Power distribution network fault rapid positioning method and system
By employing triple undersampling, residual term secondary decomposition, and dynamic similarity distance calculation, the problems of data imbalance and inaccurate understanding of interrelationships in distribution network fault location are solved, achieving more efficient and accurate fault location.
Patent Information
- Application Number
- CN202410524890.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-29
- Publication Date
- 2026-01-27
AI Technical Summary
In existing methods for fault location in power distribution networks, imbalanced raw data, excessive data volume, and severe noise data have led to poor generalization ability of the location model, low robustness of the location results, imprecise data segmentation, inability to extract more data features, and inaccurate understanding of the relationships between data points, resulting in low location accuracy.
We employ a triple undersampling process, residual term secondary decomposition, dynamic similarity distance calculation, and agglomerative hierarchical clustering. By injecting unlabeled data, selecting instances, and cleaning up neighborhoods, we increase data diversity. We also use intrinsic mode functions and dynamic similarity distance matrices to finely decompose the data, thereby improving the efficiency and accuracy of fault location.
It improves the efficiency and accuracy of fault location, enhances the diversity and representativeness of data, extracts more signal components and features, accurately describes the similarity between time series, and improves the location speed and accuracy.
Smart Images
Figure CN121410429A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power distribution network technology, specifically to a method and system for rapid fault location in power distribution networks. Background Technology
[0002] Fault location in distribution networks is a crucial part of power systems. It primarily refers to the precise location of fault points using various technical means when a fault occurs in the distribution network, enabling timely repair and restoration of grid operation. However, common fault location methods suffer from several drawbacks: imbalanced raw data, excessive data volume, and significant impact of noisy data on the location results, leading to poor generalization ability and low robustness of the location model; insufficient data segmentation preventing the extraction of more data features; and inaccurate understanding of the relationships between data points resulting in poor identification of latent patterns and cluster structures, ultimately leading to low location accuracy. Summary of the Invention
[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a method and system for rapid fault location in distribution networks. Addressing the problems of imbalanced raw data, excessive data volume, and severe influence of noisy data on location results in general distribution network fault location methods, leading to poor generalization ability and low robustness of the location model, this solution employs triple undersampling processing of the raw data based on injecting unlabeled data, instance selection, and neighborhood cleaning. This increases data diversity and representativeness, reduces data volume, better distinguishes minority class instances, and improves fault location efficiency. Furthermore, addressing the problem of insufficient data decomposition in general fault location methods, resulting in the inability to extract more data features, this solution performs a secondary decomposition of the residual term into intrinsic mode functions and residuals, enabling more precise data decomposition and extraction of more features. This approach identifies numerous signal components and characteristics; intrinsic mode functions (EMFs) represent signal variations at different frequencies, helping to identify important patterns and trends in the data, thus providing a better understanding of the data's intrinsic structure. Addressing the issue of inaccurate understanding of data point relationships in general fault location methods, leading to poor identification of latent patterns and clustering structures and ultimately low location accuracy, this solution employs a dynamic similarity distance calculation method to more accurately describe the similarity between two time series. Considering the relationships between each data point in the series, the dynamically calculated similarity distance matrix has higher information content and reliability. Hierarchical clustering using the dynamic similarity distance matrix fully utilizes the dynamic similarity information between data points. Fault location is based on dual labels, improving both location speed and accuracy.
[0004] The technical solution adopted by this invention is as follows: This invention provides a method for rapid fault location in a power distribution network, which includes the following steps:
[0005] Step S1: Data Acquisition;
[0006] Step S2: Triple undersampling processing;
[0007] Step S3: Secondary decomposition;
[0008] Step S4: Calculate the dynamic similarity distance matrix;
[0009] Step S5: Agglomerated hierarchical clustering.
[0010] Further, in step S1, the data acquisition involves collecting historical data and real-time data; both historical and real-time data include node current, node voltage, node power, grid frequency, node temperature, ambient humidity, and time data; the historical data also includes operational status data; the operational status data includes distribution network operational faults and distribution network normal operation; the real-time data also includes node location data; the operational status data is used as the historical data label; the node location data is used as the real-time data label; the label dimension does not participate in specific clustering operations, but is only used for fault location; the data is converted into vectors and standardized; historical and real-time data jointly participate in clustering.
[0011] Furthermore, in step S2, the triple undersampling process specifically includes the following steps:
[0012] Step S21: Inject unlabeled data. All injected unlabeled data is removed after clustering and then fed into the balancing method; specifically including:
[0013] Step S211: Inject unlabeled data into the training data and apply the K-Means clustering algorithm to the injected data;
[0014] Step S212: For each cluster, for each data point in the cluster, if the point belongs to the unlabeled dataset, remove the point from the cluster and perform random undersampling on the removed clusters.
[0015] Step S213: Merge the undersampled clusters into the updated training data one;
[0016] Step S22: Instance selection. After processing the training data, only instances close to the cluster centers and boundaries are used after clustering. Before performing sample balancing, some instances not close to the boundaries or cluster centers are removed from each cluster. Specifically, this includes:
[0017] Step S221: Apply the k-Means clustering algorithm to the dataset and divide it into k clusters;
[0018] Step S222: Traverse each cluster;
[0019] Step S2221: For each cluster C i Initialize dataset D i ; and add each data point in the cluster to dataset D. i middle;
[0020] Step S2222: Transfer dataset D i The data points in the dataset are sorted according to their distance from the centroid; and the sorted dataset D is then used to sort the data points. i Undersampling is performed on the first and last quarter of the data points.
[0021] Step S2223: Perform undersampling operation on the selected undersampled samples;
[0022] Step S223: Merge the undersampled datasets from each cluster to form the updated training data two;
[0023] Step S23: Neighborhood cleaning. This involves processing the training data and removing majority class instances within a distance threshold from each minority class instance's neighborhood. Specifically, this includes:
[0024] Step S231: Apply the k-Means clustering algorithm to the input dataset;
[0025] Step S232: Perform the following steps for each cluster:
[0026] Step S2321: Initialize the minority class instance set and the majority class instance set to empty sets;
[0027] Step S2322: Iterate through each data point in the cluster; if the data point is a minority class instance, add it to the minority class instance set; otherwise, add it to the majority class instance set.
[0028] Step S2323: Initialize the temporary dataset as a collection of majority class instances;
[0029] Step S2324: For each point p in the minority class instance set j and each point p in the majority class instance set m Iterate through; if p j and p m If the distance to the point is less than a set threshold, then remove point p from the temporary dataset. m ;
[0030] Step S2325: Merge the temporary dataset and the minority class instance set, and perform undersampling on the merged dataset;
[0031] Step S233: Merge the undersampled cluster datasets into the final training dataset.
[0032] Furthermore, in step S3, the secondary decomposition specifically includes the following steps:
[0033] Step S31: Initial decomposition of distribution network monitoring data Y v The components are decomposed into trend components, seasonal components, and residual components, as shown below:
[0034] ;
[0035] In the formula, T v It is the trend component, that is, the overall trend of data changes over time; S v It is a seasonal component, that is, a recurring periodic variation in the data; R v It is the residual component, that is, the part that cannot be explained by the trend and seasonal components;
[0036] Step S32: Calculate the sample entropy. A sample entropy threshold is preset. If the sample entropy of the residual component is higher than the sample entropy threshold, data decomposition is performed. The formula used is as follows:
[0037] ;
[0038] In the formula, SampEn(·) is the sample entropy, and m, r, and N represent the number of dimensions, similarity tolerance, and time series length, respectively; A m (r) and B m (r) is the matching probability of two time series data under a given m-dimensionality and r-similarity tolerance;
[0039] Step S33: Quadratic decomposition, decompose the residual term into intrinsic mode functions and residuals, using the following formula:
[0040] ;
[0041] Where x(t) is the residual component of the original distribution network data, IMF i (t) represents the k intrinsic mode functions from high frequency to low frequency, which are components obtained through signal decomposition. k represents the total number of IMF components, which determines the number of components that are decomposed in the signal decomposition process; res is the residual.
[0042] Further, in step S4, calculating the dynamic similarity distance matrix specifically includes the following steps:
[0043] Step S41: Define the dynamic similarity distance, which describes the similarity between two time series, denoted by A and B. The formula used is as follows:
[0044] ;
[0045] In the formula, D(i,j) represents the similarity between the first data point to the i-th data point in sequence A and the first data point to the j-th data point in sequence B; D(i-1,j) represents the similarity between the first data point to the (i-1)-th data point in sequence A and the first data point to the j-th data point in sequence B; D(i,j-1) represents the similarity between the first data point to the i-th data point in sequence A and the first data point to the (j-1)-th data point in sequence B; D(i-1,j-1) represents the similarity between the first data point to the (i-1)-th data point in sequence A and the first data point to the (j-1)-th data point in sequence B; d(A i B j ) is the Euclidean distance from the i-th data point in sequence A to the j-th data point in sequence B;
[0046] Step S42: Recursive calculation. The dynamic similarity distance elements are calculated recursively. First, the boundary conditions are initialized, and D(0,0)=d(A0,B0), D(i,0) and D(0,j) are initialized to positive infinity. For D(·) parameters less than 0, the default correction is 0. Second, the matrix elements {i=1,2,...,n} and {j=1,2,...,m} are traversed, and D(i,j) is calculated. Finally, the dynamic similarity distance matrix is obtained.
[0047] Furthermore, in step S5, the agglomerative hierarchical clustering specifically includes the following steps:
[0048] Step S51: Calculate the dynamic similarity distance matrix between each component in the quadratic decomposition;
[0049] Step S52: Hierarchical clustering. The dynamic similarity distance matrix is used as the similarity matrix of the power distribution network data for hierarchical clustering. Hierarchical clustering is a bottom-up clustering method. At the beginning, each time series is regarded as a separate cluster. Then, the closest pair of clusters are merged together. The merging operation is repeated until all clusters are merged into a single cluster containing all time series.
[0050] Step S53: Determine the number of clusters. After clustering is completed, use the silhouette coefficient to evaluate the clustering effect and determine the optimal number of clusters accordingly. Select the number of clusters with the largest silhouette coefficient as the final number of clusters. Finally, obtain the clustering results.
[0051] Step S54: Fault location; Select distribution network operation fault or distribution network operation normal as cluster labels based on the number of data of each operating status in the historical data labels; take the label of the real-time data of distribution network operation fault as the node location as the final fault location result.
[0052] The present invention provides a rapid fault location system for power distribution networks, comprising a data acquisition module, a triple undersampling processing module, a secondary decomposition module, a dynamic similarity distance matrix calculation module, and an agglomerative hierarchical clustering module;
[0053] The data acquisition module collects historical and real-time data and sends the data to the triple undersampling processing module;
[0054] The triple undersampling processing module performs triple undersampling processing on the original data based on injecting unlabeled data, instance selection, and neighborhood cleaning, and then sends the data to the secondary decomposition module.
[0055] The secondary decomposition module decomposes the residual term into intrinsic mode functions and residuals; and sends the data to the dynamic similarity distance matrix calculation module.
[0056] The dynamic similarity distance matrix accurately describes the similarity between two time series using the dynamic similarity distance calculation module, and sends the data to the agglomerative hierarchical clustering module;
[0057] The agglomerative hierarchical clustering module performs hierarchical clustering of distribution network data based on a dynamic similarity distance matrix and performs fault location of the distribution network based on dual labels.
[0058] The beneficial effects achieved by the present invention using the above solution are as follows:
[0059] (1) In view of the problems that the general distribution network fault location method has imbalanced original data, excessive data volume, and serious influence of noise data on the location results, resulting in poor generalization ability of the location model and low robustness of the location results, this scheme performs triple undersampling processing on the original data based on injecting unlabeled data, instance selection and neighborhood cleaning, so as to increase the diversity and representativeness of the data, reduce the data volume, better distinguish minority instances, and improve the fault location efficiency.
[0060] (2) In view of the problem that the general fault location method is not fine enough in data splitting, which makes it impossible to extract more data features, this scheme decomposes the residual term into intrinsic mode function and residual, which can decompose the data more finely and extract more signal components and features; the intrinsic mode function represents the change of the signal at different frequencies, which can help identify important patterns and trends in the data, thereby better understanding the internal structure of the data.
[0061] (3) In view of the problem that the general fault location method has inaccurate understanding of the relationship between data points, resulting in poor recognition of potential patterns and clustering structures in the data and low final location accuracy, this scheme adopts the dynamic similarity distance calculation method to more accurately describe the similarity between two time series. Considering the relationship between each data point in the series, the dynamic similarity distance matrix obtained by recursive calculation has higher information content and reliability. Using the dynamic similarity distance matrix for hierarchical clustering can make full use of the dynamic similarity information between data. Fault location is based on dual labels to improve the location speed and accuracy. Attached Figure Description
[0062] Figure 1 A flowchart illustrating a method for rapid fault location in a power distribution network provided by the present invention;
[0063] Figure 2 This is a schematic diagram of a rapid fault location system for a power distribution network provided by the present invention;
[0064] Figure 3 This is a flowchart illustrating step S3;
[0065] Figure 4 This is a flowchart of step S5.
[0066] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation
[0067] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0068] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.
[0069] Example 1, see Figure 1 The present invention provides a method for rapid fault location in a power distribution network, the method comprising the following steps:
[0070] Step S1: Data acquisition, collecting historical and real-time data;
[0071] Step S2: Triple undersampling processing, which involves injecting unlabeled data, instance selection, and neighborhood cleaning to perform triple undersampling processing on the original data;
[0072] Step S3: Secondary decomposition, decompose the residual term into intrinsic mode functions and residuals;
[0073] Step S4: Calculate the dynamic similarity distance matrix. The dynamic similarity distance calculation method is used to accurately describe the similarity between two time series.
[0074] Step S5: Agglomerated hierarchical clustering, hierarchical clustering of distribution network data based on dynamic similarity distance matrix, and fault location of distribution network based on dual labels.
[0075] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, both the historical data and the real-time data include node current, node voltage, node power, grid frequency, node temperature, ambient humidity, and time data. The historical data also includes operating status data, which includes distribution network operation faults and distribution network operation normality. The real-time data also includes node location data. The operating status data is used as the historical data label. The node location data is used as the real-time data label. The label dimension does not participate in the specific clustering operation but is only used for fault location. The data is converted into vectors and standardized. The historical data and real-time data jointly participate in clustering.
[0076] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, the triple undersampling processing specifically includes the following steps:
[0077] Step S21: Inject unlabeled data. All injected unlabeled data is removed after clustering and then fed into the balancing method; specifically including:
[0078] Step S211: Inject unlabeled data into the training data and apply the K-Means clustering algorithm to the injected data;
[0079] Step S212: For each cluster, for each data point in the cluster, if the point belongs to the unlabeled dataset, remove the point from the cluster and perform random undersampling on the removed clusters.
[0080] Step S213: Merge the undersampled clusters into the updated training data one;
[0081] Step S22: Instance selection. After processing the training data, only instances close to the cluster centers and boundaries are used after clustering. Before performing sample balancing, some instances not close to the boundaries or cluster centers are removed from each cluster. Specifically, this includes:
[0082] Step S221: Apply the k-Means clustering algorithm to the dataset and divide it into k clusters;
[0083] Step S222: Traverse each cluster;
[0084] Step S2221: For each cluster C i Initialize dataset D i ; and add each data point in the cluster to dataset D. i middle;
[0085] Step S2222: Transfer dataset D i The data points in the dataset are sorted according to their distance from the centroid; and the sorted dataset D is then used to sort the data points. i Undersampling is performed on the first and last quarter of the data points.
[0086] Step S2223: Perform undersampling operation on the selected undersampled samples;
[0087] Step S223: Merge the undersampled datasets from each cluster to form the updated training data two;
[0088] Step S23: Neighborhood cleaning. This involves processing the training data and removing majority class instances within a distance threshold from each minority class instance's neighborhood. Specifically, this includes:
[0089] Step S231: Apply the k-Means clustering algorithm to the input dataset;
[0090] Step S232: Perform the following steps for each cluster:
[0091] Step S2321: Initialize the minority class instance set and the majority class instance set to empty sets;
[0092] Step S2322: Iterate through each data point in the cluster; if the data point is a minority class instance, add it to the minority class instance set; otherwise, add it to the majority class instance set.
[0093] Step S2323: Initialize the temporary dataset as a collection of majority class instances;
[0094] Step S2324: For each point p in the minority class instance set jand each point p in the majority class instance set m Iterate through; if p j and p m If the distance to the point is less than a set threshold, then remove point p from the temporary dataset. m ;
[0095] Step S2325: Merge the temporary dataset and the minority class instance set, and perform undersampling on the merged dataset;
[0096] Step S233: Merge the undersampled cluster datasets into the final training dataset.
[0097] By performing the above operations, this solution addresses the problems of unbalanced raw data, excessive data volume, and severe impact of noisy data on the location results in general power distribution network fault location methods, which lead to poor generalization ability and low robustness of the location model. This solution performs triple undersampling processing on the raw data based on injecting unlabeled data, instance selection, and neighborhood cleaning. This increases the diversity and representativeness of the data, reduces the data volume, better distinguishes minority instances, and improves fault location efficiency.
[0098] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S3, the secondary decomposition specifically includes the following steps:
[0099] Step S31: Initial decomposition of distribution network monitoring data Y v The components are decomposed into trend components, seasonal components, and residual components, as shown below:
[0100] ;
[0101] In the formula, T v It is the trend component, that is, the overall trend of data changes over time; S v It is a seasonal component, that is, a recurring periodic variation in the data; R v It is the residual component, that is, the part that cannot be explained by the trend and seasonal components;
[0102] Step S32: Calculate the sample entropy. A sample entropy threshold is preset. If the sample entropy of the residual component is higher than the sample entropy threshold, data decomposition is performed. The formula used is as follows:
[0103] ;
[0104] In the formula, SampEn(·) is the sample entropy, and m, r, and N represent the number of dimensions, similarity tolerance, and time series length, respectively; A m (r) and B m(r) is the matching probability of two time series data under a given m-dimensionality and r-similarity tolerance;
[0105] Step S33: Quadratic decomposition, decompose the residual term into intrinsic mode functions and residuals, using the following formula:
[0106] ;
[0107] Where x(t) is the residual component of the original distribution network data, IMF i (t) represents the k intrinsic mode functions from high frequency to low frequency, which are components obtained through signal decomposition. k represents the total number of IMF components, which determines the number of components that are decomposed in the signal decomposition process; res is the residual.
[0108] By performing the above operations, this solution addresses the problem of general fault location methods not being able to extract more data features due to insufficient data segmentation. It decomposes the residual term into intrinsic mode functions and residuals, which can more finely decompose the data and extract more signal components and features. The intrinsic mode function represents the changes of the signal at different frequencies, which can help identify important patterns and trends in the data, thereby better understanding the internal structure of the data.
[0109] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, calculating the dynamic similarity distance matrix specifically includes the following steps:
[0110] Step S41: Define the dynamic similarity distance, which describes the similarity between two time series, denoted by A and B. The formula used is as follows:
[0111] ;
[0112] In the formula, D(i,j) represents the similarity between the first data point to the i-th data point in sequence A and the first data point to the j-th data point in sequence B; D(i-1,j) represents the similarity between the first data point to the (i-1)-th data point in sequence A and the first data point to the j-th data point in sequence B; D(i,j-1) represents the similarity between the first data point to the i-th data point in sequence A and the first data point to the (j-1)-th data point in sequence B; D(i-1,j-1) represents the similarity between the first data point to the (i-1)-th data point in sequence A and the first data point to the (j-1)-th data point in sequence B; d(A i B j ) is the Euclidean distance from the i-th data point in sequence A to the j-th data point in sequence B;
[0113] Step S42: Recursive calculation. The dynamic similarity distance elements are calculated recursively. First, the boundary conditions are initialized, and D(0,0)=d(A0,B0), D(i,0) and D(0,j) are initialized to positive infinity. For D(·) parameters less than 0, the default correction is 0. Second, the matrix elements {i=1,2,...,n} and {j=1,2,...,m} are traversed, and D(i,j) is calculated. Finally, the dynamic similarity distance matrix is obtained.
[0114] Example 6, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S5, agglomerative hierarchical clustering specifically includes the following steps:
[0115] Step S51: Calculate the dynamic similarity distance matrix between each component in the quadratic decomposition;
[0116] Step S52: Hierarchical clustering. The dynamic similarity distance matrix is used as the similarity matrix of the power distribution network data for hierarchical clustering. Hierarchical clustering is a bottom-up clustering method. At the beginning, each time series is regarded as a separate cluster. Then, the closest pair of clusters are merged together. The merging operation is repeated until all clusters are merged into a single cluster containing all time series.
[0117] Step S53: Determine the number of clusters. After clustering is completed, use the silhouette coefficient to evaluate the clustering effect and determine the optimal number of clusters accordingly. Select the number of clusters with the largest silhouette coefficient as the final number of clusters. Finally, obtain the clustering results.
[0118] Step S54: Fault location; Select distribution network operation fault or distribution network operation normal as cluster labels based on the number of data of each operating status in the historical data labels; take the label of the real-time data of distribution network operation fault as the node location as the final fault location result.
[0119] By performing the above operations, this solution addresses the problem of inaccurate understanding of the relationships between data points in general fault location methods, leading to poor identification of potential patterns and clustering structures in the data and ultimately low location accuracy. Instead, it employs a dynamic similarity distance calculation method to more accurately describe the similarity between two time series. Considering the relationships between each data point in the series, the dynamic similarity distance matrix obtained through recursive calculation has higher information content and reliability. Hierarchical clustering using the dynamic similarity distance matrix fully utilizes the dynamic similarity information between data points. Fault location is based on dual labels, improving both location speed and accuracy.
[0120] Example 7, see Figure 2Based on the above embodiments, this embodiment provides a rapid fault location system for power distribution networks, including a data acquisition module, a triple undersampling processing module, a secondary decomposition module, a dynamic similarity distance matrix calculation module, and an agglomerative hierarchical clustering module.
[0121] The data acquisition module collects historical and real-time data and sends the data to the triple undersampling processing module;
[0122] The triple undersampling processing module performs triple undersampling processing on the original data based on injecting unlabeled data, instance selection, and neighborhood cleaning, and then sends the data to the secondary decomposition module.
[0123] The secondary decomposition module decomposes the residual term into intrinsic mode functions and residuals; and sends the data to the dynamic similarity distance matrix calculation module.
[0124] The dynamic similarity distance matrix accurately describes the similarity between two time series using the dynamic similarity distance calculation module, and sends the data to the agglomerative hierarchical clustering module;
[0125] The agglomerative hierarchical clustering module performs hierarchical clustering of distribution network data based on a dynamic similarity distance matrix and performs fault location of the distribution network based on dual labels.
[0126] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0127] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.
[0128] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.
Claims
1. A method for rapid fault location in a distribution network, characterized in that: The method includes the following steps: Step S1: Data Acquisition; Step S2: Triple undersampling processing, which involves injecting unlabeled data, instance selection, and neighborhood cleaning to perform triple undersampling processing on the original data; Step S3: Secondary decomposition, decompose the residual term into intrinsic mode functions and residuals; Step S4: Calculate the dynamic similarity distance matrix. The dynamic similarity distance calculation method is used to accurately describe the similarity between two time series. Step S5: Agglomerated hierarchical clustering, hierarchical clustering of distribution network data based on dynamic similarity distance matrix, and fault location of distribution network based on dual labels.
2. The method for rapid fault location in a distribution network according to claim 1, characterized in that: In step S2, the triple undersampling process specifically includes the following steps: Step S21: Inject unlabeled data. All injected unlabeled data is removed after clustering and then fed into the balancing method; specifically including: Step S211: Inject unlabeled data into the training data and apply the k-Means clustering algorithm to the injected data; Step S212: For each cluster, for each data point in the cluster, if the point belongs to the unlabeled dataset, remove the point from the cluster and perform random undersampling on the removed clusters. Step S213: Merge the undersampled clusters into the updated training data one; Step S22: Instance selection. After processing the training data, only instances close to the cluster centers and boundaries are used after clustering. Before performing sample balancing, some instances not close to the boundaries or cluster centers are removed from each cluster. Specifically, this includes: Step S221: Apply the k-Means clustering algorithm to the dataset and divide it into k clusters; Step S222: Traverse each cluster; Step S2221: For each cluster C i Initialize dataset D i ; and add each data point in the cluster to dataset D. i middle; Step S2222: Transfer dataset D i The data points in the dataset are sorted according to their distance from the centroid; and the sorted dataset D is then used to sort the data points. i Undersampling is performed on the first and last quarter of the data points. Step S2223: Perform undersampling operation on the selected undersampled samples; Step S223: Merge the undersampled datasets from each cluster to form the updated training data two; Step S23: Neighborhood cleaning. This involves processing the training data and removing majority class instances within a distance threshold from each minority class instance's neighborhood. Specifically, this includes: Step S231: Apply the k-Means clustering algorithm to the input dataset; Step S232: Perform the following steps for each cluster: Step S2321: Initialize the minority class instance set and the majority class instance set to empty sets; Step S2322: Iterate through each data point in the cluster; if the data point is a minority class instance, add it to the minority class instance set; otherwise, add it to the majority class instance set. Step S2323: Initialize the temporary dataset as a collection of majority class instances; Step S2324: For each point p in the minority class instance set j and each point p in the majority class instance set m Iterate through; if p j and p m If the distance to the point is less than a set threshold, then remove point p from the temporary dataset. m ; Step S2325: Merge the temporary dataset and the minority class instance set, and perform undersampling on the merged dataset; Step S233: Merge the undersampled cluster datasets into the final training dataset.
3. The method for rapid fault location in a distribution network according to claim 1, characterized in that: In step S4, calculating the dynamic similarity distance matrix specifically includes the following steps: Step S41: Define the dynamic similarity distance, which describes the similarity between two time series, denoted by A and B. The formula used is as follows: ； In the formula, D(i,j) represents the similarity between the first data point to the i-th data point in sequence A and the first data point to the j-th data point in sequence B; D(i-1,j) represents the similarity between the first data point to the (i-1)-th data point in sequence A and the first data point to the j-th data point in sequence B; D(i,j-1) represents the similarity between the first data point to the i-th data point in sequence A and the first data point to the (j-1)-th data point in sequence B; D(i-1,j-1) represents the similarity between the first data point to the (i-1)-th data point in sequence A and the first data point to the (j-1)-th data point in sequence B; d(A i B j ) is the Euclidean distance from the i-th data point in sequence A to the j-th data point in sequence B; Step S42: Recursive calculation. The dynamic similarity distance elements are calculated recursively. First, the boundary conditions are initialized, and D(0,0)=d(A0,B0), D(i,0) and D(0,j) are initialized to positive infinity. For D(·) parameters less than 0, the default correction is 0. Second, the matrix elements {i=1,2,...,n} and {j=1,2,...,m} are traversed, and D(i,j) is calculated. Finally, the dynamic similarity distance matrix is obtained.
4. The method for rapid fault location in a distribution network according to claim 1, characterized in that: In step S3, the secondary decomposition specifically includes the following steps: Step S31: Initial decomposition of distribution network monitoring data Y v The components are decomposed into trend components, seasonal components, and residual components, as shown below: ； In the formula, T v It is the trend component, that is, the overall trend of data changes over time; S v It is a seasonal component, that is, a recurring periodic variation in the data; R v It is the residual component, that is, the part that cannot be explained by the trend and seasonal components; Step S32: Calculate the sample entropy. A sample entropy threshold is preset. If the sample entropy of the residual component is higher than the sample entropy threshold, data decomposition is performed. The formula used is as follows: ； In the formula, SampEn(·) is the sample entropy, and m, r, and N represent the number of dimensions, similarity tolerance, and time series length, respectively; A m (r) and B m (r) is the matching probability of two time series data under a given m-dimensionality and r-similarity tolerance; Step S33: Quadratic decomposition, decompose the residual term into intrinsic mode functions and residuals, using the following formula: ； Where x(t) is the residual component of the original distribution network data, IMF i (t) represents the k intrinsic mode functions from high frequency to low frequency, which are components obtained through signal decomposition. k represents the total number of IMF components, which determines the number of components that are decomposed in the signal decomposition process; res is the residual.
5. The method for rapid fault location in a distribution network according to claim 1, characterized in that: In step S5, the agglomerative hierarchical clustering specifically includes the following steps: Step S51: Calculate the dynamic similarity distance matrix between each component in the quadratic decomposition; Step S52: Hierarchical clustering. The dynamic similarity distance matrix is used as the similarity matrix of the power distribution network data for hierarchical clustering. Hierarchical clustering is a bottom-up clustering method. At the beginning, each time series is regarded as a separate cluster. Then, the closest pair of clusters are merged together. The merging operation is repeated until all clusters are merged into a single cluster containing all time series. Step S53: Determine the number of clusters. After clustering is completed, use the silhouette coefficient to evaluate the clustering effect and determine the optimal number of clusters accordingly. Select the number of clusters with the largest silhouette coefficient as the final number of clusters. Finally, obtain the clustering results. Step S54: Fault location; Select distribution network operation fault or distribution network operation normal as cluster labels based on the number of data of each operating status in the historical data labels; take the label of the real-time data of distribution network operation fault as the node location as the final fault location result.
6. The method for rapid fault location in a distribution network according to claim 1, characterized in that: In step S1, the data acquisition involves collecting historical and real-time data. Both historical and real-time data include node current, node voltage, node power, grid frequency, node temperature, ambient humidity, and time data. The historical data also includes operational status data, encompassing both distribution network operational faults and normal operation. The real-time data further includes node location data. Operational status data is used as the label for historical data. Node location data is used as the label for real-time data. The label dimension does not participate in specific clustering operations but is only used for fault location. The data is converted into vectors and standardized. Historical and real-time data jointly participate in clustering.
7. A rapid fault location system for a distribution network, used to implement a rapid fault location method for a distribution network as described in any one of claims 1-6, characterized in that: It includes a data acquisition module, a triple undersampling processing module, a secondary decomposition module, a dynamic similarity distance matrix calculation module, and an agglomerative hierarchical clustering module; The data acquisition module collects historical and real-time data and sends the data to the triple undersampling processing module; The triple undersampling processing module performs triple undersampling processing on the original data based on injecting unlabeled data, instance selection, and neighborhood cleaning, and then sends the data to the secondary decomposition module. The secondary decomposition module decomposes the residual term into intrinsic mode functions and residuals; and sends the data to the dynamic similarity distance matrix calculation module. The dynamic similarity distance matrix accurately describes the similarity between two time series using the dynamic similarity distance calculation module, and sends the data to the agglomerative hierarchical clustering module; The agglomerative hierarchical clustering module performs hierarchical clustering of distribution network data based on a dynamic similarity distance matrix and performs fault location of the distribution network based on dual labels.