Method for Classifying Resistive Current Data of Lightning Arrester and Method for Judging Operating State of Lightning Arrester

Abstract

The present invention relates to the field of nuclear power safety operation, and particularly to a method for classifying resistive current data of lightning arresters and a method for judging the working state of lightning arresters. The classification method includes: obtaining humidity data and resistive current data of a lightning arrester to form an original data set; improving a bandwidth formula, and generating B resistive current reference data sets according to a KDE kernel density estimation function, where the resistive current reference data sets and the humidity data form a reference data set; calculating the sum of squared classification distances of all data sets, i.e., the original data set and the generated reference data set after processing, under different numbers of clustering centers k, and calculating a Gap statistic to select an optimal K best value; performing K-means clustering analysis on the original data set based on the optimal K best value to divide it into a "rain" class and a "sunny" class. The classification of the present invention is accurate, effectively improves the online monitoring accuracy of lightning arresters, avoids misjudging the working state of lightning arresters, and realizes the stable operation of the power system.

Description

Technical Field

[0001] The present invention relates to the field of nuclear power safety operation, and particularly to a method for classifying resistive current data of a lightning arrester and a method for judging the working state of a lightning arrester. Background Art

[0002] The monitoring of resistive current is an important means for detecting faults of zinc oxide lightning arresters, and the change of resistive current directly affects the working state of the on-line monitoring system of lightning arresters. During on-site operation, the resistive current is easily affected by external humidity factors. When encountering weather such as rain, snow, or fog, the air humidity increases, directly affecting the wetting condition of the surface contamination layer of the lightning arrester, changing the leakage current, and thus causing the change of resistive current; when the weather is clear and the humidity is low, the resistive current will return to normal. This phenomenon causes false alarms in the on-line monitoring system of lightning arresters and affects its normal operation. Therefore, the resistive current data can be divided into two categories according to the humidity change situation. One category is the "clear" category, that is, the resistive current data is not affected by the humidity change. In this case, if the change rate of the resistive current increases significantly, it indicates that the lightning arrester has an abnormal situation; the other category is the "rain" category, that is, the resistive current data is affected by the humidity change, interfering with the judgment of the monitoring state result of the lightning arrester. In this category, the abnormal change of the resistive current does not mean that there is a problem with the lightning arrester equipment.

[0003] For resistive current data with interference characteristics, using cluster analysis for state recognition has become an important technical means. However, simply dividing the resistive current data into two categories affects the accurate distinction of the state of the lightning arrester; considering the complexity of the resistive current data, the K-means cluster analysis method is more efficient. In practical applications, it is very important to select a suitable K value. If the K value is selected improperly, it may lead to information loss or the generation of meaningless small classes, that is, the problems of underfitting and overfitting. The Gap statistic quantifies the clustering effect under different K values by calculating the difference between the clustering error of the real data set and the clustering error of the reference data set, avoids the deviation problem of subjective judgment, reduces the dependence on a single data set, and improves the reliability of the division result. The Gap statistic depends on the distribution of the generated reference data. The distribution of resistive current data of different lightning arresters in a substation is different, and the general uniform distribution, normal distribution, and spherical distribution are not applicable, resulting in inaccurate selection of the K value. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for classifying resistive current data of a lightning arrester and a method for judging the working state of a lightning arrester, which can accurately classify the data with obvious change of the resistive current of the lightning arrester caused by excessive humidity, effectively improve the on-line monitoring accuracy of the lightning arrester, avoid misjudging the working state of the lightning arrester, and realize the stable operation of the power system.

[0005] The present invention provides a method for classifying resistive current data of a lightning arrester, including:

[0006] Step S1: Obtain the humidity data of the lightning arrester RH and the resistive current data I R , and form the original data set X R ;

[0007] Step S2: Improve the bandwidth h formula, and generate B reference data sets of resistive current according to the KDE kernel density estimation function I ref , the resistive current reference data set I ref and the humidity data RH form the reference data set X ref ;

[0008] Step S3: Calculate the sum of squared classification distance errors of the original data set X R and the generated reference data set after processing X ref under different numbers of clustering centers k, calculate the Gap statistic, and select the optimal K best value;

[0009] Step S4: Based on the optimal K best value, perform K-means clustering analysis on the original data set X R to divide it into "rain" class and "sunny" class.

[0010] In a specific embodiment of the present invention, the step S2 specifically includes:

[0011] Initialize and set the number B of reference data sets, perform KDE kernel density estimation on the original resistive current data , according to Scott's rule, add skewness and the adjustment parameter α, calculate the bandwidth h and the density function , randomly sample n data points from it to generate a reference data set, and repeat B times to generate a total of B reference data sets .

[0012] In a specific embodiment of the present invention, the KDE kernel density estimation formula is as follows:

[0013]

[0014]

[0015] Among them, n is the same as the number of resistive current data samplings, x represents a point in the reference dataset, x i represents the i-th point in the resistive current, skewness the skewness coefficient of the original resistive current data, is the Gaussian kernel function.

[0016] In a specific embodiment of the present invention, the step S3 specifically includes:

[0017] Normalize the resistive current data I R and the generated resistive current reference dataset I ref and the humidity data RH , map the values to the range [-1, 1] to eliminate the scale difference; set the number of clustering centers k ∈ [2, K], and the maximum number of classifications does not exceed K. For the normalized original dataset X R_norm and B reference datasets X ref_norm , perform K-means clustering respectively, calculate the sum of squared classification distance errors under different k values in sequence, and calculate the Gap statistic, and select Gap(k) the one with the largest K best value as the optimal number of classifications.

[0018] In a specific embodiment of the present invention, in the step S3,

[0019] when the number of clustering centers is k, the sum of squared classification distance errors J (k) is as follows:

[0020]

[0021] Among them, C i The i-th class, y is C i All sample points in the class, m i is C i the centroid of the class;

[0022] Gap(k) and the optimal number of classifications K best are calculated as follows:

[0023]

[0024] .

[0025] In a specific embodiment of the present invention, the number k of clustering centers is 2.

[0026] In a specific embodiment of the present invention, step S4 specifically includes: performing k = 2 mean classification on the centroids K best obtained by K-means clustering analysis, that is, obtaining the centroids of the "rain" class and the "sunny" class, and then dividing the "rain" class and the "sunny" class according to the m R -means classification result. K best

[0027] In a specific embodiment of the present invention, in step S2, the value of B is 5 to 10.

[0028] In a specific embodiment of the present invention, the maximum number of classifications is 5 to 8.

[0029] The present invention provides a method for judging the working state of a lightning arrester, including the following steps:

[0030] Classifying the resistive current data and humidity data by using the classification method described above;

[0031] Collecting the resistive current value of the lightning arrester to be measured and the humidity value during the test;

[0032] Comparing the collected resistive current and humidity values with the classification results;

[0033] According to the resistive current data of the "sunny" category obtained, it is determined that the change in the resistive current is normal; according to the resistive current data of the "rain" category, it is determined that the obvious change in the resistive current is caused by excessive humidity, and the working state of the lightning arrester is normal.

[0034] Compared with the prior art, the method for classifying the resistive current data of the lightning arrester and the method for judging the working state of the lightning arrester according to the present invention, by adding a skewness coefficient and adjusting parameters, helps to improve the reliability of density estimation, enables the reference data set generated by the KDE kernel function estimation method to accurately reflect the complexity of the original resistive current, and retains the distribution characteristics of the resistive data. Based on this, the clustering quality of the K-means algorithm is improved.

[0035] By clustering multiple reference data sets simultaneously, the dependence on a single original resistive data set is reduced, and the reliability of the classification result can be enhanced; the Gap statistic is used to evaluate the clustering effects of different K values, and the best K value is selected, which can effectively avoid the overfitting or underfitting problems caused by improper selection of the K value.

[0036] Using K-means clustering can accurately classify the resistive current data of the "rain" category, determine that the change in resistive current is caused by excessive humidity, avoid misjudgment of the arrester on-line monitoring system, and realize the normal operation of the on-line monitoring system. It has important practical significance for the arrester on-line monitoring system. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It represents a flowchart of a method for classifying the resistive current data of an arrester based on K-means clustering;

[0038] Figure 2 It represents the original data graph of humidity and resistive current;

[0039] Figure 3 It represents the calculation result graph of the K-means clustering Gap statistic for different K values;

[0040] Figure 4 It represents the "clear" data graph of humidity and resistive current;

[0041] Figure 5 It represents the "rain" data graph of humidity and resistive current. DETAILED DESCRIPTION OF THE INVENTION

[0042] To further understand the present invention, the implementation scheme of the present invention will be described below in conjunction with embodiments. However, it should be understood that these descriptions are only for further explaining the features and advantages of the present invention, rather than limiting the present invention.

[0043] The embodiment of the present invention discloses a method for classifying the resistive current data of an arrester based on K-means clustering, as Figure 1 shown, which specifically includes the following steps: Step S1: Obtain the humidity data RH and the resistive current data I R , and form an original data set X R ;

[0044] Step S2: Improve the bandwidth h formula, and generate B reference data sets of resistive current according to the KDE kernel density estimation function I ref , and the resistive current reference data set I ref and the humidity data RH form a reference data set X ref ;

[0045] Specifically, it includes:

[0046] Initialize and set the number B of reference data sets, and for the original resistive current data Perform KDE kernel density estimation, according to Scott's rule, add skewness And the adjustment parameter α, the bandwidth h and density function can be calculated , randomly sample n data points from it to generate a reference data set, repeat B times, and generate a total of B reference data sets , the KDE kernel density estimation formula is as follows:

[0047]

[0048]

[0049] Where n is the same as the number of resistive current data samples, and x represents the point in the reference data set. x i represents the i-th point in the resistive current, skewness The skewness coefficient of the original resistive current data, is the Gaussian kernel function.

[0050] Step S3: Calculate the original dataset of all datasets X R and the reference dataset generated after processing X ref The sum of squares of classification distance errors under different cluster center numbers k is calculated, and the Gap statistic is calculated to select the optimal K best value;

[0051] Includes: Normalized resistive current data I R , generated resistive current reference data set I ref , humidity data RH , map the values to the range of [-1, 1] to eliminate scale differences; set the number of cluster centers k∈[2,K], the maximum number of classifications does not exceed K, and standardize the original data set X R_norm , B reference datasets X ref_norm Perform K-means clustering respectively, calculate the sum of squares of classification distance errors under different k values, and calculate the Gap statistic, and select Gap(k) The largest value K best as the optimal number of categories.

[0052] When the number of cluster centers is k, the sum of squares of classification distance errors J (k) The formula is as follows:

[0053]

[0054] Among them, C i the i nth y class C i is all the sample points in the m i class, C i and

[0055] Gap(k) the optimal number of classes K best is calculated as follows:

[0056]

[0057]

[0058] Step S4: Based on the optimal K best value, perform K-means clustering analysis on the original data set X R to divide it into "rain" class and "sunny" class.

[0059] Specifically, it includes: performing k = 2 mean classification on the K best centroids obtained from K-means clustering analysis, that is, obtaining the centroids of the "rain" class and the "sunny" class, and then dividing the "rain" class and the "sunny" class according to the K -means classification result. best Based on the obtained resistive current data of the "sunny" class, it is determined that the change of the resistive current is normal; based on the resistive current data of the "rain" class, it is determined that the obvious change of the resistive current is caused by excessive humidity, and the working state of the arrester is normal.

[0060] By adjusting the bandwidth formula, the present invention can adapt to the complex distribution of resistive current data, use KDE kernel density estimation to generate a reference data set consistent with the complexity of the original resistive current data, improve the clustering quality of its application of the K-means algorithm, and select the most suitable K value by calculating the Gap statistic, which can accurately distinguish normal conditions and the resistive current data of the arrester caused by excessive humidity, and has important practical significance for the on-line monitoring system of the arrester.

[0061] The present invention also provides a method for judging the working state of an arrester, including the following steps:

[0062] Classify the resistive current data and humidity data by using the classification method described above;

[0063] Use the classification method described above to classify the resistive current data and humidity data;

[0064] Collect the resistive current value of the arrester to be measured and the humidity value during the test;

[0065] Compare the collected resistive current and humidity values with the classification results;

[0066] Based on the obtained resistive current data of the "sunny" category, it is determined that the change in the resistive current is normal; based on the resistive current data of the "rain" category, it is determined that the obvious change in the resistive current is caused by excessive humidity, and the working state of the arrester is normal.

[0067] To further understand the present invention, the following will combine embodiments to elaborate in detail on the method for classifying the resistive current data of arresters based on K-means clustering provided by the present invention. The protection scope of the present invention is not limited by the following embodiments.

[0068] Embodiment 1

[0069] The method for classifying the resistive current of an arrester includes the following steps:

[0070] S1. Obtain the humidity data RH and the resistive current data I R , and form the original data set X R ;

[0071] Figure 2 is the original data graph of humidity and resistive current. It can be seen from the graph that under low humidity conditions, the calculated value of the resistive current is relatively stable; under high humidity conditions, the calculated value of the resistive current has a large change range and is relatively scattered; the greater the humidity, the resistive current does not necessarily increase, but may also decrease.

[0072] S2. Initialize and set the number of reference data sets to 10 and the adjustment parameter α. Perform KDE kernel density estimation on the original resistive current data I R . According to Scott's rule, add skewness and the adjustment parameter α, calculate the bandwidth h and the density function , randomly sample n data points from it to generate a reference data set, and repeat 10 times to generate 10 reference data sets . The KDE kernel density estimation formula is as follows:

[0073]

[0074]

[0075] Among them, n is the same as the number of samples of the resistive current data, x represents the points in the reference data set, x i represents the i-th point in the resistive current,skewness The skewness coefficient of the original resistive current data is the Gaussian kernel function.

[0076] S3. Calculate all data sets and the sum of squared classification distance errors at different numbers of clustering centers k, calculate the Gap statistic, and select the optimal K best value;

[0077] Normalize the resistive current data I R , the generated resistive current reference data set I ref , humidity data RH , map the values to the range [-1, 1] to eliminate scale differences; set the number of clustering centers k ∈ [2, K], and in the embodiments of the present application, k ∈ [2, 8]. Normalize the original data set X R_norm , 10 reference data sets X ref_norm are respectively subjected to K-means clustering, and the sum of squared classification distance errors at different k values are calculated in turn; finally, calculate the Gap statistic Gap(k), to make Gap(k) the largest K best as the optimal number of classifications.

[0078] When the number of clustering centers is k, calculate the sum of squared classification distance errors J (k) , that is, the sum of the squares of the position distances from each data point to the center of its category. The formula is as follows:

[0079]

[0080] where C i the i th category y is C i all sample points in the category m i is C i the centroid of the category;

[0081] Gap(k) Calculation formula and optimal number of classifications The formula is as follows:

[0082]

[0083]

[0084] It can be seen from Figure 3 that when K best = 5, the Gap statistic value is the largest.

[0085] S4. Perform K-means clustering analysis on the selected clustering values. The specific process of dividing into the "rain" class and the "sunny" class is as follows: The K best centroids m R obtained by K-means clustering analysis are classified by the mean with k = 2, that is, the centroids of the "rain" class and the "sunny" class are obtained, and then the "rain" class and the "sunny" class are divided according to the K best -means classification results.

[0086] According to the obtained resistive current data of the "sunny" category, it is determined that the change of the resistive current is normal; according to the resistive current data of the "rain" category, it is determined that the obvious change of the resistive current is caused by excessive humidity, and the working state of the arrester is normal.

[0087] Figure 4 is the graph of humidity and resistive current "sunny" data. The resistive current data is stable. Under high humidity conditions, the change of the resistive current is small, and the calculation of the resistive current of the arrester is normal. Figure 5 is the graph of humidity and resistive current "rain" data. The resistive current is greatly affected by humidity and changes significantly. The resistive current of the arrester mutates due to excessive humidity, but the actual arrester works normally. The present invention can accurately classify the "rain" class data and provide guarantee for the stable operation of the arrester online monitoring system.

[0088] The description of the above embodiments is only used to help understand the method and its core idea of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

[0089] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for classifying resistive current data of a lightning arrester, characterized in that, including: Step S1: Obtain the humidity data RH and resistive current data I of the arrester R , and form the original data set X R ; Step S2: Initialize and set the number B of reference data sets, and perform KDE kernel density estimation on the original resistive current data I R According to Scott's rule, add skewness and adjustment parameter α, calculate the bandwidth h and the density function Randomly sample n data points from it to generate a reference data set, and repeat B times to generate a total of B reference data sets the skewness coefficient of the original resistive current data; Step S3: Calculate the original dataset X of all datasets R and the generated reference dataset X after processing ref The sum of squared classification distance errors under different numbers of clustering centers k, calculate the Gap statistic, and select the optimal K best value; Step S4: Based on the optimal K best value, perform K-means clustering analysis on the original dataset X R to divide it into "rain" class and "sunny" class.

2. The method for classifying resistive current data of a lightning arrester according to claim 1, wherein the KDE kernel density estimation formula is as follows: where n is the same as the number of resistive current data samplings, x represents a point in the reference dataset, and x i represents the i-th point in the resistive current, and Y(·) is the Gaussian kernel function.

3. The method for classifying resistive current data of a lightning arrester according to claim 1, characterized in that the specific steps of step S3 include: Normalized resistive current data I R and the generated resistive current reference data set I ref Humidity data RH, map the values to the range [-1, 1] to eliminate scale differences; set the number of cluster centers k ∈ [2, K], the maximum number of classifications does not exceed K, and the original data set X after normalization R_norm and B reference data sets X ref_norm Perform K-means clustering respectively, calculate the sum of squared classification distances under different k values in turn, and calculate the Gap statistic, and select the K with the largest Gap(k) value best as the optimal number of classifications.

4. The method for classifying resistive current data of a lightning arrester according to claim 3, characterized in that, in step S3, When the number of cluster centers is k, the sum of squared classification distance errors J (k) The formula is as follows: Among them, C i is the i-th class, and y is all the sample points in class C i ; m i is the centroid of class C i ; Gap(k) and the optimal number of clusters K best The calculation formula is as follows:

5. The method for classifying resistive current data of a lightning arrester according to claim 1, characterized in that, the number k of the cluster centers is 2.

6. The method for classifying resistive current data of a lightning arrester according to claim 5, characterized in that, The specific steps of step S4 include: taking the K centroids m obtained by K-means clustering analysis best and performing mean classification with k = 2, that is, obtaining the centroids of the "rain" class and the "sunny" class, and then dividing the "rain" class and the "sunny" class according to the K R -means classification results. best ​ 7. The method for classifying resistive current data of a lightning arrester according to claim 1, wherein in step S2, the value of B ranges from 5 to 10.

8. The method for classifying resistive current data of a lightning arrester according to claim 4, characterized in that the maximum number of classifications ranges from 5 to 8.

9. A method for judging the working state of a lightning arrester, characterized in that, including the following steps: classifying the resistive current data and humidity data by using the classification method according to any one of claims 1 to 8; collecting the resistive current value of the arrester to be measured and the humidity value during the test; comparing the collected resistive current and humidity values with the classification results; judging that the change of the resistive current is normal according to the "sunny" category resistive current data obtained; judging that the obvious change of the resistive current is caused by excessive humidity and the working state of the arrester is normal according to the "rain" category resistive current data.

Images