A transformer fault diagnosis method, device and equipment based on vibration signal
By constructing a two-dimensional comparison and longitudinal analysis of the vibration signal characteristics of transformers over multiple time periods, the problem of lack of long-term data and poor interpretability in transformer fault diagnosis is solved, and accurate fault identification is achieved even without complete samples.
Patent Information
- Application Number
- CN202211182047.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-09-27
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-09-27
AI Technical Summary
Existing technologies lack long-term vibration data characteristics and patterns in transformer fault diagnosis, resulting in poor interpretability of machine learning diagnostic models. Furthermore, they require complete fault training samples, making it difficult to achieve accurate online monitoring and condition assessment.
By acquiring vibration signal characteristics of transformers over multiple time periods, including amplitude, odd-even harmonic ratio, high-frequency energy ratio, and spectral complexity, two-dimensional features are constructed and compared to identify candidate transformers. Faulty transformers are then determined through longitudinal comparison. This combined approach of lateral and longitudinal comparisons enables accurate diagnosis even without complete fault training samples.
In the absence of complete fault training samples, it improves the accuracy and reliability of transformer fault diagnosis, and can effectively identify faulty transformers when machine learning algorithms are unavailable.
Smart Images

Figure CN115754507B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of fault diagnosis, and in particular to a transformer fault diagnosis method, device and equipment based on vibration signals. Background Art
[0002] Power transformers play the role of converting, transmitting, distributing electric energy and connecting power lines of various voltage levels in the power system. They are the core equipment of the power grid and are of great significance to the safe and reliable operation of the power system. Once a fault occurs, it will have a huge impact on the power system. The transformer winding and core are the two most important components of the transformer. According to relevant statistics, the transformer winding and core are also the two most faulty components of the transformer. The transformer winding and core will vibrate due to the magnetostrictive effect and the electromotive force when the transformer is running. When the winding and core are defective or faulty, the corresponding vibration signal will also change. Therefore, it is feasible and of great significance to detect transformer faults using vibration signals.
[0003] At present, in the research on using vibration signals to detect the status of transformers, some studies collect transformer vibration data and extract features, then summarize and analyze them, and determine the distribution range by observing the corresponding vibration characteristic values. Features outside this range are considered to be abnormal values, and thus the transformer is determined to be an abnormal transformer. For this method, since researchers have long collected transformer vibration signals through short-term detection, there is a lack of long-term vibration data characteristics and rules, which makes it difficult to provide support for on-site transformer online monitoring and status assessment.
[0004] Some studies have also collected vibration data in normal and abnormal conditions in advance, used machine learning to train diagnostic models, and used the high processing power of computers to distinguish abnormal and normal vibration signals. However, such methods have weak interpretability, which is not conducive to assisting operation and maintenance personnel in making decisions, and usually require more complete sample types and more balanced data samples to obtain a mature model that can be used commercially. Although the diagnostic accuracy of machine learning algorithms is high, in the past two years, domestic transformer voiceprint monitoring projects have been in the ascendant, and fault samples will not be able to support the implementation of deep learning models in the future. Summary of the invention
[0005] In view of this, the embodiments of the present invention provide a transformer fault diagnosis method, device and equipment based on vibration signals, which improve the accuracy of transformer fault diagnosis without perfect fault training samples.
[0006] According to a first aspect, an embodiment of the present invention provides a transformer fault diagnosis method based on vibration signals, the method comprising: obtaining vibration signal characteristics corresponding to a number of transformers, the vibration signal characteristics comprising sub-vibration signal characteristics of a number of different time periods; comparing the vibration signal characteristics of each transformer, and identifying a candidate transformer from each transformer based on an outlier vibration signal characteristic in a first comparison result; comparing the sub-vibration signal characteristics of a current candidate transformer, and performing fault diagnosis on the current candidate transformer based on a second comparison result.
[0007] Optionally, the obtaining of vibration signal features corresponding to several transformers includes: obtaining the amplitude, even-odd harmonic ratio, high-frequency energy ratio and spectrum complexity of the current transformer in several time periods; combining the amplitude and even-odd harmonic ratio of the same time period into an amplitude-even-odd harmonic ratio two-dimensional feature at the corresponding time; combining the amplitude and high-frequency energy ratio of the same time period into an amplitude-high-frequency energy ratio two-dimensional feature at the corresponding time; combining the amplitude and spectrum complexity of the same time period into an amplitude-spectrum complexity two-dimensional feature at the corresponding time; wherein the amplitude, even-odd harmonic ratio, high-frequency energy ratio, spectrum complexity, amplitude-even-odd harmonic ratio two-dimensional feature, amplitude-high-frequency energy ratio two-dimensional feature and amplitude-spectrum complexity two-dimensional feature corresponding to each time period are used as the sub-vibration signal features of each time period.
[0008] Optionally, the method also includes: taking the average value of the numerical points corresponding to the same amplitude in the amplitude-odd-even subharmonic ratio two-dimensional feature, the amplitude-high-frequency energy ratio two-dimensional feature and the amplitude-spectral complexity two-dimensional feature, and using the average value point to update the amplitude-odd-even subharmonic ratio two-dimensional feature, the amplitude-high-frequency energy ratio two-dimensional feature and the amplitude-spectral complexity two-dimensional feature.
[0009] Optionally, the vibration signal characteristics come from several different signal acquisition channels, and the vibration signal characteristics of each transformer are compared, and a candidate transformer is identified from each transformer based on the outlier vibration signal characteristics in the first comparison result, including: comparing the sub-vibration signal characteristics of the current transformer in all time periods collected from the current signal acquisition channel with the sub-vibration signal characteristics of other transformers in all time periods collected from the current signal acquisition channel, one by one according to feature types, to obtain the first comparison result; based on the outliers in the sub-vibration signal characteristics of all time periods in the first comparison result, determining whether the current signal acquisition channel is an abnormal channel; if the current signal acquisition channel is an abnormal channel, recording the current transformer as the candidate transformer.
[0010] Optionally, the sub-vibration signal features of the current candidate transformer are compared, and fault diagnosis of the current candidate transformer is performed based on the second comparison result, including: comparing the sub-vibration signal features of each time period collected by the current abnormal channel of the current candidate transformer according to the same feature type, and if a migrated sub-vibration signal feature appears, marking the current abnormal channel as a fault; traversing all abnormal channels of the current candidate transformer to perform sub-vibration signal feature comparison, if each of the abnormal channels is marked with the fault mark after comparison, and all signal acquisition channels of the current candidate transformer are the abnormal channels, then the current candidate transformer is determined to be a faulty transformer; if no migrated sub-vibration signal feature appears in all abnormal channels of the current candidate transformer after feature comparison, then the current candidate transformer is determined to be a normal transformer.
[0011] Optionally, the sub-vibration signal features of each time period collected by the current abnormal channel of the current candidate transformer are compared according to the same feature type, including: curve fitting the two-dimensional sub-vibration signal features of each time period collected by the current abnormal channel of the current candidate transformer to obtain several two-dimensional feature curves of each feature type in each time period; determining a reference time period for the feature type currently being compared; successively calculating the average deviation between the two-dimensional feature curves of other time periods and the two-dimensional feature curve of the reference time period; and using the average deviation corresponding to the two-dimensional feature curves of other time periods to determine whether migration occurs in the sub-vibration signal features of other time periods in the feature type currently being compared.
[0012] Optionally, the method further includes: for the sub-vibration signal characteristics of the current characteristic type of each transformer, traversing and executing the steps of sequentially calculating the average deviation between the two-dimensional characteristic curves of other time periods and the two-dimensional characteristic curve of the reference time period; and determining the migration threshold of the current characteristic type using the average deviation corresponding to the two-dimensional characteristic curves of other time periods of each transformer obtained.
[0013] According to a second aspect, an embodiment of the present invention provides a transformer fault diagnosis device based on vibration signals, the device comprising: a feature extraction module, used to obtain vibration signal features corresponding to a number of transformers, the vibration signal features comprising sub-vibration signal features of a number of different time periods; a horizontal comparison module, used to compare the vibration signal features of each transformer, and identify a candidate transformer from each transformer based on the outlier vibration signal features in a first comparison result; a vertical comparison module, used to compare the sub-vibration signal features of a current candidate transformer, and perform fault diagnosis on the current candidate transformer based on a second comparison result.
[0014] According to the third aspect, an embodiment of the present invention provides a transformer fault diagnosis device based on a vibration signal, comprising: a memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method described in the first aspect or any optional implementation manner of the first aspect by executing the computer instructions.
[0015] According to a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the method described in the first aspect or any optional implementation manner of the first aspect.
[0016] The technical solution provided by this application has the following advantages:
[0017] The technical solution provided by the present application first obtains the vibration signal characteristics that change significantly when the transformer fails, and the vibration signal characteristics of each transformer are generated by vibration signals collected in multiple different signal collection time periods, thereby including sub-vibration signal characteristics of several different time periods. Then, the transformers are compared, and the comparison parameters are the vibration signal characteristics of each transformer, that is, the sub-vibration signal characteristics of each transformer in the full time period are compared. According to the first comparison result, the vibration signal characteristics of the outliers are found, so that one or more candidate transformers with obvious differences from other transformers are identified according to the vibration signal characteristics of the outliers. Afterwards, considering that the reason for the abnormality of the transformer vibration signal may be caused by the uniqueness of the transformer itself in addition to the transformer failure, this embodiment compares the sub-vibration signal characteristics of each candidate transformer in multiple time periods according to the time dimension again. If the abnormality of the transformer vibration signal is caused by its own uniqueness, the difference between the sub-vibration signal characteristics of different time periods will not be very large. If the abnormality of the transformer vibration signal is caused by a fault, there will be obvious differences between the sub-vibration signal characteristics of different time periods, so that through longitudinal comparison, the faulty transformer that actually fails can be further determined from the candidate transformers. Through the above steps, accurate diagnosis of transformer faults was achieved without complete fault training samples and without the help of machine learning algorithms. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] The features and advantages of the present invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the present invention in any way. In the accompanying drawings:
[0019] Figure 1 A schematic diagram of the steps of a transformer fault diagnosis method based on vibration signals in one embodiment of the present invention is shown;
[0020] Figure 2 A schematic diagram showing the comparison of the amplitude-odd-even subharmonic ratio two-dimensional characteristics with and without averaging in one embodiment of the present invention is shown;
[0021] Figure 3 A schematic diagram showing a comparison of fitting two-dimensional characteristic curves of amplitude-odd-even subharmonic ratio two-dimensional characteristics by averaging and not averaging in one embodiment of the present invention;
[0022] Figure 4 A schematic flow chart of a transformer fault diagnosis method based on vibration signals in one embodiment of the present invention is shown;
[0023] Figure 5 A schematic structural diagram of a transformer fault diagnosis device based on vibration signals in one embodiment of the present invention is shown;
[0024] Figure 6 A schematic structural diagram of a transformer fault diagnosis device based on vibration signals in one embodiment of the present invention is shown. DETAILED DESCRIPTION
[0025] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.
[0026] See also Figure 1 In one embodiment, a transformer fault diagnosis method based on vibration signals specifically comprises the following steps:
[0027] Step S101: Acquire vibration signal characteristics corresponding to a number of transformers, where the vibration signal characteristics include sub-vibration signal characteristics in a number of different time periods.
[0028] Step S102: comparing the vibration signal features of each transformer, and identifying a candidate transformer from each transformer based on the outlier vibration signal features in the first comparison result.
[0029] Step S103: comparing the sub-vibration signal features of the current candidate transformer, and performing fault diagnosis on the current candidate transformer based on the second comparison result.
[0030] Specifically, firstly, the vibration signal characteristics that change more obviously when the transformer fails are obtained, and the vibration signal characteristics of each transformer are generated by the collected vibration signals of multiple different signal collection time periods, thereby including sub-vibration signal characteristics of several different time periods. For example: the vibration signals of 4 time periods are collected in two years, and the length of each time period is 3 months. The vibration signals collected in each time period can extract sub-vibration signal characteristics, and the sub-vibration signal characteristics of the 4 time periods constitute the vibration signal characteristics of the full time period. Then the first comparison is performed. The first comparison is a horizontal comparison between transformers. The comparison parameter is the vibration signal characteristics of each transformer, that is, the sub-vibration signal characteristics of each transformer in the full time period are compared. If a transformer has a fault, the vibration signal characteristics of the faulty transformer will stand out due to the outlier. According to the first comparison result, the outlier vibration signal characteristics are found, so as to identify one or more candidate transformers with obvious differences from other transformers according to the outlier vibration signal characteristics.
[0031] Afterwards, considering that the reason for the abnormal transformer vibration signal may be caused by the uniqueness of the transformer itself in addition to the transformer failure, this embodiment, for each candidate transformer, performs a second longitudinal comparison of the sub-vibration signal features of each candidate transformer in multiple time periods according to the time dimension. For example: for the current candidate transformer, the sub-vibration signal features of the four time periods of the current candidate transformer are compared in pairs, and the longitudinal comparison method is compared according to the time dimension. The specific comparison method is only used as an example and is not limited to this. If the abnormality of the transformer vibration signal is caused by its own uniqueness, the difference between the sub-vibration signal features of different sampling time periods will not be very large. If the abnormality of the transformer vibration signal is caused by a fault, there will be obvious differences between the sub-vibration signal features of different time periods, so that through longitudinal comparison, the faulty transformer that actually fails is further determined from the candidate transformer based on the second comparison result. Through the above steps, accurate diagnosis of transformer faults is achieved without complete fault training samples and without the help of machine learning algorithms.
[0032] Specifically, in one embodiment, the above step S101 specifically includes the following steps:
[0033] Step 1: Obtain the amplitude, odd-even harmonic ratio, high-frequency energy ratio and spectrum complexity of the current transformer in several time periods.
[0034] Step 2: The amplitude and odd-even subharmonic ratio of the same time period are combined into a two-dimensional feature of amplitude-odd-even subharmonic ratio according to the corresponding time.
[0035] Step 3: The amplitude and high-frequency energy ratio of the same time period are combined into a two-dimensional feature of amplitude-high-frequency energy ratio according to the corresponding time.
[0036] Step 4: The amplitude and spectrum complexity of the same time period are combined into amplitude-spectrum complexity two-dimensional features according to the corresponding time.
[0037] Among them, the amplitude, even-odd subharmonic ratio, high-frequency energy ratio, spectrum complexity, amplitude-even-odd subharmonic ratio two-dimensional feature, amplitude-high-frequency energy ratio two-dimensional feature and amplitude-spectrum complexity two-dimensional feature corresponding to each time period are used as sub-vibration signal features of each time period.
[0038] Specifically, the embodiment of the present invention decides to use amplitude, odd-even subharmonic ratio, high frequency energy ratio and spectrum complexity as features.
[0039] The calculation formula of odd-even harmonic ratio R1 is as follows:
[0040]
[0041] Where A i represents the amplitude of the ith harmonic of 50 Hz, and N represents the number of all 50 Hz harmonics within 2000 Hz.
[0042] Whether it is the core or the winding, when they are excited and vibrate, the vibration signal generated should theoretically be 100Hz, and the harmonics should also only have even harmonics. Therefore, odd harmonics themselves mean that there are other vibration sources or abnormalities in the core and windings at the source of the vibration. Although in practice, due to the complex structure of the power transformer, even a transformer in good condition often has certain odd harmonics in the vibration signal transmitted from the inside, this does not hinder the scientific nature of using the ratio of odd harmonics to even harmonics as a characteristic. Research has found that when DC bias magnetization and loose core occur, 50Hz odd harmonics begin to appear and increase, so the odd-even harmonic ratio has the function of reflecting the vibration characteristics of the transformer.
[0043] The high frequency energy ratio R2 is calculated as follows:
[0044]
[0045] Where A i represents the amplitude of the ith harmonic of 50 Hz, and N represents the number of all 50 Hz harmonics within 2000 Hz.
[0046] The transformer vibration signal is mainly within 1000Hz, and the components above 1000Hz are basically attenuated to 0, which is difficult to be sensed by the sensor. This feature is also verified in the data collected on site. However, in practice, some high-order harmonics occasionally exist. Therefore, this embodiment decides to limit the frequency range to within 2000Hz.
[0047] In addition, when using the high-frequency energy ratio parameter, the existing technology often divides high frequencies into frequencies of 750 Hz and above. However, according to the on-site spectrum, it can be found that from 700 Hz upwards, the energy has basically decayed to 0. After measurement, if 750 Hz is used as the standard, the high-frequency energy ratio will drop to 10 -2 The order of magnitude is not conducive to data statistics and analysis. The present application has found that when a fault such as a loose core clamp occurs, the energy of frequencies of 400 Hz and above will increase. Therefore, the embodiment of the present invention defines high frequency as frequencies of 400 Hz and above, that is, i=8, and performs statistics and analysis on the vibration data.
[0048] The formula for calculating spectral complexity is as follows:
[0049]
[0050]
[0051] Where R i is the amplitude proportion of the ith component at 50 Hz, H is the spectrum complexity, and the meanings of the other parameters are the same as before.
[0052] Spectral complexity describes the degree of dispersion of the spectrum. The smaller the value, the more concentrated the spectrum energy. When there is only a 100Hz component in the signal, R i The second component in is 1, the rest are 0, and the spectrum complexity H is 0. During normal operation, according to the data collected on site, compared with the two parameters of odd-even harmonic ratio and high-frequency energy ratio, the spectrum complexity is generally larger, which also reflects that under the normal operation of the power transformer, the spectrum is generally relatively scattered. When various faults occur in the transformer, with the increase of odd harmonics and the increase of high-frequency energy, the characteristics of the spectrum will also change. Therefore, the spectrum complexity can be used to characterize the vibration characteristics of the power transformer.
[0053] At the same time, considering that the amplitude can represent the energy at this time to a certain extent, if an abnormality occurs, the spectrum under the amplitude will change, thereby causing changes in the other three characteristic quantities. If the amplitude and the other characteristic quantities are used to form two-dimensional data, the distribution of the two-dimensional data will also change. Therefore, in addition to using the amplitude as the sub-vibration signal feature, the embodiment of the present invention also observes the distribution formed by the three groups of two-dimensional data of amplitude-odd harmonic ratio, amplitude-high frequency energy ratio and amplitude-spectrum complexity (hereinafter referred to as two-dimensional features) to observe the data, and observe the distribution of the other three characteristic quantities at the same amplitude point of different transformers.
[0054] Taking the amplitude-odd harmonic ratio two-dimensional feature as an example, the vibration signal in the time domain changes with time and the odd-even harmonic ratio is also a numerical point distribution about time. Then, at the same sampling moment, each odd-even harmonic ratio will correspond to the amplitude at that moment. Then, with the amplitude as the horizontal coordinate and the odd-even harmonic ratio as the vertical coordinate, a two-dimensional feature of the odd-even harmonic ratio changing with the amplitude is established. The amplitude-high-frequency energy ratio two-dimensional feature and the amplitude-spectral complexity are similar, and are both feature distributions about the amplitude change. The one-dimensional feature (one-dimensional feature refers to the distribution of amplitude, odd-even harmonic ratio, high-frequency energy ratio and spectrum complexity changing with sampling time) and two-dimensional feature selected and created by this embodiment can deeply analyze the changes in the vibration signal when the transformer fails, and has a clear and explainable physical meaning, and is also convenient for technicians to conduct subsequent research on fault tracing, fault location, etc. during data analysis.
[0055] Specifically, in one embodiment, the transformer fault diagnosis method based on vibration signals provided by the embodiment of the present invention further includes the following steps:
[0056] Step 5: Take the average value of the numerical points corresponding to the same amplitude in the amplitude-odd-even subharmonic ratio two-dimensional feature, the amplitude-high-frequency energy ratio two-dimensional feature and the amplitude-spectral complexity two-dimensional feature, and use the average value point to update the amplitude-odd-even subharmonic ratio two-dimensional feature, the amplitude-high-frequency energy ratio two-dimensional feature and the amplitude-spectral complexity two-dimensional feature.
[0057] Specifically, since the amplitude points of the vibration signal have repeated values, for example, the amplitude is 1m / s 2 It will be detected multiple times in the vibration signal collection, and each time it is detected, the other three feature quantities will also be detected. Therefore, the distribution of the amplitude-odd-even harmonic ratio two-dimensional feature, the amplitude-high-frequency energy ratio two-dimensional feature and the amplitude-spectral complexity two-dimensional feature shown in the figure is that there are multiple points (corresponding to multiple vertical coordinate values) under the same amplitude point (one horizontal axis). In order to more clearly show the change trend relationship between the feature quantities, the embodiment of the present invention takes the average value of the multiple feature value points corresponding to each amplitude point to achieve this effect. The comparison before and after taking the average value is as follows: Figure 2 As shown, taking the average does not affect the overall distribution.
[0058] Specifically, in one embodiment, the vibration signal features collected by the embodiment of the present invention come from several different signal collection channels, and the above step S102 specifically includes the following steps:
[0059] Step six: Compare the sub-vibration signal features of all time periods collected by the current transformer from the current signal acquisition channel with the sub-vibration signal features of all time periods collected by other transformers from the current signal acquisition channel one by one according to feature types to obtain a first comparison result.
[0060] Step 7: Based on the outliers in the sub-vibration signal features of all time periods in the first comparison result, determine whether the current signal acquisition channel is an abnormal channel.
[0061] Step 8: If the current signal acquisition channel is an abnormal channel, the current transformer is recorded as a candidate transformer.
[0062] Specifically, in this embodiment, in order to further improve the accuracy of transformer fault diagnosis, multiple signal acquisition channels are set for each transformer, so as to collect vibration signals at different positions of the transformer, comprehensively analyze the signals of each channel for fault diagnosis, and improve the reliability of diagnosis. Assuming that the embodiment of the present invention adopts two signal acquisition channels, the specific method of the first horizontal comparison is as follows:
[0063] First, compare the sub-vibration signal features of each transformer with the sub-vibration signal features of the same type in the full time period. If it is a one-dimensional feature, such as an amplitude feature, splice the amplitudes of the four sampled time periods into the amplitude of the full time period to obtain the amplitude distribution of the time change of the full time period. For example: Channel 1 of transformer No. 1, the four sampling time periods are August 23, 2021 to September 24, 2021, October 8, 2021 to October 21, 2021, November 28, 2021 to December 16, 2021, and December 29, 2021 to March 5, 2022. After splicing, the amplitude features from August 23, 2021 to March 5, 2022 are obtained. Then, the amplitude collected by channel 1 of each transformer is processed as above to obtain the amplitude characteristics of each transformer in the whole time period, and then the amplitude characteristics of each transformer are compared with each other to find out the outlier amplitude characteristics, and it is considered that the transformer corresponding to the outlier characteristics and the acquisition channel are abnormal. For example, the amplitude characteristics of channel 1 of transformer No. 1, after being compared with the amplitude characteristics of channel 1 of other transformers, it is found that the amplitude characteristics of channel 1 of transformer No. 1 are outliers, then channel 1 of transformer No. 1 is an abnormal channel, transformer No. 1 is a candidate transformer for further diagnosis, and the processing method of channel No. 2 is similar, and the processing of odd-even harmonic ratio, high-frequency energy ratio and one-dimensional characteristics of spectrum complexity is similar to amplitude, which will not be repeated. The embodiment of the present invention identifies a certain channel of a certain transformer, and once an outlier characteristic appears, the channel is marked as abnormal, and the transformer is marked as a candidate transformer.
[0064] If the features involved in the comparison are two-dimensional features, the two-dimensional features of the same type in each time period are superimposed to obtain the two-dimensional features of the entire time period, and then the comparison is performed. For example, channel 1 of transformer 1 obtains 4 sets of amplitude-odd harmonic ratio two-dimensional features through feature extraction in 4 sampling time periods, and then the 4 sets of amplitude-odd harmonic ratio two-dimensional features are superimposed according to the corresponding amplitude relationship. After superposition, the multiple odd and even harmonic ratios corresponding to each amplitude point are averaged, and the amplitude-odd harmonic ratio two-dimensional features of channel 1 of transformer 1 in the entire time period are obtained. Similarly, the amplitude-odd harmonic ratio two-dimensional features of channel 1 of other transformers are processed in the same way. Then the amplitude-odd harmonic ratio two-dimensional features of each transformer are compared with each other. If an outlier feature appears, the transformer corresponding to the outlier feature is considered to be a candidate transformer, and channel 1 of the transformer currently involved in the comparison is an abnormal channel. For example, after comparing the amplitude-odd harmonic ratio 2D feature of channel 1 of transformer 1 with the amplitude-odd harmonic ratio 2D feature of channel 1 of other transformers, if the amplitude-odd harmonic ratio 2D feature of channel 1 of transformer 1 is outliers, then channel 1 of transformer 1 is an abnormal channel and transformer 1 is a candidate transformer. The comparison process of channel 2 is similar, and the comparison process of amplitude-high frequency energy ratio 2D feature and amplitude-spectral complexity 2D feature is similar to the processing comparison of the above amplitude-odd harmonic ratio 2D feature.
[0065] Through the results of the first horizontal comparison of this embodiment, one or more candidate transformers are selected from a number of transformers, and the signal acquisition channels in which each candidate transformer has abnormalities are recorded. In subsequent steps, based on the current identification results, the signal characteristics of each channel are further integrated to determine whether the candidate transformer is an actual faulty transformer, so as to improve the accuracy of fault diagnosis.
[0066] Specifically, in one embodiment, the above step S103 specifically includes the following steps:
[0067] Step nine: compare the sub-vibration signal features of each time period collected by the current abnormal channel of the current candidate transformer according to the same feature type. If a migrated sub-vibration signal feature appears, mark the current abnormal channel as a fault.
[0068] Step 10: Traverse all abnormal channels of the current candidate transformer and perform sub-vibration signal feature comparison. If each abnormal channel is marked with a fault mark after comparison, and all signal acquisition channels of the current candidate transformer are abnormal channels, the current candidate transformer is determined to be a faulty transformer.
[0069] Step 11: If all abnormal channels of the current candidate transformer do not show the migrated sub-vibration signal feature after feature comparison, the current candidate transformer is determined to be a normal transformer.
[0070] Specifically, this embodiment further performs a longitudinal comparison of the transformer data based on the horizontal comparison between the transformers, so as to determine whether the candidate transformer selected in the above steps 6 to 8 is a real faulty transformer. The specific method is as follows:
[0071] For the abnormal channels of each candidate transformer, the sub-vibration signal features of the same type and different time periods of an abnormal channel are compared with each other. If feature migration occurs, the current abnormal channel is determined to be a faulty channel. For example, assuming that transformer No. 1 is a candidate transformer, channel No. 1 of transformer No. 1 is marked as an abnormal channel after steps six to eight. Taking the amplitude-odd harmonic ratio two-dimensional feature as an example, the amplitude-odd harmonic ratio two-dimensional features of the four sampling time periods of channel No. 1 are compared with each other. If the amplitude-odd harmonic ratio two-dimensional feature of a certain sampling time period has a feature migration phenomenon, channel No. 1 of transformer No. 1 is determined to be a faulty channel, and the comparison process of other sub-vibration signal features of channel No. 1 is the same. If channel No. 2 of transformer No. 1 is also an abnormal channel, it is also marked as a faulty channel after the longitudinal comparison of the above steps, that is, all two signal sampling channels of transformer No. 1 are marked as faulty channels, then it is considered that the signal abnormality of transformer No. 1 is caused by a fault, and transformer No. 1 is a faulty transformer, thereby improving the reliability of fault diagnosis. If both Channel 1 and Channel 2 of Transformer 1 are not faulty channels after longitudinal comparison, it is considered that the abnormal signal of Transformer 1 is caused by its own uniqueness and the transformer is a normal transformer. If only some channels of Transformer 1 are marked as faulty channels and other channels are not marked as faulty channels, it is considered difficult to determine whether Transformer 1 is faulty and requires on-site inspection by technicians.
[0072] Specifically, in one embodiment, the above step nine specifically includes the following steps:
[0073] Step 12: curve fitting is performed on the two-dimensional sub-vibration signal characteristics of each time period collected by the current abnormal channel of the current candidate transformer to obtain a number of two-dimensional characteristic curves of each characteristic type in each time period.
[0074] Step 13: Determine the benchmark time period for the feature type currently being compared.
[0075] Step 14: Calculate the average deviation between the two-dimensional characteristic curves of other time periods and the two-dimensional characteristic curve of the reference time period in sequence.
[0076] Step 15: Using the average deviation corresponding to the two-dimensional characteristic curves of other time periods, determine whether the sub-vibration signal features of other time periods in the feature type currently being compared have migrated.
[0077] Specifically, it can be seen from the analysis part of step nine to step eleven that, with the passage of time, the characteristic quantity of some transformer vibration data will also change. This longitudinal change is combined with the analysis of the outlier abnormal value obtained by horizontal comparison of multiple transformers to verify whether the feature is indeed an abnormal value. However, it is difficult to achieve quantitative analysis if only the difference analysis of the image is performed. Therefore, under the condition that only two-dimensional features are selected to participate in the longitudinal comparison, the embodiment of the present invention performs curve fitting on the two-dimensional sub-vibration signal features of each time period collected by the current abnormal channel of each candidate transformer. When the two-dimensional features of each time period are longitudinally compared, a reference time period is selected for a certain type of feature (the signal and signal features of the reference time period are closest to the normal signal and the normal signal features), and then the average deviation between the two-dimensional feature curves of other time periods and the two-dimensional feature curves of the reference time period is calculated. Finally, the average deviations calculated in each other time period are used to determine whether the sub-vibration signal features of other time periods in the feature type currently being compared are migrated. If migration occurs, it is considered that the two-dimensional feature of the current type is abnormal, and the corresponding abnormal channel is a fault channel. For example: Channel No. 1 of transformer No. 1 is currently performing a longitudinal comparison of the amplitude-odd harmonic ratio two-dimensional characteristics. First, curve fitting is performed on the amplitude-odd harmonic ratio two-dimensional characteristics of the four time periods to obtain four two-dimensional characteristic curves. Then, one of the curves is the curve corresponding to the reference time period, and the average deviations of the remaining three curves and the curve of the reference time period are calculated respectively. The calculated average deviation can be used to quantify which curve has undergone feature migration.
[0078] The formula for calculating the average deviation Rc is as follows:
[0079]
[0080] In the formula, x i is the characteristic quantity corresponding to a certain amplitude in the reference time period, y i is the characteristic quantity corresponding to the same amplitude in some other time period, and N is the number of amplitude points.
[0081] For example, if the average deviation of some other time period exceeds a preset threshold, it is considered that feature migration has occurred. Through the steps of this embodiment, the accuracy of the longitudinal comparison is improved.
[0082] Specifically, the process of fitting the two-dimensional characteristic curves of each time period in this embodiment is as follows:
[0083] Taking the amplitude-odd harmonic ratio two-dimensional feature as an example, for the amplitude-odd harmonic ratio two-dimensional feature after taking the average value, such as Figure 3 As shown in the figure, the regression analysis method is used to fit the two-dimensional characteristics of amplitude-odd harmonic ratio. It is not difficult to find that Figure 3(a) The averaged amplitude-odd and even harmonics are compared with the curve of the two-dimensional feature fitting. Figure 3 (b) There is no obvious difference between the curves without averaging, but the fitting calculation amount after averaging is smaller and the fitting efficiency is higher.
[0084] In addition, in this embodiment, general regression analysis is linear fitting, but often the linear relationship does not quite conform to the characteristics of the data, so sometimes it is necessary to increase the number of fitting models. The embodiment of the present invention evaluates the fitting effect by the mean square error to determine the best number of fitting times. The mean square error (MSE) formula is:
[0085]
[0086] Where m is the number of fitted data points, i is the i-th fitted data point, and y i is the fitted value, x i is the actual value.
[0087] Through 15 times of model fitting, it was found that when the number of times the fitting model was used increased, the corresponding mean square error also decreased. After the 4th time, the trend of the mean square error reduction slowed down. At the same time, taking the average did not affect the corresponding trend. Therefore, in order to avoid overfitting, the embodiment of the present invention decided to set the number of fitting models to 4, and use a 4th-order polynomial to perform curve fitting on the two-dimensional features.
[0088] Specifically, in a real-time example, a transformer fault diagnosis method based on a vibration signal provided by an embodiment of the present invention further includes the following steps:
[0089] Step 16: For each sub-vibration signal feature of the current feature type of each transformer, traverse and execute step 14.
[0090] Step 17: Determine the migration threshold of the current feature type by using the average deviation corresponding to the two-dimensional characteristic curves of each transformer in other time periods.
[0091] Specifically, the embodiment of the present invention further uses the average deviation calculated during the longitudinal comparison to set the migration threshold, thereby further realizing the quantification of the longitudinal comparison. In the identification, each abnormal channel data of each transformer is longitudinally compared. If the average deviation calculated for each channel exceeds the set migration threshold, it is considered that the transformer is faulty.
[0092] This embodiment obtains the two-dimensional features of each transformer, and then executes the above step 14 for the same feature type to obtain the average deviation of the two-dimensional feature curves of each transformer in other time periods relative to the reference time period, and then uses multiple average deviations according to a series of statistical algorithms such as average value and mode to determine the migration threshold of the current feature type. Therefore, in each longitudinal comparison, if the average deviation of the two-dimensional features of a feature type of the current signal channel exceeds its corresponding migration threshold, the current signal channel is considered to be a fault channel.
[0093] Specifically, in order to more clearly describe the technical solution provided by the embodiment of the present invention, the above steps of the present application are explained below with an embodiment of a specific scenario application:
[0094] like Figure 4 As shown, it is a flow chart of the transformer fault diagnosis method based on vibration signal of this embodiment. In this embodiment, 8 transformers are selected for testing, and each transformer is sampled using 2 data channels. The vibration data obtained are from 2021.8.23-2021.9.24, 2021.10.9-2021.10.19, 2021.11.29-2021.12.12 and 2021.12.20-2022.3.5. Then the vibration data of each time period is calculated separately to obtain the sub-vibration signal features of the four time periods, which are amplitude, odd-even subharmonic ratio, high-frequency energy ratio and spectrum complexity. Four one-dimensional features, amplitude-odd-even subharmonic ratio two-dimensional features, amplitude-high-frequency energy ratio two-dimensional features and amplitude-spectrum complexity two-dimensional features.
[0095] Then, for channel 1, the 8 transformers were compared horizontally. The horizontal comparison included one-dimensional feature comparison between different transformers, specifically, 8 sets of amplitude features collected by channel 1 for all 8 transformers, and then the 8 sets of amplitude features were compared with each other. Similarly, the odd-even subharmonic ratio of channel 1 of the 8 transformers, the high-frequency energy ratio of channel 1 of the 8 transformers, and the spectrum complexity of channel 1 of the 8 transformers were also compared. Through comparison, it was found that the distribution range of the amplitude signals of each transformer was roughly similar, the odd-even subharmonic ratio of transformer No. 6 was higher than that of other transformers, and there was an outlier phenomenon. The spectrum complexity of transformer No. 1 was significantly higher than that of other transformers, and there was an outlier phenomenon.
[0096] The horizontal comparison also includes the comparison of two-dimensional features between different transformers, specifically including 8 groups of amplitude-odd harmonic ratio two-dimensional features created by using the amplitude and odd-even harmonic ratio collected by channel 1 for all 8 transformers, and comparing the 8 groups of amplitude-odd harmonic ratio two-dimensional features with each other. Similarly, the comparison of two-dimensional features also includes the comparison of the amplitude-high-frequency energy ratio two-dimensional features of channel 1 of 8 transformers and the comparison of the amplitude-spectrum complexity two-dimensional features of channel 1 of 8 transformers. After comparison, it was found that the amplitude-odd harmonic ratio two-dimensional features of transformers No. 1, No. 4 and No. 6 were higher than those of other transformers, and outliers appeared. The amplitude-high-frequency energy ratio two-dimensional features of transformers No. 4 and No. 6 were higher than those of other transformers, and outliers appeared. The amplitude-spectrum complexity two-dimensional features of transformers No. 7 and No. 8 were significantly lower than those of other transformers, reflecting that the vibration signals of transformers No. 7 and No. 8 were more concentrated, and combined with the other two figures, it was believed that there were more even harmonics distributed below 400Hz in the two channels, which was more in line with the theoretical spectrum characteristics of normal transformers. Therefore, it was believed that the probability of failure of transformers No. 7 and No. 8 was low.
[0097] Similarly, a horizontal comparison of features was performed between the 8 transformers for channel 2, including a one-dimensional feature comparison between different transformers, including comparison of the amplitude of channel 2 of the 8 transformers, comparison of the odd-even harmonic ratio of channel 2 of the 8 transformers, comparison of the high-frequency energy ratio of channel 2 of the 8 transformers, and comparison of the spectrum complexity of channel 2 of the 8 transformers. It was found that for the vibration signal from channel 2, except that the amplitude and odd-even harmonic ratio of transformer No. 6 were significantly higher than those of other transformers, and the spectrum complexity was slightly lower, the distribution range of the characteristic quantities of the vibration signals of the other transformers were relatively similar, so only transformer No. 6 showed an outlier phenomenon.
[0098] Then, the two-dimensional feature comparison between different transformers was carried out, including the comparison of the two-dimensional features of the amplitude-odd harmonic ratio of channel 2 of 8 transformers, the comparison of the two-dimensional features of the amplitude-high-frequency energy ratio of channel 2 of 8 transformers, and the comparison of the two-dimensional features of the amplitude-spectral complexity of channel 2 of 8 transformers. It was found that the two-dimensional feature of the amplitude-odd harmonic ratio of transformer No. 6 was significantly larger than that of other transformers, while the spectral complexity was smaller. At the same time, the two-dimensional feature of the amplitude-high-frequency energy ratio of transformer No. 6 was also relatively small among the same features of various transformers, and except for transformer No. 6, the distribution of each feature quantity of the other transformers was similar.
[0099] In summary, the result of the horizontal comparison is that the comparison results of transformers No. 1, No. 4 and No. 6 show outliers. Therefore, transformers No. 1, No. 4 and No. 6 are candidate transformers, among which the abnormal channel of transformers No. 1 and No. 4 is channel 1, and channel 1 and channel 2 of transformer No. 6 are both abnormal channels.
[0100] Although the transformers are from the same manufacturer and have the same voltage level, the vibration characteristics of each transformer will be more or less different due to the uniqueness of the data caused by manual production of some processes in the transformer production and the different installation positions of the sensors. Therefore, in order to further overcome the impact of such differences, the sub-vibration signal characteristics of the above-mentioned transformers No. 1, No. 4 and No. 6 are longitudinally compared to observe the changes in the two-dimensional distribution in different time periods. If there is no significant change, it can be considered that the probability of failure is small.
[0101] A longitudinal comparison is conducted on the abnormal channels of transformers No. 1, No. 4 and No. 6, comparing the differences of the same transformer, the same channel and the same feature type in different time periods.
[0102] After comparison, it was found that the distribution range of each characteristic distribution of channel 1 of transformer No. 1 did not show obvious changes over time and was roughly distributed within a range; the characteristic quantities such as amplitude-odd harmonic ratio two-dimensional characteristics, amplitude-high-frequency energy ratio two-dimensional characteristics and amplitude-spectrum complexity two-dimensional characteristics of channel 1 of transformers No. 4 and No. 6 obviously increased over time, and characteristic migration occurred; the distribution range of amplitude-high-frequency energy ratio and amplitude-odd harmonic ratio of channel 2 of transformer No. 6 increased over time, and characteristic migration occurred, while amplitude-spectrum complexity remained basically unchanged.
[0103] Since the longitudinal vibration of channel 1 of transformer No. 1 does not change much, it is considered difficult to distinguish whether the outlier value of the vibration signal of channel 1 of transformer No. 1 is due to the uniqueness of the transformer itself or the existence of a fault. Since the vibration data of channel 1 of transformer No. 4 and No. 6 and channel 2 of transformer No. 6 have migration phenomena, it is considered that the probability of fault is relatively high. Therefore, channel 1 of transformer No. 4 and No. 6 and channel 2 of transformer No. 6 are marked as fault channels.
[0104] In summary, it was found that there were problems with the vibration signals of Channel 1 and Channel 2 of Transformer No. 6, and both channels were marked as faulty channels, so Transformer No. 6 was considered to be a faulty transformer. As for Transformer No. 4, only Channel 1 showed abnormality in the horizontal and vertical comparisons. At the same time, it was found that the distribution of Channel 2 of Transformers No. 1 and No. 4 did not show a significant trend of change over time, except that the high-frequency energy ratio of Channel 2 of Transformer No. 4 decreased slightly over time. Therefore, it is believed that the horizontal abnormal value of Channel 1 of Transformer No. 1 is caused by its own uniqueness, while it is difficult to judge whether Transformer No. 4 is abnormal or not.
[0105] The above longitudinal comparison process is a conclusion obtained through intuitive comparison of images. In order to further quantify the longitudinal comparison process during the longitudinal comparison, the longitudinal comparison process is quantified in the following manner:
[0106] The average of multiple characteristic values corresponding to each amplitude point of the three two-dimensional features, namely, the amplitude-odd harmonic ratio two-dimensional feature, the amplitude-high frequency energy ratio two-dimensional feature, and the amplitude-spectral complexity two-dimensional feature, is used to obtain the averaged two-dimensional feature. After averaging, regression analysis is used to fit the data, and the fitting polynomial is a 4th-order polynomial, so as to obtain the two-dimensional characteristic curve corresponding to each two-dimensional feature.
[0107] Before making a longitudinal comparison, by observing the time domain comparison chart, it is found that in the data of the time period of 2021.8.23-2021.9.24, the difference in vibration characteristics of most transformers is relatively small. Then, as time goes by, the vibration characteristics of some transformers begin to shift. Therefore, it is believed that the data in the time period of 2021.8.23-2021.9.24 is likely to be normal data, so the data in this time period is used as the reference time period. The two-dimensional characteristic curves of other time periods are compared with the two-dimensional characteristic curves of the reference characteristic time period to try to divide the normal and abnormal ranges. For example, assuming that the current two-dimensional characteristics of the amplitude-odd harmonic ratio of channel 1 of transformer No. 6 are compared longitudinally, then the two-dimensional characteristic curves of the amplitude-odd harmonic ratio in other time periods are compared with the two-dimensional characteristic curves of the reference time period, and the three average biases corresponding to the three time periods are calculated.
[0108] For the eight transformers, two-dimensional features are calculated, and then each transformer performs longitudinal comparison on channel 1 and channel 2 according to the reference time period, so that the average deviation of the two channels is calculated for each transformer, and three average biases are calculated for each channel. Then, for channel 1 and channel 2, the migration threshold of each feature type is comprehensively determined using the average deviation between the eight transformers.
[0109] During the period from 2021.10.9 to 2021.10.19, due to equipment anomalies, some transformers have no data for this time period, so the vibration amount in the time period from 2021.8.23 to 2021.9.24 is used as the reference time period, and its average deviation from time period 1 (2021.11.29 to 2021.12.12) and time period 2 (2021.12.20 to 2022.3.5) is calculated respectively. From the previous analysis, it can be seen that the probability of transformer No. 6 being abnormal is relatively high, and the anomaly happens to occur within the range of time period 2, so it is used as the standard to divide the average deviation threshold. For the vibration data of the 8 transformers in the substation, 0.4 was set as the threshold of the amplitude-odd-even harmonic ratio; for the amplitude-high-frequency energy ratio, the No. 4 and No. 6 transformers of channel 1 and the No. 4 and No. 7 transformers of channel 2 were more prominent than other transformers, so 0.15 was set as the threshold; for the amplitude-spectral complexity, it was found that the spectral complexity of the 8 transformers did not change much, so it was considered that the spectral complexity of the 8 transformers was within the normal range, and 0.4 and 0.35 were set as the thresholds for channel 1 and channel 2, respectively.
[0110] For example, we now start the longitudinal comparison of the amplitude-odd harmonic ratio two-dimensional features of channel 1 of transformer No. 6. First, the amplitude-odd harmonic ratio two-dimensional features of each time period are fitted into two-dimensional feature curves to obtain four two-dimensional feature curves corresponding to the four time periods. Then, the average deviation is calculated using the curves of the three other time periods and the curve of the reference time period. Then, we check whether there is a deviation exceeding the threshold of 0.4 in the three average deviations. If so, channel 1 of transformer No. 6 is considered to be the fault channel.
[0111] Through the above steps, the technical solution provided by the present application first obtains the vibration signal characteristics that change more obviously when the transformer fails, and the vibration signal characteristics of each transformer are generated by the vibration signals collected in multiple different signal collection time periods, thereby including several sub-vibration signal characteristics of different time periods. Then, the transformers are compared, and the comparison parameters are the vibration signal characteristics of each transformer, that is, the sub-vibration signal characteristics of each transformer in the full time period are compared. According to the first comparison result, the vibration signal characteristics of the outliers are found, so that one or more candidate transformers with obvious differences from other transformers are identified according to the vibration signal characteristics of the outliers. Afterwards, considering that the reason for the abnormality of the transformer vibration signal may be caused by the uniqueness of the transformer itself in addition to the transformer failure, therefore, in this embodiment, for each candidate transformer, the sub-vibration signal characteristics of each candidate transformer in multiple time periods are compared longitudinally again according to the time dimension. If the abnormality of the transformer vibration signal is caused by its own uniqueness, the difference between the sub-vibration signal characteristics of different time periods will not be very large. If the abnormality of the transformer vibration signal is caused by a fault, there will be obvious differences between the sub-vibration signal characteristics of different time periods, so that through longitudinal comparison, the faulty transformer with a real fault is further determined from the candidate transformers. Through the above steps, accurate diagnosis of transformer faults was achieved without complete fault training samples and without the help of machine learning algorithms.
[0112] like Figure 5 As shown, this embodiment also provides a transformer fault diagnosis device based on vibration signals, the device comprising:
[0113] The feature extraction module 101 is used to obtain vibration signal features corresponding to a number of transformers, wherein the vibration signal features include sub-vibration signal features in a number of different time periods. For details, please refer to the relevant description of step S101 in the above method embodiment, which will not be repeated here.
[0114] The horizontal comparison module 102 is used to compare the vibration signal characteristics of each transformer and identify a candidate transformer from each transformer based on the outlier vibration signal characteristics in the first comparison result. For details, please refer to the relevant description of step S102 in the above method embodiment, which will not be repeated here.
[0115] The longitudinal comparison module 103 is used to compare the sub-vibration signal characteristics of the current candidate transformer and perform fault diagnosis on the current candidate transformer based on the second comparison result. For details, please refer to the relevant description of step S103 in the above method embodiment, which will not be repeated here.
[0116] The transformer fault diagnosis device based on vibration signals provided in an embodiment of the present invention is used to execute the transformer fault diagnosis method based on vibration signals provided in the above embodiment. Its implementation method and principle are the same. For details, please refer to the relevant description of the above method embodiment and will not be repeated here.
[0117] Figure 6 A transformer fault diagnosis device based on vibration signals according to an embodiment of the present invention is shown. The device includes a processor 901 and a memory 902, which can be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.
[0118] The processor 901 may be a central processing unit (CPU). The processor 901 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.
[0119] The memory 902 is a non-transitory computer-readable storage medium that can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as program instructions / modules corresponding to the methods in the above method embodiments. The processor 901 executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions and modules stored in the memory 902, that is, implementing the methods in the above method embodiments.
[0120] The memory 902 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required by at least one function; the data storage area may store data created by the processor 901, etc. In addition, the memory 902 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 902 may optionally include a memory remotely arranged relative to the processor 901, and these remote memories may be connected to the processor 901 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0121] One or more modules are stored in the memory 902 , and when executed by the processor 901 , the method in the above method embodiment is executed.
[0122] The specific details of the above-mentioned transformer fault diagnosis device based on vibration signals can be understood by referring to the corresponding descriptions and effects in the above-mentioned method embodiment, and will not be repeated here.
[0123] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the implemented program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.
[0124] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.
Claims
1. A transformer fault diagnosis method based on vibration signals, characterized in that: The method comprises: obtaining vibration signal features corresponding to a plurality of transformers, wherein the vibration signal features include sub-vibration signal features in a plurality of different time periods; comparing the vibration signal features of the respective transformers, and identifying a candidate transformer from the respective transformers based on an outlier vibration signal feature in a first comparison result; comparing the sub-vibration signal features of a current candidate transformer, and performing fault diagnosis on the current candidate transformer based on a second comparison result; The obtaining of vibration signal features corresponding to a number of transformers includes: obtaining the amplitude, odd-even subharmonic ratio, high-frequency energy ratio and spectrum complexity of the current transformer in a number of time periods; combining the amplitude and odd-even subharmonic ratio of the same time period into an amplitude-odd-even subharmonic ratio two-dimensional feature at the corresponding time; combining the amplitude and high-frequency energy ratio of the same time period into an amplitude-high-frequency energy ratio two-dimensional feature at the corresponding time; combining the amplitude and spectrum complexity of the same time period into an amplitude-spectrum complexity two-dimensional feature at the corresponding time; wherein the amplitude, odd-even subharmonic ratio, high-frequency energy ratio, spectrum complexity, amplitude-odd-even subharmonic ratio two-dimensional feature, amplitude-high-frequency energy ratio two-dimensional feature and amplitude-spectrum complexity two-dimensional feature corresponding to each time period are used as the sub-vibration signal features of each time period; The method further includes: averaging the numerical points corresponding to the same amplitude in the amplitude-odd-even subharmonic ratio two-dimensional feature, the amplitude-high-frequency energy ratio two-dimensional feature, and the amplitude-spectral complexity two-dimensional feature, and updating the amplitude-odd-even subharmonic ratio two-dimensional feature, the amplitude-high-frequency energy ratio two-dimensional feature, and the amplitude-spectral complexity two-dimensional feature using the average value point; The vibration signal features come from several different signal acquisition channels, and the vibration signal features of each transformer are compared, and a candidate transformer is identified from each transformer based on the outlier vibration signal features in the first comparison result, including: comparing the sub-vibration signal features of the current transformer in all time periods collected from the current signal acquisition channel with the sub-vibration signal features of other transformers in all time periods collected from the current signal acquisition channel, one by one according to feature types, to obtain the first comparison result; based on the outliers of the sub-vibration signal features of all time periods in the first comparison result, determining whether the current signal acquisition channel is an abnormal channel; if the current signal acquisition channel is an abnormal channel, recording the current transformer as the candidate transformer; The method of comparing the sub-vibration signal features of the current candidate transformer and performing fault diagnosis on the current candidate transformer based on the second comparison result includes: comparing the sub-vibration signal features of each time period collected by the current abnormal channel of the current candidate transformer according to the same feature type, and if a migrated sub-vibration signal feature appears, marking the current abnormal channel as a fault; traversing all abnormal channels of the current candidate transformer to perform sub-vibration signal feature comparison, and if each of the abnormal channels is marked with the fault mark after comparison, and all signal collection channels of the current candidate transformer are the abnormal channels, determining that the current candidate transformer is a faulty transformer; if no migrated sub-vibration signal feature appears in all abnormal channels of the current candidate transformer after feature comparison, determining that the current candidate transformer is a normal transformer; The sub-vibration signal features of each time period collected by the current abnormal channel of the current candidate transformer are compared according to the same feature type, including: curve fitting of the two-dimensional sub-vibration signal features of each time period collected by the current abnormal channel of the current candidate transformer to obtain a number of two-dimensional feature curves of each feature type in each time period; determining a reference time period for the feature type currently being compared; sequentially calculating the average deviation between the two-dimensional feature curves of other time periods and the two-dimensional feature curve of the reference time period; and using the average deviation corresponding to the two-dimensional feature curves of other time periods to determine whether migration occurs in the sub-vibration signal features of other time periods in the feature type currently being compared.
2. A transformer fault diagnosis device based on vibration signals, characterized in that: The device comprises: The feature extraction module is used to obtain the vibration signal features corresponding to several transformers, wherein the vibration signal features include sub-vibration signal features of several different time periods; the method of obtaining the vibration signal features corresponding to several transformers includes: obtaining the amplitude, odd-even harmonic ratio, high-frequency energy ratio and spectrum complexity of the current transformer in several time periods; combining the amplitude and odd-even harmonic ratio of the same time period into an amplitude-odd-even harmonic ratio two-dimensional feature at the corresponding time; combining the amplitude and high-frequency energy ratio of the same time period into an amplitude-high-frequency energy ratio two-dimensional feature at the corresponding time; combining the amplitude and spectrum complexity of the same time period into an amplitude-spectrum complexity Two-dimensional features; wherein the amplitude, even-odd subharmonic ratio, high-frequency energy ratio, spectrum complexity, amplitude-even-odd subharmonic ratio two-dimensional features, amplitude-high-frequency energy ratio two-dimensional features and amplitude-spectrum complexity two-dimensional features corresponding to each time period are used as the sub-vibration signal features of each time period; the average value of the numerical points corresponding to the same amplitude in the amplitude-even-odd subharmonic ratio two-dimensional features, the amplitude-high-frequency energy ratio two-dimensional features and the amplitude-spectrum complexity two-dimensional features is taken, and the average value point is used to update the amplitude-even-odd subharmonic ratio two-dimensional features, the amplitude-high-frequency energy ratio two-dimensional features and the amplitude-spectrum complexity two-dimensional features; A lateral comparison module, for comparing the vibration signal features of each transformer, and identifying a candidate transformer from each transformer based on the outlier vibration signal features in the first comparison result; the vibration signal features come from several different signal acquisition channels, and the vibration signal features of each transformer are compared, and the candidate transformer is identified from each transformer based on the outlier vibration signal features in the first comparison result, including: comparing the sub-vibration signal features of the current transformer in all time periods collected from the current signal acquisition channel with the sub-vibration signal features of other transformers in all time periods collected from the current signal acquisition channel, one by one according to feature types, to obtain the first comparison result; based on the outliers of the sub-vibration signal features of all time periods in the first comparison result, determining whether the current signal acquisition channel is an abnormal channel; if the current signal acquisition channel is an abnormal channel, recording the current transformer as the candidate transformer; A longitudinal comparison module is used to compare the sub-vibration signal features of the current candidate transformer and perform fault diagnosis on the current candidate transformer based on a second comparison result; the comparison of the sub-vibration signal features of the current candidate transformer and the fault diagnosis on the current candidate transformer based on the second comparison result include: comparing the sub-vibration signal features of each time period collected by the current abnormal channel of the current candidate transformer according to the same feature type, if a migrated sub-vibration signal feature appears, marking the current abnormal channel as a fault; traversing all abnormal channels of the current candidate transformer to perform sub-vibration signal feature comparison, if each of the abnormal channels is marked with the fault mark after comparison, and all signal acquisition channels of the current candidate transformer are the abnormal channels, then determining that the current candidate transformer is a faulty transformer; if If all abnormal channels of the current candidate transformer are subjected to feature comparison and no sub-vibration signal feature of migration appears, the current candidate transformer is determined to be a normal transformer; the sub-vibration signal features of each time period collected by the current abnormal channel of the current candidate transformer are compared according to the same feature type, including: curve fitting of the two-dimensional sub-vibration signal features of each time period collected by the current abnormal channel of the current candidate transformer to obtain a number of two-dimensional feature curves of each feature type in each time period; determining a reference time period for the feature type currently being compared; calculating the average deviation between the two-dimensional feature curves of other time periods and the two-dimensional feature curve of the reference time period in turn; and determining whether the sub-vibration signal features of other time periods in the feature type currently being compared appear to migrate by using the average deviation corresponding to the two-dimensional feature curves of other time periods.
3. A transformer fault diagnosis device based on vibration signals, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method as claimed in claim 1 by executing the computer instructions.
4. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to cause the computer to execute the method according to claim 1.
Citation Information
Patent Citations
Online state monitoring method for on-load tap changer of converter transformer
CN106124973A
High voltage circuit breaker running state real-time monitoring system based on CAN bus
CN108120925A
Transformer on-load tap changer and winding deformation fault detection method
CN109856501A
On-site verification method for application effect of transformer voiceprint monitoring device
CN113009399A
Front-end framework for monitoring oil-filled transformer in real time
CN113327741A
