CN117630611B - Full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating method and system - Google Patents
Full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating method and system Download PDFInfo
- Publication number
- CN117630611B CN117630611B CN202410085528.7A CN202410085528A CN117630611B CN 117630611 B CN117630611 B CN 117630611B CN 202410085528 A CN202410085528 A CN 202410085528A CN 117630611 B CN117630611 B CN 117630611B
- Authority
- CN
- China
- Prior art keywords
- partial discharge
- signal
- prpd
- signals
- frequency partial
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 239000013598 vector Substances 0.000 claims abstract description 57
- 230000001360 synchronised effect Effects 0.000 claims abstract description 17
- 238000001514 detection method Methods 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 238000005070 sampling Methods 0.000 claims description 74
- 239000011159 matrix material Substances 0.000 claims description 30
- 238000004422 calculation algorithm Methods 0.000 claims description 16
- 238000004458 analytical method Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 11
- 238000005516 engineering process Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 9
- 230000009467 reduction Effects 0.000 claims description 8
- 238000013461 design Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 6
- 238000007637 random forest analysis Methods 0.000 claims description 5
- 238000001914 filtration Methods 0.000 claims description 4
- 238000012795 verification Methods 0.000 claims description 4
- 238000009499 grossing Methods 0.000 claims description 3
- 238000003909 pattern recognition Methods 0.000 claims description 3
- 238000012847 principal component analysis method Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000001228 spectrum Methods 0.000 claims 6
- 238000007689 inspection Methods 0.000 abstract description 2
- 230000000737 periodic effect Effects 0.000 abstract description 2
- 238000011176 pooling Methods 0.000 description 9
- 230000004913 activation Effects 0.000 description 8
- 238000012423 maintenance Methods 0.000 description 6
- 238000005457 optimization Methods 0.000 description 6
- 238000000605 extraction Methods 0.000 description 5
- 230000004927 fusion Effects 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000013139 quantization Methods 0.000 description 3
- 230000035945 sensitivity Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 239000003086 colorant Substances 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000012634 fragment Substances 0.000 description 2
- 230000002068 genetic effect Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 239000002245 particle Substances 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 201000004569 Blindness Diseases 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013144 data compression Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000012850 discrimination method Methods 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 239000012212 insulator Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000004806 packaging method and process Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 230000008707 rearrangement Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012502 risk assessment Methods 0.000 description 1
- 238000009666 routine test Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Abstract
The invention discloses a full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating method and system, comprising the following steps: laying high-frequency partial discharge sensors, receiving original signals of at least N high-frequency partial discharge sensors, and preprocessing to obtain processed signals; reading the processed signals, analyzing and calculating to obtain a full-bandwidth high-frequency partial discharge PRPD map, and pre-storing the map; constructing a map recognition module, and extracting feature vectors from a full-bandwidth high-frequency partial discharge PRPD map; constructing and training an integrated learning model, and taking a feature vector as input to obtain a high-frequency partial discharge type; and sending the full-bandwidth high-frequency partial discharge PRPD map and the high-frequency partial discharge type into a display module for display. The invention improves the detection rate of partial discharge pulse information, does not capture an intermittent period, can be used for periodic inspection of high-voltage cables, establishes a cable withstand voltage synchronous partial discharge scene and the like, and has wide application prospect.
Description
Technical Field
The invention relates to the field of high-voltage cable state monitoring, in particular to a full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating method and system.
Background
As the power transmission network is continuously developed in an intelligent direction, the offline routine test and the regular maintenance method which are adopted in the past have the defects of high blindness, frequent operation, excessive maintenance and the like, so that the reliability of equipment is reduced, and the equipment is increasingly inconsistent with the development of the power transmission network; on the other hand, the state detection technology of the power equipment is continuously improved through the efforts of various research departments, universities and usage units, and the state maintenance of the equipment based on the state detection technology is greatly valued by the national grid company. The purpose of power cable state detection and risk assessment is to improve the power supply reliability and the equipment use efficiency of a power system as much as possible based on a state maintenance means.
For this reason, the skilled person has explored and provided solutions, but still has some technical problems. For example, in the aspect of high-frequency partial discharge spectrogram generation technology, the conventional technology at present is to collect and capture pulse waveforms, mark phases of the pulse waveforms, then combine the pulse waveforms into a data packet, upload the data packet to a terminal computer, analyze the pulse in the computer, acquire peak values from the pulse, and combine the peak values and the phases into a PRPD spectrogram. Because the frequency range of the high-frequency partial discharge signal is 3-30 MHz, the sampling rate of an ADC is 100MHz, and taking a 16-bit ADC as an example, a single data point needs 2 bytes, and if the sampling is carried out at a full speed of 100MHz, the data rate can reach 200MB/s, namely 1.6Gbit/s. Whereas normally partial discharge requires the simultaneous acquisition of 3 channels, its total data rate would reach 4.8Gbit/s. The excessive data flow brings great pressure to the post-processing circuit and system, so the conventional practice is intermittent operation, i.e. acquisition for a period of time, then transmission for a period of time, and then acquisition is started again. The mode can give consideration to the post-treatment pressure and ensure that the system operates smoothly. However, due to the processing mode, the device has a dead zone in detecting the partial discharge signal, namely the partial discharge signal generated in the intermittent period cannot be acquired, so that the effectiveness and the real-time performance of the device are seriously affected. The method can not collect occasional partial discharge signals in time, so that signal omission is caused, the reliability of equipment can be reduced, when a newly built cable is subjected to synchronous voltage-resistant partial discharge, the partial discharge signals can not be detected when the cable breaks down due to signal omission, fault cables can not be identified, breakdown accidents are caused, and the safety of a cable system and operation and maintenance personnel is seriously influenced.
Accordingly, further developments and innovations are needed to address the above-identified problems with the prior art.
Disclosure of Invention
The invention aims to provide a full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating method and system, so as to solve the problems in the prior art.
According to one aspect of the application, the full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating method comprises the following steps:
step S1, arranging high-frequency partial discharge sensors, receiving original signals of at least N high-frequency partial discharge sensors, and preprocessing to obtain processed signals; n is a natural number;
s2, reading the processed signals, analyzing and calculating to obtain a full-bandwidth high-frequency partial discharge PRPD map, and pre-storing the map;
s3, constructing a pattern recognition module, and extracting feature vectors from a full-bandwidth high-frequency partial discharge PRPD pattern;
s4, constructing and training an integrated learning model, and taking the feature vector as input to obtain a high-frequency partial discharge type;
and S5, sending the full-bandwidth high-frequency partial discharge PRPD map and the high-frequency partial discharge type into a display module for display.
According to one aspect of the application, the step S1 is further:
s11, invoking a GIS module and design requirements, generating and arranging the position of a high-frequency partial discharge sensor;
Step S12, setting sampling threshold values according to signal characteristics, forming signal paragraphs according to the sampling threshold values, setting sampling system parameters for each signal paragraph, and determining sampling rate and sampling bit number; the signal characteristics include amplitude and frequency;
step S13, collecting an original signal aiming at each signal section, dividing according to time segments to obtain signal segments which are continuous in time domain, and rearranging the sequence of the signal segments according to a preset time domain interleaving mode to obtain an interleaved signal;
step S14, performing compressed sensing sampling on the interleaved signals at a low sampling rate to obtain sampling values, taking the sampling values at the low sampling rate as input, and recovering the broadband high-frequency original signals by using an OMP compressed sensing reconstruction algorithm;
and S15, aligning signals of different paragraphs by using the synchronous signals or the time scale signals, re-splicing the signals of the different paragraphs together according to the sequence of the original signals to obtain processed signals, and re-combining the signals of the different paragraphs to obtain processed signals.
According to one aspect of the present application, the step S2 is further:
step S21, for each signal section, reading the processed signal, and analyzing and calculating signal parameters according to the characteristics and the rules of the partial discharge signal, wherein the signal parameters comprise discharge phase, discharge amplitude, discharge times and discharge energy;
S22, generating a PRPD map according to the signal parameters calculated by analysis, wherein a discharge phase is taken as an abscissa, a discharge amplitude is taken as an ordinate, discharge times or discharge energy is taken as a color shade, and a distribution map of a partial discharge signal is drawn;
and S23, selecting a preset storage device and mode according to the data format and the size of the PRPD map, and pre-storing the PRPD map.
According to one aspect of the present application, the step S3 is further:
s31, smoothing the PRPD map by adopting mean filtering, positioning characteristic points in the PRPD map based on a local maximum detection method, and obtaining characteristic point coordinates;
step S32, extracting a feature matrix of the PRPD map by a gray level co-occurrence matrix method based on the PRPD map and the feature points;
step S33, calculating the characteristic value of the characteristic matrix to obtain the characteristic value and the characteristic vector used for representing the contrast, energy and entropy of the PRPD map characteristic;
step S34: and analyzing and optimizing the feature vector by using a partial discharge waveform feature analysis method based on a statistical signal processing technology.
According to one aspect of the present application, the step S4 is further:
s41, constructing an integrated learning model based on a random forest or gradient lifting tree;
Step S42, using pre-stored partial discharge data as a training set, and using cross verification or grid search to train parameters of a model;
s43, evaluating the performance of the model by using a test set of partial discharge data through a confusion matrix or ROC curve method to obtain classification accuracy, recall and F1 value;
and S44, taking the feature vector of the PRPD map as the income of the integrated learning model to obtain the high-frequency partial discharge type.
According to one aspect of the present application, the step S5 is further:
step S51, calling a display unit and setting display parameters according to the PRPD map, the data format and the size of the high-frequency partial discharge type and the characteristics and the requirements of the display module;
and step S52, outputting the PRPD map and the high-frequency partial discharge type to a display module according to the set display parameters.
According to one aspect of the present application, the step S21 is further:
step S211, dividing the processed signal according to the window length for each signal section to obtain a plurality of data sections;
step S212, extracting characteristic waveforms from each data segment through wave crest and wave trough detection to obtain a starting point and an ending point;
step S213, for each characteristic waveform, calculating the phase angle from positive peak to next negative peak;
Step S214, a phase angle of each characteristic waveform is calculated and stored.
According to an aspect of the application, the step S34 is further:
step S341, calculating the statistical characteristics of the signals according to the waveform data of the signals, wherein the statistical characteristics comprise mean, variance, skewness and kurtosis;
step S342, normalizing the feature vector according to the statistical characteristics of the signals to enable the feature vector to accord with standard normal distribution;
step S343, according to the feature importance of the signal, the feature vector is subjected to dimension reduction, redundant or irrelevant features are removed, and main features are reserved so as to reduce the dimension and complexity of the feature vector;
step S344, selecting the feature vector according to the classification target of the signal, screening out the features that are most helpful to distinguish different types of partial discharge signals, and obtaining the optimal feature subset.
According to an aspect of the present application, in step S344, the process of obtaining the optimal feature subset is specifically:
step S34a, selecting an information gain or principal component analysis method according to the dimension and the characteristics of waveform characteristic data, and obtaining a characteristic vector with highest correlation of partial discharge types;
step S34b, calculating the importance of each feature vector of the waveform features in the PRPD map;
Step S34c, selecting a feature subset according to the value and distribution of the importance of the feature vector;
step S34d, according to the dimension and the characteristics of the feature subset, evaluating the effectiveness and the optimality of the feature subset, wherein the effectiveness and the optimality comprise classification accuracy, dimension reduction rate and feature redundancy; an optimal feature subset is obtained.
According to another aspect of the application, a full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating system is provided, which is configured to perform the method according to any one of the above technical schemes, and includes a high-frequency partial discharge sensor, a power frequency synchronous sensor, a programmable gate array FPGA, and a PRPD display module.
The method has the beneficial effects that the high-speed capturing and displaying of the PRPD map can be realized, the detection capability of the equipment to the partial discharge of the cable is improved, the situation that any partial discharge pulse information is not missed in the working process of the equipment, the capturing intermittence period is avoided, the method can be used for periodic inspection of high-voltage cables, and the method has wide application prospect in the newly-built cable withstand voltage synchronous partial discharge measurement and other scenes.
Drawings
Fig. 1 is a flow chart of the present invention.
Fig. 2 is a flowchart of step S1 of the present invention.
Fig. 3 is a flow chart of step S2 of the present invention.
Fig. 4 is a flowchart of step S3 of the present invention.
Fig. 5 is a flowchart of step S4 of the present invention.
Fig. 6 is a system architecture topology of the present invention.
Detailed Description
As shown in fig. 1, according to an aspect of the present application, the full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating method includes the following steps:
step S1, arranging high-frequency partial discharge sensors, receiving original signals of at least N high-frequency partial discharge sensors, and preprocessing to obtain processed signals; n is a natural number, preferably a natural number of 2 or more;
s2, reading the processed signals, analyzing and calculating to obtain a full-bandwidth high-frequency partial discharge PRPD map, and pre-storing the map;
s3, constructing a pattern recognition module, and extracting feature vectors from a full-bandwidth high-frequency partial discharge PRPD pattern;
s4, constructing and training an integrated learning model, and taking the feature vector as input to obtain a high-frequency partial discharge type;
and S5, sending the full-bandwidth high-frequency partial discharge PRPD map and the high-frequency partial discharge type into a display module for display.
In this embodiment, a high-frequency partial discharge sensor is used to collect partial discharge signals inside the power equipment, a full-bandwidth high-frequency partial discharge PRPD map is generated through a series of signal processing and analysis methods, characteristic information of partial discharge is extracted from the partial discharge PRPD map, an integrated learning model is used to identify the type of the partial discharge, and finally the PRPD map and the type of the partial discharge are displayed so as to evaluate and diagnose the state of the power equipment. Specifically, the compressed sensing technology is adopted to realize the low sampling rate sampling and reconstruction of the broadband high-frequency partial discharge signal, so that the data volume and the storage space are effectively reduced, and meanwhile, the integrity and the quality of the signal are ensured. By adopting the time domain interleaving mode, rearrangement and splicing of signals of different paragraphs are realized, the signal-to-noise ratio and resolution of the signals are effectively improved, and meanwhile, distortion and interference of the signals are avoided. The extraction of the feature matrix of the PRPD is realized by adopting the gray level co-occurrence matrix method, so that the texture and structural features of the PRPD are effectively described, and the robustness and stability of the features are enhanced. By adopting the partial discharge waveform characteristic analysis method based on the statistical signal processing technology, the analysis and optimization of the characteristic vector are realized, the statistical characteristics of the partial discharge signal are effectively extracted, and the dimension and the complexity of the characteristic are reduced. By adopting the integrated learning model based on random forests or gradient lifting trees, the classification of the partial discharge types is realized, the classification accuracy and efficiency are effectively improved, and the limitation and the deficiency of a single model are overcome.
As shown in fig. 2, according to an aspect of the present application, the step S1 is further:
s11, invoking a GIS module and design requirements, generating and arranging the position of a high-frequency partial discharge sensor;
and calling the GIS module, inputting information such as the model, specification, structure and position of the power equipment, generating a three-dimensional model of the power equipment, and marking the part which is likely to generate partial discharge on the model, such as an insulator, a cable joint, a switch cabinet and the like. According to design requirements, suitable high-frequency partial discharge sensors, such as ultra-high frequency (UHF) sensors, ultrasonic (US) sensors, high-frequency current (HFCT) sensors and the like, are selected, and parameters of the number, the type, the sensitivity, the frequency response and the like of the sensors are determined. According to the three-dimensional model generated by the GIS module and parameters of the sensor, an optimization algorithm such as a Genetic Algorithm (GA), a particle swarm algorithm (PSO), an ant colony Algorithm (ACO) and the like is used for calculating the optimal position of the sensor, and the optimal position is displayed on the model. According to the calculated position of the sensor, the sensor is actually arranged and connected to a signal acquisition and processing system such as a digital oscilloscope, a signal analyzer, a computer and the like, and the sensor is calibrated and tested, so that the sensor can work normally and can effectively receive partial discharge signals. The technical effects achieved by the steps are as follows: the position of the sensor can be automatically generated according to the three-dimensional model of the power equipment, the sensor is distributed and calibrated, the manual intervention and errors are reduced, and the working efficiency and quality are improved.
Step S12, setting sampling threshold values according to signal characteristics, forming signal paragraphs according to the sampling threshold values, setting sampling system parameters for each signal paragraph, and determining sampling rate and sampling bit number; the signal characteristics include amplitude and frequency;
for each high frequency partial discharge sensor, the received partial discharge signal is read, and the amplitude and frequency distribution of the signal, such as maximum value, minimum value, average value, standard deviation, peak frequency, bandwidth and the like, are calculated.
According to the distribution of the amplitude and the frequency of the signal, a sampling method based on compressed sensing is used for dynamically adjusting a sampling threshold value so as to ensure the sparsity and the integrity of the signal. Specifically, the dynamic range of the signal is determined according to the maximum value and the minimum value of the amplitude of the signal, the spectral range of the signal is determined according to the peak frequency and the bandwidth of the frequency of the signal, and then the sparseness of the signal, namely the proportion of non-zero elements in the signal, and the integrity of the signal, namely the proportion of important information in the signal, are calculated according to the dynamic range and the spectral range of the signal. According to the sparsity and the integrity of the signals, a sampling threshold is set, so that the sampled signals can retain the main characteristics of the signals, and meanwhile, the redundancy and noise of the signals are reduced.
According to the sampling threshold value, the signal is segmented, the signal is divided into a plurality of signal segments, each signal segment comprises a certain number of signal samples, and the amplitude and the frequency of the signal samples of each signal segment are within the range of the sampling threshold value, so that the uniformity and the consistency of the signal segments are ensured.
For each signal segment, setting sampling system parameters according to the amplitude and frequency distribution of the signal, and determining the sampling rate and the sampling bit number. Specifically, according to the maximum value and the minimum value of the amplitude of the signal, determining the quantization level of the signal, namely the sampling bit number, so as to ensure the quantization precision of the signal; and determining the sampling frequency, namely the sampling rate of the signal according to the peak frequency and the bandwidth of the frequency of the signal so as to ensure the frequency domain resolution of the signal. Parameters of the sampling system, such as reference voltage of an analog-to-digital converter (ADC), clock frequency, etc., are set according to the quantization level and sampling frequency of the signal.
The sampling threshold can be dynamically adjusted by using a sampling method based on compressed sensing according to the distribution of the amplitude and the frequency of the signal to form signal paragraphs, sampling system parameters are set for each signal paragraph, the sampling rate and the sampling bit number are determined, the low sampling rate sampling and reconstruction of the signal are realized, the data volume and the storage space are effectively reduced, and meanwhile, the integrity and the quality of the signal are ensured.
Step S13, collecting an original signal aiming at each signal section, dividing according to time segments to obtain signal segments which are continuous in time domain, and rearranging the sequence of the signal segments according to a preset time domain interleaving mode to obtain an interleaved signal;
for each signal segment, a digital oscilloscope or other signal acquisition equipment is used for acquiring an original signal according to the set sampling system parameters, and the acquired signal samples are stored in a memory or a hard disk for subsequent processing.
For each signal segment, dividing an original signal according to a preset time segment length to obtain a plurality of time-domain continuous signal segments, wherein each signal segment comprises a certain number of signal samples, and the time length of each signal segment is the same so as to ensure the synchronism and comparability of the signal segments.
And rearranging the sequence of the signal fragments according to a preset time domain interleaving mode for each signal fragment to obtain an interleaved signal. Specifically, according to the distribution of the amplitude and the frequency of the signal, a proper time domain interleaving mode such as random interleaving, cyclic interleaving, packet interleaving and the like is selected, and according to the rule of the interleaving mode, the sequence of the signal segments is disordered or reorganized, so that the interleaved signal is obtained. The purpose of interleaving is to improve the signal-to-noise ratio and resolution of the signal while avoiding distortion and interference of the signal.
The method can select a proper time domain interleaving mode according to the amplitude and frequency distribution of the signals, rearrange the sequence of the signal segments to obtain interleaved signals, effectively improve the signal-to-noise ratio and the resolution of the signals, and simultaneously avoid the distortion and the interference of the signals.
Step S14, performing compressed sensing sampling on the interleaved signals at a low sampling rate to obtain sampling values, taking the sampling values at the low sampling rate as input, and recovering the broadband high-frequency original signals by using an OMP compressed sensing reconstruction algorithm;
and for each interleaved signal, performing compressed sensing sampling at a sampling rate lower than the Nyquist sampling rate to obtain a sampling value. Specifically, the sampling rate of the compressed sensing samples is determined according to the sparsity of the signal, so that the sampling rate meets the sparsity requirement, i.e., the sampling rate is greater than the proportion of non-zero elements in the signal, so as to ensure the reconfigurability of the signal. Then, the signal is subjected to linear transformation by using a random matrix or other compression matrix, so as to obtain a compressed signal, namely a sampling value. The compressed sensing sampling aims to reduce the sampling rate of signals, reduce the data volume and the storage space, and simultaneously reserve the main information of the signals, thereby being beneficial to the reconstruction and analysis of the signals.
And taking the sampling value with the low sampling rate as input, and recovering the broadband high-frequency original signal by using an OMP compressed sensing reconstruction algorithm. Specifically, the signals are represented as linear combinations of sparse bases, i.e., the signals are equal to the sparse bases multiplied by the sparse coefficients, according to the sparse bases of the signals, such as wavelet bases, fourier bases, dictionary bases, and the like. And then, using an OMP algorithm to iteratively select the most relevant base vector from the sparse base, and updating the sparse coefficient until a preset reconstruction error or sparsity is reached, so as to obtain a reconstructed signal, namely an original signal. The OMP compressed sensing reconstruction algorithm aims to recover complete signals from a small number of sampling values, improve the quality and the integrity of the signals and facilitate the analysis and the identification of the signals.
The OMP compressed sensing reconstruction algorithm can be utilized to recover the original signal with broadband and high frequency from the sampling value with low sampling rate, so that the quality and the integrity of the signal are effectively improved, and the data volume and the storage space of the signal are reduced.
And S15, aligning signals of different paragraphs by using the synchronous signals or the time scale signals, re-splicing the signals of the different paragraphs together according to the sequence of the original signals to obtain processed signals, and re-combining the signals of the different paragraphs to obtain processed signals.
And collecting power frequency voltage or current signals of the power equipment by using a power frequency synchronous sensor or other synchronous equipment as synchronous signals or time scale signals for aligning signals of different paragraphs. Specifically, according to the periodicity and stability of the power frequency signal, the sampling rate and the sampling bit number of the synchronous signal or the time scale signal are determined so as to ensure the precision and the reliability of the synchronous signal or the time scale signal. Then, a digital oscilloscope or other signal acquisition device is used to acquire the synchronization signal or time scale signal, and the acquired synchronization signal or time scale signal samples are stored in a memory or a hard disk for subsequent processing.
And aligning signals of different paragraphs by using the synchronous signals or the time scale signals, and re-splicing the signals of the different paragraphs together according to the sequence of the original signals to obtain the processed signals. Specifically, according to the periodicity and stability of the synchronization signal or the time scale signal, a synchronization point, such as a zero crossing point, a peak point, etc., of the synchronization signal or the time scale signal is determined as an alignment reference point of the signal. Then, according to the synchronization point, the signals of different paragraphs are aligned in time, so that the starting point and the ending point of the signals of each paragraph are matched with the synchronization point, and the synchronism and the continuity of the signals are ensured. And finally, the signals of all the aligned paragraphs are spliced together again according to the sequence of the original signals, so that the processed signals, namely the processed signals, are obtained.
The synchronous signals or the time scale signals can be utilized to align the signals of different paragraphs, the signals of the different paragraphs are spliced together again according to the sequence of the original signals, the processed signals are obtained, the synchronism and the continuity of the signals are effectively improved, and meanwhile, the dislocation and the breakage of the signals are avoided.
As shown in fig. 3, according to an aspect of the present application, the step S2 is further:
step S21, for each signal section, reading the processed signal, and analyzing and calculating signal parameters according to the characteristics and the rules of the partial discharge signal, wherein the signal parameters comprise discharge phase, discharge amplitude, discharge times and discharge energy;
the processed signal is divided into a plurality of signal segments at regular time intervals, each signal segment comprising one or more partial discharge pulses.
For each signal segment, a peak detection algorithm or a threshold discrimination method is used to identify the peak position and amplitude of the partial discharge pulse.
And calculating the corresponding discharge phase, namely the phase angle of the applied voltage when the pulse occurs, according to the peak position of the partial discharge pulse. The discharge phase ranges from 0 deg. to 360 deg., and can be segmented or grouped with a certain accuracy.
According to the amplitude of the partial discharge pulse, the corresponding discharge energy, namely the electric energy released by the pulse in unit time, is calculated. The discharge energy is calculated by the formula e=u 2 R, wherein U is the pulse amplitude and R is the measurement resistance.
The number of partial discharge pulses, i.e. the number of discharges, in each signal segment is counted. The number of discharges reflects the degree of activity of the partial discharge.
S22, generating a PRPD map according to the signal parameters calculated by analysis, wherein a discharge phase is taken as an abscissa, a discharge amplitude is taken as an ordinate, discharge times or discharge energy is taken as a color shade, and a distribution map of a partial discharge signal is drawn;
and determining the scales and the resolution of the horizontal axis and the vertical axis of the PRPD map according to the range and the precision of the signal parameters. For example, the horizontal axis may be a section of 12 ° or 18 °, and the vertical axis may be a section of 5 mV or 10 mV, and the PRPD pattern is divided into a plurality of small squares.
For each signal segment, its position in the PRPD pattern, i.e. the corresponding square, is determined according to its discharge phase and discharge amplitude. And then determining the color of the grid in the PRPD map, namely the color shade of the corresponding small grid according to the discharge times or discharge energy. The shades may be mapped in a certain ratio or gradation, e.g. the greater the number of discharges or the discharge energy, the darker the color and vice versa.
And (3) superposing the colors of all the signal segments in the PRPD map to obtain a final distribution map of the partial discharge signals. The PRPD pattern can reflect the type, intensity and rule of partial discharge, and has important significance for diagnosis and analysis of partial discharge.
And S23, selecting a preset storage device and mode according to the data format and the size of the PRPD map, and pre-storing the PRPD map.
Converting the data format of the PRPD pattern into a format suitable for storage, for example, the color information of the PRPD pattern may be converted into a gray value or an RGB value, which is then saved as an image file, such as PNG or JPG format; the data information of the PRPD map may also be saved as a text file, such as a TXT or CSV format, where each row represents a parameter of a signal segment, including discharge phase, discharge amplitude, number of discharges, or discharge energy, etc.
According to the data size of the PRPD map, a proper storage device and mode are selected, for example, physical media such as a usb disk, a hard disk, an optical disk and the like can be used for storage, and virtual media such as a cloud, a network, a database and the like can also be used for storage. The safety and the integrity of the data should be paid attention to during storage, and the loss or damage of the data is avoided.
In order to facilitate subsequent retrieval and use, the stored PRPD patterns should be named and classified reasonably, for example, the stored PRPD patterns may be named and classified according to information such as test time, test object, test condition, etc., or may be identified and managed by using encoding modes such as bar codes, two-dimensional codes, etc.
As shown in fig. 4, according to an aspect of the present application, the step S3 is further:
s31, smoothing the PRPD map by adopting mean filtering, positioning characteristic points in the PRPD map based on a local maximum detection method, and obtaining characteristic point coordinates;
the PRPD pattern is used as a two-dimensional matrix, and each element represents the color shade of a small square, namely the discharge times or the discharge energy.
And (3) carrying out mean value filtering on the matrix of the PRPD map, namely replacing the value of each element by the average value of the elements around the element so as to eliminate noise and details in the map and highlight main features in the map.
And detecting local maxima of the matrix of the PRPD map, namely finding out local maxima in the matrix, namely small squares with the largest discharge times or discharge energy, and taking the local maxima as characteristic points in the PRPD map.
Step S32, extracting a feature matrix of the PRPD map by a gray level co-occurrence matrix method based on the PRPD map and the feature points;
If a convolutional neural network based on deep learning is used, the PRPD map can be used as input, and the characteristic vector of the image can be obtained through multi-layer convolution, pooling, activation, full connection and other operations, and the characteristic vector can reflect the global and local characteristics of the image and the abstraction degrees of different layers. Specifically, the following steps may be used:
and converting the PRPD pattern into a gray image, namely converting the color shade of each small square into a gray value, and obtaining a gray matrix.
And (3) carrying out convolution operation on the gray matrix by using a convolution kernel to obtain a feature map, namely the output of the convolution layer. The convolution operation can extract low-level features such as edges, textures and the like of the image.
And (3) using a pooling core to perform pooling operation on the feature map to obtain a pooling map, namely, the output of a pooling layer. The pooling operation can reduce the size of the image, reduce the calculation amount and simultaneously retain the main characteristics of the image.
And (3) performing an activation operation on the pooling graph by using an activation function to obtain an activation graph, namely, the output of the activation layer. The activation operation can increase the nonlinearity of the image and enhance the expression capability of the image.
Repeating the steps, and constructing a plurality of convolution layers, pooling layers and activation layers by using a plurality of convolution kernels, pooling kernels and activation functions to form a deep convolution neural network. The output of each layer can be used as input of the next layer, so that the characteristics of different layers of the image can be extracted.
And (3) using a full-connection layer to perform full-connection operation on the output of the last layer to obtain a feature vector, namely the output of the full-connection layer. The full-connection operation can integrate the features of the images to obtain a vector with a fixed length, and the vector is used as the feature representation of the images.
Step S33, calculating the characteristic value of the characteristic matrix to obtain the characteristic value and the characteristic vector used for representing the contrast, energy and entropy of the PRPD map characteristic;
for each feature extraction method, the obtained feature vectors are formed into a feature matrix, each row represents the feature vector of a signal paragraph, and each column represents a feature dimension.
For each feature matrix, the feature values and feature vectors of the feature matrix are calculated using Singular Value Decomposition (SVD) or other feature value decomposition methods. The eigenvalues reflect the principal information of the eigenvalues and the eigenvectors reflect the principal directions of the eigenvalues.
And calculating the characteristic values and the characteristic vectors used for representing the contrast, the energy and the entropy of the PRPD map characteristics according to the characteristic values and the characteristic vectors. Specifically, the contrast reflects the gray level variation degree of the PRPD pattern, the energy reflects the signal intensity of the PRPD pattern, and the entropy reflects the signal complexity of the PRPD pattern. The formulas for contrast, energy and entropy are as follows:
Contrast ratio: c=sum {i,j} （p (i,j) (i-j) 2 ) Wherein p is (i,j) I and j are the elements of the feature matrix and are the indices of the feature dimensions.
Energy: e=sum {i,j} p (i,j) 2 Wherein p is (i,j) I and j are the elements of the feature matrix and are the indices of the feature dimensions.
Entropy: h= -sum {i,j} p (i,j) log（p (i,j) ) Wherein p(i,j) I and j are the elements of the feature matrix and are the indices of the feature dimensions.
Step S34: and analyzing and optimizing the feature vector by using a partial discharge waveform feature analysis method based on a statistical signal processing technology.
As shown in fig. 5, according to an aspect of the present application, the step S4 is further:
s41, constructing an integrated learning model based on a random forest or gradient lifting tree;
and selecting a random forest or gradient lifting tree as a basic classifier, and determining the number and types of the basic classifier according to the feature dimension and the category number of the partial discharge data.
For each basic classifier, randomly extracting a part of samples and features from partial discharge data to serve as a training set, and constructing a classification model by using a decision tree algorithm to obtain an output result of each basic classifier.
And voting or weighted averaging is carried out on the output results of all the basic classifiers for each test sample, so that the final output result of the integrated learning model is obtained.
Step S42, using pre-stored partial discharge data as a training set, and using cross verification or grid search to train parameters of a model; dividing the pre-stored partial discharge data into a training set and a verification set, and determining the search space and the step length of the parameters according to the parameter types and the range of the integrated learning model. Cross-validation or grid search is selected as a parameter optimization method, and an optimal parameter combination is found in a parameter search space according to an objective function (such as classification accuracy, recall or F1 value) and an optimization algorithm (such as Bayesian optimization, genetic algorithm or particle swarm optimization).
S43, evaluating the performance of the model by using a test set of partial discharge data through a confusion matrix or ROC curve method to obtain classification accuracy, recall and F1 value; and predicting the test set of the partial discharge data by using the trained integrated learning model to obtain a prediction label and a prediction probability of each test sample. And constructing a confusion matrix according to the prediction label and the real label, calculating evaluation indexes such as classification accuracy, recall rate, F1 value and the like, and evaluating the performance of the model. And drawing an ROC curve according to the prediction probability and the real label, calculating evaluation indexes such as an AUC value and the like, and evaluating the performance of the model.
And S44, taking the feature vector of the PRPD map as the income of the integrated learning model to obtain the high-frequency partial discharge type. For each partial discharge signal, a PRPD map generation method is used to convert it into two-dimensional image data, and then feature vectors of the image are extracted as inputs to the ensemble learning model. And predicting each feature vector by using an ensemble learning model to obtain probability distribution of the corresponding partial discharge type, and determining the high-frequency partial discharge type according to the size of the probability value. And counting the occurrence times and the occurrence proportion of each high-frequency partial discharge type, analyzing the change rule of each high-frequency partial discharge type under different working conditions, and providing basis for fault diagnosis and prevention.
According to one aspect of the present application, the step S5 is further:
step S51, calling a display unit and setting display parameters according to the PRPD map, the data format and the size of the high-frequency partial discharge type and the characteristics and the requirements of the display module;
and selecting a proper display module, and determining the data format and size of the display module according to the specification parameters, such as an 8-bit or 4-bit parallel data port, the number of data bits, data coding and the like. According to the PRPD map and the data characteristics of the high-frequency partial discharge type, such as data range, data precision, data resolution, data change rate and the like, proper display parameters are selected, such as display modes, display resolution, display colors, display contrast, display brightness and the like. The display unit is invoked to connect the display module to a data source, such as via a data line, control line, etc. Display parameters are set, such as by writing control commands, adjusting potentiometers, etc.
And step S52, outputting the PRPD map and the high-frequency partial discharge type to a display module according to the set display parameters. The PRPD pattern and high frequency partial discharge type data are read, such as by reading data lines, reading data registers, etc. The data is converted into a data format required by the display module, such as by data encoding, data compression, data scaling, etc. Data is written to the display module, e.g. by writing data lines, writing data registers, etc. The PRPD pattern and the high frequency partial discharge type are displayed on a display module, such as by dot matrix display, graphic display, etc.
According to an aspect of the present application, the step S21 may further adopt the following scheme, specifically:
step S211, dividing the processed signal according to the window length for each signal section to obtain a plurality of data sections;
step S212, extracting characteristic waveforms from each data segment through wave crest and wave trough detection to obtain a starting point and an ending point;
step S213, for each characteristic waveform, calculating the phase angle from positive peak to next negative peak;
step S214, a phase angle of each characteristic waveform is calculated and stored.
The phase angle-based partial discharge signal extraction and identification method can effectively eliminate noise and interference and improve the signal-to-noise ratio and resolution of the partial discharge signal. The method utilizes the periodicity and phase characteristics of the partial discharge signals, and can accurately extract the characteristic waveforms of the partial discharge signals through wave crest and wave trough detection and phase angle calculation and convert the characteristic waveforms into phase angle distribution diagrams, thereby realizing classification and identification of the partial discharge signals. The method has higher sensitivity and robustness, and can adapt to different partial discharge types and working conditions. At the same time, the cost and complexity of partial discharge detection can be reduced. Only the partial discharge signal is required to be basically filtered, divided, detected and calculated, so that the calculation resource and time can be saved, and the efficiency and accuracy of partial discharge detection can be improved. The position and the type of the partial discharge can be found and positioned in time, so that a basis is provided for maintenance and repair of the power equipment, faults and accidents caused by the partial discharge are avoided, and normal operation and power supply of a power system are ensured. The service life and performance of the power equipment can be prolonged, the power loss and pollution are reduced, the energy and resources are saved, and the energy conservation and the environmental protection of the power system are promoted.
According to an aspect of the application, the step S34 is further:
step S341, calculating the statistical characteristics of the signals according to the waveform data of the signals, wherein the statistical characteristics comprise mean, variance, skewness and kurtosis;
step S342, normalizing the feature vector according to the statistical characteristics of the signals to enable the feature vector to accord with standard normal distribution;
step S343, according to the feature importance of the signal, the feature vector is subjected to dimension reduction, redundant or irrelevant features are removed, and main features are reserved so as to reduce the dimension and complexity of the feature vector;
step S344, selecting the feature vector according to the classification target of the signal, screening out the features that are most helpful to distinguish different types of partial discharge signals, and obtaining the optimal feature subset.
The embodiment can effectively reduce the dimension and complexity of data and improve the classification performance of partial discharge signals. The characteristic vector is obtained by utilizing waveform data of the partial discharge signals and calculating statistical characteristics such as mean value, variance, skewness, kurtosis and the like of the partial discharge signals, and then the characteristic representation is optimized by operations such as normalization, dimension reduction, selection and the like, so that the characteristic extraction and selection of the partial discharge signals are realized.
According to an aspect of the present application, in step S344, the process of obtaining the optimal feature subset is specifically:
step S34a, selecting an information gain or principal component analysis method according to the dimension and the characteristics of waveform characteristic data, and obtaining a characteristic vector with highest correlation of partial discharge types;
step S34b, calculating the importance of each feature vector of the waveform features in the PRPD map;
step S34c, selecting a feature subset according to the value and distribution of the importance of the feature vector;
step S34d, according to the dimension and the characteristics of the feature subset, evaluating the effectiveness and the optimality of the feature subset, wherein the effectiveness and the optimality comprise classification accuracy, dimension reduction rate and feature redundancy; an optimal feature subset is obtained.
The embodiment can effectively reduce the dimension and complexity of data and improve the classification performance of partial discharge signals. The waveform data of the partial discharge signals are utilized to obtain the feature vectors by calculating the statistical characteristics of the mean value, the variance, the skewness, the kurtosis and the like, and then the feature representation is optimized by the operations of normalization, the dimension reduction, the selection and the like, so that the feature extraction and the selection of the partial discharge signals are realized, and the sensitivity and the robustness are higher.
As shown in fig. 6, according to another aspect of the present application, a full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating system is provided, so as to perform the method according to any one of the foregoing technical solutions, including a high-frequency partial discharge sensor, a power frequency synchronous sensor, a programmable gate array FPGA, and a PRPD display module.
The hardware part mainly comprises:
in the circuit, 3 high-speed partial discharge acquisition channels with the sampling rate of 100MHz and 1 power frequency synchronous acquisition channel are designed; the digital signals converted by the high-speed ADC enter the FPGA. The method comprises the steps of realizing a 10us timing module, a peak value capturing module, a phase generating module, a data fusion module, a DMA controller, a RAM internal memory, a central processing unit, an Ethernet controller and the like in an FPGA. Uploading the encoded spectrogram data to a terminal computer through a network port, decoding the data in the terminal computer and displaying the data in a display module.
10us timing module design: the FPGA works at the frequency of 100MHz, each period is 10ns, 10us timing pulses are needed, and only one pulse is needed to be generated every 1000 counts. Peak capture module design: find the maximum value (maximum absolute value) within 10us, and output to the next module when the timing pulse arrives. Phase generation module design: and the current phase can be obtained by analyzing the period of the power frequency synchronizing signal and determining the ratio of the phase counter to the period counter, and the current phase is continuously output to the data fusion module. And (3) designing a data fusion module: and packaging and encoding the pulse amplitude and phase information to integrate the pulse amplitude and phase information into a data packet. DMA controller: and the data packet after fusion is transferred to an internal memory of the RAM. RAM internal memory: the data packet can be directly read by the CPU. CPU central processing unit: and the data is responsible for controlling the DMA controller, reading the data of the internal memory of the RAM and sending the data to the Ethernet controller. The net opening is as follows: and sending the data sent by the Ethernet module to a terminal computer. And the terminal computer: and receiving and decoding the data packet, and sending the data packet to a full-bandwidth PRPD display module for display. The invention carries out high-speed acquisition on signals of a high-frequency current sensor (HFCT), acquires peak value and phase information once every 10us, and transmits the acquired data to a CPU Central Processing Unit (CPU) at high speed by using a Direct Memory Access (DMA), and then uploads the data to a terminal computer. The signal acquisition process is completely realized by FPGA programming, the operation process of the FPGA is multichannel and pipelined, the RAM internal memory is used as data buffer, the acquisition intermittence period is avoided, and the defect of direct sampling by using a processor is overcome. The intermittent acquisition of the high-frequency partial discharge signals is realized, the detection capability of the equipment is greatly improved, and the effectiveness of the equipment is improved.
The algorithm modules, which are configured in the respective modules in the FPGA, have a workflow identical to the above-described method items, and will not be described in detail herein.
The preferred embodiments of the present invention have been described in detail above, but the present invention is not limited to the specific details of the above embodiments, and various equivalent changes can be made to the technical solution of the present invention within the scope of the technical concept of the present invention, and all the equivalent changes belong to the protection scope of the present invention.
Claims (8)
1. The full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating method is characterized by comprising the following steps of:
step S1, arranging high-frequency partial discharge sensors, receiving original signals of at least N high-frequency partial discharge sensors, and preprocessing to obtain processed signals; n is a natural number;
s2, reading the processed signals, analyzing and calculating to obtain a full-bandwidth high-frequency partial discharge PRPD map, and pre-storing the map;
s3, constructing a pattern recognition module, and extracting feature vectors from a full-bandwidth high-frequency partial discharge PRPD pattern;
s4, constructing and training an integrated learning model, and taking the feature vector as input to obtain a high-frequency partial discharge type;
s5, sending the full-bandwidth high-frequency partial discharge PRPD map and the high-frequency partial discharge type into a display module for display;
The step S1 is further:
s11, invoking a GIS module and design requirements, generating and arranging the position of a high-frequency partial discharge sensor;
step S12, setting sampling threshold values according to signal characteristics, forming signal paragraphs according to the sampling threshold values, setting sampling system parameters for each signal paragraph, and determining sampling rate and sampling bit number; the signal characteristics include amplitude and frequency;
step S13, collecting an original signal aiming at each signal section, dividing according to time segments to obtain signal segments which are continuous in time domain, and rearranging the sequence of the signal segments according to a preset time domain interleaving mode to obtain an interleaved signal;
step S14, performing compressed sensing sampling on the interleaved signals at a low sampling rate to obtain sampling values, taking the sampling values at the low sampling rate as input, and recovering the broadband high-frequency original signals by using an OMP compressed sensing reconstruction algorithm;
step S15, aligning signals of different paragraphs by using a synchronous signal or a time mark signal, re-splicing the signals of different paragraphs together according to the sequence of original signals to obtain processed signals, and re-combining the signals of different paragraphs to obtain processed signals;
the step S2 is further:
Step S21, for each signal section, reading the processed signal, and analyzing and calculating signal parameters according to the characteristics and the rules of the partial discharge signal, wherein the signal parameters comprise discharge phase, discharge amplitude, discharge times and discharge energy;
s22, generating a PRPD map according to the signal parameters calculated by analysis, wherein a discharge phase is taken as an abscissa, a discharge amplitude is taken as an ordinate, discharge times or discharge energy is taken as a color shade, and a distribution map of a partial discharge signal is drawn;
and S23, selecting a preset storage device and mode according to the data format and the size of the PRPD map, and pre-storing the PRPD map.
2. The full-bandwidth high-frequency partial discharge PRPD spectrum capture generation method according to claim 1, wherein the step S3 further comprises:
s31, smoothing the PRPD map by adopting mean filtering, positioning characteristic points in the PRPD map based on a local maximum detection method, and obtaining characteristic point coordinates;
step S32, extracting a feature matrix of the PRPD map by a gray level co-occurrence matrix method based on the PRPD map and the feature points;
step S33, calculating the characteristic value of the characteristic matrix to obtain the characteristic value and the characteristic vector used for representing the contrast, energy and entropy of the PRPD map characteristic;
Step S34: and analyzing and optimizing the feature vector by using a partial discharge waveform feature analysis method based on a statistical signal processing technology.
3. The full-bandwidth high-frequency partial discharge PRPD spectrum capture generation method according to claim 2, wherein the step S4 further comprises:
s41, constructing an integrated learning model based on a random forest or gradient lifting tree;
step S42, using pre-stored partial discharge data as a training set, and using cross verification or grid search to train parameters of a model;
s43, evaluating the performance of the model by using a test set of partial discharge data through a confusion matrix or ROC curve method to obtain classification accuracy, recall and F1 value;
and S44, taking the feature vector of the PRPD map as the income of the integrated learning model to obtain the high-frequency partial discharge type.
4. The full-bandwidth high-frequency partial discharge PRPD spectrum capture generation method of claim 3, wherein the step S5 further comprises:
step S51, calling a display unit and setting display parameters according to the PRPD map, the data format and the size of the high-frequency partial discharge type and the characteristics and the requirements of the display module;
and step S52, outputting the PRPD map and the high-frequency partial discharge type to a display module according to the set display parameters.
5. The full-bandwidth high-frequency partial discharge PRPD spectrum capture generation method of claim 4, wherein the step S21 further comprises:
step S211, dividing the processed signal according to the window length for each signal section to obtain a plurality of data sections;
step S212, extracting characteristic waveforms from each data segment through wave crest and wave trough detection to obtain a starting point and an ending point;
step S213, for each characteristic waveform, calculating the phase angle from positive peak to next negative peak;
step S214, a phase angle of each characteristic waveform is calculated and stored.
6. The full-bandwidth high-frequency partial discharge PRPD spectrum capture generation method of claim 5, wherein the step S34 further comprises:
step S341, calculating the statistical characteristics of the signals according to the waveform data of the signals, wherein the statistical characteristics comprise mean, variance, skewness and kurtosis;
step S342, normalizing the feature vector according to the statistical characteristics of the signals to enable the feature vector to accord with standard normal distribution;
step S343, according to the feature importance of the signal, the feature vector is subjected to dimension reduction, redundant or irrelevant features are removed, and main features are reserved so as to reduce the dimension and complexity of the feature vector;
Step S344, selecting the feature vector according to the classification target of the signal, screening out the features that are most helpful to distinguish different types of partial discharge signals, and obtaining the optimal feature subset.
7. The full-bandwidth high-frequency partial discharge PRPD spectrum capture generation method of claim 6, wherein in step S344, the process of obtaining the optimal feature subset specifically includes:
step S34a, selecting an information gain or principal component analysis method according to the dimension and the characteristics of waveform characteristic data, and obtaining a characteristic vector with highest correlation of partial discharge types;
step S34b, calculating the importance of each feature vector of the waveform features in the PRPD map;
step S34c, selecting a feature subset according to the value and distribution of the importance of the feature vector;
step S34d, according to the dimension and the characteristics of the feature subset, evaluating the effectiveness and the optimality of the feature subset, wherein the effectiveness and the optimality comprise classification accuracy, dimension reduction rate and feature redundancy; an optimal feature subset is obtained.
8. The full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating system for executing the method of any one of claims 1 to 7, comprising a high-frequency partial discharge sensor, a power frequency synchronous sensor, a programmable gate array FPGA, and a PRPD display module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410085528.7A CN117630611B (en) | 2024-01-22 | 2024-01-22 | Full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410085528.7A CN117630611B (en) | 2024-01-22 | 2024-01-22 | Full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117630611A CN117630611A (en) | 2024-03-01 |
| CN117630611B true CN117630611B (en) | 2024-04-12 |
Family
ID=90035855
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410085528.7A Active CN117630611B (en) | 2024-01-22 | 2024-01-22 | Full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117630611B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117892067A (en) * | 2024-03-15 | 2024-04-16 | 国网上海市电力公司 | Low-frequency current partial discharge monitoring anti-interference method, device, equipment and medium |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100853725B1 (en) * | 2007-06-20 | 2008-08-22 | (주) 피에스디테크 | Gis partial discharge diagnostic method and system thereof with prps algorithm |
| CN109116203A (en) * | 2018-10-31 | 2019-01-01 | 红相股份有限公司 | Power equipment partial discharges fault diagnostic method based on convolutional neural networks |
| KR102007287B1 (en) * | 2018-05-09 | 2019-08-06 | 한국전력공사 | Method and device for detecting partial discharge signal of stand alone type underground cable |
| CN111220882A (en) * | 2019-12-31 | 2020-06-02 | 国网辽宁省电力有限公司电力科学研究院 | Method for extracting characteristic quantity of GIS optical partial discharge spectrum |
| CN111967338A (en) * | 2020-07-27 | 2020-11-20 | 广东电网有限责任公司广州供电局 | Method and system for distinguishing partial discharge pulse interference signal based on mean shift clustering algorithm |
| CN112034312A (en) * | 2020-08-07 | 2020-12-04 | 中国南方电网有限责任公司超高压输电公司检修试验中心 | Power equipment insulation defect mode identification method |
| CN112149549A (en) * | 2020-09-18 | 2020-12-29 | 国网山东省电力公司泰安供电公司 | GIS partial discharge type identification method based on depth residual error network |
| CN113780308A (en) * | 2021-08-27 | 2021-12-10 | 吉林省电力科学研究院有限公司 | GIS partial discharge mode identification method and system based on kernel principal component analysis and neural network |
| KR102547467B1 (en) * | 2022-12-23 | 2023-06-26 | 주식회사 솔라리치 | Distribution board with partial discharge diagnosis function using ultra high frequency signal |
| KR20230104526A (en) * | 2021-12-30 | 2023-07-10 | 엘에스전선 주식회사 | Partial Discharge Monitoring System And Partial Discharge Monitoring Method |
| CN116910648A (en) * | 2023-07-07 | 2023-10-20 | 湖南五凌电力科技有限公司 | GIS equipment partial discharge spectrum analysis method, system and medium |
| CN117242487A (en) * | 2022-06-27 | 2023-12-15 | 上海格鲁布科技有限公司 | Separation and identification method for multisource mixed ultrahigh frequency partial discharge spectrum |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9689909B2 (en) * | 2012-01-11 | 2017-06-27 | Cablewise Techimp Limited | System for analyzing and locating partial discharges |
-
2024
- 2024-01-22 CN CN202410085528.7A patent/CN117630611B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100853725B1 (en) * | 2007-06-20 | 2008-08-22 | (주) 피에스디테크 | Gis partial discharge diagnostic method and system thereof with prps algorithm |
| KR102007287B1 (en) * | 2018-05-09 | 2019-08-06 | 한국전력공사 | Method and device for detecting partial discharge signal of stand alone type underground cable |
| CN109116203A (en) * | 2018-10-31 | 2019-01-01 | 红相股份有限公司 | Power equipment partial discharges fault diagnostic method based on convolutional neural networks |
| CN111220882A (en) * | 2019-12-31 | 2020-06-02 | 国网辽宁省电力有限公司电力科学研究院 | Method for extracting characteristic quantity of GIS optical partial discharge spectrum |
| CN111967338A (en) * | 2020-07-27 | 2020-11-20 | 广东电网有限责任公司广州供电局 | Method and system for distinguishing partial discharge pulse interference signal based on mean shift clustering algorithm |
| CN112034312A (en) * | 2020-08-07 | 2020-12-04 | 中国南方电网有限责任公司超高压输电公司检修试验中心 | Power equipment insulation defect mode identification method |
| CN112149549A (en) * | 2020-09-18 | 2020-12-29 | 国网山东省电力公司泰安供电公司 | GIS partial discharge type identification method based on depth residual error network |
| CN113780308A (en) * | 2021-08-27 | 2021-12-10 | 吉林省电力科学研究院有限公司 | GIS partial discharge mode identification method and system based on kernel principal component analysis and neural network |
| KR20230104526A (en) * | 2021-12-30 | 2023-07-10 | 엘에스전선 주식회사 | Partial Discharge Monitoring System And Partial Discharge Monitoring Method |
| CN117242487A (en) * | 2022-06-27 | 2023-12-15 | 上海格鲁布科技有限公司 | Separation and identification method for multisource mixed ultrahigh frequency partial discharge spectrum |
| KR102547467B1 (en) * | 2022-12-23 | 2023-06-26 | 주식회사 솔라리치 | Distribution board with partial discharge diagnosis function using ultra high frequency signal |
| CN116910648A (en) * | 2023-07-07 | 2023-10-20 | 湖南五凌电力科技有限公司 | GIS equipment partial discharge spectrum analysis method, system and medium |
Non-Patent Citations (2)
| Title |
|---|
| An Investigation on PRPD Patterns Generated by the UHF Measurement Technique for Power Transformers;Mohammad Akbari Azirani等;2017 25th Iranian Conference on Electrical Engineering (ICEE);20171231;1144-1147 * |
| 基于放电图谱仿真的局部放电检测仪性能评估方法及系统研制;胡岳;Chiampi M;Crotti G;张千千;;电气工程学报;20150525(第05期);91-97 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117630611A (en) | 2024-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN117630611B (en) | Full-bandwidth high-frequency partial discharge PRPD spectrogram capturing and generating method and system | |
| CN110792563B (en) | Wind turbine generator blade fault audio monitoring method based on convolution generation countermeasure network | |
| CN108573225B (en) | Partial discharge signal pattern recognition method and system | |
| CN112034310A (en) | Partial discharge defect diagnosis method and system for combined electrical appliance | |
| US11443137B2 (en) | Method and apparatus for detecting signal features | |
| CN110161388A (en) | A kind of the fault type recognition method and its system of high-tension apparatus | |
| CN112507479B (en) | Oil drilling machine health state assessment method based on manifold learning and softmax | |
| CN113850330A (en) | Power distribution network fault cause detection method based on short-time Fourier transform and convolutional neural network | |
| CN113805018A (en) | Intelligent identification method for partial discharge fault type of 10kV cable of power distribution network | |
| CN111860447A (en) | Method for accurately identifying telemetry time sequence data mode of random phase shift satellite | |
| CN116520068A (en) | Diagnostic method, device, equipment and storage medium for electric power data | |
| CN114492146B (en) | Bolt group loosening positioning and quantitative analysis method and system based on transfer learning | |
| CN115659271A (en) | Sensor abnormality detection method, model training method, system, device, and medium | |
| CN115880472A (en) | Intelligent diagnosis and analysis system for electric power infrared image data | |
| CN111597957B (en) | Transformer winding fault diagnosis method based on morphological image processing | |
| Chang et al. | Partial discharge pattern recognition for underground cable joints using convolutional neural network | |
| Ahmed | Development of a continuous condition monitoring system based on probabilistic modelling of partial discharge data for polymeric insulation cables | |
| CN113762131A (en) | Method and terminal for detecting deformation position of oil casing double-layer pipe column of high-sulfur-content gas field | |
| CN112560674B (en) | Method and system for detecting sound signal quality | |
| Dopierała | Fault detection method for energy measurement systems equipped with a Rogowski coil using the coil's response to a unit voltage jump and a fully convolutional neural network | |
| CN117609874B (en) | Rock fault friction microseismic detection method and system based on integrated deep learning | |
| Chang et al. | Application of Convolutional Neural Networks on Insulation Status Prediction Using Partial discharges for Underground Cable Joints | |
| CN116660761B (en) | Lithium ion battery detection method and system | |
| CN113743355B (en) | Switch device state checking method, device, system and computer equipment | |
| TWI779649B (en) | Testing system, judging device, testing method for testing insulation of a circuit board, and computer-readable recording medium thereof |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |