CN115602193A - Multi-information fusion power plant electrical equipment classification and abnormality detection method - Google Patents
Multi-information fusion power plant electrical equipment classification and abnormality detection method Download PDFInfo
- Publication number
- CN115602193A CN115602193A CN202211214141.4A CN202211214141A CN115602193A CN 115602193 A CN115602193 A CN 115602193A CN 202211214141 A CN202211214141 A CN 202211214141A CN 115602193 A CN115602193 A CN 115602193A
- Authority
- CN
- China
- Prior art keywords
- electrical equipment
- fault
- sound
- neural network
- confidence
- 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
- 230000005856 abnormality Effects 0.000 title claims abstract description 17
- 230000004927 fusion Effects 0.000 title claims abstract description 17
- 238000001514 detection method Methods 0.000 title claims description 11
- 230000002159 abnormal effect Effects 0.000 claims abstract description 49
- 238000000034 method Methods 0.000 claims abstract description 43
- 238000013528 artificial neural network Methods 0.000 claims abstract description 41
- 238000012549 training Methods 0.000 claims description 32
- 238000012544 monitoring process Methods 0.000 claims description 25
- 238000011176 pooling Methods 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 15
- 230000009466 transformation Effects 0.000 claims description 13
- 238000007781 pre-processing Methods 0.000 claims description 12
- 238000013527 convolutional neural network Methods 0.000 claims description 11
- 238000004451 qualitative analysis Methods 0.000 claims description 4
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 abstract description 13
- 238000005516 engineering process Methods 0.000 description 14
- 238000010801 machine learning Methods 0.000 description 14
- 238000012545 processing Methods 0.000 description 12
- 238000007635 classification algorithm Methods 0.000 description 11
- 230000005236 sound signal Effects 0.000 description 7
- 238000013135 deep learning Methods 0.000 description 5
- 238000003745 diagnosis Methods 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 230000007613 environmental effect Effects 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000001228 spectrum Methods 0.000 description 3
- 101100136092 Drosophila melanogaster peng gene Proteins 0.000 description 2
- 230000003042 antagnostic effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000012896 Statistical algorithm Methods 0.000 description 1
- 235000017606 Vaccinium vitis idaea Nutrition 0.000 description 1
- 244000077923 Vaccinium vitis idaea Species 0.000 description 1
- 208000031975 Yang Deficiency Diseases 0.000 description 1
- 238000005311 autocorrelation function Methods 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000007257 malfunction Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000012567 pattern recognition method Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method 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
- G10L25/27—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)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a method for classifying and detecting abnormality of electrical equipment of a power plant by multi-information fusion. The method comprises the following steps: acquiring multivariate information of electrical equipment; judging the state of the electrical equipment; the method comprises the steps that a preset sound sensor obtains the audio frequency of the operation of the electrical equipment; performing state recognition through the trained deep neural network to obtain a state recognition confidence score; if the state recognition confidence coefficient is less than or equal to the confidence coefficient threshold value, determining that the electrical equipment has a fault, otherwise, completing the state sound classification of the electrical equipment; carrying out known fault abnormity identification through the trained deep neural network to obtain a fault identification confidence score; if the fault identification confidence score is less than or equal to the confidence threshold, the electrical equipment is determined to have unknown fault, otherwise, the known fault abnormal identification of the electrical equipment is completed; and identifying unknown faults of the electrical equipment, and adding the samples into a fault sound sample library of the electrical equipment as new abnormal samples. The invention solves the problems of the current algorithm.
Description
Technical Field
The invention relates to the field of electric power, in particular to a sound classification and abnormality detection method for electrical equipment of a power plant with multi-information fusion.
Background
The computer hearing technology mainly aims at the recognition and processing of non-human voice, including music signal processing combined with music field knowledge and audio information processing of other field knowledge. The computer hearing technology is a subject for understanding and analyzing the content of digital sound and music based on audio signal processing and machine learning, and is a cross subject for understanding and analyzing the content of digital sound and music by using a calculation method, and the main basic subjects are audio signal processing and machine learning. Currently, the development directions of computer hearing technology are mainly divided into 2 categories: music-oriented computer hearing technology and ambient sound-oriented computer hearing technology (which may be referred to as general audio-based computer hearing technology).
The application characteristics of the computer hearing technology are mainly reflected in that: on the basis of traditional audio signal processing, audio features are extracted, and state monitoring and fault diagnosis are completed by combining a machine learning algorithm (mainly a pattern recognition method). In the industrial field, there are also many studies and applications of computer hearing technology, and fault diagnosis is performed based on device operation noise, and mainly, application of acoustic target recognition is performed.
In the field of electric power, in the operation process of electric power equipment, sound is generated due to mechanical vibration, the sound under normal operation generally has certain regularity, but when some fault occurs in the equipment, the sound changes along with the change of the operation state or the operation structure, for example, when the mechanical fault occurs, the vibration characteristic or vibration energy in a part of frequency band of the equipment changes, and simultaneously harsh or sharp noise is accompanied. In addition, overload operation or other malfunctions of the equipment may also cause abnormal sound changes. Therefore, the sound signal of the power equipment contains a lot of operation state information, and an engineer with abundant experience can judge whether the equipment is in an abnormal operation state or not according to the abnormal sound of the field equipment through the change of audio characteristics such as tone, volume, pitch and the like, and even judge the type and severity of the fault.
In the prior art, the sound abnormality detection method for the electrical equipment includes a traditional sound monitoring method for the electrical equipment and a machine learning-based sound monitoring method for the electrical equipment.
The traditional electrical equipment sound monitoring method specifically comprises the following steps:
since a target sound signal is mixed with a large amount of noise signals due to conditions such as the operating state and the operating environment of the rotating equipment, when the computer hearing technology is used for state monitoring of the rotating equipment, most of the traditional audio processing methods such as spectrum analysis or wavelet transformation are combined to judge whether equipment faults exist, then the complex algorithm of audio feature extraction such as WPT, EMD or MFCC is used for processing, and finally, traditional classifiers such as HMM or SVM are used for completing state monitoring and fault early warning tasks. The acoustic signal processing of the rotating device is generally combined with other monitoring parameters such as the traditional vibration signal, and the result of the audio signal analysis is used for assisting judgment, so that the result based on the vibration signal analysis is more reliable. However, at present, there is no unified and reliable device acoustic signal processing algorithm that can be applied to all devices of a power system, different acoustic signal feature extraction methods need to be selected in combination with the operating characteristics and working states of the devices themselves, and a traditional audio signal processing technology is mostly adopted in application.
Compared with rotating equipment, the non-rotating equipment mainly comprises small-sized equipment of a power plant or primary and secondary equipment of a transformer substation, and has the characteristics of small vibration amplitude, small environmental noise and the like due to the fact that no high-power rotating device runs. From the above analysis, it can be seen that the non-rotating device performing the state monitoring through the device acoustic signal can complete the state monitoring task without the help of other auxiliary information; meanwhile, most of the non-rotating equipment runs in an open environment, and the processing difficulty is relatively simple, so that the audio processing of the non-rotating equipment only needs traditional time-frequency domain algorithms such as FFT, STFT or WA to extract features, and the basic monitoring requirements can be met by performing state classification through mathematical statistical algorithms such as VQ, autocorrelation coefficients or fuzzy clustering. However, the application of the current algorithm reduces the combination degree of the audio processing technology and the machine learning algorithm, and is not beneficial to improving the state monitoring recognition rate and the stability.
The electrical equipment sound monitoring method based on machine learning specifically comprises the following steps:
the sound-based on-line monitoring is a simple and reliable non-invasive monitoring method, does not interfere with the normal operation of the electrical equipment, and can well reflect the working state and abnormal conditions of the electrical equipment. Electrical equipment is various in types, complex in structure and various in fault types, and health conditions are difficult to infer directly from sound. The machine learning method is widely concerned about the acoustic anomaly monitoring of the electrical equipment, the deep learning technology is generally used for training more data than the traditional machine learning, but the training difficulty is not obviously increased due to the fact that the data scale and the feature dimension are too large, only the forward propagation of a neural network is needed during online operation, and the calculation efficiency is high. However, applying deep learning techniques to unsupervised/semi-supervised anomaly monitoring is still in the initial stage. The current deep learning anomaly detection method comprises a self-encoder, a variational self-encoder, a single-target generation countermeasure active learning and a multi-target generation countermeasure active learning. Principal component analysis is similar in principle to the self-encoder, where principal component analysis is a vector linear combination using linear algebraic techniques, and the auto-encoder is a vector non-linear combination using deep neural network techniques. The variational self-encoder uses a probability method when calculating the abnormal score, and is more interpretable compared with a self-encoder. Multi-target generative antagonistic active learning uses multiple generators on a single target generative antagonistic active learning basis to improve performance.
In the prior art, the sound classification algorithm of the electrical equipment includes a traditional sound classification algorithm and a sound classification algorithm based on machine learning.
The traditional sound classification algorithm specifically comprises the following steps:
the operation of electrical equipment often produces specific sound, and the traditional sound classification algorithm mainly utilizes an artificial feature extraction method and then utilizes the difference of features to classify the sound. Extracting the time domain characteristics such as short-time energy, short-time average amplitude, short-time zero-crossing rate, short-time autocorrelation function and the like, combining the frequency domain characteristics such as Fourier transform, spectrogram, cepstrum, chromatogram and the like, comprehensively judging by methods such as a similarity matrix and the like to realize sound classification,
the sound classification algorithm based on machine learning specifically comprises the following steps:
most of sound classification algorithms based on machine learning are applied to the aspects of voice recognition, environmental sound classification and the like, and less deep learning classification algorithms are applied to the running sound of power equipment, and the traditional machine learning is to firstly extract sound features (such as Mel frequency spectrum cepstrum xishu), and then utilize k-nearest neighbor algorithm, support vector machine, dictionary learning ([ 4] Sun Yuwei, luo Lingen, chen Jingde, wang Hui, cheng Ge, jiang Xiuchen, circuit breaker mechanical fault diagnosis method based on sound features and improved sparse representation classification), BP neural network ([ 3] Wang Jianjiang, peng Tao. Intelligent operation and maintenance assistant decision technology [ J ] electronic test) based on transformer running sound recognition and the like
The electric equipment operation sound classification algorithm based on deep learning is still in the starting stage at present, and a convolutional neural network and a long-time memory network are utilized to identify the partial discharge of a switch cabinet and identify and classify the voiceprint of a transformer (5 Revqi, research on a high-voltage switch cabinet partial discharge identification method based on ultrasonic signals). The prior art has realized fault diagnosis of motorcycle engine sound through convolutional neural and cyclic neural networks ([ 1] Nie Huilan. Equipment sound fault diagnosis model research based on deep neural networks [ D ]. Hunan Tan university),
most of the existing electrical equipment sound monitoring technologies adopt a traditional sound monitoring method, the efficiency is low, the difficulty is high, and a machine learning method needs large data scale and many training times and is still in a starting stage.
The sound classification algorithm has been widely used in the fields of speech recognition, environmental sound recognition and the like, and is currently less applied to the operation sound classification of electrical equipment. In the prior art, a Mel frequency spectrum feature and a reverse Mel frequency spectrum feature are fused to be used as a sound data set, a multi-classification support vector machine is used for classifying the sound of a power plant, and the fault is not diagnosed and identified. ([ 2] Di Yongjie, peng Yani, yang Xu, hu Dongyang, wang Xinying power plant equipment voice recognition algorithm incorporating MFCC and IMFCC features)
However, the current classification algorithm often fixes the object to be detected first, and detects the existing abnormality of prior, so that the strong learning power of machine learning cannot be fully exerted.
Disclosure of Invention
The invention provides a multi-information fusion power plant electrical equipment classification and abnormity detection method by combining the processes of fault discovery and post-artificial analysis, and solves the problems that only a single object is detected, only known abnormity can be detected, and the operation state of equipment is not classified in the conventional algorithm.
The purpose of the invention is realized by at least one of the following technical solutions.
A multi-information fusion power plant electrical equipment classification and abnormality detection method solves the problems that the existing sound monitoring algorithm only detects a single object, can only detect known abnormality and does not classify the running state of equipment, and comprises the following steps:
s1, acquiring multivariate information of electrical equipment through a power plant monitoring system;
s2, judging the state of the electrical equipment according to the comprehensive criterion;
s3, a preset sound sensor acquires the audio frequency of the operation of the electrical equipment;
s4, training a deep neural network by adopting an electrical equipment state sound sample library, and performing state recognition through the trained deep neural network to obtain a state recognition confidence score;
s5, if the state recognition confidence coefficient is less than or equal to the confidence coefficient threshold value, determining that the electrical equipment has a fault, and executing the step S6, otherwise, finishing the state sound classification of the electrical equipment;
s6, training a deep neural network by adopting an electrical equipment fault sound sample library, and performing known fault abnormity identification through the trained deep neural network to obtain a fault identification confidence score;
s7, if the fault identification confidence score is not greater than the confidence threshold, the electrical equipment is determined to have unknown faults, and the step S8 is executed, otherwise, the known fault abnormal identification of the electrical equipment is completed;
and S8, identifying unknown faults of the electrical equipment, and adding the unknown faults as new abnormal samples into a fault sound sample library of the electrical equipment.
Further, in step S1, the multivariate information includes a rotation speed, an active power output, a reactive power output, a guide vane opening degree, and a water head.
Further, in step S2, the operation state of the electrical device and the corresponding multivariate information set are counted in advance, the correspondence between the operation state and the multivariate information is extracted and obtained as a comprehensive criterion, and the operation state of the electrical device is determined according to the comprehensive criterion and the multivariate information of the electrical device obtained in step S1.
Further, in step S3, by presetting the sound sensor position, the electric device operation sound of the specific object is collected and stored.
Further, in step S4, firstly, training a deep neural network through an electrical device state sound sample library, wherein the deep neural network in the training process adopts a LeNet convolution neural network structure;
the training process comprises sound preprocessing, time-frequency transformation and convolutional neural network training; the sound preprocessing is time length cutting, and time-frequency transformation converts a time domain signal into a spectrogram;
and secondly, acquiring the current running state of the electrical equipment based on the comprehensive criterion of the multivariate information, and performing state recognition on the acquired audio by using the trained deep neural network to obtain the confidence coefficient of the specific running state recognition.
Further, in step S5, the confidence of the identification of the specific operation state of the electrical device is compared with a preset confidence threshold, and if the confidence is greater than the confidence threshold, the identification of the operation state of the electrical device is considered to be normal; if the confidence coefficient is less than or equal to the confidence coefficient threshold value, the audio is regarded as abnormal audio.
Further, in step S6, firstly, training a deep neural network through an electrical device fault sound sample library, where the deep neural network in the training process adopts a LeNet convolution neural network structure;
the training process comprises sound preprocessing, time-frequency transformation and convolutional neural network training; the sound preprocessing is time length cutting, and time-frequency transformation converts a time domain signal into a spectrogram;
and carrying out known fault recognition on the abnormal audio by using the trained deep neural network to obtain an abnormal fault recognition result and a fault recognition confidence coefficient thereof.
Further, in step S7, comparing the confidence of the fault identification with a preset confidence threshold, and if the confidence is greater than the confidence threshold, determining that the known fault identification of the electrical device is normal and the known fault occurs in the electrical device; if the confidence coefficient is less than or equal to the confidence coefficient threshold value, the audio is regarded as unknown abnormal audio.
Further, in step S8, an unknown abnormal audio is identified, and qualitative analysis is performed on the unknown abnormal audio in a manual manner by analyzing the multivariate information data of the electrical device in the abnormal state at this time, and a fault label is printed on the unknown abnormal audio, and the unknown abnormal audio is added to the electrical device fault sound sample library.
Compared with the prior art, the invention has the advantages that:
the method provided by the invention solves the problems of great difficulty and low efficiency caused by deep excavation of sound characteristics of different electrical equipment operation states in the traditional sound monitoring method;
2 the method provided by the invention simultaneously realizes the effects of classification of the running state of the electrical equipment, identification of the audio frequency of the known fault and identification of the audio frequency of the unknown fault, and solves the problems that the current algorithm only detects a single object, only detects the known abnormity and does not classify the running state of the equipment.
Drawings
FIG. 1 is a flowchart illustrating steps of a method for classifying and detecting abnormality of electrical equipment of a power plant with multivariate information fusion according to an embodiment of the present invention;
FIG. 2 is a diagram illustrating a training process of a convolutional neural network according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a LeNet network in the embodiment of the present invention.
Detailed Description
The above combination of identification and tracking methods is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited to the above embodiment, and any other modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be regarded as equivalent replacements within the protection scope of the present invention.
Example (b):
a multi-information fusion power plant electrical equipment classification and abnormality detection method solves the problems that the existing sound monitoring algorithm only detects a single object, can only detect known abnormality and does not classify the running state of equipment, and comprises the following steps as shown in figure 1:
s1, acquiring multivariate information of electrical equipment through a power plant monitoring system;
the multivariate information comprises rotating speed, active power output, reactive power output, guide vane opening and water head.
S2, judging the state of the electrical equipment according to the comprehensive criterion;
and (2) counting the running state of the electrical equipment and the corresponding multivariate information set in advance, extracting to obtain the corresponding relation between the running state and the multivariate information as a comprehensive criterion, and determining the running state of the electrical equipment according to the comprehensive criterion and the multivariate information of the electrical equipment obtained in the step (S1).
S3, a preset sound sensor acquires the audio frequency of the operation of the electrical equipment;
through presetting the position of the sound sensor, the electric equipment operation sound of a specific object is collected and stored in a single-channel WAV form.
S4, training a deep neural network by adopting an electrical equipment state sound sample library, and performing state recognition through the trained deep neural network to obtain a state recognition confidence score;
firstly, training a deep neural network through an electrical equipment state sound sample library, wherein the deep neural network in the training process adopts a LeNet convolution neural network structure;
the training process comprises sound preprocessing, time-frequency transformation and convolutional neural network training; the sound preprocessing is time length cutting, and time-frequency transformation is used for converting a time domain signal into a spectrogram;
and secondly, acquiring the current running state of the electrical equipment based on the comprehensive criterion of the multivariate information, and performing state recognition on the acquired audio by using the trained deep neural network to obtain the confidence coefficient of the specific running state recognition.
S5, if the state recognition confidence coefficient is less than or equal to the confidence coefficient threshold value, determining that the electrical equipment has a fault, and executing the step S6, otherwise, finishing the state sound classification of the electrical equipment;
comparing the confidence coefficient of the specific operation state identification of the electrical equipment with a preset confidence coefficient threshold, and if the confidence coefficient is greater than the confidence coefficient threshold, judging that the operation state identification of the electrical equipment is normal; if the confidence coefficient is less than or equal to the confidence coefficient threshold value, the audio is regarded as abnormal audio.
S6, training a deep neural network by adopting an electrical equipment fault sound sample library, and performing known fault abnormity identification through the trained deep neural network to obtain a fault identification confidence score;
firstly, training a deep neural network through an electrical equipment fault sound sample library, wherein the deep neural network in the training process adopts a LeNet convolution neural network structure
The training process comprises sound preprocessing, time-frequency transformation and convolutional neural network training; the sound preprocessing is time length cutting, and time-frequency transformation converts a time domain signal into a spectrogram;
and carrying out known fault recognition on the abnormal audio by using the trained deep neural network to obtain an abnormal fault recognition result and a fault recognition confidence coefficient thereof.
S7, if the fault identification confidence score is less than or equal to the confidence threshold, the electrical equipment is considered to have unknown faults, and the step S8 is executed, otherwise, the known fault abnormal identification of the electrical equipment is completed;
comparing the confidence coefficient of the fault identification with a preset confidence coefficient threshold, and if the confidence coefficient is greater than the confidence coefficient threshold, determining that the known fault identification of the electrical equipment is normal and the known fault occurs in the electrical equipment; if the confidence coefficient is less than or equal to the confidence coefficient threshold value, the audio is regarded as unknown abnormal audio.
S8, identifying unknown faults of the electrical equipment, carrying out qualitative analysis on the unknown abnormal audio in a manual mode by analyzing the multivariate information data of the electrical equipment in the abnormal state at the moment, and adding the unknown abnormal audio as a new abnormal sample into an electrical equipment fault sound sample library;
and analyzing the multivariate information data of the electrical equipment in the abnormal state at the moment, carrying out qualitative analysis on the unknown abnormal audio in a manual mode, marking a fault label on the unknown abnormal audio, and adding the unknown abnormal audio into a fault sound sample library of the electrical equipment.
And obtaining the operation data of the electrical equipment through the monitoring system, storing the operation data in a wav format, and obtaining the operation state label of the electrical equipment according to the comprehensive criterion. As shown in fig. 3, the LeNet network includes an input layer, a first convolutional layer, a first max-pooling layer, a second convolutional layer, a second max-pooling layer, a flattening layer, a first fully-connected layer, and a second fully-connected layer (i.e., an output layer). The first convolution layer is provided with convolution kernels 5*5, and the number of the convolution kernels is 20; the first largest pooling layer is provided with a window of 2*2 in size, and is sampled downwards; the second convolution layer is provided with convolution kernels 5*5, and the number of the convolution kernels is 50; the setting window size of the second pooling layer is 2*2, and downward sampling is carried out; the length of the first full connection layer setting list is 500; and setting the length of the list to be the number of the states of the electrical equipment to be monitored by the second full-connection layer.
Obtaining equipment sound data through a sound sensor, classifying the collected sound data by utilizing a pre-trained LeNet network, and obtaining classification labels and confidence degrees thereof as shown in the following table:
TABLE 1 LeNet Sound Classification results and confidence thereof
By setting the confidence threshold of normal state recognition to be 85%, STFT-1.WAV, STFT-2.WAV and STFT-3.WAV can be known to be sound data in the first normal operation state, STFT-4.WAV and STFT-5.WAV are sound data in the second normal operation state, and STFT-6.WAV, STFT-7.WAV and STFT-8.WAV are abnormal sound data, so that the classification of the normal state of the equipment is realized.
Example 2:
in this embodiment, the abnormal sound data STFT-6.Wav, STFT-7.Wav, and STFT-8.Wav are obtained by the device normal state classification at the previous stage. The LeNet network comprises an input layer, a first convolution layer, a first maximum pooling layer, a second convolution layer, a second maximum pooling layer, a flattening layer, a first full-connection layer and a second full-connection layer (namely an output layer). The first convolution layer is provided with convolution kernels of 5*5, and the number of the convolution kernels is 20; the first largest pooling layer setting window has the size of 2*2, and is sampled downwards; the second convolution layer is provided with convolution kernels of 5*5, and the number of the convolution kernels is 50; the second pooling layer is provided with a window of 2*2 in size, and is sampled downwards; the length of the first full connection layer setting list is 500; the second full connection layer setting list has the length of the known fault state number of the electrical equipment needing to be monitored.
The pre-trained LeNet network is used for carrying out known fault classification on abnormal sound data obtained in the last stage, and the obtained classification labels and the confidence coefficients thereof are shown in the following table:
TABLE 2 LeNet Sound Fault Classification results and confidence thereof
| Sound data | LeNet classification status | Confidence level |
| STFT-6.WAV | Knock fault | 94.935％ |
| STFT-7.WAV | Knock fault | 95.375％ |
| STFT-8.WAV | Knock fault | 52.363％ |
By setting a fault state identification confidence threshold of 80%, it can be known that STFT-6.WAV and STFT-7.WAV are knock fault (known fault) sound data, and STFT-8.WAV is unknown fault abnormal sound data, so that sound classification of equipment under the known fault is realized, and unknown fault of the electrical equipment is found in a time period of collecting STFT-8.WAV.
Example 3:
in this embodiment, the electrical device stored in the MP3 format runs the sound data. The LeNet network comprises an input layer, a first convolution layer, a first maximum pooling layer, a second convolution layer, a second maximum pooling layer, a flattening layer, a first full-connection layer and a second full-connection layer (namely an output layer). The first convolution layer is provided with convolution kernels 5*5, and the number of the convolution kernels is 20; the first largest pooling layer setting window has the size of 2*2, and is sampled downwards; the second convolution layer is provided with convolution kernels 5*5, and the number of the convolution kernels is 50; the setting window size of the second pooling layer is 2*2, and downward sampling is carried out; the length of the first full connection layer setting list is 500; and the length of the second full-connection layer setting list is the number of the states of the electrical equipment to be monitored.
Obtaining equipment sound data through a sound sensor, classifying the collected sound data by utilizing a pre-trained LeNet network, and obtaining classification labels and confidence degrees thereof as shown in the following table:
TABLE 3 LeNet Sound Classification results and confidence thereof
| Sound data | LeNet classification status | Confidence level |
| STFT-1.MP3 | Normal operation state one | 96.258％ |
| STFT-2.MP3 | Normal operation state one | 98.658％ |
| STFT-3.MP3 | Normal operation state two | 93.499％ |
| STFT-4.MP3 | Normal operation state two | 97.957％ |
| STFT-5.MP3 | Normal operation state two | 73.602％ |
| STFT-6.MP3 | Normal operation state two | 70.934％ |
| STFT-7.MP3 | Normal operation state two | 81.759％ |
By setting the confidence threshold of normal state recognition to be 90%, it can be known that STFT-1.MP3 and STFT-2.MP3 are sound data in the normal operation state, STFT-3.MP3 and STFT-4.MP3 are sound data in the normal operation state, and STFT-5.MP3, STFT-6.MP3 and STFT-7.MP3 are abnormal sound data, and the classification of the normal state of the equipment is realized.
And carrying out fault classification identification on the STFT-5.MP3, the STFT-6.MP3 and the STFT-7. MP3. The LeNet network comprises an input layer, a first convolution layer, a first maximum pooling layer, a second convolution layer, a second maximum pooling layer, a flattening layer, a first full-connection layer and a second full-connection layer (namely an output layer). The first convolution layer is provided with convolution kernels 5*5, and the number of the convolution kernels is 20; the first largest pooling layer is provided with a window of 2*2 in size, and is sampled downwards; the second convolution layer is provided with convolution kernels 5*5, and the number of the convolution kernels is 50; the setting window size of the second pooling layer is 2*2, and downward sampling is carried out; the length of the first full connection layer setting list is 500; the second full-connectivity layer set-up list has a length of the number of known fault conditions of the electrical device to be monitored.
The pre-trained LeNet network is used for carrying out known fault classification on abnormal sound data obtained in the last stage, and obtained classification labels and confidence coefficients thereof are shown in the following table:
TABLE 4 LeNet Sound Fault Classification results and confidence thereof
| Sound data | LeNet classification status | Confidence level |
| STFT-5.MP3 | Knock fault | 86.688％ |
| STFT-6.MP3 | Knock fault | 87.843％ |
| STFT-7.MP3 | Knock fault | 89.090％ |
By setting the fault state identification confidence threshold value of 80%, it can be known that STFT-5.MP3 and STFT-6.MP3 are knock fault (known fault) sound data, and STFT-7.MP3 is unknown fault abnormal sound data, so that sound classification of the equipment under the known fault is realized, and the unknown fault of the electrical equipment is found in the time period of collecting the STFT-7. MP3.
Claims (10)
1. A multi-information fusion power plant electrical equipment classification and abnormality detection method is characterized by comprising the following steps:
s1, acquiring multivariate information of electrical equipment through a power plant monitoring system;
s2, judging the state of the electrical equipment according to the comprehensive criterion;
s3, a preset sound sensor acquires the audio frequency of the operation of the electrical equipment;
s4, training a deep neural network by adopting an electrical equipment state sound sample library, and performing state recognition through the trained deep neural network to obtain a state recognition confidence score;
s5, if the state recognition confidence coefficient is less than or equal to the confidence coefficient threshold value, the electrical equipment is considered to have a fault, the step S6 is executed, and otherwise, the state sound classification of the electrical equipment is completed;
s6, training a deep neural network by adopting an electrical equipment fault sound sample library, and performing known fault abnormity identification through the trained deep neural network to obtain a fault identification confidence score;
s7, if the fault identification confidence score is not greater than the confidence threshold, the electrical equipment is determined to have unknown faults, and the step S8 is executed, otherwise, the known fault abnormal identification of the electrical equipment is completed;
and S8, identifying unknown faults of the electrical equipment, and adding the unknown faults as new abnormal samples into a fault sound sample library of the electrical equipment.
2. The method for classifying and detecting the abnormality of the electrical equipment of the power plant with the multivariate information fusion as claimed in claim 1, wherein in the step S1, the multivariate information comprises a rotating speed, an active power output, a reactive power output, a guide vane opening degree and a water head.
3. The multivariate information fusion power plant electrical equipment classification and abnormality detection method as claimed in claim 1, characterized in that in step S2, the electrical equipment operation state and the corresponding multivariate information set are counted in advance, the correspondence between the operation state and the multivariate information is extracted as a comprehensive criterion, and the electrical equipment operation state is determined according to the comprehensive criterion and the multivariate information of the electrical equipment obtained in step S1.
4. The method for classifying and detecting the electrical equipment of the power plant with the multivariate information fusion as claimed in claim 1, wherein in the step S3, the operation sound of the electrical equipment of the specific object is collected and stored by presetting the position of the sound sensor.
5. The method for detecting the classification and the abnormality of the electrical equipment of the power plant integrated with the multivariate information according to claim 1, wherein in step S4, a deep neural network is trained through an electrical equipment state sound sample library, and the deep neural network in the training process adopts a convolutional neural network structure;
the training process comprises sound preprocessing, time-frequency transformation and convolutional neural network training; the sound preprocessing is time length cutting, and time-frequency transformation converts a time domain signal into a spectrogram;
and secondly, acquiring the current running state of the electrical equipment based on the comprehensive criterion of the multivariate information, and performing state recognition on the acquired audio by using the trained deep neural network to obtain the confidence coefficient of the specific running state recognition.
6. The method for classifying and detecting the electrical equipment of the power plant with the multivariate information fusion as claimed in claim 1, wherein in the step S5, the confidence of the identification of the specific operation state of the electrical equipment is compared with a preset confidence threshold, and if the confidence is greater than the confidence threshold, the identification of the operation state of the electrical equipment is considered to be normal; if the confidence coefficient is less than or equal to the confidence coefficient threshold value, the audio is regarded as abnormal audio.
7. The method for classifying and detecting the electrical equipment of the power plant with the multivariate information fusion as claimed in claim 1, wherein in step S6, a deep neural network is trained through an electrical equipment fault sound sample library, and the deep neural network in the training process adopts a convolutional neural network structure;
the training process comprises sound preprocessing, time-frequency transformation and convolutional neural network training; the sound preprocessing is time length cutting, and time-frequency transformation converts a time domain signal into a spectrogram;
and carrying out known fault recognition on the abnormal audio by using the trained deep neural network to obtain an abnormal fault recognition result and a fault recognition confidence coefficient thereof.
8. The method for classifying and detecting the abnormality of the electrical equipment of the power plant with the multivariate information fusion according to claim 5 or 7, wherein the convolutional neural network comprises an input layer, a convolutional layer, a pooling layer and an output layer.
9. The method for classifying and detecting the electrical equipment of the power plant with the multivariate information fusion as claimed in claim 1, wherein in step S7, the confidence of the fault identification is compared with a preset confidence threshold, and if the confidence is greater than the confidence threshold, the known fault identification of the electrical equipment is considered to be normal, and the known fault occurs in the electrical equipment; if the confidence coefficient is less than or equal to the confidence coefficient threshold value, the audio is regarded as unknown abnormal audio.
10. The multivariate information fusion power plant electrical equipment classification and abnormality detection method as claimed in claim 1, wherein in step S8, unknown abnormal audio is identified, qualitative analysis is performed on the unknown abnormal audio manually by analyzing multivariate information data of the electrical equipment in an abnormal state at the time, a fault label is printed on the unknown abnormal audio, and the unknown abnormal audio is added to an electrical equipment fault sound sample library.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2022/135716 WO2023245990A1 (en) | 2022-06-24 | 2022-11-30 | Classification and abnormality detection method for electrical devices of power plant on basis of multi-information fusion |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210727865 | 2022-06-24 | ||
| CN2022107278652 | 2022-06-24 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115602193A true CN115602193A (en) | 2023-01-13 |
Family
ID=84845469
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211214141.4A Pending CN115602193A (en) | 2022-06-24 | 2022-09-30 | Multi-information fusion power plant electrical equipment classification and abnormality detection method |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN115602193A (en) |
| WO (1) | WO2023245990A1 (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116359642A (en) * | 2023-03-10 | 2023-06-30 | 湖南金烽信息科技有限公司 | Transformer running state 5G intelligent monitoring system and method |
| CN116754934A (en) * | 2023-05-22 | 2023-09-15 | 杭州轨物科技有限公司 | Mechanical characteristic fault diagnosis method for high-voltage circuit breaker |
| CN117711436A (en) * | 2024-02-05 | 2024-03-15 | 中国电子科技集团公司第十五研究所 | Far-field sound classification method and device based on multi-sensor fusion |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117893947A (en) * | 2024-01-23 | 2024-04-16 | 国网内蒙古东部电力有限公司 | Operation dynamic management and control data processing method based on electric power operation recorder |
| CN117914003B (en) * | 2024-03-19 | 2024-05-24 | 沈阳智帮电气设备有限公司 | Intelligent monitoring auxiliary method and system for box-type transformer based on cloud edge cooperation |
| CN118364271B (en) * | 2024-06-19 | 2024-08-30 | 国网山东省电力公司邹城市供电公司 | GIS breaker operation state monitoring method and system based on sound and image characteristics |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200233397A1 (en) * | 2019-01-23 | 2020-07-23 | New York University | System, method and computer-accessible medium for machine condition monitoring |
| CN111370027B (en) * | 2020-03-02 | 2023-04-07 | 乐鑫信息科技（上海）股份有限公司 | Off-line embedded abnormal sound detection system and method |
| CN111666982B (en) * | 2020-05-19 | 2023-04-18 | 上海核工程研究设计院股份有限公司 | Electromechanical equipment fault diagnosis method based on deep neural network |
| CN114323647A (en) * | 2020-09-29 | 2022-04-12 | 斯凯孚公司 | Method and system for fault diagnosis through bearing noise detection |
| CN114530166A (en) * | 2022-01-29 | 2022-05-24 | 国网福建省电力有限公司电力科学研究院 | Transformer on-load tap-changer fault diagnosis method based on background sound texture |
-
2022
- 2022-09-30 CN CN202211214141.4A patent/CN115602193A/en active Pending
- 2022-11-30 WO PCT/CN2022/135716 patent/WO2023245990A1/en unknown
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116359642A (en) * | 2023-03-10 | 2023-06-30 | 湖南金烽信息科技有限公司 | Transformer running state 5G intelligent monitoring system and method |
| CN116359642B (en) * | 2023-03-10 | 2024-01-23 | 湖南金烽信息科技有限公司 | Transformer running state 5G intelligent monitoring system and method |
| CN116754934A (en) * | 2023-05-22 | 2023-09-15 | 杭州轨物科技有限公司 | Mechanical characteristic fault diagnosis method for high-voltage circuit breaker |
| CN116754934B (en) * | 2023-05-22 | 2024-02-23 | 杭州轨物科技有限公司 | Mechanical characteristic fault diagnosis method for high-voltage circuit breaker |
| CN117711436A (en) * | 2024-02-05 | 2024-03-15 | 中国电子科技集团公司第十五研究所 | Far-field sound classification method and device based on multi-sensor fusion |
| CN117711436B (en) * | 2024-02-05 | 2024-04-09 | 中国电子科技集团公司第十五研究所 | Far-field sound classification method and device based on multi-sensor fusion |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2023245990A1 (en) | 2023-12-28 |
| WO2023245990A8 (en) | 2024-01-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115602193A (en) | Multi-information fusion power plant electrical equipment classification and abnormality detection method | |
| CN111626153B (en) | Integrated learning-based partial discharge fault state identification method | |
| CN109034046B (en) | Method for automatically identifying foreign matters in electric energy meter based on acoustic detection | |
| JP7199608B2 (en) | Methods and apparatus for inspecting wind turbine blades, and equipment and storage media therefor | |
| CN110044623B (en) | Intelligent rolling bearing fault identification method based on empirical mode decomposition residual signal characteristics | |
| CN112201260B (en) | Transformer running state online detection method based on voiceprint recognition | |
| CN108169639B (en) | Method for identifying switch cabinet fault based on parallel long-time and short-time memory neural network | |
| CN116825131A (en) | Power plant equipment state auditory monitoring method integrating frequency band self-downward attention mechanism | |
| CN112397054B (en) | Power dispatching voice recognition method | |
| CN102789779A (en) | Speech recognition system and recognition method thereof | |
| CN116230013A (en) | Transformer fault voiceprint detection method based on x-vector | |
| CN116778964A (en) | Power transformation equipment fault monitoring system and method based on voiceprint recognition | |
| Poorna et al. | Emotion recognition using multi-parameter speech feature classification | |
| CN116776075A (en) | Fan blade health monitoring system and monitoring method thereof | |
| CN114352486A (en) | Wind turbine generator blade audio fault detection method based on classification | |
| CN112268615A (en) | Electromechanical equipment vibration signal feature extraction method | |
| CN115510909A (en) | Unsupervised algorithm for DBSCAN to perform abnormal sound features | |
| Qi et al. | Feature classification method of frequency cepstrum coefficient based on weighted extreme gradient boosting | |
| CN118091341A (en) | Cable partial discharge detection method and system based on multi-voiceprint feature spectrogram recognition | |
| Song et al. | A new hybrid method for bearing fault diagnosis based on CEEMDAN and ACPSO-BP neural network | |
| Yang et al. | Transient fault diagnosis of track circuit based on MFCC-DTW | |
| CN116386633A (en) | Intelligent terminal equipment control method and system suitable for noise condition | |
| CN114492543A (en) | Partial discharge classification and identification method based on time-frequency domain characteristics | |
| CN114898527A (en) | Wearable old man falling detection system and method based on voice assistance | |
| Zheng et al. | Circulation pump bearing fault diagnosis research based on BAM-CNN model and CWT processing time-frequency map |
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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20230412 Address after: 511400 No.858, Lianhua Avenue West, Donghuan Street, Panyu District, Guangzhou, Guangdong Applicant after: Energy Storage Research Institute of China Southern Power Grid Peak Regulation and Frequency Regulation Power Generation Co.,Ltd. Address before: 511400 room 208, 100 Dongxing Road, Donghuan street, Panyu District, Guangzhou City, Guangdong Province Applicant before: SOUTHERN POWER GRID PEAK LOAD AND FREQUENCY REGULATION POWER GENERATION Co.,Ltd. |
|
| TA01 | Transfer of patent application right |