Circuit Breaker Fault Detection Method Based on Electrical Parameter Analysis

Abstract

This application relates to the technical field of digital data processing, specifically to a circuit breaker fault detection method based on electrical parameter analysis. The method includes: obtaining monitoring data and constructing a parameter matrix; screening suspicious elements based on the prediction deviation of elements in the parameter matrix and the mutation point detection result, determining a first eigenvalue according to the starting moment determined by the suspicious elements and the degree of regularity of data fluctuations within the fluctuation duration of the suspicious elements, and then combining the consistency analysis result of irregular data fluctuations in the parameter matrices of different types of monitoring data to determine a fault interval. Furthermore, target data is screened from the residual term, and a plurality of clustering trees are obtained based on the target data using the agglomerative hierarchical clustering algorithm. The fault detection result of the circuit breaker is determined based on the differences between different clustering trees. In this application, residual components are used to replace the initial data, reducing the amount of clustering data while improving the clustering effect of the clustering number on monitoring data that is sensitive to different faults.

Description

Technical Field

[0001] This application relates to the technical field of digital data processing, and specifically relates to a circuit breaker fault detection method based on electrical parameter analysis. Background Art

[0002] A circuit breaker is a device widely used in high-voltage and low-voltage power systems, which can automatically cut off the power supply in case of overload, short circuit or other faults. The circuit breaker uses an electromagnetic mechanism to drive the switch to act, and can quickly cut off the circuit to protect the equipment from damage caused by faults. Therefore, the safe operation of the circuit breaker itself is a prerequisite for the power system to maintain normal operation.

[0003] At present, significant progress has been made in the fault diagnosis technology of circuit breakers based on data-driven methods. Whether a circuit breaker has a fault and the type of the fault are diagnosed through the analysis results of vibration signals, electrical signals, or electrical parameters. Among them, the method based on data clustering is one of the commonly used methods for fault diagnosis using the electrical parameters of circuit breakers. Such methods have good robustness when facing a large amount of electrical parameter data collected on circuit breakers. However, when dealing with a large scale of electrical parameters, the time overhead is large, the timeliness of the fault diagnosis of the circuit breaker is poor, and it will cause a large amount of memory consumption. Summary of the Invention

[0004] In order to solve the technical problems that when using the data clustering method to diagnose the electrical parameters of a circuit breaker, the large scale of parameter quantity leads to poor detection timeliness and high computational memory consumption, this application provides a circuit breaker fault detection method based on electrical parameter analysis, and the specific technical solutions adopted are as follows:

[0005] This application provides a circuit breaker fault detection method based on electrical parameter analysis, and this method includes the following steps:

[0006] Obtain the monitoring data of the circuit breaker and construct a parameter matrix for each type of monitoring data;

[0007] Screen suspicious elements based on the prediction deviation of the elements in the parameter matrix and the mutation point detection result, and determine the first eigenvalue of the suspicious elements according to the starting moment determined by the suspicious elements in different row elements of the parameter matrix and the degree of regularity of the data fluctuation within the fluctuation duration of the suspicious elements;

[0008] Based on the first eigenvalue of the suspicious elements in each row element of the parameter matrix, combine the consistency analysis result of the irregular data fluctuations in the parameter matrices of different types of monitoring data to determine the fault interval for circuit breaker fault detection;

[0009] Screen target data from the residual terms according to the fault interval for circuit breaker fault detection, obtain several clustering trees based on the target data using the agglomerative hierarchical clustering algorithm, and determine the circuit breaker fault detection result based on the differences between different clustering trees.

[0010] Preferably, the screening of suspicious elements includes:

[0011] Using a data prediction algorithm, obtain the predicted value of each element based on several elements in the same row as each element in the parameter matrix and located before each element, and take the absolute value of the difference between the predicted value and each element as the prediction deviation of each element;

[0012] Obtain the mutation points in each row of elements in the parameter matrix by using the method of mutation point detection;

[0013] For any non-mutation point in each row of elements, calculate the sum of the time intervals between the element of each non-mutation point and the adjacent left and right mutation points as the fluctuation duration of each non-mutation point;

[0014] Take the non-mutation point with the largest prediction deviation and the smallest fluctuation duration in each row of elements as the suspicious element in each row of elements.

[0015] Preferably, the determination of the first eigenvalue of the suspicious element includes:

[0016] Take the acquisition moment of the suspicious element as the starting moment, and calculate the cumulative result on the parameter matrix of the time intervals between the starting moments determined by the suspicious elements in each row of elements in the parameter matrix and the starting moments determined by the suspicious elements in the remaining rows of elements;

[0017] Take the ratio of the cumulative result to the fluctuation duration of the suspicious element in each row of elements as the fault eigenvalue of the suspicious element in each row of elements;

[0018] Obtain the first eigenvalue of the suspicious element according to the prediction deviation of all elements, the discrete eigenvalue of all elements, and the fault eigenvalue within the fluctuation duration of the suspicious element.

[0019] Preferably, the acquisition of the first eigenvalue of the suspicious element includes:

[0020] Take the sum of the discrete eigenvalues of all elements within the fluctuation duration of the suspicious element and the tuning constant as the denominator;

[0021] Take the ratio of the product of the mean value of the prediction deviations of all elements within the fluctuation duration of the suspicious element and the fault eigenvalue of the suspicious element to the denominator as the first eigenvalue of the suspicious element.

[0022] Preferably, the determination of the fault interval for circuit breaker fault detection includes:

[0023] Perform differential measurement on the fluctuation durations of the suspicious elements determined by each circuit breaker in all parameter matrices to obtain the measurement result;

[0024] Obtain the correlation analysis results between the characteristic sequences of each circuit breaker on different two sets of monitoring data by using the method of correlation analysis;

[0025] Determine the second eigenvalue of the suspicious element by using the correlation analysis results, measurement results, and the first eigenvalue of the suspicious element between the characteristic sequences;

[0026] Obtain the second eigenvalue of the suspicious element of each circuit breaker in all parameter matrices, obtain the mode of all the second eigenvalues, and take the union of the fluctuation durations of the suspicious elements corresponding to the mode as the suspected fault interval of each circuit breaker;

[0027] Take the union of the suspected fault intervals of all circuit breakers as the fault interval for circuit breaker fault detection.

[0028] Preferably, the obtaining method of the characteristic sequence is: Take the sequence composed of the elements within the fluctuation duration of the corresponding suspicious element of each circuit breaker in the parameter matrix as the characteristic sequence of each circuit breaker on the corresponding monitoring data in the parameter matrix.

[0029] Preferably, the determining of the second eigenvalue of the suspicious element includes:

[0030] Calculate the mean value of the correlation analysis results between the characteristic sequences of each circuit breaker on any two different sets of monitoring data;

[0031] The second eigenvalue of the suspicious element consists of three parts: the first eigenvalue of the suspicious element, the measurement result, and the mean value; among them, the second eigenvalue is positively correlated with the first eigenvalue and the mean value respectively; the second eigenvalue is negatively correlated with the measurement result.

[0032] Preferably, the screening of the target data includes:

[0033] Take each row element in the parameter matrix as the input, and use the time series decomposition STL algorithm to obtain the residual term of each row element;

[0034] Denote the set composed of the residual components within the fault interval for circuit breaker fault detection in each residual term decomposed from each parameter matrix as the target data of each residual term;

[0035] Take the set composed of the target data of the residual terms of the corresponding row elements of each circuit breaker from all parameter matrices as the target data set of each circuit breaker.

[0036] Preferably, the determining of the circuit breaker fault detection result includes:

[0037] Use the hierarchical clustering algorithm to obtain the clustering tree of each circuit breaker based on the target data set of each circuit breaker;

[0038] Determine the status difference of each circuit breaker according to the height difference between the clustering trees of different circuit breakers and the difference in the number of nodes at the same height;

[0039] Use an anomaly detection algorithm to detect the outliers in the status differences of all circuit breakers, and regard the circuit breaker corresponding to the outlier as the faulty circuit breaker.

[0040] Preferably, determining the fault status of each circuit breaker includes:

[0041] Take the clustering tree of each circuit breaker as the target clustering tree respectively;

[0042] Calculate the height difference between the target clustering tree and the clustering tree of any other circuit breaker;

[0043] Calculate the cumulative result of the differences in the number of nodes at the same height between the target clustering tree and the clustering tree of any other circuit breaker at all the same heights;

[0044] Take the cumulative result of the product of the height difference and the cumulative result on the clustering trees of all circuit breakers as the status difference of the circuit breaker corresponding to the target clustering tree.

[0045] This application has at least the following beneficial effects:

[0046] First, the present application performs time series decomposition on the initial breaker voltage data to obtain residual components. Residual terms are obtained by performing time series decomposition on each row element in each parameter matrix. Compared with anomaly detection using monitoring data, the use of residual terms can better reflect the irregular data fluctuations caused by the breaker's own faults. Secondly, by analyzing the differences between the data fluctuations caused by the power system where the breaker is located and the irregular data fluctuations caused by the breaker's faults, suspicious elements that may be in the fault interval and the first eigenvalue of the suspicious elements are obtained, realizing a preliminary evaluation of the monitoring data in the breaker's fault state. After that, the second eigenvalue of the suspicious elements is determined using the correlation between the data fluctuations of different types of monitoring data during the breaker's own faults, and then the suspected fault interval of the breaker is determined. Considering that there are certain differences in the sensitivity of different types of monitoring data to faults during the breaker's faults, the subsequent clustering tree can process monitoring data with various sensitivities. After that, the union of all the suspected fault intervals is obtained as the fault interval for breaker fault detection. The acquisition of the fault interval takes into account that on the one hand, the noise sensitivities of different sensors are different; on the other hand, the breakers at different positions in the power system are affected differently. Taking the union can use more data for subsequent detection, eliminate the influence of noise data, and improve the accuracy of subsequent fault detection. After that, the target data set of each breaker is obtained using the fault interval for breaker fault detection and used as the input data for the agglomerative hierarchical clustering algorithm to generate a clustering tree, reducing the amount of clustering data while maintaining the clustering tree's pair. In the present application, the residual components caused by faults are used to replace the initial data, which not only reduces the data volume and improves the running efficiency of the algorithm, but also improves the clustering effect of the clustering tree on monitoring data with different fault sensitivities, making the final fault detection result more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] To more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0048] Figure 1 It is a flowchart of the breaker fault detection method based on electrical parameter analysis provided by the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] To further elaborate on the technical means and effects adopted by this application to achieve the intended invention purpose, the following specifically describes, in conjunction with the accompanying drawings and preferred embodiments, the specific implementation manner, structure, features, and effects of the circuit breaker fault detection method based on electrical parameter analysis proposed according to this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs.

[0051] The following specifically describes, in conjunction with the accompanying drawings, the specific solution of the circuit breaker fault detection method based on electrical parameter analysis provided by this application.

[0052] A circuit breaker fault detection method based on electrical parameter analysis provided by an embodiment of this application, specifically, provides the following circuit breaker fault detection method based on electrical parameter analysis. Please refer to Figure 1 . The method includes the following steps:

[0053] Step S001, obtain the real-time monitoring data of the circuit breaker and preprocess the obtained data.

[0054] The online monitoring system of the circuit breaker usually consists of three parts: a monitoring front end, a monitoring host computer, and a background system. The monitoring front end online monitors the electrical parameters of the circuit breaker through data acquisition devices, analyzes and calculates the collected monitoring data through the monitoring host computer, provides data support for diagnosing circuit breaker faults, and uses the background system to record the monitoring data and fault diagnosis results. Among them, the data acquisition devices include, but are not limited to, opening current sensors, closing current sensors, Hall sensors, and voltage sensors. The monitoring data that can be collected includes, but is not limited to, opening current, closing current, operating voltage, opening speed, closing speed, and voltage.

[0055] Specifically, each type of monitoring data of the circuit breaker collected by the monitoring front end is expressed in the form of a parameter matrix for subsequent data analysis and recording. First, for each type of monitoring data, in order to avoid missing data during the collection process, the collection results are cleaned by means of data filling to ensure that specific collection results exist for all types of monitoring data at the same collection moment. Second, a parameter matrix is constructed using the processed monitoring data: each column in the parameter matrix consists of several pieces of the same type of monitoring data of the circuit breaker at the same collection moment, and each row in the parameter matrix consists of the monitoring data of one circuit breaker at different collection moments. And the constructed parameter matrix is respectively denoted as 、 , which respectively represent the parameter matrices of the first type, the second type, and the nth type of monitoring data.

[0056] Among them, data filling is a well-known technology in the field of data processing, and this application will not elaborate further. Common data filling methods include, but are not limited to, mean filling, nearest neighbor filling, and interpolation filling. This application does not impose special restrictions on the specific methods of data filling. Preferably, as an embodiment of this application, the acquisition results are cleaned by using the nearest neighbor filling method.

[0057] Thus, the parameter matrix of the circuit breaker is obtained for analyzing the condition of the circuit breaker.

[0058] Step S002: Comprehensively analyze the fluctuation degree of the circuit breaker in different types of monitoring data and the predictability during the data fluctuation period, screen out the suspicious elements in the parameter matrix, and determine the first eigenvalue of the suspicious elements.

[0059] Agglomerative Hierarchical Clustering is a data processing method that constructs a clustering hierarchy by gradually merging similar samples or clusters. The basic idea of agglomerative hierarchical clustering is to start from each sample or cluster, gradually merge the most similar samples or clusters until all samples or clusters are merged into a large cluster or a predetermined stopping condition is reached. This algorithm is relatively robust, but due to its high computational complexity, agglomerative hierarchical clustering has large space and time overheads when dealing with large-scale datasets, resulting in long calculation times and large memory consumption, etc. Therefore, in this application, the parameter matrix obtained by collecting the circuit breaker is optimized to obtain less data that can reflect whether the circuit breaker is in a faulty state as the input data of the agglomerative hierarchical clustering algorithm, so as to reduce the running time and complexity of the algorithm on the premise of being able to detect circuit breaker faults.

[0060] First, perform time series analysis on the monitoring data of all circuit breakers in the entire power system. Taking the monitoring matrix of the nth electrical parameter as an example, each row element of is respectively used as the input of the seasonal-trend decomposition procedure based on loess (STL) time series decomposition algorithm, and each row element is decomposed into a trend term, a residual term, and a seasonal term. The seasonal smoothing parameter, trend smoothing parameter, and estimation window length in the algorithm are respectively set to the empirical values 11, 17, and 25. Among them, time series decomposition is a well-known technology in the field of data processing, and this application will not elaborate further.

[0061] Secondly, for each row element in the parameter matrix, the residual term represents the data characteristics that cannot be explained by the trend and seasonal terms. The residual term includes outliers, noise, and irregular data changes in the monitoring data of the circuit breaker over time. Therefore, when a circuit breaker fails, the data changes in the monitoring data during the failure are significantly different from the data trend of the monitoring data when the circuit breaker is normal, and cannot be reflected by the trend term and seasonal term, but can be characterized by the residual term.

[0062] Due to various power conditions such as power scheduling and harmonic oscillation in the power system, there are certain fluctuations in the monitoring data on the circuit breaker, which also exist in the decomposed residual term. For example, the inrush current and overvoltage generated when the circuit breaker is closed will directly cause current and voltage fluctuations; when there is a latent fault in the power system where the circuit breaker is located, there will be contact problems caused by short-circuit current, which may also cause current fluctuations.

[0063] In summary, the decomposed residual term not only contains the data changes caused by the circuit breaker failure, but also may have data fluctuations caused by various factors in the power system where the circuit breaker is located. Therefore, it is necessary to conduct in-depth analysis on the residual term.

[0064] Specifically, when the circuit breaker fails, various monitoring data of the circuit breaker will change synchronously. And in a short period of time after the failure, that is, the same type of monitoring data at adjacent several acquisition moments will have irregular data fluctuations, and this kind of data fluctuation is unpredictable; while the data fluctuations in the parameter matrix caused by the above power conditions in the power system are usually predictable, and generally there will be mutation situations only in a few or even one type of monitoring data.

[0065] First, for each element in the parameter matrix take the elements in the same row as each element and the first K elements before each element as the input of the prediction algorithm, use the prediction model to output the predicted value of each element, and take the absolute value of the difference between the predicted value and each element as the prediction deviation of each element. The larger the prediction deviation, the greater the amplitude of the data change of the element, and the more likely it is caused by the circuit breaker failure.

[0066] It should be noted that data prediction is a well-known technology in the field of data processing, and will not be elaborated in this application. Common data predictions include, but are not limited to, prediction methods based on support vector machines, exponential smoothing method, autoregressive integrated moving average model ARIMA, prediction models based on neural networks. This application does not make special restrictions on data prediction methods. Preferably, as an embodiment of this application, the autoregressive integrated moving average model ARIMA is used to obtain the prediction results of each element.

[0067] Secondly, use the method of mutation point detection to obtain the parameter matrix The mutation points in each line of elements, where corresponding to these mutation points are the elements with changed monitoring data. After obtaining all the mutation points in each line of elements. For any non-mutation point in each line of elements, calculate the sum of the time intervals between the element at each non-mutation point and its adjacent left and right mutation points as the fluctuation duration of each non-mutation point. The smaller the fluctuation duration, the more it represents that the element is in a relatively intense short-term data fluctuation, which may be caused by instantaneous faults such as partial discharge and overvoltage. If it is a data fluctuation caused by influencing factors in the power system, the duration of data change is relatively long.

[0068] Specifically, the elements to the left of the first mutation point and the elements to the right of the last mutation point in each line of elements do not participate in the calculation of the time interval. This is because if an element is the monitoring data during a circuit breaker fault, then the mutation point corresponding to the moment when the circuit breaker fault resumes normal must be to the right of this element.

[0069] Furthermore, take the non-mutation point with the largest prediction deviation and the smallest fluctuation duration in each line of elements as the suspicious element that is highly likely to be during a circuit breaker fault, and take the acquisition moment of the suspicious element as the starting moment of the irregular data fluctuation. In the parameter matrix For the starting moment determined according to the above process for each line of elements in it, if the starting moments in multiple lines of elements belong to the same moment, it indicates that there is probably a composite fault such as short circuit and phase break superimposition in the power system where the circuit breaker is located, triggering the linkage protection mechanism in the power system, resulting in the synchronous operation of circuit breakers at different positions, rather than an irregular data fluctuation caused by the fault of the circuit breaker itself.

[0070] Here, calculate the first eigenvalue of the suspicious element to evaluate the continuous occurrence of irregular data fluctuations caused by the circuit breaker fault at several consecutive acquisition moments after the starting moment. The specific calculation formula is as follows:

[0071]

[0072] In the formula, is the fault eigenvalue of the suspicious element in the first line of elements, M is the number of rows of the parameter matrix, 、 are respectively the starting moments determined by the suspicious element in the first row and the m-th row in the parameter matrix, is the fluctuation duration of the suspicious element in the first line of elements.

[0073] Based on the fault eigenvalue and combined with the degree of regularity of the data fluctuation within the fluctuation duration of the suspicious element, comprehensively determine the first eigenvalue of the suspicious element:

[0074]

[0075] In the formula, is the first eigenvalue of the suspicious element among the elements in the first row, is the mean value of the prediction deviations of all elements within the fluctuation duration of the suspicious element among the elements in the first row, is the discrete eigenvalue of all elements within the fluctuation duration of the suspicious element among the elements in the first row. is a parameter adjustment constant used to avoid the influence of a zero denominator on the calculation. In this application, takes a positive value less than 0.01. Preferably, in an embodiment of this application, the empirical value 0.001 is taken.

[0076] It should be noted that the discrete eigenvalue is used to evaluate the irregular data fluctuations of all elements within the fluctuation duration of the suspicious element. The calculation methods of the discrete eigenvalue include, but are not limited to, the distribution variance and the coefficient of variation. This application does not impose special restrictions on the specific calculation method. Preferably, in an embodiment of this application, the coefficient of variation of all elements within the fluctuation duration of the suspicious element is calculated as the discrete eigenvalue.

[0077] It should be further noted that the greater the time deviation of the start moment of the data change determined by the monitoring data of different circuit breakers in the parameter matrix, the lower the possibility of the residual term caused by the compound fault in the power system; at the same time, the higher the degree of data irregularity, the greater the predictability, and the higher the degree of irregularity of the data change, the greater the correlation between the data characteristics expressed in the residual term and the circuit breaker fault. That is the greater the value of

[0078] the greater the possibility that the suspicious element in the first row is abnormal data caused by a circuit breaker fault, and the more likely the residual term of the elements within the fluctuation duration of the suspicious element in the first row reflects the irregular data fluctuation characteristics caused by the circuit breaker fault.

[0079] S003. Based on the first eigenvalue of the suspicious element in each row of the parameter matrix, combined with the consistency analysis result of the irregular data fluctuations in the parameter matrices of different types of monitoring data, determine the fault interval for circuit breaker fault detection.

[0080] In the above steps, the first eigenvalue of the suspicious elements in each row of the parameter matrix is determined based on the fluctuation duration of the same type of monitoring data of the sensor and the synchronization characteristics of the fluctuations of the circuit breaker, and a preliminary evaluation is carried out on whether different data in the residual term are caused by the circuit breaker failure. However, it is still not sufficient to obtain the data for circuit breaker failure detection only through a single parameter. Therefore, in this application, further analysis is carried out on the parameter matrix of different types of monitoring data at the starting moment determined based on each suspicious element in the parameter matrix to determine the fault interval for circuit breaker failure detection.

[0081] Specifically, according to the above process, the suspicious elements in each row of elements are obtained in the parameter matrix of each type of monitoring data respectively. For any circuit breaker, taking the a-th circuit breaker as an example for description, after determining the suspicious elements in the monitoring data of the a-th circuit breaker in all parameter matrices, the fluctuation duration and the starting moment of each suspicious element are further obtained respectively.

[0082] On the one hand, when the a-th circuit breaker fails, within the time period from the failure of the a-th circuit breaker to the return to the normal state, the duration of data changes in different types of monitoring data on the a-th circuit breaker should be relatively consistent, that is, the fluctuation durations of the suspicious elements in different parameter matrices are basically close. On the other hand, if the data fluctuation is caused by the circuit breaker failure, then for different types of monitoring data of the same circuit breaker, for example, when the closing current of the circuit breaker increases, it will cause difficulty in closing, and the closing speed will change significantly; when the closing resistance of the circuit breaker fails, the closing current will increase, resulting in the failure of inrush current suppression and possibly accompanied by voltage fluctuations. That is, the failure of the circuit breaker itself will inevitably lead to a certain degree of correlation between different types of monitoring data. Therefore, the elements of different types of monitoring data within the fluctuation duration can be subjected to correlation analysis, and the time range of the circuit breaker failure can be further screened according to the correlation analysis result.

[0083] Specifically, the suspicious elements corresponding to the rows in all parameter matrices of the a-th circuit breaker are obtained respectively, and the sequence composed of the elements within the fluctuation duration of each suspicious element is used as the characteristic sequence of the corresponding monitoring data in the parameter matrix. Using the method of correlation analysis, the correlation between different characteristic sequences of the a-th circuit breaker is analyzed, and the correlation between the characteristic sequences of the a-th circuit breaker on the i-th type of monitoring data and the j-th type of monitoring data is denoted as , and the mean value of the correlations between the characteristic sequences of the a-th circuit breaker on all types of monitoring data is denoted as .

[0084] Embodiment 1:

[0085] Taking the characteristic sequence of any one of the n kinds of monitoring data of the a-th circuit breaker as the dependent variable respectively, and the characteristic sequences of the remaining kinds of monitoring data as the independent variables, the correlation analysis algorithm is used to analyze the correlation between the independent variable and the dependent variable, and the correlation between different characteristic sequences is determined.

[0086] Embodiment 2:

[0087] Taking the characteristic sequences of any 2 kinds of the n kinds of monitoring data of the a-th circuit breaker as the independent variables respectively, and the characteristic sequences of the remaining kinds of monitoring data as the dependent variables respectively, the correlation analysis algorithm is used to analyze the correlation between the independent variable and the dependent variable, and the correlation between different characteristic sequences is determined.

[0088] It should be noted that correlation analysis is a well-known technology in the field of data processing. Commonly used correlation analysis algorithms include, but are not limited to, grey relational analysis, response surface analysis, and multi-factor analysis. This application does not impose special restrictions on the specific method. Preferably, as an embodiment of this application, the grey relational analysis method is used to obtain the correlation between the different characteristic sequences.

[0089] Here, the second eigenvalue of the suspicious element is calculated to evaluate whether there is a certain correlation between the irregular data fluctuations that occur during the fluctuation duration of the circuit breaker. First, a difference evaluation is performed on the fluctuation durations determined in all parameter matrices of the a-th circuit breaker. The greater the difference in the lengths of all the fluctuation durations, the greater the difference in the fluctuation conditions of different kinds of monitoring data, and the greater the influence of other factors in the power system. Among them, the purpose of performing the difference evaluation is to measure the difference between different fluctuation durations. Therefore, on the premise of being able to achieve this purpose, the difference between the fluctuation durations can be measured by means including, but not limited to, distribution variance and coefficient of variation.

[0090] Secondly, the second eigenvalue of the suspicious element consists of the first eigenvalue of the suspicious element, the difference measurement result of the fluctuation duration, and three parts. Among them, the second eigenvalue is positively correlated with the first eigenvalue and respectively, and the second eigenvalue is negatively correlated with the difference measurement result of the fluctuation duration.

[0091] It should be noted that positive correlation means that when one variable increases, the other variable also increases, and the change directions of the two variables are the same. When one variable changes from large to small or from small to large, the other variable also changes from large to small or from small to large; the specific relationship is determined by the actual application, and this application does not impose special restrictions.

[0092] It should be noted that negative correlation means that when one variable increases, the other variable decreases accordingly. The change directions of the two variables are opposite. When one variable changes from large to small or from small to large, the other variable also changes from small to large or from large to small. The specific relationship is determined by the actual application, and no special restrictions are imposed in this application.

[0093] Preferably, in an embodiment of the present application, the specific calculation formula for the second eigenvalue of the suspicious element is as follows:

[0094]

[0095] In the formula, is the second eigenvalue of the suspicious element of the a-th circuit breaker in the j-th parameter matrix, is the first eigenvalue of the suspicious element of the a-th circuit breaker in the j-th parameter matrix, is the mean value of the correlation between the corresponding eigen-sequences of the a-th circuit breaker in all parameter matrices, is the mean value of the difference between the corresponding fluctuation durations of the a-th circuit breaker in all parameter matrices.

[0096] Furthermore, according to the above steps, the second eigenvalues of the suspicious elements of the a-th circuit breaker in all types of parameter matrices are respectively obtained, the mode of all the second eigenvalues is obtained, and the union of the fluctuation durations of the suspicious elements corresponding to the mode is used as the suspected fault interval when the a-th circuit breaker fails. The reason for taking the mode here is that when the circuit breaker fails, there will be certain differences in the sensitivity of different types of monitoring data to the fault. To train a detection model with excellent detection performance, it should be able to process monitoring data with various sensitivities.

[0097] According to the above steps, the suspected fault intervals of all circuit breakers are respectively determined, and the union of all the suspected fault intervals is obtained as the fault interval for circuit breaker fault detection. The reason for taking the union in this way is that on the one hand, the noise sensitivities of different sensors are different; on the other hand, the circuit breakers at different positions in the power system are affected differently. Taking the union can use more data for subsequent detection, eliminate the influence of noise data, and improve the accuracy of subsequent fault detection.

[0098] S004, Screen target data from the residual term according to the fault interval of circuit breaker fault detection, obtain several clustering trees based on the target data using the agglomerative hierarchical clustering algorithm, and determine the fault detection result of the circuit breaker based on the differences between different clustering trees.

[0099] After obtaining the fault interval for the breaker fault detection through the above steps, a set composed of the residual components belonging to the fault interval in each residual term is extracted from each parameter matrix, which is denoted as the target data of each residual term; the set composed of the target data of the residual terms of the corresponding row elements of each breaker from all parameter matrices is used as the target data set of each breaker.

[0100] Further, after obtaining the target data sets of all breakers according to the above process, the target data sets of each breaker are respectively used as the input data of the agglomerative hierarchical clustering algorithm. When clustering, the average linkage algorithm is used to calculate the similarity between clusters, and no clustering threshold is set. The clustering tree of each target data set is obtained in turn, and the number of nodes of each clustering tree at each height is respectively counted. The more the number of nodes, the worse the clustering effect at the corresponding height.

[0101] Secondly, the abnormal data in the target data sets of all parameter matrices is evaluated according to the height difference between different clustering trees and the difference in the number of nodes at the same height. For the monitoring data with greater changes when the breaker itself fails, the residual components collected from the monitoring data in the fault interval can better reflect the characteristics of the monitoring data during breaker failure, and the higher the irregularity degree of the residual components. Therefore, compared with the other clustering trees, the clustering tree formed by the target data of the parameter matrix of the faulty breaker has a higher irregularity degree of the residual components at the same height and a worse aggregation during the clustering process.

[0102]

[0103] In the formula, is the state difference of the a-th breaker, M is the number of breakers, is the height difference between the clustering trees obtained from the target data of the a-th breaker and the b-th breaker, is the cumulative result of the difference in the number of nodes at all the same heights between the clustering trees obtained from the target data of the a-th breaker and the b-th breaker.

[0104] Specifically, if the heights of two clustering trees are inconsistent, when calculating the difference in the number of nodes at the same height between the two clustering trees, the clustering tree with the maximum height is used as the standard, and the number of nodes of the clustering tree with the minimum height at the height difference between the two clustering trees is recorded as 0. For example, if the heights of two clustering trees are 10 and 8 respectively, for the clustering tree with a height of 8, the number of nodes at heights 9 and 10 is recorded as 0.

[0105] According to the above steps, the state differences of all breakers are obtained respectively, and the outlier detection algorithm is used to detect the outliers, and the breaker corresponding to each outlier is used as the faulty breaker.

[0106] It should be noted that anomaly detection is a well-known technology in the field of data processing, and the specific process will not be elaborated here. Common anomaly detection methods include, but are not limited to, LOF anomaly detection and isolation forest detection. This application does not impose special restrictions on the anomaly detection algorithm. Preferably, as an embodiment of this application, the LOF anomaly detection algorithm is used to obtain the outliers among all the state differences.

[0107] It should be noted that the above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included in the protection scope of this application.

[0108] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other, and the key point of each embodiment is to illustrate the differences from other embodiments.

Claims

1. A circuit breaker fault detection method based on electrical parameter analysis, characterized in that: The method comprises the following steps: Obtain monitoring data of circuit breakers and construct parameter matrix for each monitoring data; Using a data prediction algorithm to obtain a predicted value for each element based on a number of elements in the parameter matrix that are in the same row as each element and are located before each element, the absolute value of the difference between the predicted value and each element is used as the predicted deviation of each element; The mutation point in each row of the parameter matrix is ​​obtained by using the mutation point detection method; For any non-mutation point in each row of elements, the sum of the time intervals between each non-mutation point and its adjacent left and right mutation points is calculated as the fluctuation duration of each non-mutation point; The non-mutation point with the largest prediction deviation and the smallest fluctuation duration in each row of elements is taken as the suspicious element in each row of elements; the first eigenvalue of the suspicious element is determined according to the starting time determined by the suspicious element in different rows of elements in the parameter matrix and the regularity of the data fluctuation within the fluctuation duration of the suspicious element; Performing difference measurement on the fluctuation duration of the suspicious elements determined in all parameter matrices of each circuit breaker to obtain the measurement result; The correlation analysis method is used to obtain the correlation analysis results between the characteristic sequences of each circuit breaker on two different monitoring data; Determine the second eigenvalue of the suspicious element by using the correlation analysis result between the feature sequences, the measurement result, and the first eigenvalue of the suspicious element; Obtain the second eigenvalue of the suspicious element of each circuit breaker in all parameter matrices, obtain the mode of all the second eigenvalues, and use the union of the fluctuation durations of the suspicious elements corresponding to the modes as the suspected fault interval of each circuit breaker; The union of the suspected fault intervals of all circuit breakers is used as the fault interval for circuit breaker fault detection; The target data is screened from the residual items according to the fault interval of circuit breaker fault detection, and several clustering trees are obtained based on the target data using an agglomerative hierarchical clustering algorithm. The fault detection result of the circuit breaker is determined based on the difference between different clustering trees.

2. The circuit breaker fault detection method based on electrical parameter analysis according to claim 1, characterized in that: The determining of the first characteristic value of the suspicious element includes: Taking the acquisition time of the suspicious element as the starting time, calculating the cumulative result of the time interval between the starting time determined by the suspicious element in each row of elements in the parameter matrix and the starting time determined by the suspicious elements in the remaining rows of elements on the parameter matrix; The ratio of the accumulated result to the fluctuation duration of the suspicious element in each row of elements is used as the fault characteristic value of the suspicious element in each row of elements; The first characteristic value of the suspicious element is obtained according to the prediction deviations of all elements within the fluctuation duration of the suspicious element, the discrete characteristic values ​​of all elements, and the fault characteristic value.

3. The circuit breaker fault detection method based on electrical parameter analysis according to claim 2, characterized in that: The obtaining of the first characteristic value of the suspicious element includes: The sum of the discrete eigenvalues ​​of all elements within the fluctuation duration of the suspicious element and the parameter adjustment constant is used as the denominator; The ratio of the product of the mean of the predicted deviations of all elements within the fluctuation time of the suspicious element and the fault characteristic value of the suspicious element to the denominator is taken as the first characteristic value of the suspicious element.

4. The circuit breaker fault detection method based on electrical parameter analysis according to claim 1, characterized in that: The characteristic sequence is obtained by taking a sequence composed of elements within the fluctuation duration of the suspicious element corresponding to each circuit breaker in the parameter matrix as a characteristic sequence of each circuit breaker on the monitoring data corresponding to the parameter matrix.

5. The circuit breaker fault detection method based on electrical parameter analysis according to claim 1, characterized in that: The determining the second characteristic value of the suspicious element comprises: Calculate the mean of the correlation analysis results between the characteristic sequences of each circuit breaker on any two different monitoring data; The second eigenvalue of the suspicious element consists of the first eigenvalue of the suspicious element, the measurement result, and the mean; wherein the second eigenvalue is positively correlated with the first eigenvalue and the mean, respectively; and the second eigenvalue is negatively correlated with the measurement result.

6. The circuit breaker fault detection method based on electrical parameter analysis according to claim 1, characterized in that: The screening target data includes: Take each row element in the parameter matrix as input and use the time series decomposition STL algorithm to obtain the residual term of each row element; A set of residual components belonging to the fault interval of circuit breaker fault detection in each residual item obtained by decomposing each parameter matrix is ​​recorded as target data of each residual item; The set consisting of the target data of the residual items of the corresponding row elements of each circuit breaker from all the parameter matrices is taken as the target data set of each circuit breaker.

7. The circuit breaker fault detection method based on electrical parameter analysis according to claim 1, characterized in that: The determining of the fault detection result of the circuit breaker comprises: A clustering tree of each circuit breaker is obtained based on the target data set of each circuit breaker using a hierarchical clustering algorithm; Determine the status difference of each circuit breaker based on the height difference between the clustering trees of different circuit breakers and the difference between the number of nodes at the same height; An abnormal value in the state difference of all circuit breakers is detected by using an abnormal detection algorithm, and the circuit breaker corresponding to the abnormal value is regarded as a faulty circuit breaker.

8. The circuit breaker fault detection method based on electrical parameter analysis according to claim 7, characterized in that: Determining the fault status of each circuit breaker includes: The clustering tree of each circuit breaker is used as the target clustering tree; Calculate the height difference between the target clustering tree and the clustering tree of any other circuit breaker; Calculate the cumulative result of the difference between the number of nodes at the same height between the target clustering tree and the clustering tree of any other circuit breaker at the same height; The product of the height difference value and the accumulation result is accumulated on the clustering trees of all circuit breakers as the state difference of the circuit breaker corresponding to the target clustering tree.

Images