steps: acquiring and preprocessing sound time domain data; converting the time domain signal into a frequency domain by utilizing Fourier transformation, simultaneously carrying out frequency domain feature selection and algorithm training based on an embedding method, and selecting an optimal feature subset; extracting time domain signal characteristics of the original sound signal by using a packaging method; filtering and fusing the time domain features and the frequency domain features by a mutual information method; and establishing a neural network, and identifying the sound state based on the time-frequency domain fusion signal. According to the invention, based on effective fusion of time domain information and frequency domain information, on one hand, important characteristic information of the time domain and the frequency domain is reserved, the data dimension is reduced through an embedding method and a packaging method, and on the other hand, three states of normal work, normal air leakage and abnormal air leakage are effectively identified through a machine learning algorithm, so that the identification accuracy is improved.
Description
Technical Field
The invention relates to the technical field of electrical equipment state voice recognition, in particular to an electrical equipment state voice recognition method considering time-frequency domain feature fusion.
Background
The power plant equipment is often huge in size and densely distributed, the defects can not be detected by the inspection of inspectors in the past, the full period and the full range of monitoring can not be ensured, and the micro faults and the defects can be detected immediately by arranging sensors and a machine learning recognition algorithm, so that serious production accidents and economic losses caused by further development are prevented.
The current equipment voice recognition method mainly aims at electric main equipment such as transformers, cables and circuit breakers, and lacks voice recognition research on power plant sealing equipment. The current commonly used classification and identification method adopts neural network identification, and the southeast university Deng Aidong et al propose a wind power bearing fault diagnosis method based on a time-frequency domain convolution network and a depth forest, obtain fault data of vibration, working conditions, rotating speed and load, extract characteristics according to the time-frequency domain convolution network, and complete fault diagnosis through a two-layer depth forest model (Deng Aidong, liu Dongchuan, yang Hongjiang, fan Yongsheng. Wind power bearing fault diagnosis method based on the time-frequency domain convolution network and the depth forest [ P ]. Jiangsu province: CN114964780A, 2022-08-30.); the patent Yun Xiao and the like of Beijing university propose a rolling bearing fault diagnosis method based on time-frequency domain multidimensional vibration feature fusion, firstly, wavelet denoising treatment is carried out on vibration signals, time domain feature parameters are obtained through feature extraction, an energy matrix is obtained through wavelet packet decomposition and energy moment calculation, most of the energy matrix is synthesized, and the bearing states are judged according to index distances (the patent Yun Xiao, gu Limin, lv Jinsong, ji Changxu, yao Dechen, li Qian, lu Yong. The rolling bearing fault diagnosis method based on time-frequency domain multidimensional vibration feature fusion [ P ]. Beijing: CN104655423A, 2015-05-27.); the power company of Jiangsu province, overhaul division Tan Fenglei et al, proposed a method for diagnosing latent defects of a transformer based on sound monitoring, firstly, according to the installation position of a sound sensor based on a noise attenuation rule, judging whether the transformer has latent defects or not based on characteristic frequency and defect evaluation indexes (Tan Fenglei, zhu Chao, chen, deng Kai, gao Shiyu, chenlong. A method for diagnosing latent defects of a transformer based on sound monitoring [ P ]. Jiangsu province: CN113253156A, 2021-08-13.); hunan university of science and technology Wu Xiaowen et al propose a method and system for diagnosing transformer faults using voice feature codes, which judge the results of transformer faults according to voice feature code rules and combinations (Wu Xiaowen, lu Ming, chen Chaoyang, he, xie, tan Zhuangxi, ying, cao Hao; the Shandong and intelligent science and technology limited company Yang Wenjiang et al propose a high-voltage sleeve digitalized evaluation method and system based on time-frequency domain feature fusion, and the state of the high-voltage sleeve is judged by utilizing the analysis results of time domain or frequency domain evaluation units in different sampling intervals (Yang Wenjiang, chen Xin, zhao Fei, liu Peng and Feng Xu. The high-voltage sleeve digitalized evaluation method and system based on time-frequency domain feature fusion [ P ]. Shandong province: CN115015684A, 2022-09-06.); none of the above inventive methods relate to monitoring of the leakage condition of the device.
Disclosure of Invention
The invention provides a method for recognizing the state sound of an electrical device by considering time-frequency domain feature fusion, which completes the time-frequency domain feature fusion of the sound of the device according to the data acquisition and feature extraction of a sensor, combines the strong learning ability of a neural network, can simultaneously recognize the normal working condition, the normal air release working condition and the abnormal air leakage working condition of the device, and improves the accuracy of sound recognition.
The invention has the innovation points that the time domain information and the frequency domain information are screened and fused by utilizing an embedding method, a packaging method and mutual information entropy, and the fusion of a time domain and a frequency domain is considered when the sound state of equipment is identified, so that on one hand, the data reduction and feature extraction is completed, and further, the neural network is utilized for identification to obtain a better identification effect.
The object of the invention is achieved by at least one of the following technical solutions.
An electrical equipment state voice recognition method considering time-frequency domain feature fusion comprises the following steps:
s1, acquiring and preprocessing sound time domain data;
s2, converting the time domain signal into a frequency domain by utilizing Fourier transformation, simultaneously carrying out frequency domain feature selection and algorithm training based on an embedding method, and selecting an optimal feature subset;
s3, extracting time domain signal characteristics of the original sound signal by using a packaging method;
s4, filtering and fusing the time domain features and the frequency domain features by a mutual information method;
and S5, establishing a neural network, and identifying the sound state based on the time-frequency domain fusion signal.
Further, in step S1, the sound time domain data is a sound time domain signal obtained according to a sound sensor installed in the electrical device;
the preprocessing is to slice the acquired sound data according to time length;
N e ＝V/l e
wherein l e For the time domain signal length obtained by slicing the sound data in time length, N E The number of segments, V, is the sound data acquired by the sensor.
Further, in step S2, the time domain signal is converted into the frequency domain by using fourier transform, and the frequency domain feature selection and the algorithm training are simultaneously performed based on the embedding method, so as to select an optimal feature subset, which includes the following steps:
s2.1, converting an original sound signal into a frequency domain by utilizing Fourier transformation;
s2.2, carrying out frequency domain feature selection and algorithm training simultaneously based on an embedding method.
Further, in step S2.1, the original sound signal is converted to the frequency domain by fourier transform, specifically as follows:
F(ω)＝∫l e (x)×e -j2πωx dx
f (omega) is sound frequency domain information, the converted sound frequency domain information is formed into a frequency domain set, and the corresponding sound tag is formed into a target set.
Further, in step S2.2, frequency domain feature selection and algorithm training are performed simultaneously based on the embedding method, specifically as follows:
training and evaluating effects on the frequency domain set and the target set by using a random forest algorithm model, filtering the feature set according to the training effects, and searching and traversing all features each time;
firstly, establishing a random forest algorithm model, and carrying out instance initialization, namely setting the number of the evaluators in the random forest algorithm model for evaluating the model effect, and setting a proper number of evaluators to achieve balance between training difficulty and the model effect;
then, carrying out feature selection SelectFromModel instantiation, namely inputting a random forest algorithm model initialized in the last step, setting a super-parameter evaluation threshold value, and then inputting a frequency domain set and a target set to start iterative solution;
and reserving a feature set meeting a threshold value as a subsequent frequency domain feature set according to the weight ranking of the features.
Further, in step S3, the time domain signal feature extraction is performed on the original sound signal by using the packing method:
firstly, initializing a random forest algorithm by an example, and setting the number of estimators;
carrying out instantiation of feature selection, namely inputting an initialized random forest algorithm, setting an objective function as a recursive feature elimination method, setting reserved feature numbers and setting the feature number removed in each iteration; then inputting a time domain set and a target set for solving, wherein the time domain set is formed by an original sound data set, and the target set is formed by a corresponding sound tag;
and sorting the features in each iteration according to the importance of the features, and selecting the optimal features until a feature set meeting the retention feature number is selected as a subsequent time domain feature set.
Further, in step S4, the time domain features and the frequency domain features are filtered and fused by adopting a mutual information method:
solving mutual information entropy of each characteristic variable:
wherein R is mutual information entropy, t is time domain feature sequence number, N T For the number of the time domain feature number, i.e. the number of the time domain feature sets, f is the frequency domain feature sequence number, N F For the number of frequency domain feature numbers, i.e., frequency domain feature sets, p (t, f) is the joint distribution of the objects, p (t) is the edge distribution of the time domain features to the objects, and p (f) is the edge distribution of the frequency domain features to the objects.
Further, if the information entropy is 1, the two variables are completely related, and one of the variables needs to be removed; if the information entropy is not 1, fusing according to the mode that the time domain features are before and the frequency domain features are after.
Further, in step S5, a neural network is established, and the sound state is identified based on the time-frequency domain fusion signal:
the neural network comprises an input layer, a first hidden layer and a second hidden layer which are connected in sequence;
the input of the input layer is the characteristic frequency band screened in the step S4; the output of the first hidden layer and the second hidden layer is the state of the electrical equipment, the error is obtained by utilizing forward propagation, and the weight is reversely updated through the partial derivative and the learning rate;
the inter-layer propagation of the neural network is as follows:
g＝wc+b
wherein g is the number of neurons of the second hidden layer, c is the number of neurons of the first hidden layer, b is a bias term, and w is a weight.
Further, the weight update formula is:
wherein w' is the updated weight, L is the neural network error, and alpha is the learning rate.
Compared with the prior art, the invention has the advantages that:
the invention provides a state voice recognition method of electrical equipment considering time-frequency domain feature fusion, which effectively solves the problems of insufficient practicability, low reliability and the like of the current voice monitoring technology through gas monitoring and physical detection, and utilizes a feature screening method to complete time-frequency feature fusion and utilizes time-frequency domain information as an input neural network to achieve an accurate recognition effect.
Drawings
Fig. 1 is a flowchart of a step of an electrical device state voice recognition method considering time-frequency domain feature fusion in an embodiment of the present invention;
FIG. 2 is a flowchart illustrating steps of an algorithm for performing frequency domain feature selection based on an embedding method according to an embodiment of the present invention;
fig. 3 is a flowchart illustrating steps of an algorithm for performing time domain signal feature extraction on a sound original signal by using a packaging method according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1:
an electrical equipment state voice recognition method considering time-frequency domain feature fusion, as shown in fig. 1, comprises the following steps:
s1, acquiring and preprocessing sound time domain data;
the sound time domain data are sound time domain signals obtained according to a sound sensor arranged on the electrical equipment;
the preprocessing is to slice the acquired sound data according to time length;
N e ＝V/l e
wherein l e For the time domain signal length obtained by slicing the sound data in time length, N E The number of segments, V, is the sound data acquired by the sensor.
S2, converting a time domain signal into a frequency domain by utilizing Fourier transformation, and simultaneously carrying out frequency domain feature selection and algorithm training based on an embedding method to select an optimal feature subset, wherein the method comprises the following steps of:
s2.1, converting the original sound signal into a frequency domain by utilizing Fourier transformation:
F(ω)＝∫l e (x)×e -j2πωx dx
wherein F (ω) is sound frequency domain information;
then the converted sound frequency domain information is formed into a frequency domain set, and the corresponding sound label is formed into a target set
S2.2, carrying out frequency domain feature selection and algorithm training simultaneously based on an embedding method:
training and evaluating effects on the frequency domain set and the target set by using a random forest algorithm model, filtering the feature set according to the training effects, and searching and traversing all features each time;
firstly, establishing a random forest algorithm model, and carrying out instance initialization, namely setting the number of the evaluators in the random forest algorithm model for evaluating the model effect, and setting a proper number of evaluators to achieve balance between training difficulty and the model effect;
then, carrying out feature selection SelectFromModel instantiation, namely inputting a random forest algorithm model initialized in the last step, setting a super-parameter evaluation threshold value, and then inputting a frequency domain set and a target set to start iterative solution;
and according to the weight ranking of the features, reserving a feature set meeting a threshold value as a subsequent frequency domain feature set, wherein the algorithm flow is shown in fig. 2.
S3, extracting time domain signal characteristics of the original sound signal by using a packaging method:
firstly, initializing a random forest algorithm by an example, and setting the number of estimators;
carrying out instantiation of feature selection, namely inputting an initialized random forest algorithm, setting an objective function as a recursive feature elimination method, setting reserved feature numbers and setting the feature number removed in each iteration; then inputting a time domain set and a target set for solving, wherein the time domain set is formed by an original sound data set, and the target set is formed by a corresponding sound tag;
and sorting the features in each iteration according to the importance of the features, selecting the optimal features until a feature set meeting the retention feature number is selected and used as a subsequent time domain feature set, wherein the algorithm flow is shown in fig. 3.
S4, filtering and fusing the time domain features and the frequency domain features by adopting a mutual information method:
solving mutual information entropy of each characteristic variable:
wherein R is mutual information entropy, t is time domain feature sequence number, N T For the number of the time domain feature number, i.e. the number of the time domain feature sets, f is the frequency domain feature sequence number, N F P (t, f) is the joint distribution of the frequency domain feature number, namely the number of the frequency domain feature sets, p (t) is the edge distribution of the time domain feature to the target, and p (f) is the edge distribution of the frequency domain feature to the target; if the information entropy is 1, the two variables are completely phaseClosing, wherein one of the two is required to be removed; if the information entropy is not 1, fusing according to the mode that the time domain features are before and the frequency domain features are after.
S5, a neural network is established, and the sound state is identified based on the time-frequency domain fusion signal;
the neural network comprises an input layer, a first hidden layer and a second hidden layer which are connected in sequence;
the input of the input layer is the characteristic frequency band screened in the step S4; the output of the first hidden layer and the second hidden layer is the state of the electrical equipment, the error is obtained by utilizing forward propagation, and the weight is reversely updated through the partial derivative and the learning rate;
the inter-layer propagation of the neural network is as follows:
g＝wc+b
wherein g is the number of neurons of the second hidden layer, c is the number of neurons of the first hidden layer, b is a deviation term, and w is a weight;
the weight update formula is:
wherein w' is the updated weight, L is the neural network error, and alpha is the learning rate.
In this embodiment, sound data of 3 states of the electrical sealing device are collected, namely, normal working, wav, normal air leakage, wav and abnormal air leakage, wav, a sound time domain sample is obtained through 50ms slicing, a data dimension is [960,9600], a sound frequency domain sample is obtained through fourier frequency domain conversion processing, a data dimension is [960,96000], the number of reserved package method features is set to be 100, an embedding method threshold is set to be 0.005, a time-frequency domain fusion sample space is obtained based on model evaluation and mutual information method feature screening of the embedding method and the packaging method, and the time-frequency domain feature dimension is [960,100].
The neural network input layer is provided with 80 nodes, the hidden layer is provided with 10 neurons, the full-connection layer is utilized to output predicted label values, the overall accuracy is 98.12% after 100 times of training, and the test set results are shown in table 1.
TABLE 1
| Category(s) | Normal operation | Abnormal air leakage | Normally let out air |
| The method has the accuracy | 98.3％ | 99.1％ | 97.3％ |
Example 2:
collecting 3 states of sound data of the electric sealing equipment, namely normal operation, MP3, normal air leakage, MP3, abnormal air leakage, MP3, obtaining a sound time domain sample by 50ms slicing, obtaining a data dimension [960,9600], obtaining the sound frequency domain sample by Fourier frequency domain conversion processing, the data dimension is [960,96000], the feature retention quantity of the packaging method is set to be 100, the threshold value of the embedding method is set to be 0.005, the model evaluation based on the embedding method and the packaging method and the mutual information method feature screening are carried out, the obtained time-frequency domain fused sample space is [960,130], and the time domain feature dimension is [960,100].
The neural network input layer is provided with 80 nodes, the hidden layer is provided with 10 neurons, the predicted label value is output by using the full-connection layer, the overall accuracy is 97.81% after 100 times of training, and the test set result is shown in table 2.
TABLE 2
| Category(s) | Normal operation | Abnormal air leakage | Normally let out air |
| The method has the accuracy | 98.1％ | 99.7％ | 95.3％ |
Example 3:
in this embodiment, sound data of 3 states of the electrical sealing device are collected, namely, normal working, wav, normal air leakage, wav and abnormal air leakage, wav, a sound time domain sample is obtained through 50ms slicing, a data dimension is [960,9600], a sound frequency domain sample is obtained through fourier frequency domain conversion processing, a data dimension is [960,96000], the number of package method feature retention is 130, an embedding method threshold is set to be 0.002, a sample space fused with a time domain is obtained based on model evaluation and mutual information method feature screening of the embedding method and the packaging method is [960,160], and a time domain feature dimension is [960,130].
The neural network input layer is provided with 80 nodes, the hidden layer is provided with 10 neurons, the full-connection layer is utilized to output predicted label values, the overall accuracy is 99.27% after 100 times of training, and the test set results are shown in table 3.
TABLE 3 Table 3
| Category(s) | Normal operation | Abnormal air leakage | Normally let out air |
| The method has the accuracy | 99.1％ | 99.0％ | 99.1％ |
The above-mentioned identification and tracking method combination is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited by the above-mentioned examples, and any other modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention should be equivalent and are included in the scope of the present invention.
Claims (10)
1. The electrical equipment state voice recognition method considering time-frequency domain feature fusion is characterized by comprising the following steps of:
s1, acquiring and preprocessing sound time domain data;
s2, converting the time domain signal into a frequency domain by utilizing Fourier transformation, simultaneously carrying out frequency domain feature selection and algorithm training based on an embedding method, and selecting an optimal feature subset;
s3, extracting time domain signal characteristics of the original sound signal by using a packaging method;
s4, filtering and fusing the time domain features and the frequency domain features by a mutual information method;
and S5, establishing a neural network, and identifying the sound state based on the time-frequency domain fusion signal.
2. The method for recognizing the state sound of the electrical equipment in consideration of the fusion of the time-frequency domain features according to claim 1, wherein in the step S1, the sound time domain data is a sound time domain signal obtained according to a sound sensor installed in the electrical equipment;
the preprocessing is to slice the acquired sound data according to time length;
N e ＝V/l e
wherein l e For the time domain signal length obtained by slicing the sound data in time length, N E The number of segments, V, is the sound data acquired by the sensor.
3. The method for recognizing the state sound of the electrical equipment by considering the fusion of time-frequency domain features according to claim 1, wherein in the step S2, the time domain signal is converted into the frequency domain by using fourier transform, the frequency domain feature selection and the algorithm training are simultaneously performed based on the embedding method, and the best feature subset is selected, comprising the following steps:
s2.1, converting an original sound signal into a frequency domain by utilizing Fourier transformation;
s2.2, carrying out frequency domain feature selection and algorithm training simultaneously based on an embedding method.
4. A method for recognizing a state sound of an electrical device in consideration of time-frequency domain feature fusion according to claim 3, wherein in step S2.1, the original sound signal is converted into the frequency domain by fourier transform, specifically as follows:
F(ω)＝∫l e (x)×e -j2πωx dx
f (omega) is sound frequency domain information, the converted sound frequency domain information is formed into a frequency domain set, and the corresponding sound tag is formed into a target set.
5. The method for recognizing the state sound of the electrical equipment by considering the fusion of the time-frequency domain features according to claim 4, wherein in the step S2.2, the frequency domain feature selection and the algorithm training are simultaneously performed based on the embedding method, specifically comprising the following steps:
training and evaluating effects on the frequency domain set and the target set by using a random forest algorithm model, filtering the feature set according to the training effects, and searching and traversing all features each time;
firstly, establishing a random forest algorithm model, and initializing an instance, namely setting the number of evaluators in the random forest algorithm model for evaluating the model effect;
then, carrying out feature selection SelectFromModel instantiation, namely inputting a random forest algorithm model initialized in the last step, setting a super-parameter evaluation threshold value, and then inputting a frequency domain set and a target set to start iterative solution;
and reserving a feature set meeting a threshold value as a subsequent frequency domain feature set according to the weight ranking of the features.
6. The method for recognizing state sound of electrical equipment in consideration of time-frequency domain feature fusion according to claim 1, wherein in step S3, time domain signal feature extraction is performed on the original sound signal by using a packaging method:
firstly, initializing a random forest algorithm by an example, and setting the number of estimators;
carrying out instantiation of feature selection, namely inputting an initialized random forest algorithm, setting an objective function as a recursive feature elimination method, setting reserved feature numbers and setting the feature number removed in each iteration; then inputting a time domain set and a target set for solving, wherein the time domain set is formed by an original sound data set, and the target set is formed by a corresponding sound tag;
and sorting the features in each iteration according to the importance of the features, and selecting the optimal features until a feature set meeting the retention feature number is selected as a subsequent time domain feature set.
7. The method for recognizing the state sound of the electrical equipment by considering the fusion of the time-frequency domain features according to claim 1, wherein in the step S4, the time-frequency domain features and the frequency domain features are filtered and fused by adopting a mutual information method:
solving mutual information entropy of each characteristic variable:
wherein R is mutual information entropy, t is time domain feature sequence number, N T For the number of the time domain feature number, i.e. the number of the time domain feature sets, f is the frequency domain feature sequence number, N F For the number of frequency domain feature numbers, i.e., frequency domain feature sets, p (t, f) is the joint distribution of the objects, p (t) is the edge distribution of the time domain features to the objects, and p (f) is the edge distribution of the frequency domain features to the objects.
8. The method for recognizing the state sound of the electrical equipment taking the fusion of the time-frequency domain characteristics into consideration as set forth in claim 7, wherein if the information entropy is 1, it is explained that two variables are completely related, and one of the variables needs to be removed; if the information entropy is not 1, fusing according to the mode that the time domain features are before and the frequency domain features are after.
9. The method for recognizing the state of the electrical equipment by considering the fusion of the time-frequency domain features according to claim 1, wherein in the step S5, a neural network is established to recognize the state of the sound based on the time-frequency domain fusion signal:
the neural network comprises an input layer, a first hidden layer and a second hidden layer which are connected in sequence;
the input of the input layer is the characteristic frequency band screened in the step S4; the output of the first hidden layer and the second hidden layer is the state of the electrical equipment, the error is obtained by utilizing forward propagation, and the weight is reversely updated through the partial derivative and the learning rate;
the inter-layer propagation of the neural network is as follows:
g＝wc+b
wherein g is the number of neurons of the second hidden layer, c is the number of neurons of the first hidden layer, b is a bias term, and w is a weight.
10. The method for recognizing the state sound of the electrical equipment taking into consideration the fusion of time-frequency domain features according to claim 9, wherein the weight updating formula is as follows:
wherein w' is the updated weight, L is the neural network error, and alpha is the learning rate.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211686454.XA CN116092525A (en) | 2022-12-27 | 2022-12-27 | Electrical equipment state voice recognition method considering time-frequency domain feature fusion |
| PCT/CN2023/092574 WO2024138995A1 (en) | 2022-12-27 | 2023-05-06 | Electrical device state sound identification method considering fusion of time-domain and frequency-domain features |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211686454.XA CN116092525A (en) | 2022-12-27 | 2022-12-27 | Electrical equipment state voice recognition method considering time-frequency domain feature fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116092525A true CN116092525A (en) | 2023-05-09 |
Family
ID=86207501
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211686454.XA Pending CN116092525A (en) | 2022-12-27 | 2022-12-27 | Electrical equipment state voice recognition method considering time-frequency domain feature fusion |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN116092525A (en) |
| WO (1) | WO2024138995A1 (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2024138995A1 (en) * | 2022-12-27 | 2024-07-04 | 南方电网调峰调频发电有限公司储能科研院 | Electrical device state sound identification method considering fusion of time-domain and frequency-domain features |
| CN118399609A (en) * | 2024-06-24 | 2024-07-26 | 中能拾贝(广州)科技有限公司 | Generator set monitoring device based on micro-distance distributed optical fiber temperature sensor |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104655423B (en) * | 2013-11-19 | 2017-09-15 | 北京交通大学 | A kind of Fault Diagnosis of Roller Bearings based on time-frequency domain multi-dimensional vibration Fusion Features |
| US11726223B2 (en) * | 2019-12-10 | 2023-08-15 | Origin Rose Llc | Spectral analysis and machine learning to detect offset well communication using high frequency acoustic or vibration sensing |
| CN113253156B (en) * | 2021-05-17 | 2023-01-06 | 国网江苏省电力有限公司检修分公司 | Sound monitoring-based latent defect diagnosis method for transformer |
| CN113866572B (en) * | 2021-09-06 | 2024-05-28 | 西安理工大学 | DC fault arc detection and positioning method under multi-power electronic device access condition |
| CN114527410B (en) * | 2021-12-31 | 2024-09-24 | 湖南科技大学 | Method and system for diagnosing transformer faults by utilizing sound feature codes |
| CN114964780A (en) * | 2022-05-30 | 2022-08-30 | 东南大学 | Wind power bearing fault diagnosis method based on time-frequency domain convolutional network and deep forest |
| CN115015684B (en) * | 2022-08-10 | 2022-11-29 | 山东和兑智能科技有限公司 | High-voltage bushing digital evaluation method and system based on time-frequency domain feature fusion |
| CN116092525A (en) * | 2022-12-27 | 2023-05-09 | 南方电网调峰调频发电有限公司储能科研院 | Electrical equipment state voice recognition method considering time-frequency domain feature fusion |
-
2022
- 2022-12-27 CN CN202211686454.XA patent/CN116092525A/en active Pending
-
2023
- 2023-05-06 WO PCT/CN2023/092574 patent/WO2024138995A1/en unknown
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2024138995A1 (en) * | 2022-12-27 | 2024-07-04 | 南方电网调峰调频发电有限公司储能科研院 | Electrical device state sound identification method considering fusion of time-domain and frequency-domain features |
| CN118399609A (en) * | 2024-06-24 | 2024-07-26 | 中能拾贝(广州)科技有限公司 | Generator set monitoring device based on micro-distance distributed optical fiber temperature sensor |
| CN118399609B (en) * | 2024-06-24 | 2024-09-06 | 中能拾贝(广州)科技有限公司 | Generator set monitoring device based on micro-distance distributed optical fiber temperature sensor |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2024138995A1 (en) | 2024-07-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116092525A (en) | Electrical equipment state voice recognition method considering time-frequency domain feature fusion | |
| CN112254964A (en) | Rolling bearing fault diagnosis method based on rapid multi-scale convolution neural network | |
| CN112257530B (en) | Rolling bearing fault diagnosis method based on blind signal separation and support vector machine | |
| CN103389430B (en) | A kind of oil-filled transformer fault detection method based on Bayesian discrimination theory | |
| CN103018629A (en) | Method for analyzing power system fault recording data based on Marla algorithm | |
| CN113391207A (en) | Motor fault detection method, medium and system | |
| CN108562821A (en) | A kind of method and system determining Single-phase Earth-fault Selection in Distribution Systems based on Softmax | |
| CN112305388B (en) | On-line monitoring and diagnosing method for insulation partial discharge faults of generator stator winding | |
| CN109214522B (en) | Equipment performance degradation evaluation method based on multiple attributes | |
| CN116089813A (en) | Wind turbine generator set transmission chain fault diagnosis method based on machine learning | |
| CN117607672A (en) | Intelligent monitoring method and system for GIS circuit breaker | |
| CN111983365A (en) | Transformer winding deformation detection method based on oscillatory wave multistage decomposition | |
| CN111860241A (en) | Power equipment discharge fault identification method based on wavelet packet analysis | |
| CN114048787B (en) | Method and system for intelligently diagnosing bearing fault in real time based on Attention CNN model | |
| CN117191396A (en) | Gear box fault diagnosis method based on two-stage migration | |
| CN112883639A (en) | GIS equipment service life prediction device and method based on machine learning | |
| CN117633588A (en) | Pipeline leakage positioning method based on spectrum weighting and residual convolution neural network | |
| CN114136604B (en) | Rotary equipment fault diagnosis method and system based on improved sparse dictionary | |
| CN116227172A (en) | Rolling bearing performance degradation evaluation method based on convolutional neural network | |
| CN114742098A (en) | Fault diagnosis method of rolling bearing based on time-frequency diagram and multi-scale CNN | |
| CN109917245B (en) | Ultrasonic detection partial discharge signal mode identification method considering phase difference | |
| CN112651147B (en) | Ocean platform fault diagnosis method based on Hilbert-Huang transform and support vector machine | |
| Zheng et al. | Wavelet packet decomposition and neural network based fault diagnosis for elevator excessive vibration | |
| CN114970044B (en) | Rolling bearing fault diagnosis method and system based on threshold convolutional neural network | |
| CN118883068A (en) | Bearing fault diagnosis method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |