CN112201260A - An online detection method of transformer operating state based on voiceprint recognition - Google Patents

Abstract

The invention provides an on-line detection method for transformer operating state based on voiceprint recognition, belonging to the technical field of transformer fault detection. The method includes: S101, performing frame-by-frame processing on the collected voiceprint signals during normal operation of the transformer, and obtaining a feature vector of each frame of the voiceprint signal; S102, fusing each feature value in the feature vector to obtain a comprehensive The evaluation index and the weight of each eigenvalue; S103, if the comprehensive evaluation index obeys the normal distribution, use the 3σ criterion in statistics to obtain the alarm line for the comprehensive evaluation index; S104, for the transformer of unknown operating state, according to the obtained The weight of the characteristic value is used to calculate the corresponding comprehensive evaluation index. If the calculated comprehensive evaluation index exceeds the alarm line for many times in a row, it is determined that the transformer in the unknown operating state is abnormal. By adopting the invention, the detection rate of abnormally operating transformers can be improved, the false detection rate is reduced, and a large number of fault samples are not required.

Description

An online detection method of transformer operating state based on voiceprint recognition

technical field

The invention relates to the technical field of transformer fault detection, in particular to an online detection method of transformer operating state based on voiceprint recognition.

Background technique

In recent years, with the rapid development of UHV backbone grids, the "strong, straight and alternating" power grid structure has brought enormous pressure to the safe and stable operation of the power grid. Some UHV equipment has high technical complexity and is still in a period of unstable quality. Once a failure occurs, it will have a great impact on production and life. The traditional detection methods mainly include: oil chromatography detection method, ultrasonic detection method, vibration detection method, etc.; among them,

The oil chromatography detection method judges the type and severity of the fault by detecting the content of the gas components in the oil. Since there is a certain time from the occurrence of the fault to the change of the gas components in the oil, this method is not timely;

Ultrasonic detection method is currently mainly used to detect and locate partial discharge faults, and only has a good ability to identify partial discharge faults, and has limited ability to detect other fault modes;

The vibration detection method extracts the vibration signal of the transformer through the sensor attached to the surface of the transformer. This method is a contact detection method. The installation position and number of the sensor directly affect the detection effect of the transformer operating state.

To sum up, due to the limitations of traditional detection methods such as incomplete detection and inconvenience, it is necessary to adopt a new detection method to realize the online detection of transformer operating status.

During the operation of the transformer, voiceprint signals of different frequencies will be generated and radiated outward through the air medium, mainly including the voiceprint signal of the transformer body structure, the voiceprint signal of the cooling device and the sound generated by the on-load tap changer. pattern signal, etc. Among these voiceprint signals, the dominant voiceprint signal of the transformer body has very obvious spectral characteristics, that is, the fundamental frequency of the 100Hz waveform is mixed with high-order harmonic components during normal operation. The frequency of the voiceprint signal caused by the operation of the cooling system is generally concentrated in the low frequency range within 100Hz. The voiceprint signal generated when the on-load tap-changer works is generally a high-frequency component. Therefore, the voiceprint signal of the transformer contains important information that can reflect the operating state of the transformer. By analyzing the voiceprint characteristics of the transformer under different operating states, the fault of the transformer can be effectively detected. The voiceprint signal can be collected by the acoustic sensor. The acoustic sensor is a non-contact sensor, which is more flexible than the vibration acceleration sensor. The voiceprint signal can be collected without touching the transformer body.

The existing feature extraction methods mainly include extracting the time-frequency domain features of signals such as characteristic frequency, kurtosis, and mean value, extracting energy features by using wavelet decomposition or wavelet packet decomposition, and extracting Mel-Frequency Cepstral Coefficients (Mel-FrequencyCepstral Coefficients, MFCCs) eigenvectors, etc. These methods can effectively extract the effective information of the transformer vibrating sound signal, but the number and selection of eigenvalues is a difficult point. Too many extracted eigenvalues will reduce the sensitivity of the system, resulting in a low fault detection rate; if too few extracted eigenvalues, it will be impossible to fully grasp the signal characteristics of the vibro-acoustic signal, resulting in poor system stability and false positives. The alarm rate is too high. On the other hand, in feature recognition, the methods used in existing research and patents mainly include vector quantization (VQ), hidden Markov model (HMM), support vector machine (SVM), expert analysis system, correlation analysis, etc. , the biggest problem of these methods is: a large number of fault samples must be used as the premise to build a reasonable classification model, but the fault samples of transformers are often less, in the case of unbalanced data sets, how to carry out effective fault identification, It is the core key problem that needs to be solved urgently.

SUMMARY OF THE INVENTION

The embodiments of the present invention provide an online detection method for transformer operating status based on voiceprint recognition. Through weight analysis, the detection rate of abnormally operating transformers can be improved, the false detection rate can be reduced, and a large number of fault samples are not required, and only the training transformer is required to be normal. The voiceprint signal samples at runtime are sufficient, so this method is effective for all abnormal states (or fault types), greatly improving the scope of application. The technical solution is as follows:

On the one hand, an online detection method of transformer operating state based on voiceprint recognition is provided, and the method includes:

S101, perform frame-by-frame processing on the collected voiceprint signals when the transformer is in normal operation, and obtain a feature vector of each frame of voiceprint signals;

S102, fusing each eigenvalue in the eigenvector to obtain a comprehensive evaluation index and a weight of each eigenvalue;

S103, judging whether the obtained comprehensive evaluation index obeys the normal distribution, if it obeys the normal distribution, then use the 3σ criterion in statistics to obtain an alarm line for the comprehensive evaluation index;

S104, for a transformer in an unknown operating state, calculate its corresponding comprehensive evaluation index according to the weights of the obtained eigenvalues. If the calculated comprehensive evaluation index exceeds the alarm line several times in a row, it is determined that the transformer in the unknown operating state is abnormal .

Further, performing frame-by-frame processing on the collected voiceprint signals during normal operation of the transformer, and obtaining the feature vector of each frame of voiceprint signals includes:

The voiceprint signal collected when the transformer is in normal operation is divided into frames at a preset time interval, and the mean, variance, peak value, peak-to-peak value, root mean square value and skewness of each frame of voiceprint signal are obtained. A total of 13 eigenvalues including index, kurtosis index, margin index, pulse index, peak index, pass frequency value, spectrum energy and main frequency constitute the eigenvector.

Further, the comprehensive evaluation index and the weight of each eigenvalue obtained by fusing each eigenvalue in the eigenvector include:

A1, normalize the m×n-dimensional data matrix composed of m frames of voiceprint signals obtained after frame division processing, where n represents the number of eigenvalues in the eigenvector;

A2, determine the information entropy value e _j of the jth eigenvalue in the normalized data matrix, and determine the information utility value d _{j of the jth eigenvalue according to the obtained information entropy value e j ;}

A3, according to the obtained information utility value d _j , determine the weight W _j of the jth eigenvalue;

A4: Determine a comprehensive evaluation index according to the weights of all the obtained eigenvalues.

Further, the normalized expression is:

where x _ij is the j-th eigenvalue of the i-th frame voiceprint signal before normalization, y _ij is the j-th eigenvalue of the i-th frame voiceprint signal after normalization, x _{j min} , x _{j max} represents the minimum and maximum value of the j-th eigenvalue in the data matrix, respectively.

Further, the information entropy value e _j is expressed as:

表示y_ij占归一化处理后m帧声纹信号第j个特征值之和的比重，y_ij为归一化处理后的第i帧声纹信号的第j个特征值。in,

Indicates the proportion of y _ij to the sum of the j-th eigenvalues of the m-frame voiceprint signal after normalization processing, and y _ij is the j-th eigenvalue of the i-th frame voiceprint signal after normalization processing.

Further, the information utility value _dj is expressed as:

d _j =1-e _j .

Further, the weight W _j is expressed as:

Further, if it obeys the normal distribution, then using the 3σ criterion in statistics to obtain the alarm line for the comprehensive evaluation index includes:

If it obeys the normal distribution, the 3σ criterion in statistics is used to obtain the 2σ and 3σ values for the comprehensive evaluation index, and the 2σ and 3σ values are used as the first-level alarm line and the second-level alarm line respectively.

Further, the method also includes:

If it does not obey the normal distribution, the comprehensive evaluation index will be transformed normally, and the 2σ and 3σ values of the comprehensive evaluation index after normal transformation are obtained by using the 3σ criterion in statistics, and the 2σ and 3σ values will be regarded as the first-level values respectively. Alarm line and secondary alarm line.

The beneficial effects brought by the technical solutions provided by the embodiments of the present invention include at least:

In the embodiment of the present invention, through weight analysis, more important feature values can be highlighted, so that they occupy a higher proportion, thereby helping to improve the detection rate of abnormally operating transformers and reducing the false detection rate; The 3σ criterion in science is used to obtain the alarm line for the comprehensive evaluation index. In this way, a large number of fault samples are not needed, but only the voiceprint signal samples when the transformer is in normal operation. Therefore, this method is suitable for all abnormal states (or fault types) All are effective, greatly improving the scope of application.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

1 is a schematic flowchart of an online detection method for transformer operating state based on voiceprint recognition provided by an embodiment of the present invention;

2 is a schematic structural diagram of an online detection system for transformer operating status based on voiceprint recognition provided by an embodiment of the present invention;

3 is a schematic diagram of a training and testing process of an online detection method for transformer operating status based on voiceprint recognition provided by an embodiment of the present invention;

4 is a schematic diagram of an alarm line of a 220kV outdoor transformer in a substation provided by an embodiment of the present invention;

FIG. 5 is a schematic diagram of a detection result of a simulated iron core looseness fault provided by an embodiment of the present invention;

FIG. 6 is a schematic diagram of a detection result of a simulated winding deformation fault provided by an embodiment of the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

As shown in FIG. 1 , an embodiment of the present invention provides an online detection method for a transformer operating state based on voiceprint recognition, and the method includes:

S101, perform frame-by-frame processing on the collected voiceprint signals when the transformer is in normal operation, and obtain a feature vector of each frame of voiceprint signals;

S102, fusing each eigenvalue in the eigenvector to obtain a comprehensive evaluation index and a weight of each eigenvalue;

S103, judging whether the obtained comprehensive evaluation index obeys the normal distribution, if it obeys the normal distribution, then use the 3σ criterion in statistics to obtain an alarm line for the comprehensive evaluation index;

S104, for a transformer in an unknown operating state, calculate its corresponding comprehensive evaluation index according to the weights of the obtained eigenvalues. If the calculated comprehensive evaluation index exceeds the alarm line several times in a row, it is determined that the transformer in the unknown operating state is abnormal .

The on-line detection method for transformer operating state based on voiceprint recognition according to the embodiment of the present invention can highlight more important feature values through weight analysis and make them occupy a higher proportion, thereby helping to improve the abnormal operation of transformers. The detection rate reduces the false detection rate; and the 3σ criterion in statistics is used to obtain the alarm line for the comprehensive evaluation index. In this way, a large number of fault samples are not required, and only the voiceprint signal samples of the normal operation of the transformer need to be trained. Therefore, the method is effective for all abnormal states (or fault types), greatly improving the scope of application.

The online detection method for transformer operating state based on voiceprint recognition provided by the embodiment of the present invention is implemented by the online detection system for transformer operating state based on voiceprint recognition shown in FIG. and a computer, wherein the computer is provided with a data storage and analysis module programmed by labview for executing S101-S104.

In this embodiment, the model of the acoustic sensor can be GRAS 40PP CCP Free-field QCMicrophone, which is used to collect the voiceprint signal of the transformer, so as to realize real-time and effective detection of the operating state of the transformer in a non-contact way; The sampling frequency is 51.2kHz and the precision is 24 bits. The acoustic sensors are respectively arranged around the transformer, are connected to the input end of the data acquisition card, and use multi-channel synchronous acquisition. The output end of the data acquisition card is connected to the network port of the computer for data display, storage and analysis.

In the specific implementation of the above-mentioned method for online detection of transformer operating state based on voiceprint recognition, further, the collected voiceprint signals during normal operation of the transformer are processed in frames, and the average value of each frame of voiceprint signals is obtained. The eigenvectors include:

The voiceprint signal collected when the transformer is in normal operation is divided into frames at a preset time interval (for example, 1s), and the mean, variance, peak value, peak-to-peak value, and mean square of the voiceprint signal of each frame are obtained. A total of 13 eigenvalues including root value, skewness index, kurtosis index, margin index, pulse index, peak index, pass frequency value, spectral energy and dominant frequency constitute an eigenvector.

In this embodiment, S101-S103 execute the training process shown in FIG. 3 . First, a large number of voiceprint signals during normal operation of the transformer are collected as training samples, and then the training set data composed of the training samples is divided into a frame with 1s as a frame. Frame processing to obtain the feature vector composed of the above 13 feature values. In this way, the sub-frame interval can be shortened, the amount of data can be increased, and richer voiceprint information can be obtained without losing spectral features. Assuming that m seconds of transformer voiceprint signals are collected in total, there are m frames of voiceprint signals in total, and the above 13 eigenvalues are calculated for each frame, so that an m×n data matrix can be obtained, where n represents the feature in the eigenvector. Number of values, n=13.

In this embodiment, the importance of different eigenvalues is different, which means that the contribution value of different eigenvalues to evaluating the operating state of the transformer is different, so the importance of different eigenvalues needs to be evaluated, and the evaluation criteria can be used The larger the weight, the greater the contribution of the eigenvalue to the evaluation of the operating state of the transformer, and vice versa. In this application, the entropy weight method can be used to analyze each eigenvalue (ie: 13 eigenvalues) in the eigenvector to obtain the weight of the 13 eigenvalues. Finally, according to the obtained weight of each eigenvalue, all the eigenvalues Fusion into a comprehensive evaluation index (MVF), which can include the following steps:

A1, normalize the m×n-dimensional data matrix composed of m frames of voiceprint signals obtained after frame division processing, where n represents the number of eigenvalues in the eigenvector;

In this embodiment, the normalized expression is:

where x _ij is the j-th eigenvalue of the i-th frame voiceprint signal before normalization, y _ij is the j-th eigenvalue of the i-th frame voiceprint signal after normalization, x _{j min} , x _{j max} represents the minimum value and maximum value of the jth eigenvalue in the data matrix, respectively, the maximum value of i is m, and the maximum value of j is 13.

A2, determine the information entropy value e _j of the jth eigenvalue in the normalized data matrix, and determine the information utility value d _{j of the jth eigenvalue according to the obtained information entropy value e j ;}

In this embodiment, the information entropy value e _j is expressed as:

表示y_ij占归一化处理后m帧声纹信号第j个特征值之和的比重，y_ij为归一化处理后的第i帧声纹信号的第j个特征值。in,

Indicates the proportion of y _ij to the sum of the j-th eigenvalues of the m-frame voiceprint signal after normalization processing, and y _ij is the j-th eigenvalue of the i-th frame voiceprint signal after normalization processing.

In this embodiment, the information utility value d _j is expressed as:

d _j =1-e _j .

A3, according to the obtained information utility value d _j , determine the weight W _j of the jth eigenvalue;

In this embodiment, the weight W _j is expressed as:

It can be seen from the expression of the weight W _j that the cumulative sum of the weight W _j is 1.

A4, according to the weights of all the obtained eigenvalues, determine the comprehensive evaluation index (MVF), where the MVF is expressed as:

In the specific embodiment of the above-mentioned method for online detection of transformer operating state based on voiceprint recognition, further, if the normal distribution is obeyed, using the 3σ criterion in statistics to obtain the alarm line for the comprehensive evaluation index includes:

If it obeys the normal distribution, the 3σ criterion in statistics is used to obtain the 2σ and 3σ values for the comprehensive evaluation index, and the 2σ and 3σ values are used as the first-level alarm line and the second-level alarm line respectively.

In this embodiment, the K-S test method can be used to judge whether the obtained comprehensive evaluation index (MVF) obeys the normal distribution. If it obeys the normal distribution, the 3σ criterion in statistics is used to obtain the 2σ and 3σ values of the comprehensive evaluation index. , and take the 2σ and 3σ values as the first-level alarm line and the second-level alarm line respectively;

If it does not obey the normal distribution, the comprehensive evaluation index will be transformed normally, and the 2σ and 3σ values of the comprehensive evaluation index after normal transformation are obtained by using the 3σ criterion in statistics, and the 2σ and 3σ values will be regarded as the first-level values respectively. Alarm line and secondary alarm line.

The 220kV outdoor transformer alarm line of a substation obtained according to the above method is shown in Figure 4, in which high and low loads can be determined according to the load diagram of the transformer.

In this embodiment, through S101-S103, an alarm line has been established for the normal operation state of the transformer. For the transformer with unknown operating state, perform the test process as shown in Figure 3: collect the voiceprint signal of the transformer in the unknown operating state as the test sample, and divide the test set data composed of the test sample into a frame with 1s as a frame. 13 eigenvalues of the test set data, and then fuse the 13 eigenvalues into one comprehensive evaluation index according to the weight of the training set. If the calculated comprehensive evaluation index exceeds the first-level alarm line for many times, it will enter the observation state. If the secondary alarm line is exceeded for many times, it is determined that the transformer in the unknown operating state is abnormal.

In order to verify the effectiveness of the online detection method and system for transformer operating state based on voiceprint recognition provided by the embodiment of the present invention, the method and system are applied to the detection of iron core looseness faults and winding deformation fault detection, and the simulation results of detection of iron core looseness faults are obtained. As shown in Figure 5, the simulation results of winding deformation fault detection are shown in Figure 6.

To sum up, the method and system for online detection of transformer operating status based on voiceprint recognition provided by the embodiments of the present invention have the following effective effects:

1) The detection method provided by the present invention is real-time, capable of feeding back abnormal information in time when an abnormality occurs in the transformer, so as to avoid additional losses and major disasters;

2) The detection method provided by the present invention can perform real-time and effective detection on the operating state of the transformer without contacting the transformer body, can replace manual inspection to a certain extent, and reduce human errors while improving work efficiency;

3) The detection method provided by the present invention can highlight more important eigenvalues through weight analysis and make them occupy a higher proportion, thereby improving the detection rate of abnormally operating transformers, reducing the false detection rate, and improving the stability of the system;

4) The detection method based on the 3σ criterion provided by the present invention does not require a large number of fault samples, but only needs to train the voiceprint signal samples during normal operation of the transformer. Therefore, the method is effective for all abnormal states (or fault types), and greatly The applicability of the method has been improved.

The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within the range.

Claims (9)

1. a transformer operating state online detection method based on voiceprint recognition, is characterized in that, comprises: S101, perform frame-by-frame processing on the collected voiceprint signals when the transformer is in normal operation, and obtain a feature vector of each frame of voiceprint signals; S102, fusing each eigenvalue in the eigenvector to obtain a comprehensive evaluation index and a weight of each eigenvalue; S103, judging whether the obtained comprehensive evaluation index obeys the normal distribution, if it obeys the normal distribution, then use the 3σ criterion in statistics to obtain an alarm line for the comprehensive evaluation index; S104, for a transformer in an unknown operating state, calculate its corresponding comprehensive evaluation index according to the weights of the obtained eigenvalues. If the calculated comprehensive evaluation index exceeds the alarm line several times in a row, it is determined that the transformer in the unknown operating state is abnormal . 2. the transformer operating state online detection method based on voiceprint recognition according to claim 1, is characterized in that, the described voiceprint signal during normal operation of the collected transformer is divided into frames, and each frame of sound is obtained. The eigenvectors of the wrinkle signal include: The voiceprint signal collected when the transformer is in normal operation is divided into frames at a preset time interval, and the mean, variance, peak value, peak-to-peak value, root mean square value and skewness of each frame of voiceprint signal are obtained. A total of 13 eigenvalues including index, kurtosis index, margin index, pulse index, peak index, pass frequency value, spectrum energy and main frequency constitute the eigenvector. 3. The online detection method of transformer operating state based on voiceprint recognition according to claim 2, wherein the weight of each eigenvalue obtained by fusing each eigenvalue in the eigenvector to obtain a comprehensive evaluation index and each eigenvalue comprises: A1, normalize the m×n-dimensional data matrix composed of m frames of voiceprint signals obtained after frame division processing, where n represents the number of eigenvalues in the eigenvector; A2, determine the information entropy value e _j of the jth eigenvalue in the normalized data matrix, and determine the information utility value d _{j of the jth eigenvalue according to the obtained information entropy value e j ;} A3, according to the obtained information utility value d _j , determine the weight W _j of the jth eigenvalue; A4: Determine a comprehensive evaluation index according to the weights of all the obtained eigenvalues. 4. the transformer operating state online detection method based on voiceprint recognition according to claim 3, is characterized in that, the expression of normalization processing is:

Among them, x _ij is the j-th eigenvalue of the i-th frame voiceprint signal before normalization processing, y _ij is the j-th eigenvalue of the i-th frame voiceprint signal after normalization processing, x _jmin , x _jmax Represent the minimum and maximum values of the jth eigenvalue in the data matrix, respectively. 5. the transformer operating state online detection method based on voiceprint recognition according to claim 3 is characterized in that, information entropy value e _j is represented as:

in,

Indicates the proportion of y _ij to the sum of the j-th eigenvalues of the m-frame voiceprint signal after normalization processing, and y _ij is the j-th eigenvalue of the i-th frame voiceprint signal after normalization processing. 6. The transformer operating state online detection method based on voiceprint recognition according to claim 5, is characterized in that, the information utility value d _j is expressed as: d _j =1-e _j . 7. The transformer operating state online detection method based on voiceprint recognition according to claim 6, is characterized in that, weight W _j is expressed as:

8. The on-line detection method of transformer operating state based on voiceprint recognition according to claim 1, characterized in that, if the described obeys a normal distribution, then using the 3σ criterion in statistics to obtain the alarm line for the comprehensive evaluation index comprises the following steps: : If it obeys the normal distribution, the 3σ criterion in statistics is used to obtain the 2σ and 3σ values for the comprehensive evaluation index, and the 2σ and 3σ values are used as the first-level alarm line and the second-level alarm line respectively. 9. The method for online detection of transformer operating state based on voiceprint recognition according to claim 8, wherein the method further comprises: If it does not obey the normal distribution, the comprehensive evaluation index will be transformed normally, and the 2σ and 3σ values of the comprehensive evaluation index after normal transformation are obtained by using the 3σ criterion in statistics, and the 2σ and 3σ values will be regarded as the first-level values respectively. Alarm line and secondary alarm line.

Images