A multi-type partial discharge identification method based on XGBOOST algorithm

Through the combined method of wavelet denoising and feature extraction combined with XGBOOST algorithm, the problem of difficulty in distinguishing localized release types in traditional methods is solved, and high accuracy recognition of HFCT signals is achieved, especially in improving computing efficiency and recognition accuracy, which has achieved significant results. Results.

CN118885799BActive Publication Date: 2024-12-06NANJING FUHUA XINNENG TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411328948.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-24
Publication Date
2024-12-06
Estimated Expiration
2044-09-24

AI Technical Summary

Technical Problem

Traditional signal processing methods are difficult to accurately distinguish different types of locally distributed signals, especially because the time domain waveforms of different types of locally distributed signals are relatively close, resulting in low recognition accuracy.

Method used

The wavelet denoising algorithm is used to preprocess the HFCT signal, calculate the short-time energy and zero-crossing rate in frames, intercept the discharge pulse waveform, extract the equivalent duration, equivalent bandwidth and fuzzy entropy as feature vectors, and classify it using the XGBOOST algorithm. Identify that only the fuzzy entropy of the discharge pulse is calculated to improve efficiency.

Benefits of technology

It realizes high accuracy recognition of three types of locally distributed signals: tip, suspension and edge surface, improving the operating efficiency and accuracy of the identification system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118885799B_ABST
    Figure CN118885799B_ABST
Patent Text Reader

Abstract

The present invention provides a method for multi-type identification of partial discharge based on HFCT signals, comprising the following steps: Step 1, setting up an experimental environment and collecting HFCT partial discharge signals. Step 2, using an endpoint detection method based on short-time energy and zero-crossing rate to intercept discharge pulses in the collected HFCT data of different discharge types. Step 3, extracting three features of equivalent duration, equivalent bandwidth and fuzzy entropy of the discharge pulse to form a feature vector. Step 4, selecting the xgboost algorithm as a classifier to classify and identify the extracted feature quantities. The method can effectively identify three different discharge types: tip, suspension and surface. Step 5, using the trained model to classify and identify the discharge pulses in the partial discharge data with a duration of 1s and statistically classify the classification results, and taking the type with the largest number of pulses as the discharge type of the data. The method can effectively determine whether partial discharge occurs and its discharge type in the detection of various electrical equipment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of partial discharge detection of electrical equipment and relates to a multi-type partial discharge identification method based on an XGBOOST algorithm. Background Art

[0002] Although the traditional PRPD diagram can better reflect whether partial discharge has occurred, it is difficult to accurately distinguish the type of partial discharge from the PRPD diagram in actual engineering applications. In addition, the time domain waveforms of partial discharge signals of different partial discharge types are relatively similar, and traditional signal processing methods are unable to cope with this problem. Summary of the invention

[0003] Purpose of the invention: The technical problem to be solved by the present invention is that it is difficult for traditional signal processing methods to effectively distinguish the types of discharges, and a method is provided for effectively identifying and distinguishing different types of partial discharge signals.

[0004] The method of the present invention specifically comprises the following steps:

[0005] Step 1: Establish a partial discharge experimental platform, collect three types of HFCT partial discharge signals, namely, tip, suspension and surface, based on a high-speed acquisition card, and then use a wavelet denoising algorithm to pre-process the collected HFCT signals.

[0006] Step 2: divide the collected HFCT signal into frames, calculate the short-time energy and zero-crossing rate of each frame, and extract the discharge pulse waveform from the time domain signal according to the set threshold.

[0007] Step 3: Calculate the equivalent duration, equivalent bandwidth and fuzzy entropy of the discharge pulse to form a feature vector.

[0008] Step 4: Use the XGBOOST algorithm to train the constructed feature vector for the classifier, where 70% of the data is used to train the model and 30% of the data is used as a test set for verification.

[0009] Step 5: Use the trained model to classify and identify each pulse in the 1s data, and count the number of pulses of each type. The type with the largest number of pulses is taken as the discharge type of the data.

[0010] The improvement of the present invention is that different types of discharge signals are not directly analyzed, but discharge pulses are intercepted from time domain signals, and different discharge types are distinguished by analyzing the discharge pulses. In particular, the calculation of fuzzy entropy no longer directly calculates the fuzzy entropy of the complete time domain signal but only calculates the fuzzy entropy value of a complete discharge pulse.

[0011] Since the time domain waveforms of different types of discharge pulses are relatively close, traditional signal features such as peak-to-peak value, mean or variance cannot distinguish different types of discharge pulses well. Therefore, the equivalent time length and equivalent bandwidth are extracted as input parameters according to the signal analysis method of the equivalent time-frequency method. The size of the fuzzy entropy is closely related to the frequency component in the signal, so the fuzzy entropy of the discharge pulse is extracted as an auxiliary feature. In practical applications, fuzzy entropy is usually used to calculate a complete time domain signal of a certain duration. However, due to the high measurement bandwidth of the HFCT signal (3-30MHz), the amount of signal data collected without distortion is large, and the calculation efficiency of the fuzzy entropy will be greatly reduced, and the entire recognition system will lose its practicality. The duration of the HFCT discharge pulse is usually in the nanosecond level, which is short and the discharge pulse still retains rich frequency information. Therefore, only calculating the fuzzy entropy of the discharge pulse can not only reflect the characteristics of different discharge types, but also greatly improve the operating efficiency of the entire recognition system.

[0012] In summary, compared with the existing recognition methods, this method is simple to implement, has high operating efficiency, and has high recognition accuracy for the three types of discharges: tip, suspension, and surface. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a flowchart of the method of the present invention.

[0014] Figure 2 It is the waveform diagram of the discharge pulse at the intercepted tip, suspension and along the surface.

[0015] Figure 3 It is the TF-map of tip, suspension and along the surface. DETAILED DESCRIPTION

[0016] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more clear.

[0017] The present invention provides a partial discharge identification method capable of distinguishing different types of discharges, such as Figure 1 As shown, the specific implementation steps are as follows:

[0018] Step 1: Use the wavelet algorithm to pre-process the collected partial discharge signal to remove noise.

[0019] Step 2: Divide the denoised partial discharge signal into frames, calculate the short-time energy and zero-crossing rate of each frame, and set the threshold value. According to the set threshold value, the discharge pulse signal is intercepted from the original signal, such as Figure 3 As shown, data with a duration of 0.2s is taken as a batch processing object, the frame length is set to 200, and the frame shift is set to 50.

[0020] Step 3: Calculate the equivalent duration, equivalent bandwidth and fuzzy entropy of the pulse signal. The calculation method of the equivalent duration and equivalent bandwidth is as follows:

[0021] Assume that the time pulse waveform is s(t), t is the sampling time, and .

[0022] First, the signal s(t) is normalized.

[0023] ,

[0024] It is the time center.

[0025] After obtaining the time center of gravity, the time domain equivalent time length of a single waveform is calculated using the standard deviation method. and frequency domain equivalent bandwidth ,

[0026] ,

[0027] ,,

[0028] f is the sampling frequency, S(f) is the Fourier transform of the standardized signal s(t).

[0029] Since the collected signals are all discrete time series, their time center of gravity is:

[0030] .

[0031] The frequency center of gravity is:

[0032] .

[0033] The equivalent duration is:

[0034] .

[0035] The equivalent bandwidth is:

[0036] .

[0037] Step 4: Equivalent duration and equivalent bandwidth have been widely studied and applied as important features for distinguishing different discharge pulses. In the experiment, it was found that the combination of equivalent duration and equivalent bandwidth with the XGBOOST algorithm can achieve a high recognition accuracy. However, if the recognition accuracy is to be further improved, new features must be added or the classification algorithm must be improved. Fuzzy entropy is an entropy feature, and its value is closely related to the frequency component in the signal. However, in signal processing, the fuzzy entropy value of the entire signal is usually calculated instead of the fuzzy entropy size of the discharge pulse. However, due to the high measurement bandwidth and large data volume of HFCT, direct calculation of fuzzy entropy will be inefficient and difficult to meet actual needs. At the same time, different types of discharge pulses still retain certain frequency information and the center frequencies of different discharge types are different. Therefore, in theory, the fuzzy entropy size of different discharge pulses will also be different. In addition, different discharge pulses have relatively similar time domain waveforms. Statistical features based on time domain waveforms such as mean and variance are difficult to play a role. Therefore, it is proposed to calculate the fuzzy entropy of the discharge pulse as an auxiliary feature.

[0038] The extracted equivalent time, equivalent bandwidth and fuzzy entropy constitute the feature vector Z=[T1,T2,T3]. 70% of the collected partial discharge data are used as training sets and 30% as test sets. Then the XGBOOST algorithm is selected as the classifier for training. The learning rate of the XGBOOST algorithm is set to 0.1, the maximum depth of the tree is 6, the number of trees is set to 300, and the objective function is softmax. According to the test set test results, the average accuracy of the classification model based on XGBOOST reached 99.7%. Compared with only extracting the two features of equivalent time and equivalent bandwidth, the accuracy has been further improved. It can effectively identify three types of HFCT partial discharge signals: tip, suspension and surface, such as Figure 3 shown.

[0039] Step 5: Use the trained model to classify and identify each discharge pulse in the data of the specified duration, then count the number of pulses of each type, and take the type with the largest number of pulses as the discharge type of the data.

Claims

1. Multi-type partial discharge identification method based on XGBOOST algorithm, Features: The following steps are involved: Step 1: De-noising preprocessing of the acquired HFCT signal; Step 2: Divide the acquired HFCT signal into frames, calculate the short-time energy and zero-crossing rate of each frame, and extract the discharge pulse waveform from the time domain signal according to the set threshold; Step 3: Calculate the equivalent duration of the discharge pulse, the equivalent bandwidth and the fuzzy entropy to construct the feature vector; Step 4: Train the model on the constructed feature vector; Step 5: Use the trained model to self-time the pulse type and perform statistics, and determine the discharge type based on the pulse statistics. In step 4, the extracted equivalent duration, equivalent bandwidth and fuzzy entropy constitute the feature vector Z=[T1, T2, T3]. The XGBOOST algorithm is used as the classifier to train the constructed feature vector, where 70% of the data is used to train the model and 30% of the data is used as a test set for verification. The specified duration data is used as the processing unit to classify and identify the discharge pulses in the time domain waveform, and then the number of different types of pulses is counted, and the pulse type with the largest number is taken as the discharge type of the data.

2. The method according to claim 1, Features: In step 1, the acquired HFCT data is preprocessed for denoising using a wavelet algorithm.

3. The method according to claim 1, Features: In step 3, the specific method is: Assume that the time pulse waveform is s(t), t is the sampling time, and , First, the signal s(t) is normalized. , That is the time center of gravity. After obtaining the time center of gravity, the time domain equivalent time length of a single waveform is calculated using the standard deviation method. and frequency domain equivalent bandwidth , , , f is the sampling frequency, S(f) is the Fourier transform of the standardized signal s(t), Since the collected signals are all discrete time series, their time center of gravity is: , The frequency center of gravity is: , The equivalent duration is: , The equivalent bandwidth is: .。 4. The method according to claim 1, Features: The learning rate of the XGBOOST algorithm is set to 0.1, the maximum depth of the tree is 6, the number of trees is set to 300, and the objective function is softmax.

5. The method according to claim 1, Features: The specified time length is within 1S.

Citation Information

Patent Citations

  • Cable joint typical defect partial discharge characterization feature mining and pattern recognition method

    CN114252738A

  • State-of-health models for lithium-silicon batteries

    US20230187710A1