Research method for partial discharge law of suspended water drops in flowing transformer oil

Abstract

The invention relates to the technical field of power equipment, and discloses a partial discharge law research method for suspended water drops in flowing transformer oil, which comprises the following steps: S1, changing the flow velocity and temperature in a model; s2, boosting; s3, capturing discharge capacity-phase data by adopting a pulse current method; s4, acquiring PD signals to a digital partial discharge acquisition system; s5, calculating skewness and steepness characteristic values in positive and negative half cycles; s6, sampling for multiple times; s7, carrying out KMO and Bartlett spherical inspection on the KMO and the Bartlett; s8, principal component analysis dimensionality reduction; s9, extracting characteristic parameters; s10, carrying out PSO-SVM verification; and S11, revealing a PD rule by the characteristic value after dimension reduction. The invention provides a method for researching the partial discharge law of suspended water drops in flowing transformer oil under large temperature difference, which comprises the following steps of: calculating characteristic values such as skewness, steepness and signal entropy in positive and negative half cycles, calculating correlation coefficients on an overall phase, taking phase symmetry as characteristic values, and calculating the partial discharge law of the suspended water drops in the flowing transformer oil from two dimensions of the overall and the positive and negative half cycles; and the discharge mechanism in the insulating oil containing the suspended water drops is revealed.

Description

Technical Field

[0001] The invention relates to the technical field of electric power equipment, and in particular to a method for studying the local discharge law of suspended water droplets in flowing transformer oil. Background Art

[0002] In the process of large-scale power equipment being put into use, the voltage level requirements are getting higher and higher. Transformers play a role in connecting the past and the future in the operation of power equipment. Among them, insulation failure is the main reason affecting the stable operation of transformers. In recent years, it has become the primary factor in high-voltage equipment accidents, accounting for up to 50%. When an insulation failure occurs in an oil-immersed transformer, a certain amount of moisture is sometimes detected in the oil. The moisture mainly exists in three states: dissolved water, suspended water, and deposited water. In particular, the large temperature difference in the Qinghai-Tibet region causes suspended water droplets in the transformer oil to precipitate at low temperatures.

[0003] The main sources of moisture are as follows: the hydroxyl bonds of the insulating oil break during partial discharge (PD), the internal and external pressure difference causes external water to intrude, and the dissolved water and deposited water precipitate and dissolve out suspended water during cooling and heating respectively. The dissolved water and deposited water have little effect on partial discharge (PD). Existing studies have not considered the partial discharge (PD) law and breakdown characteristics of suspended water droplets in oil under a flowing state, and at this time the partial discharge (PD) process can no longer be explained by the "small bridge" theory. Summary of the invention

[0004] In order to overcome or alleviate one or more of the above technical problems, the purpose of the present invention is to provide a method for studying the local discharge law of suspended water droplets in flowing transformer oil, to supplement the existing research on the local discharge (PD) law of suspended water droplets in flowing transformer oil, and to explore the discharge law in transformer oil containing suspended water droplets.

[0005] The present invention provides the following technical solutions:

[0006] A method for studying the partial discharge law of suspended water droplets in flowing transformer oil comprises the following steps:

[0007] S1: changing flow rate and temperature in the discharge defect model;

[0008] S2: using an AC booster to gradually boost the voltage; in the discharge defect model, the AC booster is used to boost the voltage at different flow rates and temperatures, the voltage increases from zero, and each step of the voltage increase is 200V;

[0009] S3: Pulse current method is used to capture discharge amount-phase data;

[0010] S4: the PD signal is collected into a digital partial discharge collection system to obtain a PRPD spectrum, and the partial discharge PD signal is collected into a digital partial discharge detection system to obtain a partial discharge phase distribution spectrum;

[0011] S5: Calculate the skewness and steepness eigenvalues ​​in the positive and negative half cycles, and calculate the correlation coefficient and phase symmetry difference as eigenvalues ​​on the overall phase. Specifically, calculate the skewness, steepness, signal entropy, peak-to-peak value, average symmetry difference and maximum discharge in the positive and negative half cycles, and calculate the correlation coefficient and phase symmetry on the overall phase. There are eight types of eigenvalues ​​and fourteen characteristic parameters;

[0012] S6: multiple sampling, i.e., executing steps S3 to S5 cyclically for a predetermined number of times;

[0013] S7: KMO and Bartlett sphericity test. Because the characteristic dimension is too high, there is signal redundancy, resulting in information overlap. Principal component analysis is performed to find a few common factor variables that reflect the common characteristics between variables. The test methods used are KMO and Bartlett sphericity test. KMO is an indicator of whether factor analysis is applicable. KMO is kept between 0.5 and 1. Bartlett sphericity test is used to examine whether the variables are independent of each other, and the significance is approximately equal to 0.

[0014] S8: Dimensionality reduction by principal component analysis; select multiple voltage samples of different types, fourteen characteristic variables form a data matrix, normalize, calculate the correlation coefficient matrix, extract the cumulative contribution rate greater than 85%, and obtain the scores of each principal component. In the component score coefficient matrix, if the eigenvalue principal component exceeds 50% and is negative, it indicates a strong negative correlation and needs to be discarded. Principal component analysis and dimensionality reduction are performed after feature extraction of the collected partial discharge phase distribution map;

[0015] S9: Extract characteristic parameters of the principal components, and extract characteristic quantities with strong positive correlation of at least 50% or more principal components;

[0016] S10: The PSO-SVM algorithm verifies that the feature parameters after dimensionality reduction can better analyze the PD law. The particle swarm optimization-support vector machine algorithm verifies that the accuracy of the feature values ​​after dimensionality reduction on the test set is much higher than that before dimensionality reduction.

[0017] S11: The eigenvalues ​​after dimensionality reduction reveal the PD law, and the eigenvalues ​​after dimensionality reduction reveal the law and characteristics of partial discharge PD of suspended water droplets in transformer oil.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] (1) The present invention proposes a method for feature extraction and law analysis of partial discharge (PD) of suspended water droplets in flowing transformer oil under large temperature difference. The characteristic values ​​such as skewness, steepness, signal entropy are calculated in the positive and negative half cycles, and the correlation coefficient and phase symmetry are calculated as characteristic values ​​on the overall phase. The discharge mechanism in insulating oil containing suspended water droplets is revealed from the two dimensions of overall and positive and negative half cycles, avoiding the previous method of only performing longitudinal and transverse analysis of discharge quantity-phase (q~Φ) in the two dimensions of "stratification" and "clustering" on the collected partial discharge phase distribution (PRPD) spectrum.

[0020] (2) The present invention is based on principal component analysis (PCA) and particle swarm optimization-support vector machine (PSO-SVM), and is explained from the dual theory of feature dimension reduction + pattern recognition. Principal component analysis (PCA) dimension reduction: reduces signal redundancy, reduces information overlap caused by too many eigenvalues, and improves the efficiency and accuracy of partial discharge (PD) law research. Particle swarm optimization-support vector machine (PSO-SVM) verification was carried out, and it was verified that the accuracy of the eigenvalues ​​after dimensionality reduction on the test set was much greater than that before dimensionality reduction. Finally, the eigenvalues ​​after dimensionality reduction were used to reveal the law and characteristics of partial discharge (PD) of suspended water droplets in transformer oil. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The present invention is a flowchart of a method for studying the partial discharge law of suspended water droplets in flowing transformer oil proposed in an embodiment of the present invention.

[0022] Figure 2 This is a partial discharge feature extraction diagram of suspended water droplets in flowing transformer oil proposed in an embodiment of the present invention.

[0023] Figure 3 This is a principal component analysis diagram of partial discharge of suspended water droplets in flowing transformer oil proposed in an embodiment of the present invention.

[0024] Figure 4 This is a dimensionality reduction diagram of the partial discharge characteristics of suspended water droplets in the flow transformer oil proposed in an embodiment of the present invention.

[0025] Figure 5 This is a test set accuracy graph of partial discharge of suspended water droplets in flowing transformer oil before dimensionality reduction proposed in an embodiment of the present invention.

[0026] Figure 6 This is the accuracy graph of the test set after dimensionality reduction of partial discharge of suspended water droplets in the flowing transformer oil proposed by the present invention. DETAILED DESCRIPTION

[0027] The present invention is described in detail below in conjunction with the embodiments and drawings, but it should be understood that the embodiments and drawings are only used to exemplify the present invention and do not constitute any limitation on the protection scope of the present invention. All reasonable changes and combinations within the scope of the inventive concept of the present invention fall within the protection scope of the present invention.

[0028] The present invention is a further study based on the existing developed discharge defect model. The internal structure of the existing discharge model is as follows: the left and right ends of the shell are respectively provided with an oil inlet and an oil outlet, and a circular upper electrode plate and a corresponding lower electrode plate are provided therein. There is a cavity between the upper and lower electrodes, the upper electrode end is at a high potential, and the lower electrode plate is at a zero potential. Please refer to the existing literature "Ma Shouxiao. Study on the local discharge characteristics and influencing factors of metal particles in insulating oil under flow state [D]. Chongqing University, 2016."

[0029] The present invention will be further described below in conjunction with the accompanying drawings.

[0030] Example 1

[0031] like Figure 1 This embodiment provides a method for studying the partial discharge law of suspended water droplets in flowing transformer oil, comprising the following steps:

[0032] S1: changing the flow rate and temperature in the developed discharge defect model;

[0033] S2: Use GDYD-20kVA / 200kV AC booster to boost the voltage. In the developed cyclic discharge defect model, the voltage is boosted by GDYD-20kVA / 200kV AC booster at different flow rates and temperatures. The voltage increases from zero, and each step increases by 200V.

[0034] S3: The pulse current method recommended by IEC60270 standard is used to capture the discharge quantity-phase (q~Φ) data;

[0035] S4: PD signal is collected into GDJF-2007 digital partial discharge acquisition system to obtain PRPD spectrum. The partial discharge (PD) signal is collected into GDJF-2007 digital partial discharge detection system to obtain partial discharge phase distribution (PhaseResolved Partial Discharge, referred to as PRPD) spectrum;

[0036] S5: Calculate eigenvalues ​​such as skewness and steepness in the positive and negative half cycles, and calculate the correlation coefficient and phase symmetry difference as eigenvalues ​​on the overall phase. Specifically, calculate skewness, kurtosis, signal entropy, peak-to-peak value, average symmetry variance, and maximum discharge (Qmax) in the positive and negative half cycles, and calculate the correlation coefficient (corr) and phase symmetry on the overall phase. There are eight types of eigenvalues ​​and fourteen characteristic parameters in total;

[0037] S6: multiple sampling, i.e., executing steps S3 to S5 cyclically for a predetermined number of times;

[0038] S7: KMO and Bartlett sphericity test. Because the characteristic dimension is too high, there is signal redundancy, resulting in information overlap. Principal component analysis (PCA) is required to find a few common factor variables that reflect the common characteristics between variables. Commonly used test methods include KMO and Bartlett sphericity test. KMO is an indicator of whether factor analysis is applicable. If KMO is kept between 0.5 and 1, it is appropriate. Bartlett sphericity test is used to examine whether the variables are independent of each other. If the significance is approximately equal to 0, it is qualified.

[0039] S8: Dimensionality reduction by principal component analysis; 80 voltage samples of 4 different types are selected, and 14 characteristic variables form a 80*14 data matrix, which is normalized, and the correlation coefficient matrix is ​​calculated. The cumulative contribution rate greater than 85% is extracted to obtain the scores of each principal component. In the component score coefficient matrix, if the eigenvalue principal component exceeds 50% and is negative, it indicates a strong negative correlation and needs to be discarded. The principal component analysis (PCA) is performed to reduce the dimension by extracting the features of the collected partial discharge phase distribution (PRPD) spectrum;

[0040] S9: Extract characteristic parameters that are at least 50% or more of the principal components, and extract characteristic quantities that are strongly positively correlated with at least 50% or more of the principal components;

[0041] S10: The PSO-SVM algorithm verifies that the feature parameters after dimensionality reduction can better analyze the PD law. The PSO-SVM algorithm verifies that the accuracy of the feature values ​​after dimensionality reduction on the test set is much higher than that before dimensionality reduction.

[0042] S11: The eigenvalues ​​after dimensionality reduction reveal the PD law. The eigenvalues ​​after dimensionality reduction reveal the law and characteristics of partial discharge (PD) of suspended water droplets in transformer oil.

[0043] In step S4, the partial discharge (PD) of suspended water droplets in transformer oil is extracted into a partial discharge phase distribution (PRPD) spectrum, and characteristic values ​​such as skewness and steepness are calculated in the positive and negative half cycles. The correlation coefficient and phase symmetry are calculated as characteristic values ​​on the overall phase, and the consideration is made from two dimensions: the overall phase and the positive and negative half cycles.

[0044] In step S8, since the characteristic dimension is too high and there is signal redundancy, principal component analysis (PCA) must be performed to find a few common factor variables that reflect the common characteristics between variables, and the partial correlation and independence of the data are analyzed by KMO and Bartlett sphericity test. Then, according to the normalization process, the correlation coefficient matrix is ​​calculated, and the cumulative contribution rate greater than 85% is extracted to obtain the scores of each principal component. In the component score coefficient matrix, if the eigenvalue principal component exceeds 50% and is negative, it indicates a strong negative correlation and needs to be discarded to achieve dimensionality reduction.

[0045] In step S10, the particle swarm optimization-support vector machine (PSO-SVM) algorithm is used to verify that the accuracy of the eigenvalues ​​after dimensionality reduction on the test set is much greater than that before dimensionality reduction. Finally, the eigenvalues ​​after dimensionality reduction are used to reveal the laws and characteristics of partial discharge (PD) of suspended water droplets in transformer oil.

[0046] The experimental process of S1 to S10 is described as follows:

[0047] S1: at flow rate 0.9m 3 / h, under AC voltage of 31kV, the partial discharge (PD) law of suspended water droplets in transformer oil at different temperatures is studied. In the developed cyclic discharge defect model, 40, 50, 60, and 70℃ are set as four gradients to simulate the real life environment;

[0048] S2: Boost voltage through GDYD-20kVA / 200kV AC booster;

[0049] S3 and S4: The pulse current method recommended by IEC60270 standard is used to collect the partial discharge (PD) signal into the GDJF-2007 digital partial discharge detection system to obtain the partial discharge phase distribution (PRPD) spectrum;

[0050] S5: Feature extraction shows that Figure 2In the experiment, the characteristics of 20 groups of partial discharge phase distribution (PRPD) spectra were extracted at four gradients at a flow rate of 0.9m3 / h and an AC voltage of 31kV. The skewness, kurtosis, entropy, peak-to-peak, average symmetry variance, and maximum discharge (Qmax) were calculated in the positive and negative half cycles respectively. The correlation coefficient (corr) and phase symmetry (phase symmetry) were calculated on the overall phase. There were eight types of eigenvalues ​​and fourteen characteristic parameters.

[0051] S6: multiple sampling;

[0052] S7: Perform KMO and Bartlett sphericity test to further reduce feature dimensionality. Figure 3 In the analysis, due to the high feature dimension, signal redundancy and information overlap, principal component analysis (PCA) is required to find a few common factor variables that reflect the common characteristics of the variables. Commonly used test methods include KMO and Bartlett sphericity test. KMO is an indicator of whether factor analysis is applicable. If KMO is kept between 0.5 and 1, it is appropriate. Bartlett sphericity test is used to examine whether the variables are independent of each other. If the significance is approximately equal to 0, it is qualified.

[0053] S8: Select 80 groups of voltage samples under four temperature gradients, and 14 characteristic variables form a 80*14 data matrix, normalize them, calculate the correlation coefficient matrix, extract the cumulative contribution rate greater than 85%, and obtain the scores of each principal component. In the component score coefficient matrix, if the eigenvalue principal component exceeds 50% and is negative, it indicates a strong negative correlation and needs to be discarded. After feature extraction of the collected partial discharge phase distribution (PRPD) spectrum and principal component analysis (PCA) dimensionality reduction, extract the feature quantity that has a strong positive correlation with at least 50% and above of the principal components;

[0054] S9: Figure 4 As shown, after comparative dimension reduction, the characteristic quantities that have a strong positive correlation with at least 50% and above of the principal components are: skewness +, steepness +, maximum discharge +-, signal entropy +-, peak-to-peak value +- (+ represents the positive half cycle, - represents the negative half cycle);

[0055] S10: Pattern recognition verification shows that Figure 5 , 6In the test set, the accuracy of the eigenvalues ​​after dimensionality reduction is much higher than that before dimensionality reduction by using the particle swarm optimization-support vector machine (PSO-SVM) algorithm. After 30 comparisons before and after dimensionality reduction, it can be seen that the accuracy after dimensionality reduction is generally above 95%, while the accuracy before dimensionality reduction is generally around 90%, indicating that the eigenvalues ​​after dimensionality reduction can more accurately analyze the laws of partial discharge (PD);

[0056] S11: Finally, the law and characteristics of partial discharge (PD) of suspended water droplets in transformer oil are revealed by the eigenvalues ​​after dimensionality reduction.

[0057] The process of this test method can supplement the study of the partial discharge (PD) law of suspended water droplets in flowing transformer oil, and supplement the discharge theory in transformer oil containing suspended water droplets. The collected partial discharge phase distribution (PRPD) spectrum is feature extracted by extracting skewness, steepness and other eigenvalues; then the principal component analysis (PCA) is performed to reduce the dimension, and the feature quantities with strong positive correlation with at least 50% and above of the principal components are extracted; the particle swarm optimization-support vector machine (PSO-SVM) algorithm is used to verify that the accuracy of the eigenvalues ​​after dimensionality reduction on the test set is much greater than that before dimensionality reduction; finally, the eigenvalues ​​after dimensionality reduction are used to reveal the law and characteristics of partial discharge (PD) of suspended water droplets in transformer oil, and further provide ideas for real-time charged detection of partial discharge (PD) of suspended water droplets in insulating oil.

[0058] The above embodiments are only preferred implementations of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A method for studying the partial discharge law of suspended water droplets in flow transformer oil, characterized in that: The following steps are involved: S1: changing flow rate and temperature in the discharge defect model; S2: using an AC booster to gradually boost the voltage; in the discharge defect model, the AC booster is used to boost the voltage at different flow rates and temperatures, the voltage increases from zero, and each step of the voltage increase is 200V; S3: Pulse current method is used to capture discharge amount-phase data; S4: the PD signal is collected into a digital partial discharge collection system to obtain a PRPD spectrum, and the partial discharge PD signal is collected into a digital partial discharge detection system to obtain a partial discharge phase distribution spectrum; S5: Calculate the skewness and steepness eigenvalues ​​in the positive and negative half cycles, and calculate the correlation coefficient and phase symmetry difference as eigenvalues ​​on the overall phase. Specifically, calculate the skewness, steepness, signal entropy, peak-to-peak value, average symmetry difference and maximum discharge in the positive and negative half cycles, and calculate the correlation coefficient and phase symmetry on the overall phase. There are eight types of eigenvalues ​​and fourteen characteristic parameters; S6: multiple sampling, i.e., executing steps S3 to S5 cyclically for a predetermined number of times; S7: KMO and Bartlett sphericity test. Because the characteristic dimension is too high, there is signal redundancy, resulting in information overlap. Principal component analysis is performed to find a few common factor variables that reflect the common characteristics between variables. The test methods used are KMO and Bartlett sphericity test. KMO is an indicator of whether factor analysis is applicable. KMO is kept between 0.5 and 1. Bartlett sphericity test is used to examine whether the variables are independent of each other, and the significance is approximately equal to 0. S8: Dimensionality reduction by principal component analysis; select multiple voltage samples of different types, fourteen characteristic variables form a data matrix, normalize, calculate the correlation coefficient matrix, extract the cumulative contribution rate greater than 85%, and obtain the scores of each principal component. In the component score coefficient matrix, if the eigenvalue principal component exceeds 50% and is negative, it indicates a strong negative correlation and needs to be discarded. Principal component analysis and dimensionality reduction are performed after feature extraction of the collected partial discharge phase distribution map; S9: Extract characteristic parameters of the principal components, and extract characteristic quantities with strong positive correlation of at least 50% or more principal components; S10: The PSO-SVM algorithm verifies that the feature parameters after dimensionality reduction can better analyze the PD law. The particle swarm optimization-support vector machine algorithm verifies that the accuracy of the feature values ​​after dimensionality reduction on the test set is much higher than that before dimensionality reduction. S11: The eigenvalues ​​after dimensionality reduction reveal the PD rules, and the eigenvalues ​​after dimensionality reduction reveal the rules and characteristics of partial discharge PD of suspended water droplets in transformer oil.

Images