Partial discharge identification method based on prior rule and number source information fusion

Abstract

The invention discloses a partial discharge identification method based on priori rule and data source information fusion, which comprises the following steps: firstly, carrying out evidence rule calculation on detection data, judging whether the detection data triggers an evidence rule, if the detection data triggers a strong evidence rule, directly outputting a discharge type as a strong evidence diagnosis result, and if the detection data triggers the strong evidence rule, directly outputting the discharge type as the strong evidence diagnosis result; the invention belongs to the technical field of discharge mode recognition, and aims at solving the problem that in the prior art, the partial discharge phenomenon has great randomness. The technical effects of the method are that two knowledge dimensions of the evidence theory and the number source information are fused, final voting fusion decision is carried out, and the most reasonable diagnosis evaluation result is given. Compared with a traditional diagnosis method, the novel expert diagnosis system provided by the invention gives consideration to expert experience evidence and actually detected data source information feature fingerprint mapping identification, has a very high error-tolerant rate, and particularly has higher type identification accuracy under the conditions of interference and insufficient data detection quantity.

Description

Technical Field

[0001] The invention relates to the technical field of discharge pattern recognition, and in particular to a local discharge recognition method based on a priori rules and digital source information fusion. Background Art

[0002] At present, the mainstream diagnostic strategies for partial discharge detection devices used in transformers mainly rely on detection spectra, including PRPD spectra, PRPS spectra, etc. These methods are essentially based on the diagnosis of digital source information. The implementation of the diagnostic method strongly depends on two factors:

[0003] ① Accuracy of spectrum detection

[0004] ②The completeness of the algorithm’s own prior graph library

[0005] The accuracy of spectrum detection is easily affected by factors such as on-site detection strategies, interference conditions, and cumulative detection time, and it is easy to produce incomplete spectra or spectra contaminated by interference, which in turn leads to the failure of the diagnostic algorithm.

[0006] The algorithm's own atlas library is generally obtained through measurements of typical laboratory models. However, there is a big difference between the laboratory model and the actual discharge conditions on site. Especially for large transformers, the discharge conditions are complex and changeable. Pure laboratory data cannot guarantee the completeness of fault data. The local discharge phenomenon itself is very random, which has a fatal impact on the accuracy of diagnosis. Summary of the invention

[0007] To this end, the present invention provides a partial discharge identification method based on a priori rules and digital source information fusion to solve the above problems in the prior art.

[0008] In order to achieve the above object, the present invention provides the following technical solutions:

[0009] According to a first aspect of the present invention, a method for identifying partial discharge based on a priori rules and digital source information fusion comprises the following steps:

[0010] S1: First, the evidence rule calculation is performed on the detection data to determine whether the detection data triggers the evidence rule; S2: If the strong evidence rule is triggered, the discharge type is directly output as a strong evidence diagnosis result;

[0011] S3: If only the weak evidence rule is triggered, the diagnosis result of the evidence rule and the diagnosis result based on the source information are voted for consistency. If the two diagnosis results are consistent, the diagnosis result is output;

[0012] S4: If the two diagnostic results are inconsistent, an unknown type result is output;

[0013] S5: If the evidence rule is not triggered, the diagnosis result of the digital source information is used as the final diagnosis output result;

[0014] Specifically, the detection data is derived from the spectral data or statistical data feature quantity of the discharge, specifically three information quantities of each discharge pulse: discharge peak value v, discharge phase φ and discharge time scale t. The discharge peak value v refers to the peak point of each discharge pulse, the discharge phase φ refers to the phase of the power frequency sine wave corresponding to the discharge pulse, and the discharge time refers to the moment information of the discharge pulse on the time axis;

[0015] The evidence rules are empirical rule tables that present the discharge signals triggered by different defect types. The diagnosis of digital source information adopts a hierarchical identification strategy at the digital source information knowledge dimension level.

[0016] The first level of diagnosis completes the identification of discharge signals and interference signals;

[0017] The second level of diagnosis completes the identification of discharge signals and discharge interference types respectively;

[0018] The third level of diagnosis completes the specific distinction and identification of gas discharge and solid insulation discharge defects;

[0019] Each level of diagnosis uses different characteristic fingerprint information to achieve the best diagnostic accuracy;

[0020] The calculation and triggering of the rules of evidence are as follows:

[0021] Q1: Perform time-division statistics on the original amplitude sequence V. Δ Take t as an example for time segment statistics, and the new amplitude sequence formed after statistics is recorded as W;

[0022] Q2: W is the amplitude sequence, dt = Δt constructs a waveform sequence, and performs FFT transformation on the sequence. The FFT transformation result is recorded as YF, where F is the frequency and Y is the amplitude corresponding to each frequency point;

[0023] Q3: Find the Y value when F=100 and F=50, record them as Y / 100 and Y / 50 respectively, Y / 100 is the 100Hz component of the original amplitude sequence V, Y / 50 is the 50Hz component of the original amplitude sequence V; Q4: Find the Y value when F=10, which is Y / 10, the 10Hz component of the original amplitude sequence V, as a reference value;

[0024] Q5: If Y / 100 ≥ 2×Y / 10, the original amplitude sequence V is considered to have a 100Hz correlation, which is considered to trigger the evidence rule;

[0025] If Y / 50≥2×Y / 10, the original amplitude sequence V is considered to have a 50Hz correlation, and the evidence rule is determined to be triggered; otherwise, the original amplitude sequence V is considered to have no 100Hz or 50Hz correlation, and the evidence rule is determined not to be triggered.

[0026] Furthermore, when the corresponding evidence rules are triggered, the diagnostic system will give the relevant corresponding types a very high decision weight. The prior evidence rules are intended to make a priori distinctions among various discharge modes based on five characteristic distributions: 50Hz / 100Hz correlation, polarity effect, "wing-shaped spectrum" characteristics, spectrum center of gravity distribution, and entropy distribution.

[0027] Furthermore, the method comprises the following specific steps:

[0028] P1: Based on the high-frequency current and ultrasonic test data obtained from the test, a V-φ scatter plot is first drawn;

[0029] P2: The DBSCAN clustering algorithm is used to obtain the separation data of the positive and negative half axes of the power frequency phase, and to detect and judge the polarity effect and wing-shaped spectrum;

[0030] P3: Draw the NV spectrum again, and calculate the amplitude entropy distribution and center of gravity distribution according to the NV spectrum distribution;

[0031] P4: Resample and divide the data into grids for statistics, and the energy of the 50Hz / 100Hz frequency can be calculated.

[0032] Furthermore, the detected PRPD spectrum, PRPS spectrum and NV spectrum are extracted according to the diagnostic criterion results, and the fingerprint feature quantity of the characteristic spectrum is extracted.

[0033] Furthermore, the hierarchical diagnosis network is used to output the diagnosis results to form the diagnosis output of the digital source information. The evidence rule diagnosis results are integrated with the diagnosis of the digital source information to form an expert diagnosis system based on the fusion decision of evidence rules and digital source information. The diagnosis system outputs the final diagnosis result according to the voting decision.

[0034] The present invention has the following advantages: the present invention integrates the two knowledge dimensions of evidence theory and digital source information, and makes a final voting fusion decision to give the most reasonable diagnostic evaluation result. Compared with the traditional diagnostic method, the new expert diagnosis system proposed in this project takes into account the expert experience evidence and the digital source information feature fingerprint mapping recognition of actual detection, and has a high fault tolerance rate, especially in the presence of interference and insufficient data detection, it has a higher type recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A roadmap for expert diagnostic methods for fusion decision making;

[0036] Figure 2 Develop a methodological roadmap for a priori evidence rules;

[0037] Figure 3 To provide a roadmap for decision-making diagnostic methods based on evidence rules and digital source information fusion;

[0038] Figure 4 This is a grid statistical diagram of the Vt sequence. DETAILED DESCRIPTION

[0039] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Example 1

[0041] like Figures 1 to 4 As shown, a partial discharge identification method based on a priori rules and digital source information fusion in an embodiment of the first aspect of the present invention comprises the following steps:

[0042] S1: First, the evidence rule calculation is performed on the detection data to determine whether the detection data triggers the evidence rule; S2: If the strong evidence rule is triggered, the discharge type is directly output as a strong evidence diagnosis result;

[0043] S3: If only the weak evidence rule is triggered, the diagnosis result of the evidence rule and the diagnosis result based on the source information are voted for consistency. If the two diagnosis results are consistent, the diagnosis result is output;

[0044] S4: If the two diagnostic results are inconsistent, an unknown type result is output;

[0045] S5: If the evidence rule is not triggered, the diagnosis result of the digital source information is used as the final diagnosis output result;

[0046] Specifically, the detection data is derived from the discharge spectrum data or statistical data feature quantity, and three information quantities of each discharge pulse: discharge peak value v, discharge phase φ and discharge time scale t. The discharge peak value v refers to the peak point of each discharge pulse, the discharge phase φ refers to the phase of the power frequency sine wave corresponding to the discharge pulse, and the discharge time refers to the moment information of the discharge pulse on the time axis;

[0047] The evidence rules are empirical rule tables that present the discharge signals triggered by different defect types. The diagnosis of digital source information adopts a hierarchical identification strategy at the digital source information knowledge dimension level.

[0048] The first level of diagnosis completes the identification of discharge signals and interference signals;

[0049] The second level of diagnosis completes the identification of discharge signals and discharge interference types respectively;

[0050] The third level of diagnosis completes the specific distinction and identification of gas discharge and solid insulation discharge defects;

[0051] Each level of diagnosis uses different characteristic fingerprint information to achieve the best diagnostic accuracy;

[0052] The calculation and triggering of the rules of evidence are as follows:

[0053] Q1: Perform time-division statistics on the original amplitude sequence V. Figure 4 Shown with Δ Take t as an example for time segment statistics, and the new amplitude sequence formed after statistics is recorded as W;

[0054] Q2: W is the amplitude sequence, dt = Δt constructs a waveform sequence, and performs FFT transformation on the sequence. The FFT transformation result is recorded as YF, where F is the frequency and Y is the amplitude corresponding to each frequency point;

[0055] Q3: Find the Y value when F=100 and F=50, record them as Y / 100 and Y / 50 respectively, Y / 100 is the 100Hz component of the original amplitude sequence V, Y / 50 is the 50Hz component of the original amplitude sequence V; Q4: Find the Y value when F=10, which is Y / 10, the 10Hz component of the original amplitude sequence V, as a reference value;

[0056] Q5: If Y / 100 ≥ 2×Y / 10, the original amplitude sequence V is considered to have a 100Hz correlation, which is considered to trigger the evidence rule;

[0057] If Y / 50≥2×Y / 10, the original amplitude sequence V is considered to have 50Hz correlation, which is considered to trigger the evidence rule; otherwise, the original amplitude sequence V is considered to have no 100Hz or 50Hz correlation, which is considered not to trigger the evidence rule;

[0058] In the above embodiment, it should be noted that the evidence rule table for different types of discharges is as follows:

[0059]

[0060]

[0061]

[0062] Example 2

[0063] like Figures 1 to 3 As shown, a partial discharge identification method based on prior rules and digital source information fusion includes all the contents of Example 1. In addition, when the corresponding evidence rule is triggered, the diagnostic system will give the relevant corresponding type a high decision weight. The prior evidence rule intends to distinguish various discharge modes a priori from five characteristic distributions: 50Hz / 100Hz correlation, polarity effect, "wing-shaped spectrum" characteristics, spectrum center of gravity distribution, and entropy value distribution;

[0064] In the above embodiment, it should be noted that for the PRPD, PRPS, and NV distribution maps of high-frequency and ultrasonic detection information, a suitable criterion function is selected based on the principle of highest recognition accuracy, and 23 phase-independent characteristic parameters are selected, and 10 characteristic parameters that are most suitable for discharge and interference type identification are selected.

[0065] Example 3

[0066] like Figures 1 to 3 As shown, a partial discharge identification method based on a priori rules and digital source information fusion includes all the contents of embodiment 2, and further includes the following specific steps:

[0067] P1: Based on the high-frequency current and ultrasonic test data obtained from the test, a V-φ scatter plot is first drawn;

[0068] P2: The DBSCAN clustering algorithm is used to obtain the separation data of the positive and negative half axes of the power frequency phase, and to detect and judge the polarity effect and wing-shaped spectrum;

[0069] P3: Draw the NV spectrum again, and calculate the amplitude entropy distribution and center of gravity distribution according to the NV spectrum distribution;

[0070] P4: Resample and divide the grids for statistics, and the 50Hz / 100Hz frequency energy can be calculated;

[0071] In the above embodiment, it should be noted that P2 to P4 together constitute the diagnostic criterion result of the evidence rule;

[0072] The test results of fault type identification using phase-independent feature quantities show that the average recognition accuracy of the BP network recognition method using a single-layer recognition method is 81.9%, and the average recognition accuracy of the fuzzy recognition method is 82.4%. The recognition accuracy of the two methods using a hierarchical recognition method is increased to 98.3% and 89.8% respectively; see the table below for details:

[0073]

[0074] Example 4

[0075] like Figures 1 to 3As shown, a partial discharge identification method based on prior rules and digital source information fusion includes all the contents of Example 3. In addition, the detected PRPD spectrum, PRPS spectrum and NV spectrum are extracted according to the diagnosis criterion results, the fingerprint feature of the characteristic spectrum is extracted, and the diagnosis result is output using a hierarchical diagnosis network to form a diagnosis output of the digital source information. The diagnosis result of the evidence rule and the diagnosis of the digital source information together constitute an expert diagnosis based on the fusion decision of the evidence rule and the digital source information. The diagnosis system outputs the final diagnosis result according to the voting decision.

Claims

1. A partial discharge identification method based on a priori rules and digital source information fusion, characterized in that: The steps include: S1: First, the evidence rule calculation is performed on the detection data to determine whether the detection data triggers the evidence rule; S2: If the strong evidence rule is triggered, the discharge type is directly output as a strong evidence diagnosis result; S3: If only the weak evidence rule is triggered, the diagnosis result of the evidence rule and the diagnosis result based on the source information are voted for consistency. If the two diagnosis results are consistent, the diagnosis result is output; S4: If the two diagnostic results are inconsistent, an unknown type result is output; S5: If the evidence rule is not triggered, the diagnosis result of the digital source information is used as the final diagnosis output result; The detection data obtains the discharge peak value v, the discharge phase φ and the discharge time scale t from the discharge spectrum data or statistical data feature quantity; Evidence rule calculation is performed based on the discharge peak value v, discharge phase φ and discharge time scale t to determine whether the evidence rule is triggered; The evidence rules are empirical rule tables that present the discharge signals triggered by different defect types. The diagnosis of digital source information adopts a hierarchical identification strategy at the digital source information knowledge dimension level. The first level of diagnosis completes the identification of discharge signals and interference signals; The second level of diagnosis completes the identification of discharge signals and discharge interference types respectively; The third level of diagnosis completes the specific distinction and identification of gas discharge and solid insulation discharge defects; Each layer of diagnosis uses different characteristic fingerprint information to achieve the best diagnostic accuracy.

2. A partial discharge identification method based on a priori rules and digital source information fusion according to claim 1, characterized in that: When the corresponding evidence rules are triggered, the diagnostic system will give the relevant corresponding types a high decision weight. The prior evidence rules are intended to make a priori distinctions between various discharge modes based on five characteristic distributions: 50Hz / 100Hz correlation, polarity effect, "wing-shaped spectrum" characteristics, spectrum center of gravity distribution, and entropy distribution.

3. The method for identifying partial discharge based on a priori rules and digital source information fusion according to claim 1, characterized in that: The following specific steps are also included: P1: Based on the high-frequency current and ultrasonic test data obtained from the test, a V-φ scatter plot is first drawn; P2: The DBSCAN clustering algorithm is used to obtain the separation data of the positive and negative half axes of the power frequency phase, and to detect and judge the polarity effect and wing-shaped spectrum; P3: Draw the NV spectrum again, and calculate the amplitude entropy distribution and center of gravity distribution according to the NV spectrum distribution; P4: Resample and divide the data into grids for statistics, and the energy of the 50Hz / 100Hz frequency can be calculated.

4. The method for identifying partial discharge based on a priori rules and digital source information fusion according to claim 3 is characterized in that: P2 to P4 together constitute the diagnostic criteria result of the evidence rule.

5. The method for identifying partial discharge based on a priori rules and digital source information fusion according to claim 3 is characterized in that: According to the diagnostic criteria, the detected PRPD spectrum, PRPS spectrum and NV spectrum are extracted, and the fingerprint feature quantity of the characteristic spectrum is extracted.

6. The method for identifying partial discharge based on a priori rules and digital source information fusion according to claim 5, characterized in that: The hierarchical diagnosis network is used to output the diagnosis results to form the diagnosis output of the data source information. The evidence rule diagnosis results are integrated with the diagnosis of the data source information to form an expert diagnosis based on the fusion decision of evidence rules and data source information. The diagnosis system outputs the final diagnosis results based on the voting decision.

Images