A visual extraction method for high-incidence areas of contact network defects
The adaptive K-Means algorithm is used to perform spatiotemporal analysis of contact network defects, solving the problem of inconvenient contact network defect data statistics, achieving accurate identification and risk assessment of high-incidence defect areas, and supporting efficient maintenance of the contact network.
Patent Information
- Application Number
- CN202210637982.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-08
- Publication Date
- 2025-09-16
- Estimated Expiration
- 2042-06-08
AI Technical Summary
In the existing technology, it is inconvenient to collect statistics on contact network defects, and the method for identifying high-incidence areas of defects is not convenient and efficient enough, making it difficult to provide intuitive warning information for railway and subway operations.
The adaptive K-Means algorithm is used to mark the spatiotemporal information of the contact network defect records. By calculating the total silhouette coefficient and Euclidean distance iteration, the high-incidence areas of defects are finally output. The risk is evaluated by combining the defect level and spatial density, and a maintenance plan is formulated.
It achieves accurate extraction of high-incidence areas of contact network defects, provides effective maintenance plan support, optimizes maintenance resource deployment, and ensures safe operation of the contact network with high accuracy and timeliness.
Smart Images
Figure CN115204587B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of catenary maintenance, and in particular to a method for visually extracting areas with high incidence of catenary defects. Background Art
[0002] The rapid development of railways and subways has made travel more convenient. However, with the dramatic increase in the scale of electrified railways, subsequent operational and maintenance conflicts have become increasingly prominent. Electrified railway contact networks operate in complex conditions, influenced by a variety of factors, resulting in increasingly diverse manifestations of defects and faults. Currently, contact network defect data is limited to statistical reporting. In summary, a convenient, efficient, and reliable method for identifying high-incidence areas of contact network defects is proposed. This method intuitively provides maintenance personnel at railway and subway operating companies with warning information about these areas, which is of great significance for improving the quality of railway and subway operations. Summary of the Invention
[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for visually extracting high-incidence areas of contact network defects, which solves the problem of inconvenient statistical extraction methods for contact network defect data.
[0004] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is: a method for visually extracting areas with high incidence of contact network defects, comprising the following steps:
[0005] S1. Mark the time and space information of the contact network defect records;
[0006] S2. Adaptively cluster the marked catenary defect data, and calculate the total silhouette coefficient for each of the clustered catenary defect data. When the total silhouette coefficient is the largest, output the high-incidence area of defects. Otherwise, proceed to step S3.
[0007] S3. Increase the number of cluster center points and return to step S2 for iteration.
[0008] Furthermore, the specific steps of step S1 are: according to the time and kilometer mark of occurrence of the contact network defect, the contact network defect records in the past X months are plotted one by one on a two-dimensional plane with the kilometer mark as the horizontal coordinate and the occurrence time as the vertical coordinate.
[0009] Furthermore, the specific steps of clustering in step S2 are:
[0010] S21, divide all defect records into K clusters, namely {C1, C2, ..., C K}, calculate the center point m of each cluster i , indicating the spatial and temporal locations where catenary defects are most likely to occur;
[0011] S22. Calculate the Euclidean distance E between the defect record and the cluster center point;
[0012] S23. Perform algorithm iteration to minimize the Euclidean distance.
[0013] Furthermore: the center point m of each cluster i The calculation formula is:
[0014]
[0015] In the above formula, x is the defect record in the cluster, C i is the i-th cluster.
[0016] Furthermore, the Euclidean distance E between the defect record and the cluster center is calculated as follows:
[0017]
[0018] In the above formula, k is the number of defect records.
[0019] Furthermore, the specific steps of the algorithm iteration in step S23 are:
[0020] S231, from the time-space distribution point set D of the defect record = {x1, x2, ..., x m} randomly select k defect records to calculate the initial k centroids {m1,m2,...,m k}, m is the number of defect records;
[0021] S322, cluster division result C={C1, C2, ..., C K Each cluster of} is initialized to an empty set C t ,
[0022]
[0023] S323. For i=1, 2, ..., m, calculate each defect record x i With each centroid m j The distance between (j=1,2,...,k) x i Marked as minimum d ij The corresponding centroid category l i , and the cluster partition result C Updated to
[0024] S324. For j=1, 2, ..., k, C j All defect records in the cluster are recalculated into new cluster center points m j ;
[0025] S325: If no cluster center point changes in two consecutive iterations, proceed to step S326; otherwise, when the number of iterations is equal to the maximum number of iterations N, the iteration ends;
[0026] S326, output the updated cluster division result C={C1, C2, ..., C k} and the temporal and spatial extents covered by each cluster.
[0027] Furthermore, the output defect high-incidence area is specifically: for a predetermined range of k, the updated cluster division result C={C1, C2, ..., C k The total silhouette coefficient of each clustering result of k value in} is calculated respectively, and the clustering result with the largest total silhouette coefficient is selected as the output of the high-incidence area of defects in time and space.
[0028] Furthermore, the total silhouette coefficient is the average of the silhouette coefficients of k defect records, and the calculation formula of the silhouette coefficient S(i) is:
[0029]
[0030] In the above formula, a(i) is the average distance between the defect record and other defect records in the same cluster on the distribution plane, and b(i) is the average distance between the defect record and other defect records in other clusters on the distribution plane.
[0031] Furthermore, the method further includes step S4, evaluating the risk of each high-defect area and formulating a corresponding maintenance plan to reduce the defect and failure risk in the area.
[0032] Furthermore, the risk calculation formula for the defect-prone area is:
[0033]
[0034] In the above formula, Risk j is the risk of the jth defect-prone spatiotemporal region, n is the number of defects in the region, k i is the grade coefficient of each defect record, and F is the median of the distance from the data points in each cluster to the corresponding cluster center.
[0035] The beneficial effects of the present invention are:
[0036] 1. Innovative method for extracting high-incidence areas of contact network defects
[0037] This paper applies the adaptive K-Means algorithm to catenary defect analysis for the first time, accurately extracting high-risk defect areas without requiring complex processing of defect records. This provides effective theoretical support for the development of catenary maintenance plans, optimizing maintenance resource deployment and ensuring safe catenary operation.
[0038] 2. Accuracy and timeliness
[0039] This method fully utilizes the temporal, spatial, and level information of catenary defects, and adaptively iterates to effectively avoid errors caused by human intervention. It adapts to the business scenarios of catenary operation and maintenance, and has high reliability. Furthermore, its computational speed allows it to mine 100,000 defect records in minutes, demonstrating excellent timeliness. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 Flowchart of the present invention;
[0041] Figure 2 Schematic diagram of contact network defect distribution in an embodiment of the present invention;
[0042] Figure 3 Schematic diagram of the defect high-incidence time prediction result in an embodiment of the present invention. DETAILED DESCRIPTION
[0043] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.
[0044] like Figure 1 As shown, a visual extraction method for high-incidence areas of contact network defects includes the following steps:
[0045] S1. Mark the time and space information of the contact network defect records;
[0046] The defect data collected on the actual site of the contact network can be regarded as a series of strings with time and space labels. According to the time and kilometer mark of the contact network defect, the contact network defect records in the past 9 months are plotted one by one on a two-dimensional plane with kilometer mark and time as the coordinate axis. That is, each contact network defect record can be represented by a two-dimensional coordinate, with the horizontal axis being the kilometer mark of the defect and the vertical axis being the time of defect discovery. The contact network defect distribution is as follows: Figure 2 shown.
[0047] S2. Adaptively cluster the marked catenary defect data, and calculate the total silhouette coefficient for each of the clustered catenary defect data. When the total silhouette coefficient is the largest, output the high-incidence area of defects. Otherwise, proceed to step S3.
[0048] Using an adaptive K-Means clustering method, defects that occurred at similar times and locations were grouped into the same cluster. The center of each cluster represents the high incidence of contact network defects at that time and spatial location.
[0049] The specific steps of clustering are:
[0050] S21, divide all defect records into K clusters, namely {C1, C2, ..., C K}, calculate the center point m of each cluster i , indicating the spatial and temporal locations where catenary defects are most likely to occur;
[0051] The center point m of each cluster i The calculation formula is:
[0052]
[0053] In the above formula, x is the defect record in the cluster, C i is the i-th cluster.
[0054] S22. Calculate the Euclidean distance E between the defect record and the cluster center point;
[0055] The calculation formula of the Euclidean distance E between the defect record and the cluster center point is:
[0056]
[0057] In the above formula, k is the number of defect records.
[0058] S23. Perform algorithm iteration to minimize the Euclidean distance.
[0059] The specific steps of algorithm iteration are:
[0060] S231, from the time-space distribution point set D of the defect record = {x1, x2, ..., x m} randomly select k defect records to calculate the initial k centroids {m1,m2,...,m k}, m is the number of defect records;
[0061] S322, cluster division result C={C1, C2, ..., C K Each cluster of} is initialized to an empty set C t ,
[0062]
[0063] S323. For i=1, 2, ..., m, calculate each defect record x i With each centroid m j The distance between (j=1,2,...,k) x i Marked as minimum d ij The corresponding centroid category l i , and the cluster partition result C Updated to
[0064] S324. For j=1, 2, ..., k, C j All defect records in the cluster are recalculated into new cluster center points m j ; Cluster center m j The calculation formula is:
[0065]
[0066] In the above formula, x is the defect record in the cluster, C j is the jth cluster.
[0067] S325: If no cluster center point changes in two consecutive iterations, proceed to step S326; otherwise, when the number of iterations is equal to the maximum number of iterations N, the iteration ends;
[0068] S326, output the updated cluster division result C={C1, C2, ..., C k} and the temporal and spatial extents covered by each cluster.
[0069] The traditional KMeans clustering algorithm requires a predefined number of cluster centers, K, which significantly impacts the algorithm's results. We propose an adaptive K value selection strategy based on the silhouette coefficient. This algorithm reduces human intervention and improves the effectiveness of mining results in high-incidence spatial and temporal regions.
[0070] The silhouette coefficient combines the two factors of cohesion and separation. It can be used to evaluate the impact of different algorithms or different algorithm operation modes on clustering results based on the same original data.
[0071] For each defect record, the silhouette coefficient is calculated as follows:
[0072]
[0073] In the above formula, a(i) is the average distance between the defect record and other defect records in the same cluster on the distribution plane, and b(i) is the average distance between the defect record and other defect records in other clusters on the distribution plane.
[0074] The average of the silhouette coefficients of all defect records is the total silhouette coefficient of this clustering result. Combined with the overhead network business, the range of k can be pre-given as [2,20]. The updated cluster division result C={C1,C2,...,C k The total silhouette coefficient of each k value clustering result in} is calculated respectively, and the clustering result with the largest total silhouette coefficient is selected as the output of the defect spatiotemporal high-incidence area, such as Figure 3 shown.
[0075] S3. Increase the number of cluster center points and return to step S2 for iteration.
[0076] Based on the above clustering results, the risk of each high-defect spatiotemporal region can be further evaluated. Taking into account the point density of the cluster and the defect level information within the cluster, the risk of each high-defect spatiotemporal region is shown as follows:
[0077]
[0078] In the above formula, Risk j is the risk of the jth defect-prone spatiotemporal region, n is the number of defects in the region, k i is the grade coefficient of each defect record, and F is the median of the distance from the data points in each cluster to the corresponding cluster center.
[0079] Based on the risk level of each defect-prone time and space area, a corresponding maintenance plan is formulated to reduce the risk of defects and failures in that area, thereby improving the quality of contact network operations.
[0080] 1. Innovative method for extracting high-incidence areas of contact network defects
[0081] This paper applies the adaptive K-Means algorithm to catenary defect analysis for the first time, accurately extracting high-risk defect areas without requiring complex processing of defect records. This provides effective theoretical support for the development of catenary maintenance plans, optimizing maintenance resource deployment and ensuring safe catenary operation.
[0082] 2. Accuracy and timeliness of the plan
[0083] This method fully utilizes the temporal, spatial, and level information of catenary defects, and adaptively iterates to effectively avoid errors caused by human intervention. It adapts to the business scenarios of catenary operation and maintenance, and has high reliability. Furthermore, its computational speed allows it to mine 100,000 defect records in minutes, demonstrating excellent timeliness.
Claims
1. A visual extraction method for high-incidence areas of contact network defects, characterized in that: The following steps are involved: S1. Mark the time and space information of the contact network defect records; S2. Adaptively cluster the marked catenary defect data, and calculate the total silhouette coefficient for each of the clustered catenary defect data. When the total silhouette coefficient is the largest, output the high-incidence area of defects. Otherwise, proceed to step S3. S3, increase the number of cluster center points and return to step S2 for iteration; The specific steps of clustering in S2 are: S21. Divide all defect records into K clusters, , calculate the center point of each cluster , indicating the spatial and temporal locations where catenary defects are most likely to occur; S22. Calculate the Euclidean distance E between the defect record and the cluster center point; S23, performing algorithm iteration to minimize the Euclidean distance; The center point of each cluster The calculation formula is: In the above formula, x is the defect record in the cluster, For the i clusters; The calculation formula of the Euclidean distance E between the defect record and the cluster center point is: In the above formula, k The number of defect records; The specific steps of the algorithm iteration in step S23 are: S231, from the time and space distribution point set of defect records Random selection k Defect records are calculated initially k centroid , m is the number of defect records; S322, cluster division results Each cluster is initialized to an empty set , ; S323, for , calculate each defect record With each centroid The distance between ,Will Mark as minimum The category of the corresponding centroid , and the cluster partition result C Updated to ; S324, for ,Will Recalculate new cluster centers for all defect records in ; S325: If no cluster center point changes in two consecutive iterations, proceed to step S326; otherwise, when the number of iterations is equal to the maximum number of iterations N, the iteration ends; S326: Output updated cluster division results The temporal and spatial extents covered by each cluster.
2. The visual extraction method for high-incidence areas of contact network defects according to claim 1 is characterized in that: The specific steps of step S1 are: according to the time and kilometer mark of the occurrence of the contact network defect, the contact network defect records in the past X months are plotted one by one on a two-dimensional plane with the kilometer mark as the horizontal coordinate and the occurrence time as the vertical coordinate.
3. The visual extraction method of high-incidence areas of contact network defects according to claim 1 is characterized in that: The output defect high-incidence area is specifically: for the pre-given k The updated cluster division results Each of k The total silhouette coefficient of the clustering results of the values is calculated respectively, and the clustering result with the largest total silhouette coefficient is selected as the output of the high-incidence area of defects in time and space.
4. The visual extraction method of high-incidence areas of contact network defects according to claim 3 is characterized in that: The total silhouette coefficient is k The mean of the silhouette coefficients of the defect records, the silhouette coefficient The calculation formula is: In the above formula, is the average distance between the defect record and other defect records in the same cluster on the distribution plane, is the average distance between the defect record and other defect records in other clusters on the distribution plane.
5. The visual extraction method of high-incidence areas of contact network defects according to claim 1 is characterized in that: The method further includes step S4, evaluating the risk of each high-defect area and formulating a corresponding maintenance plan to reduce the defect and failure risk in the area.
6. The visual extraction method of high-incidence areas of contact network defects according to claim 5 is characterized in that: The risk calculation formula for the defect-prone area is: In the above formula, For the j The risk of a high-incidence spatial and temporal region of defects, n is the number of defects in the region, is the grade coefficient of each defect record, and F is the median of the distance from the data points in each cluster to the corresponding cluster center.