A method for identifying a fault location voltage traveling wave of a secondary fusion complete equipment

Abstract

This application relates to the field of circuit fault detection, specifically to a voltage traveling wave identification method adapted to fault location in primary and secondary integrated equipment. This invention studies the time-frequency domain characteristics of fault traveling waves, uses extended feature extraction to extract different training samples, and combines sample labels with a minimum Euclidean distance classifier to achieve fault wave identification. Feature extraction is easy, feature space requirements are small, classifier computation is low, and fault wave identification can be achieved without a specific deep learning optimized MCU chip, thus having good practical value.

Description

Technical Field

[0001] This application relates to the field of circuit fault detection, specifically to a voltage traveling wave identification method adapted for fault location in primary and secondary integrated equipment. Background Technology

[0002] Currently, medium and low voltage power line carrier equipment all use voltage signals, resulting in numerous interference signals with varying amplitudes, periods, and frequencies in the power line environment. More and more fault location research companies are now focusing on anti-interference as a key research direction. For example, patent CN115308538B applies deep learning to fault wave identification and fault location. It uses collected field noise, simulated interference waveforms, and collected actual field voltage traveling waves as training data for deep feature extraction and fault wave identification. YOLOv5, however, is a single-stage target detection algorithm, meaning that target detection and location are trained and output simultaneously. This leads to high learning complexity, large computational resources, and large feature space storage requirements. Summary of the Invention

[0003] Based on the above problems, this application adopts a relatively simple least Euclidean distance classifier. By studying the distribution relationship of fault traveling waves at different scales, it proposes using the number of points passing through different amplitudes as a feature feature for identifying fault waves. This method is computationally simple, has a small feature space, and utilizes a fast screening method to distinguish between fault and non-fault waves. Combined with a time-matching method on the master station side, it can be effectively applied to distribution network environments. The technical solution is as follows:

[0004] A method for fault location voltage traveling wave identification adapted to primary and secondary integrated equipment includes the following steps:

[0005] S1. Collect noise data, carrier data and traveling wave data on the line, organize them into fault traveling wave waveform sets and non-fault traveling wave waveform sets, and perform tagging processing;

[0006] S2. Perform waveform preprocessing on the acquired fault traveling wave waveform and non-fault traveling wave waveform;

[0007] S3. Perform feature extraction on the preprocessed data, and store the extracted feature set and label set for classification;

[0008] S4. Preprocess and extract features from the actual acquired waveforms, and use the least Euclidean distance classifier to match the feature set and label set in S3.

[0009] Preferably, step S1 specifically involves the following steps:

[0010] S11. Noise data, carrier data, and traveling wave data on the line are collected using a capacitive voltage divider electronic voltage sensor;

[0011] S12. The phase mode transformation results in line mode data AB and CB, which are then digitally synthesized in the form of A-B+CB.

[0012] S13. Label and categorize the fault types.

[0013] Preferably, in step S13, the medium-voltage carrier signal, low-voltage carrier signal, fault traveling wave, and other interference noise are labeled with “0”, “1”, “2”, and “3” respectively.

[0014] Preferably, the waveform preprocessing in step S2 includes noise reduction, which uses mean filtering for noise reduction:

[0015] The number of points N in the mean filter f With sampling rate f s And related to the high-frequency components of the noise floor, the number of points N in the mean filter. f The calculation method is as follows:

[0016]

[0017] Where |·| represents the absolute value function, t w It is the window length for mean filtering;

[0018] The mean filter is calculated as follows:

[0019] ;

[0020] Among them, y i x represents the mean-filtered data at time i. i This represents the data before mean filtering at time i.

[0021] Preferably, the waveform preprocessing in step S2 includes dual-channel fitting processing:

[0022] Data is acquired through channels with different gains, and the acquired data is fitted according to the gain factor.

[0023] Assuming channel 1 is a low-gain acquisition channel and channel 2 is a high-gain acquisition channel, with the gain of channel 2 being α times that of channel 1, the low-gain channel values ​​are amplified by α times, and then compared with the high-gain channel data. The larger positive value is taken, and the smaller negative value is taken, resulting in the fitted data for both channels. This represents the data after dual-channel fitting at time i.

[0024] Preferably, the waveform preprocessing in step S2 includes logarithmic transformation: the data after dual-channel fitting is logarithmically transformed, as shown in the following formula:

[0025]

[0026] Represents the logarithmically transformed data at time i, log 10 (·) represents a base-10 logarithmic transformation function, which transforms the data after the two-channel fitting process. and logarithmically transformed data All of them are passed to the feature extraction module.

[0027] Preferably, step S3 involves feature extraction of the preprocessed data, and the specific method is as follows:

[0028] Feature extraction is represented by the number of amplitude crossing points. Assuming there are features at four amplitude levels...

[0029] (ρ1, ρ2, ρ3, ρ4), the calculation method is as follows:

[0030] ;

[0031] Where N(ρ) represents the number of crossing points when the amplitude is ρ, M represents the total number of sampling points of the normalized data, and sgn(·) represents the sign function. Represents data after logarithmic transformation Or data after two-channel fitting

[0032] Therefore, the feature set j represents the j-th data set, where the feature set and label corresponding to all data are stored.

[0033] Preferably, the formula for calculating the Euclidean distance in step S4 is as follows:

[0034]

[0035] Among them, R j S represents the Euclidean distance to the j-th feature in the feature set. e (i) represents the i-th feature value of the actual sampled data. The i-th feature value in the j-th group of the feature set represents the feature value of the i-th element.

[0036] Compared with the prior art, the beneficial effects of this application are as follows:

[0037] (1) The three-phase synthesis method replaces the feature extraction and training of the two-path model, reducing the number of data processing paths;

[0038] (2) Data preprocessing methods, including noise reduction processing, ensure signal quality and help improve the probability of fault wave identification when high impedance transition resistor is grounded.

[0039] (3) By using extended feature extraction of different training samples and combining sample labels with a minimum Euclidean distance classifier, the fault wave identification function is achieved. Feature extraction is easy, feature space requirements are small, classifier computation is low, and fault wave identification can be achieved without a specific deep learning optimized MCU chip, which has good practical value. Attached Figure Description

[0040] Figure 1 This is a flowchart of the application process;

[0041] Figure 2 This is a schematic diagram of dual-channel fitting processing. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0043] A method for fault location voltage traveling wave identification adapted to primary and secondary integrated equipment includes the following steps:

[0044] (1) Use a capacitive voltage divider electronic voltage sensor to collect noise data, carrier data and traveling wave data on the line, organize them into a fault traveling wave waveform set and a non-fault traveling wave waveform set, and then perform tagging processing.

[0045] Traditional traveling wave ranging requires acquiring three-phase voltages and then performing phase-mode transformation to obtain line-mode data. This line-mode data has two channels, AB and CB. Since the fault phase is not always known, two channels of data are used for fault wave identification, making data processing relatively complex. This invention proposes a three-phase digital synthesis method for fault wave identification, using the A-B+CB method. This method only requires one data channel for fault wave identification. Noise data, carrier data, and traveling wave data acquired by a capacitive voltage divider electronic voltage sensor are all synthesized using this method and then tagged and categorized. If only fault and non-fault traveling waves need to be identified, "0" represents a non-fault traveling wave and "1" represents a fault traveling wave. Alternatively, medium-voltage carrier signals, low-voltage carrier signals, fault traveling waves, and other interference noise can be categorized separately using "0", "1", "2", and "3" respectively.

[0046] (2) Perform waveform preprocessing on the collected fault traveling wave waveform and non-fault traveling wave waveform, including noise reduction, dual-channel fitting, and logarithmic transformation.

[0047] The noise reduction process primarily employs mean filtering because FIR filtering would disrupt the original state of the first half of the faulty traveling wave, causing the amplitude of the first half to extend forward and decrease. The number of points N in the mean filtering process is crucial. f With sampling rate f s And it is related to the high-frequency components of the noise floor, generally the window length t w The filter duration is 200ns, and the number of points is odd, meaning the number of points N in the mean filter is... f The calculation method is as follows:

[0048]

[0049] Where |·| represents the absolute value function, t w The window length for mean filtering ranges from 100ns to 300ns, which is approximately 1 / 4 of the time-domain period corresponding to the highest frequency of the traveling wave.

[0050] The mean filter is calculated as follows:

[0051] ;

[0052] Among them, y i x represents the mean-filtered data at time i. i This represents the data before mean filtering at time i.

[0053] The calculation method for dual-channel fitting is as follows:

[0054] To improve dynamic range and enhance product adaptability under varying distances and transition resistances, and to avoid issues such as small signal rejection and large signal clamping affecting recognition performance, data is acquired through channels with different gains. The acquired data is then fitted according to the gain factor, as follows:

[0055] Channel 1 is a low-gain acquisition channel, and Channel 2 is a high-gain acquisition channel. The gain of Channel 2 is α times that of Channel 1. Then, the data will be processed as follows: Figure 2 As shown.

[0056] The values ​​of the low-gain channel are amplified by a factor of α, and compared with the data from the high-gain channel, the larger positive value is taken, and the smaller negative value is taken, to obtain the data after dual-channel fitting. represent i The data is processed by fitting the two channels at time points, and then a logarithmic transformation is performed on the processed data.

[0057] The calculation method for logarithmic transformation is as follows:

[0058]

[0059] Represents the logarithmically transformed data at time i, log 10 (·) represents a base-10 logarithmic transformation function, which transforms the data after the two-channel fitting process. and logarithmically transformed data All values ​​are passed to the feature extraction module; (the purpose of logarithmic transformation is to better extract drastically changing feature values).

[0060] (3) Perform feature extraction on the preprocessed data, and store the extracted feature set and label set for classification;

[0061] Feature extraction is represented by the number of amplitude crossing points, with four amplitude values ​​ρ selected as (0.1, 0.3, 0.5, and 0.8), and their calculation methods are as follows:

[0062] ;

[0063] Where N(ρ) represents the number of crossing points when the amplitude is ρ, M represents the total number of sampling points of the normalized data, and sgn(·) represents the sign function. Represents data after logarithmic transformation Or data after two-channel fitting

[0064] Therefore, the feature set j represents the j-th data set, where the feature set and label corresponding to all data are stored.

[0065] (4) The waveforms actually collected on site are preprocessed and feature extracted, and the recognition results are obtained by matching the feature set and label set in (3) with the minimum Euclidean distance classifier.

[0066] Following steps two and three, the actual sampled data is preprocessed and features are extracted. The resulting features are then compared with the features obtained in step three using Euclidean distance calculation. The label corresponding to the smallest Euclidean distance is the recognition result for that set of data. The formula for calculating the Euclidean distance is as follows:

[0067]

[0068] Among them, R j S represents the Euclidean distance to the j-th feature in the feature set. e (i) represents the i-th feature value of the actual sampled data. The i-th feature value in the j-th group of the feature set represents the feature value of the i-th element.

[0069] The same batch of data was divided into two halves for training and the other half for success rate statistics. Compared with the YOLOv5-based distribution network fault location method, the success rates of the present invention were 98.2% and 98.5%, respectively. However, the data processing complexity of the present invention is much lower than that of the comparison document.

[0070] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0071] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A voltage traveling wave identification method adapted for fault location in primary and secondary integrated equipment, characterized in that, Includes the following steps: S1. Collect noise data, carrier data and traveling wave data on the line, organize them into fault traveling wave waveform sets and non-fault traveling wave waveform sets, and perform tagging processing; S2. Perform waveform preprocessing on the acquired fault traveling wave waveform and non-fault traveling wave waveform; S3. Perform feature extraction on the preprocessed data, and store the extracted feature set and label set for classification; S4. Preprocess and extract features from the actual acquired waveforms, and use the least Euclidean distance classifier to match the feature set and label set in S3.

2. The voltage traveling wave identification method for fault location in a primary and secondary integrated equipment as described in claim 1, characterized in that, The specific steps of step S1 are as follows: S11. Noise data, carrier data, and traveling wave data on the line are collected using a capacitive voltage divider electronic voltage sensor; S12. The phase mode transformation results in line mode data AB and CB, which are then digitally synthesized in the form of A-B+CB. S13. Label and categorize the fault types.

3. The voltage traveling wave identification method adapted to a fault-finding type primary and secondary fusion system according to claim 2, characterized in that, In step S13, the medium-voltage carrier signal, low-voltage carrier signal, fault traveling wave, and other interference noise are labeled with "0", "1", "2", and "3" respectively.

4. The voltage traveling wave identification method for fault location in a primary and secondary integrated equipment as described in claim 1, characterized in that, Step S2 waveform preprocessing includes noise reduction, which uses mean filtering: the number of points N in the mean filtering is... f With sampling rate f s And related to the high-frequency components of the noise floor, the number of points N in the mean filter. f The calculation method is as follows: Where |·| represents the absolute value function, t w It is the window length for mean filtering; The mean filter is calculated as follows: ； Among them, y i x represents the mean-filtered data at time i. i This represents the data before mean filtering at time i.

5. The voltage traveling wave identification method for fault location in a primary and secondary integrated equipment as described in claim 1, characterized in that, Step S2 waveform preprocessing includes dual-channel fitting: Data is acquired through channels with different gains, and the acquired data is fitted according to the gain factor. Assuming channel 1 is a low-gain acquisition channel and channel 2 is a high-gain acquisition channel, with the gain of channel 2 being α times that of channel 1, the low-gain channel values ​​are amplified by α times, and then compared with the high-gain channel data. The larger positive value is taken, and the smaller negative value is taken, resulting in the fitted data for both channels. This represents the data after dual-channel fitting at time i.

6. The voltage traveling wave identification method for fault location in a primary and secondary integrated equipment as described in claim 1, characterized in that, Step S2 waveform preprocessing includes logarithmic transformation: the data after dual-channel fitting is logarithmically transformed, as shown in the following formula: Represents the logarithmically transformed data at time i, log 10 (·) represents a base-10 logarithmic transformation function, which transforms the data after the two-channel fitting process. and logarithmically transformed data All of them are passed to the feature extraction module.

7. The voltage traveling wave identification method for fault location in a primary and secondary integrated equipment as described in claim 1, characterized in that, Step S3 involves feature extraction from the preprocessed data. The specific method is as follows: Feature extraction is represented by the number of amplitude crossing points. Assuming there are features at four amplitude levels... (ρ1, ρ2, ρ3, ρ4), the calculation method is as follows: ； Where N(ρ) represents the number of crossing points when the amplitude is ρ, M represents the total number of sampling points of the normalized data, and sgn(·) represents the sign function. Represents data after logarithmic transformation Or data after two-channel fitting Therefore, the feature set j represents the j-th data set, where the feature set and label corresponding to all data are stored.

8. The voltage traveling wave identification method for fault location in a primary and secondary integrated equipment as described in claim 1, characterized in that, The formula for calculating the Euclidean distance in step S4 is as follows: Among them, R j S represents the Euclidean distance to the j-th feature in the feature set. e (i) represents the i-th feature value of the actual sampled data. The i-th feature value in the j-th group of the feature set represents the feature value of the i-th element.

Images