A Method and Device for Fault Location of Metering Master Stations Based on Accelerated Fuzzy C-Means Clustering

Abstract

This invention discloses a method and apparatus for fault location of a metering master station based on accelerated fuzzy c-means clustering. The method includes: calling the accelerated fuzzy c-means clustering algorithm to process the abnormal dataset, iterating continuously until the cluster set of the abnormal dataset is less than a preset threshold or the number of iterations reaches a set upper limit, obtaining the cluster set and membership set after clustering; obtaining the classification result corresponding to the fault type and fault characteristics based on the cluster set and the membership set; comparing the classification result with the normal data in the log to locate the fault area. This invention uses a KCN network to accelerate the iteration speed of the FCM clustering network, ensuring both accuracy and speed in data classification and fault location.

Description

Technical Field

[0001] This invention relates to the field of network monitoring technology, and in particular to a method and apparatus for locating faults in metering master stations based on accelerated fuzzy c-means clustering. Background Technology

[0002] Existing fault location technologies for automated metering master stations include test-based fault location, AI-based fault location, and model-based fault location. Test-based methods assume the system is a white-box system, locating faults by testing the system's output or other relevant conditions when a failure occurs. AI-based methods analyze historical faults and observed phenomena during operation to determine the cause of the fault and provide corresponding solutions based on case studies. Model-based methods establish relationships between system structure and events to reflect behavior; the most common method is fault location based on log information.

[0003] The testing method can only perform simple failure model detection and requires the system to stop operating and enter the testing phase when a fault occurs. Metering and maintenance personnel often rely on experience-based detection methods, the accuracy of which is subjectively influenced and depends on the processing capabilities of the personnel, thus failing to meet current requirements for accuracy and efficiency. The training time for neural networks in artificial intelligence methods is too long, and they cannot provide ideal results for data not in the training set, making them unsuitable for real-time fault location. In model-based methods, the amount of log data acquired is enormous, including some invalid information. Existing testing, artificial intelligence, and model-based methods cannot effectively and efficiently classify the data required for fault location at the metering master station, nor can they guarantee that fault location can be performed at any time. Summary of the Invention

[0004] This invention provides a method and apparatus for fault location of metering master stations based on accelerated fuzzy c-means clustering. It uses a KCN network to accelerate the iteration speed of the FCM clustering network, ensuring the accuracy and speed of data classification and fault location.

[0005] To achieve the above objectives, a first aspect of this application provides a method for fault location of a metering master station based on accelerated fuzzy c-means clustering, comprising:

[0006] Collect log data from the metering master station and retain complete fault characteristics of abnormal data to form a non-redundant abnormal dataset;

[0007] The clustering process of the fuzzy c-means clustering algorithm is accelerated by using the iterative rules with learning rate in the KCN clustering network.

[0008] The accelerated fuzzy c-means clustering algorithm is called to process the abnormal dataset, and the iteration continues until the cluster set of the abnormal dataset is less than a preset threshold or the number of iterations reaches a set upper limit, so as to obtain the cluster set and membership set after clustering.

[0009] Based on the cluster set and the membership set, a classification result is obtained in which the fault type and fault feature correspond to each other;

[0010] The classification results are compared with the normal data in the logs to locate the fault area.

[0011] In one possible implementation of the first aspect, the acceleration of the clustering process of the fuzzy c-means clustering algorithm based on the iterative rules containing the learning rate in the KCN clustering network specifically includes:

[0012] The learning rate is defined based on the cluster set of the fuzzy convergence operator and the fuzzy c-means clustering algorithm;

[0013] The iteration rules of the KCN clustering network are updated according to the definition of the learning rate;

[0014] Replace the iterative rules of the fuzzy c-means clustering algorithm with the iterative rules of the KCN clustering network.

[0015] In one possible implementation of the first aspect, the objective function of the fuzzy c-means clustering algorithm is related to the difference between outlier data and each cluster, and the corresponding membership degree.

[0016] In one possible implementation of the first aspect, the accelerated fuzzy c-means clustering algorithm specifically includes the following steps:

[0017] Initialize the cluster set and fuzzy parameters based on the abnormal dataset, and calculate the initial membership degree set;

[0018] The iteration is performed according to the learning rate-inclusive iteration rules in the KCN clustering network and the membership set iteration rules of the fuzzy c-means clustering algorithm until the convergence condition is met.

[0019] In one possible implementation of the first aspect, the fault types include process anomalies, resource usage anomalies, and server load anomalies, and the fault characteristics include data traffic, TCP connection count, and server CPU utilization.

[0020] A second aspect of this application provides a metering master station fault location device based on accelerated fuzzy c-means clustering, comprising:

[0021] The data acquisition module is used to collect log data from the metering master station and retain complete abnormal data with fault characteristics to form a non-redundant abnormal dataset.

[0022] The acceleration module is used to accelerate the clustering process of the fuzzy c-means clustering algorithm based on the iterative rules containing the learning rate in the KCN clustering network.

[0023] The iteration module is used to call the accelerated fuzzy c-means clustering algorithm to process the abnormal dataset, and continue to iterate until the cluster set of the abnormal dataset is less than a preset threshold or the number of iterations reaches a set upper limit, so as to obtain the cluster set and membership set after clustering.

[0024] The classification module is used to obtain classification results that correspond to fault types and fault features based on the cluster set and the membership set;

[0025] The location module is used to compare the classification results with the normal data in the logs to locate the fault area.

[0026] In one possible implementation of the second aspect, the acceleration module is specifically used for:

[0027] The learning rate is defined based on the cluster set of the fuzzy convergence operator and the fuzzy c-means clustering algorithm;

[0028] The iteration rules of the KCN clustering network are updated according to the definition of the learning rate;

[0029] Replace the iterative rules of the fuzzy c-means clustering algorithm with the iterative rules of the KCN clustering network.

[0030] In one possible implementation of the second aspect, the objective function of the fuzzy c-means clustering algorithm is related to the difference between outlier data and each cluster, and the corresponding membership degree.

[0031] In one possible implementation of the second aspect, the accelerated fuzzy c-means clustering algorithm specifically includes the following steps:

[0032] Initialize the cluster set and fuzzy parameters based on the abnormal dataset, and calculate the initial membership degree set;

[0033] The iteration is performed according to the learning rate-inclusive iteration rules in the KCN clustering network and the membership set iteration rules of the fuzzy c-means clustering algorithm until the convergence condition is met.

[0034] In one possible implementation of the second aspect, the fault types include process anomalies, resource usage anomalies, and server load anomalies, and the fault characteristics include data traffic, TCP connection count, and server CPU utilization.

[0035] Compared to existing technologies, this invention provides a method and apparatus for fault location in metering master stations based on accelerated fuzzy c-means clustering. It extracts fault-related features from logs to prepare for subsequent clustering analysis. The processed dataset is then classified using a combination of FCM (fuzzy c-means clustering) and KCN (Knowledge-Based Clustering) to obtain classification results where fault types and fault features correspond. By comparing the results in the fault group with normal data, fault location can be accurately determined. Because the KCN iterative rules regarding learning rate are used to accelerate the original FCM clustering, the FCM clustering network achieves faster and more accurate iterative results. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating a method for locating faults in a metering master station based on accelerated fuzzy c-means clustering, provided by an embodiment of the present invention.

[0037] Figure 2 This is a schematic diagram of the clustering process of an accelerated fuzzy c-means clustering algorithm provided in an embodiment of the present invention;

[0038] Figure 3 This is a clustering result diagram of the accelerated fuzzy c-means clustering algorithm in one embodiment of the present invention;

[0039] Figure 4 This is a membership result diagram of the accelerated fuzzy c-means clustering algorithm in one embodiment of the present invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] Please see Figure 1 An embodiment of the present invention provides a method for fault location of a metering master station based on accelerated fuzzy c-means clustering, comprising:

[0042] S10. Collect log data from the metering master station and retain complete fault characteristics of abnormal data to form a non-redundant abnormal dataset.

[0043] S11. Accelerate the clustering process of the fuzzy c-means clustering algorithm based on the iterative rules containing the learning rate in the KCN clustering network.

[0044] S12. The accelerated fuzzy c-means clustering algorithm is called to process the abnormal dataset, and the iteration continues until the cluster set of the abnormal dataset is less than a preset threshold or the number of iterations reaches a set upper limit, so as to obtain the cluster set and membership set after clustering.

[0045] S13. Based on the cluster set and the membership set, obtain the classification results where fault types and fault characteristics correspond to each other.

[0046] S14. Compare the classification results with the normal data in the log to locate the fault area.

[0047] This invention addresses the limitations of existing fault location technologies for automated metering master stations by developing a more accurate and efficient fault location method based on fuzzy c-means clustering and the Kohonen network algorithm.

[0048] Fuzzy c-means clustering (FCM clustering) is one of the main methods of unsupervised machine learning. It is based on c-means clustering and assigns a fuzzy membership degree to each cluster based on the difference between each data object and the others in similarity to a certain cluster, so that data with high similarity are classified into the same category.

[0049] First, define a complete dataset X = [x1, x2, ..., xn] containing n data points. n ] = [x i Let i = 1, 2, ..., n, and a cluster set V = [v1, v2, ..., vn] containing c categories. c ] = [v j ], j=1,2,...,c, a fuzzy membership matrix U=[μ ij ], μ ij It represents the membership degree from the i-th data point to the j-th cluster. The objective function of the algorithm is:

[0050]

[0051]

[0052] Where ||*|| is the Euclidean norm, m is a fuzzy parameter, usually taken as 2, and the algorithm obtains the optimal solutions for v and μ through iterative loops.

[0053] Furthermore, we can obtain:

[0054]

[0055]

[0056] The FCM algorithm will continue to iterate until the change in the clustered dataset model is less than a given value or the number of iterations reaches a set upper limit, i.e., ||V(t)-V(t-1)|| < ε or t > t. max The classification is completed in time.

[0057] For example, the process of accelerating the clustering of the fuzzy c-means clustering algorithm based on the iterative rules containing the learning rate in the KCN clustering network specifically includes:

[0058] The learning rate is defined based on the cluster set of the fuzzy convergence operator and the fuzzy c-means clustering algorithm;

[0059] The iteration rules of the KCN clustering network are updated according to the definition of the learning rate;

[0060] Replace the iterative rules of the fuzzy c-means clustering algorithm with the iterative rules of the KCN clustering network.

[0061] Kohonen clustering network (KCN) is an artificial neural network with self-organizing feature maps. It also uses the Euclidean norm to measure the affinity between data objects and the target cluster, but cleverly introduces a competition mechanism to accelerate clustering by changing the weights. In its output layer, it forms a mapping that can reflect the association between various features, effectively projecting the clustering features of data in high-dimensional space to low-dimensional space.

[0062] KCN introduces a new iteration rule:

[0063] v j,t =v j,t +α ij,t *(x i -v j,t-1 (4)

[0064] m t =m0-t*Δm (5)

[0065] Δm=(m0-1) / t max (6)

[0066] in m is the learning rate of the algorithm. t This is the fuzzy parameter at iteration t, which decreases with the number of iterations, and its initial value is m0 = 2; t max It represents the maximum number of iterations.

[0067] To accelerate the clustering process, the learning rate α is... ij,t Redefining:

[0068]

[0069] Where μ u and μ d These are the critical values, representing the upper cut set (0.5, 1) and the lower cut set (0, 0.5), respectively; m u and m d It is a fuzzy convergence operator, where m u ∈(0,1),m d >m t .

[0070] Substituting the learning rate into the KCN algorithm iteration process above, we get:

[0071]

[0072] Replacing formula (2) with formula (8) yields a clustering algorithm combining FCM and KCN.

[0073] For example, the objective function of the fuzzy c-means clustering algorithm is related to the difference between outlier data and each cluster, and the corresponding membership degree.

[0074] For example, the accelerated fuzzy c-means clustering algorithm specifically includes the following steps in its execution:

[0075] Initialize the cluster set and fuzzy parameters based on the abnormal dataset, and calculate the initial membership degree set;

[0076] The iteration is performed according to the learning rate-inclusive iteration rules in the KCN clustering network and the membership set iteration rules of the fuzzy c-means clustering algorithm until the convergence condition is met.

[0077] Fault localization in the metering automation master station system mainly utilizes the log data generated by the current system. Common faults include process abnormalities, resource usage abnormalities, and server load abnormalities. Therefore, a large amount of relevant data, including data traffic, TCP connection count, and server CPU utilization, needs to be collected for feature extraction. In addition, the statistical data of the metering automation master station is often based on prefecture-level cities. Even with block-based and partitioned diagnosis, the problem of massive data volume still exists. In this embodiment of the invention, fuzzy clustering is used to process the log data.

[0078] For classification and fault location using the accelerated fuzzy c-means clustering algorithm, please refer to [link to relevant documentation]. Figure 2First, log files from the automated metering master station system need to be collected. These logs contain data from both normal operation and malfunctions. Data preprocessing is performed on these logs, standardizing the format, removing data with missing features and data from abnormal collection, and eliminating duplicate and redundant data. The remaining logs are then compiled into an anomaly dataset. Log files typically contain attributes such as IP addresses, variable values, status change information, and process information. Fault-related features are extracted from the logs to prepare for subsequent clustering analysis. The processed dataset is then classified using a combination of FCM clustering and KCN, yielding classification results that correspond to fault types and fault characteristics (occurrence conditions). Comparing the classification results with normal data allows for accurate fault location.

[0079] Compared to existing technologies, this invention provides a method for fault location in metering master stations based on accelerated fuzzy c-means clustering. It extracts fault-related features from logs to prepare for subsequent clustering analysis. The processed dataset is then classified using a combination of FCM (fuzzy c-means clustering) and KCN (Knowledge-Based Clustering) to obtain classification results where fault types and fault features correspond. By comparing the results in the fault group with normal data, fault location can be accurately determined. Because the KCN iterative rules regarding learning rate are used to accelerate the original FCM clustering, the FCM clustering network achieves faster and more accurate iterative results.

[0080] One embodiment of this application provides a fault location process when a two-dimensional array containing two feature values ​​is used as an input abnormal dataset.

[0081] Since the algorithm accepts a two-dimensional array as input, only two feature values ​​are selected in the example below. The actual feature values ​​will be paired up as two columns of data for X and input into the clustering network.

[0082] 5191 log files were collected, and process data and server load data were selected as feature values. The feature values ​​were extracted from the log files and normalized to obtain an anomaly dataset X. The cluster set V and fuzzy parameter m were initialized, and clustering was iteratively performed using the algorithm combining FCM clustering and KCN proposed above. The clustering results are as follows: Figure 3 , Figure 4 As shown.

[0083] Figure 3 In the middle, the top right corner shows service process crashes with 153 records, the bottom right corner shows process abnormalities with 2878 records, the bottom left corner shows resource abnormalities with 477 records, and the top left corner shows server load abnormalities with 1683 records.

[0084] according to Figure 4The results show that the process and load fault classification results are relatively good. This is because the selected feature values ​​are more relevant. For faults with unclear results, other more relevant feature values ​​will be selected for clustering to obtain better classification results, thereby locating the fault.

[0085] One embodiment of this application provides a metering master station fault location device based on accelerated fuzzy c-means clustering, including: a data acquisition module, an acceleration module, an iteration module, a classification module, and a location module.

[0086] The data acquisition module is used to collect log data from the metering master station and retain complete fault characteristics of abnormal data to form a non-redundant abnormal dataset.

[0087] The acceleration module is used to accelerate the clustering process of the fuzzy c-means clustering algorithm based on the iterative rules containing the learning rate in the KCN clustering network.

[0088] The iteration module is used to call the accelerated fuzzy c-means clustering algorithm to process the abnormal dataset, and continue to iterate until the cluster set of the abnormal dataset is less than a preset threshold or the number of iterations reaches a set upper limit, so as to obtain the cluster set and membership set after clustering.

[0089] The classification module is used to obtain classification results that correspond to fault types and fault features based on the cluster set and the membership set.

[0090] The location module is used to compare the classification results with the normal data in the logs to locate the fault area.

[0091] For example, the acceleration module is specifically used for:

[0092] The learning rate is defined based on the cluster set of the fuzzy convergence operator and the fuzzy c-means clustering algorithm;

[0093] The iteration rules of the KCN clustering network are updated according to the definition of the learning rate;

[0094] Replace the iterative rules of the fuzzy c-means clustering algorithm with the iterative rules of the KCN clustering network.

[0095] For example, the objective function of the fuzzy c-means clustering algorithm is related to the difference between outlier data and each cluster, and the corresponding membership degree.

[0096] For example, the accelerated fuzzy c-means clustering algorithm specifically includes the following steps in its execution:

[0097] Initialize the cluster set and fuzzy parameters based on the abnormal dataset, and calculate the initial membership degree set;

[0098] The iteration is performed according to the learning rate-inclusive iteration rules in the KCN clustering network and the membership set iteration rules of the fuzzy c-means clustering algorithm until the convergence condition is met.

[0099] For example, the fault types include process abnormalities, resource usage abnormalities, and server load abnormalities, and the fault characteristics include data traffic, TCP connection count, and server CPU utilization.

[0100] Compared to existing technologies, this invention provides a metering master station fault location device based on accelerated fuzzy c-means clustering. It extracts fault-related features from logs to prepare for subsequent clustering analysis. The processed dataset is then classified using a combination of FCM (fuzzy c-means clustering) and KCN (Knowledge-Based Clustering) to obtain classification results where fault types and fault features correspond. By comparing the results in the fault group with normal data, fault location can be accurately determined. Because the KCN iterative rules regarding learning rate are used to accelerate the original FCM clustering, the FCM clustering network achieves faster and more accurate iterative results.

[0101] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the positioning device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0102] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for fault location of metering master stations based on accelerated fuzzy c-means clustering, characterized in that, include: Collect log data from the metering master station and retain complete fault characteristics of abnormal data to form a non-redundant abnormal dataset; The learning rate is defined based on the cluster set of the fuzzy convergence operator and the fuzzy c-means clustering algorithm; The iteration rules of the KCN clustering network are updated according to the definition of the learning rate; The iteration rules of the KCN clustering network are replaced with the iteration rules of the fuzzy c-means clustering algorithm to accelerate the clustering process of the fuzzy c-means clustering algorithm. The accelerated fuzzy c-means clustering algorithm is called to process the abnormal dataset, and the iteration continues until the cluster set of the abnormal dataset is less than a preset threshold or the number of iterations reaches a set upper limit, so as to obtain the cluster set and membership set after clustering. Based on the cluster set and the membership set, a classification result is obtained in which the fault type and fault feature correspond to each other; The classification results are compared with the normal data in the logs to locate the fault area.

2. The metering master station fault location method based on accelerated fuzzy c-means clustering as described in claim 1, characterized in that, The objective function of the fuzzy c-means clustering algorithm is related to the difference between outlier data and each cluster, and the corresponding membership degree.

3. The metering master station fault location method based on accelerated fuzzy c-means clustering as described in claim 1, characterized in that, The accelerated fuzzy c-means clustering algorithm specifically includes the following steps in its execution: Initialize the cluster set and fuzzy parameters based on the abnormal dataset, and calculate the initial membership degree set; The iteration is performed according to the learning rate-inclusive iteration rules in the KCN clustering network and the membership set iteration rules of the fuzzy c-means clustering algorithm until the convergence condition is met.

4. The metering master station fault location method based on accelerated fuzzy c-means clustering as described in claim 1, characterized in that, The fault types include process abnormalities, resource usage abnormalities, and server load abnormalities, and the fault characteristics include data traffic, TCP connection count, and server CPU utilization.

5. A metering master station fault location device based on accelerated fuzzy c-means clustering, characterized in that, include: The data acquisition module is used to collect log data from the metering master station and retain complete abnormal data with fault characteristics to form a non-redundant abnormal dataset. An acceleration module is used to define the learning rate based on the cluster set of the fuzzy convergence operator and the fuzzy c-means clustering algorithm; update the iteration rules of the KCN clustering network according to the definition of the learning rate; and replace the iteration rules of the fuzzy c-means clustering algorithm with the iteration rules of the KCN clustering network. The iteration module is used to call the accelerated fuzzy c-means clustering algorithm to process the abnormal dataset, and continue to iterate until the cluster set of the abnormal dataset is less than a preset threshold or the number of iterations reaches a set upper limit, so as to obtain the cluster set and membership set after clustering. The classification module is used to obtain classification results that correspond to fault types and fault features based on the cluster set and the membership set; The location module is used to compare the classification results with the normal data in the logs to locate the fault area.

6. The metering master station fault location device based on accelerated fuzzy c-means clustering as described in claim 5, characterized in that, The objective function of the fuzzy c-means clustering algorithm is related to the difference between outlier data and each cluster, and the corresponding membership degree.

7. The metering master station fault location device based on accelerated fuzzy c-means clustering as described in claim 5, characterized in that, The accelerated fuzzy c-means clustering algorithm specifically includes the following steps in its execution: Initialize the cluster set and fuzzy parameters based on the abnormal dataset, and calculate the initial membership degree set; The iteration is performed according to the learning rate-inclusive iteration rules in the KCN clustering network and the membership set iteration rules of the fuzzy c-means clustering algorithm until the convergence condition is met.

8. The metering master station fault location device based on accelerated fuzzy c-means clustering as described in claim 5, characterized in that, The fault types include process abnormalities, resource usage abnormalities, and server load abnormalities. The fault characteristics include data traffic, TCP connection count, and server CPU utilization.

Images