Lightning arrester operation state data optimization method based on fuzzy clustering and dynamic weighted fusion

Abstract

The invention discloses a lightning arrester operation state data optimization method based on fuzzy clustering and dynamic weighted fusion. The method comprises the following steps: step 1, extracting and normalizing a total current signal and frequency domain characteristics of a lightning arrester; step 2, initializing a clustering center by a density peak method for setting a comprehensive decision value condition; 3, performing nonlinear mapping on the processed data by adopting a Gaussian kernel function, constructing an improved fuzzy clustering (IFCM) objective function with a regular term, and solving the objective function; 4, constructing a data comprehensive scoring model in combination with a dynamic weighted fusion mechanism of fuzzy membership and mahalanobis distance; and 5, screening abnormal data based on a self-adaptive threshold value. According to the method, the defects that a traditional fuzzy clustering algorithm is sensitive to an initial value and the Euclidean distance is limited are effectively overcome, low-quality data points which are high in volatility and interfered can be automatically recognized and eliminated, and the accuracy and robustness of a health assessment model are remarkably improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power equipment state monitoring and evaluation, and particularly relates to a lightning arrester operation state data optimization method based on fuzzy clustering and dynamic weighted fusion. BACKGROUND

[0002] The zinc oxide lightning arrester is widely used in the power transmission and distribution network system, and its operation state is directly related to the insulation level and safe and stable operation of the system. Factors affecting the operation state of the lightning arrester include contamination, environmental temperature, rainfall, temperature rise, phase-to-phase temperature difference, total current, total current change rate, resistive leakage current, leakage current change rate and the like. The main challenges currently faced in the online monitoring of the operation state of the lightning arrester are as follows:

[0003] 1. Data multi-source and high interference: the monitoring data is derived from voltage, temperature, total current and other sensors, and the data is easily affected by factors such as electromagnetic interference on site, sensor noise, instantaneous working condition changes (such as operating overvoltage, lightning strike) and the like, showing strong volatility and a large number of abnormal pulse points.

[0004] 2. Limitation of traditional FCM: the traditional fuzzy C-means (FCM) clustering is sensitive to the initial clustering center, and it is difficult to process the non-linear and non-spherical distribution of the lightning arrester operation data under complex working conditions using the Euclidean distance.

[0005] 3. Lack of objective quality quantization standard: there is a lack of objective and quantized standard for evaluating and screening the quality of a large amount of monitoring data, resulting in insufficient reliability of the data input into the health evaluation model and affecting the accuracy and reliability of the evaluation results.

[0006] Therefore, it is urgent to develop an optimization method capable of effectively fusing multi-source information, automatically identifying and removing interference data and objectively quantizing data quality, to provide a high-quality data basis for accurate state evaluation of the lightning arrester. SUMMARY

[0007] The purpose of the present application is to provide a lightning arrester operation state data optimization method based on fuzzy clustering and dynamic weighted fusion, which can effectively fuse multi-source information, automatically identify and remove interference data, objectively quantize data quality and significantly improve the accuracy and robustness of the health evaluation model.

[0008] The technical solution for achieving the purpose of the present application is as follows:

[0009] A lightning arrester operation state data optimization method based on fuzzy clustering and dynamic weighted fusion comprises the following steps:

[0010] Step S1, collect lightning arrester operation monitoring original data, perform frequency domain feature extraction and normalization processing, and construct a lightning arrester sample set;

[0011] Step S2: Based on the surge arrester sample set, initialize the cluster centers using the density peak method to determine the comprehensive decision value conditions;

[0012] Step S3: Based on the initial cluster centers, a Gaussian kernel function is used to perform nonlinear mapping on the surge arrester sample data to construct a fuzzy clustering objective function with a regularization term;

[0013] Step S4: Solve the fuzzy clustering objective function to obtain the fuzzy membership degree. Combine the dynamic weighted fusion mechanism of fuzzy membership degree and Mahalanobis distance to construct a comprehensive data scoring model and obtain the data scores of all samples.

[0014] Step S5: Remove abnormal samples with scores lower than the adaptive scoring threshold.

[0015] Furthermore, the raw data for monitoring the operation of the surge arrester includes power supply voltage, ambient temperature, humidity, total atmospheric pressure, and raw waveform signal of the full current.

[0016] Furthermore, Fourier transform is used to extract frequency domain features, obtain spectrum information and extract key feature parameters, including the peak value of the total current, the effective value of the total current, the effective value of the fundamental wave, the initial phase angle of the fundamental wave, the amplitude of the third harmonic, the effective value of the third harmonic, the amplitude of the fifth harmonic, and the effective value of the fifth harmonic.

[0017] Further, step S2 specifically includes:

[0018] S21. Calculate the Euclidean distance between any two points in the surge arrester sample set. ,in N is the total number of samples collected;

[0019] S22, Euclidean distance between all sample points Ascending order processing to form a set Determine the cutoff distance , For index value:

[0020] ;

[0021] in, This is a weighting factor for the operation of surge arresters. It is a rounding function;

[0022] S23, Based on Euclidean distance between sample points and cutoff distance Calculate local density minimum distance ;

[0023] ;

[0024] S24. Normalize the local density and minimum distance respectively to obtain... and Calculate the comprehensive decision value for each sample point: ;

[0025] S25. Arrange the comprehensive decision values ​​in descending order to form a sequence. Calculate the adjacent differences and determine the index position C corresponding to the largest difference as the total number of cluster centers. Then, determine all points in the sequence with indices less than C as the initial cluster centers. ,in .

[0026] Furthermore, the fuzzy clustering objective function with regularization in step S3 is:

[0027] ;

[0028] in, For Gaussian kernel function, As cluster center, Let be the membership matrix, indicating that the i-th sample belongs to the k-th cluster center. The degree of fuzziness, where m is the fuzzy weighting index; This is the regularization coefficient.

[0029] Furthermore, the data comprehensive scoring model in step S4 is as follows:

[0030] ;

[0031] in, Representing data points Dynamic weights, The membership matrix, .

[0032] Furthermore, the fuzzy weighting index m=1.

[0033] Furthermore, the dynamic weights for:

[0034] ;

[0035] ;

[0036] in, and They are respectively The mean and covariance; Represents the j-th sample The degree of deviation from the normal statistical pattern, It is a positive number.

[0037] Further, the adaptive scoring threshold is:

[0038]

[0039] wherein, is an adaptive coefficient, is the mean of all sample scores, is the standard deviation of all scores.

[0040] Further, the adaptive coefficient ranges from .

[0041] Compared with the prior art, the present application has the following advantages:

[0042] 1. Strong anti-noise: The introduction of the Gaussian kernel function and the regularization term can effectively suppress the influence of impulse interference caused by the instantaneous operating condition changes or electromagnetic interference on site, improving the data quality.

[0043] 2. Dynamic adaptability and high precision: The dynamic weight real-time update based on Mahalanobis distance calculation can adaptively evaluate the quality deviation of data under different operating conditions (such as changes in environmental temperature and humidity), significantly improving the accuracy of data quality evaluation.

[0044] 3. High computational efficiency: The density peak initialization strategy with set comprehensive decision value condition avoids random initialization, can quickly converge, and effectively reduces the number of iterations.

[0045] 4. Improved robustness: The optimized data is more suitable for arrester health assessment model, improving the accuracy and reliability of arrester operating state assessment. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings of the embodiments. Obviously, the drawings described in the following description only relate to some embodiments of the present application, and are not a limitation on the present application.

[0047] Figure 1 is the flow chart of the present application based on improved fuzzy clustering and dynamic weighted fusion of arrester operating state data optimization method;

[0048] Figure 2 is the result graph of the improved fuzzy clustering method in the embodiment of the present application;

[0049] Figure 3 is the box plot of the health score index before and after optimization in the embodiment of the present application;

[0050] Figure 4A comparison chart of the full current signal of the surge arrester before and after the method optimization in the embodiment of the present application. DETAILED DESCRIPTION

[0051] The technical solutions in the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments.

[0052] As shown in the drawings, Figure 1 The present application proposes a surge arrester operation state data optimization method based on improved fuzzy clustering and dynamic weighted fusion, which effectively overcomes the sensitivity of the traditional fuzzy clustering algorithm to the initial value and the limitation of the Euclidean distance, can automatically identify and eliminate strong volatility and low-quality data points disturbed by interference, and significantly improves the accuracy and robustness of the health assessment model, specifically including:

[0053] S1, collecting the operation monitoring original data of the surge arrester, performing frequency domain feature extraction and normalization processing.

[0054] Specifically, the present application collects historical operation data of the surge arrester of a monitoring station under different environmental conditions, covering full current, full current rate of change, resistive leakage current, leakage current rate of change, environmental temperature, phase-to-phase temperature difference, rainfall, pollution level, etc.

[0055] The collected original current signal is subjected to fast Fourier transform (FFT), the time domain signal is converted into a frequency domain signal, and key feature parameter values are extracted from the FFT spectrum result, including: full current peak value, full current effective value, fundamental wave effective value, fundamental wave initial phase angle, third harmonic amplitude, third harmonic effective value, fifth harmonic amplitude, and fifth harmonic effective value.

[0056] All the feature parameters (including environmental parameters and electrical feature parameters) collected and calculated are subjected to normalization processing to obtain N is the total number of collected samples; and the dimensional influence is eliminated to prepare for subsequent analysis.

[0057] S2, setting the density peak method to initialize the clustering center of the comprehensive decision value condition.

[0058] In order to overcome the defect that the traditional FCM algorithm is sensitive to the initial clustering center, the present application introduces the local density (D) and minimum distance (Dmin) of the sample based on the density peak clustering algorithm. ​The application discloses a density peak initialization method of the fuzzy C-means (FCM) algorithm. In the lightning arrester operation state data monitoring, the data points with high local density usually represent common or stable operation conditions, and the data points in the low density area may correspond to abnormal or transition states. By automatically selecting initial clustering centers, the randomness of the traditional FCM is avoided, the selected center points fall in the dense area of data distribution, and it is more likely to approach the global optimal solution. Compared with the traditional FCM, the algorithm can improve the convergence speed and calculation efficiency, enhance the accuracy of lightning arrester typical operation mode recognition, and provide a reliable basis for subsequent state division and fault diagnosis.

[0059] To more accurately select the initial clustering center, the application sets a comprehensive decision value condition on the basis of the original density peak method, that is, the relative sizes of the local density and the minimum distance are considered at the same time. The double screening mechanism ensures that the selected center is located in the dense area of data and can effectively distinguish different data clusters, thereby further improving the accuracy and robustness of the clustering algorithm.

[0060] Specifically, the density peak method is used to initialize the clustering center, and the steps include:

[0061] S21, calculating the Euclidean distance between any two points in the lightning arrester sample set . ;

[0062] S22, sorting all the Euclidean distances between sample points in ascending order to form a set . , determining the cut-off distance . is an index value:

[0063]

[0064] wherein, is a weight coefficient of the lightning arrester operation, and is generally 0.01-0.02;

[0065] S23, calculating the local density and the minimum distance .

[0066]

[0067] S24, respectively normalizing the local density and the minimum distance to obtain and , and calculating the comprehensive decision value of each point: .

[0068] S25, arranging the comprehensive decision values in descending order to form a sequence , calculate adjacent difference value and determine index position C corresponding to maximum difference value as total number of clustering center, determine all points in sequence with index less than C as initial clustering center point , .

[0069] S3, adopt Gaussian kernel function to carry out nonlinear mapping to processed data, construct improved fuzzy clustering (IFCM) objective function with regular term and solve.

[0070] In order to solve the problem of nonlinear distribution of data, the Gaussian kernel function is introduced to carry out nonlinear mapping on the sample, so that the data is more easily clustered in a high-dimensional feature space, is a bandwidth parameter, and the Gaussian kernel function formula is as follows:

[0071]

[0072] Specifically, in S3, the improved fuzzy clustering objective function with a regular term is as follows:

[0073]

[0074] Wherein, is a Gaussian kernel function, is a clustering center, is a membership matrix, indicating the degree of the i-th sample belonging to the k-th clustering center , and m is a fuzzy weighting index; is a regularization coefficient, as a smoothing factor, and the regular term is used to balance the clustering confidence and suppress the extreme bias of abnormal data points on fuzzy membership.

[0075] As shown in the following figure, the clustering result after dimension reduction of high-dimensional frequency domain features by using the PCA method is shown. Figure 2 Each data point represents a lightning arrester operation data record, and different clusters identified by the IFCM (improved fuzzy C-means) algorithm are distinguished by color. The clustering boundary is clear, and the clustering centers are uniformly distributed.

[0076] The density peak initialization and kernel function mapping strategy proposed in the application can effectively improve the clustering stability and enhance the ability to distinguish abnormal data points. The points at the edge or isolated points can be used as candidate abnormal points for subsequent elimination by a dynamic scoring mechanism.

[0077] S4, a dynamic weighted fusion mechanism combining fuzzy membership and Mahalanobis distance is constructed to build a data comprehensive scoring model.

[0078] Specifically, the S4 specific steps include:

[0079] S41, calculate the Mahalanobis distance of each feature dimension of each arrester sample set based on the historical normal operation condition data of the arrester:

[0080]

[0081] Wherein, and are the mean and covariance of ; represents the degree of deviation of the i-th sample from the normal operation statistical mode;

[0082] S42, calculate the dynamic weight of each sample data :

[0083]

[0084] Wherein, represents the credibility or weight of the data point in the quality evaluation, which has dynamic adaptability; is a very small positive number, used to ensure numerical stability, and generally takes the value of .

[0085] S43, fuse the clustering membership degree and the dynamic weight , to build the comprehensive quality score of the arrester sample set data :

[0086]

[0087] As shown in Figure 3 , the distribution of the two groups of data before and after optimization on the health score index is compared by means of box plot. The horizontal axis represents the processing state (before optimization / after optimization), and the vertical axis represents the health score (ranging from 0 to 1). From the figure, it can be seen that the data score after optimization moves upward as a whole, and the mean and median are significantly improved. At the same time, the abnormal value points are reduced, and the box is more concentrated, which indicates that the data quality has been improved, and the volatility has decreased.

[0088] S5, remove abnormal points according to the adaptive score threshold.

[0089] Specifically, according to the mean-standard deviation threshold method, calculate the adaptive score threshold , remove the low-quality data points of ;

[0090]

[0091] Wherein, is the adaptive coefficient, used to control the strictness of screening, the value range of ; is the mean of all scores, representing the average data quality level of the current data set, is the standard deviation of all scores, reflecting the fluctuation of quality, and the threshold sets the acceptable minimum quality standard; the calculation of the adaptive threshold ensures the objectivity and consistency of data screening, without relying on artificial experience to set fixed empirical values.

[0092] In the data monitoring of surge arresters, these abnormal data points are usually data records affected by electromagnetic pulse interference, temporary sensor failure or instantaneous working condition changes, and their quality scores are significantly lower than the average level. Effective elimination of the strong fluctuation pulse points can significantly improve the smoothness and continuity of the data trend, providing a high-quality and reliable data basis for subsequent health assessment models.

[0093] Figure 4 is a comparison chart of the full current signals of the surge arresters before and after the method optimization in the embodiment of the present application, and the comparison result shows that there are strong fluctuation pulse points in the original data; after fuzzy comprehensive evaluation, the pulse interference data is eliminated, which can significantly improve the smoothness and continuity of the data trend; the data after comprehensive evaluation is more suitable for health assessment models.

[0094] In summary, the innovation of the present application is:

[0095] 1. Improved fuzzy C-means clustering: The traditional fuzzy C-means (FCM) clustering is sensitive to the initial clustering center, and it is difficult to handle the non-linear and non-spherical distribution of surge arrester operating data under complex working conditions using the Euclidean distance. A new strategy needs to be introduced to overcome the randomness of traditional FCM and the limitations of distance measurement, in order to accurately identify the typical operating mode of the surge arrester.

[0096] 2. Dynamic weighted fusion mechanism: The Mahalanobis distance and data noise suppression weight are introduced to realize dynamic and objective quantification of data quality.

[0097] 3. Adaptive scoring and threshold screening: The threshold is automatically determined based on statistical characteristics, effectively identifying and eliminating abnormal fluctuation points.

[0098] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined in the present application can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing the operating state data of a surge arrester based on fuzzy clustering and dynamic weighted fusion, characterized in that, The method comprises the following steps: Step S1, collecting the original data of the arrester operation monitoring, performing frequency domain feature extraction and normalization processing, and constructing an arrester sample set; Step S2, based on the arrester sample set, determining the density peak value method initialization clustering center of the comprehensive decision value condition; Step S3, based on the initialization clustering center, using the Gaussian kernel function to perform nonlinear mapping on the arrester sample data, and constructing a fuzzy clustering objective function with a regularization term; Step S4, solving the fuzzy clustering objective function to obtain the fuzzy membership degree, combining the fuzzy membership degree and the dynamic weighted fusion mechanism of Mahalanobis distance, constructing a data comprehensive scoring model, and obtaining the data score of all samples; Step S5, removing the abnormal samples with a score less than the adaptive scoring threshold.

2. The method according to claim 1, wherein, The original data of the arrester operation monitoring includes power supply voltage, ambient temperature, humidity, total atmospheric pressure, and full current original waveform signal.

3. The method according to claim 2, wherein, Fourier transform is used for frequency domain feature extraction to obtain frequency spectrum information and extract key characteristic parameters, including full current peak value, full current effective value, fundamental wave effective value, fundamental wave initial phase angle, third harmonic amplitude, third harmonic effective value, fifth harmonic amplitude, and fifth harmonic effective value.

4. The method according to claim 1, wherein, The step S2 specifically comprises: S21, calculating the Euclidean distance between any two points in the arrester sample set wherein N is the total number of collected samples; S22, Euclidean distance between all sample points Ascending processing constitutes a set , Determine the cut-off distance , Index value: ； wherein a weight coefficient for the operation of the arrester, is a rounding function; S23, compute the local density based on the Euclidean distance between sample points and the cut-off distance , compute the local density , compute the minimum distance ; ； S24, normalize the local density and the minimum distance respectively to obtain and , calculate the comprehensive decision value of each sample point: ; S25, arranging the comprehensive decision values in descending order to form a sequence , calculating the adjacent difference values and determining the index position C corresponding to the maximum difference value as the total number of cluster centers, and determining all points in the sequence with an index less than C as initial cluster center points wherein .

5. The method according to claim 4, wherein, The fuzzy clustering objective function with a regularization term in step S3 is: ； wherein, is a Gaussian kernel function, is a cluster center, is a membership matrix, indicating the degree to which the ith sample belongs to the kth cluster center , m is a fuzzy weighting exponent; is a regularization coefficient.

6. The method according to claim 5, wherein, The data comprehensive scoring model in step S4 is: ； wherein, represents a dynamic weight of data points , is a membership matrix, .

7. The method according to claim 6, wherein, The fuzzy weighted index m=1.

8. The method according to claim 6, wherein, The dynamic weights are: ； ； wherein, and are respectively the mean and covariance; denotes the degree to which the jth sample deviates from the normal operating statistical pattern, is positive.

9. The method according to claim 1, wherein, The adaptive scoring threshold is: ； wherein, is the adaptive coefficient, is the mean of all sample scores, is the standard deviation of all scores.

10. The method according to claim 9, wherein, The adaptive coefficient The value range of the adaptive coefficient .

Images