A trajectory periodic pattern mining method based on k-means

Through the k-means-based trajectory periodic mode mining method, the traditional method has solved the problem of high requirements for trajectory data attributes and difficulty in adjusting the cluster number K, and achieved rapid and effective trajectory periodic law mining and analysis, which is suitable for aircraft and ships trajectory data of moving targets.

CN113902058BActive Publication Date: 2024-11-08THE 54TH RESEARCH INSTITUTE OF CHINA ELECTRONICS TECHNOLOGY GROUP CORPORATION
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111262119.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-10-28
Publication Date
2024-11-08
Estimated Expiration
2041-10-28

AI Technical Summary

Technical Problem

The traditional trajectory mining method has high requirements for the original trajectory data attributes, and requires experts to manually add labels. It is impossible to effectively mine the trajectory periodic pattern of mobile targets such as aircraft and ships, and it is impossible to automatically adjust the cluster number K, resulting in the inability to meet the growing trajectory data. need.

Method used

Using k-means-based trajectory periodic mode mining method, the trajectory data is obtained and preprocessed, and the physical center of mass is calculated and clustered using the k-means algorithm, and the center of mass is dynamically adjusted to avoid artificially specifying the cluster number K and exploring the data. Regional relationships, generate periodic law information.

Benefits of technology

It realizes unsupervised clustering, fast convergence, excellent clustering effect, high efficiency, easy implementation, easy debugging, easy expansion and strong compatibility, and can effectively explore and analyze trajectory cycle laws.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113902058B_ABST
    Figure CN113902058B_ABST
Patent Text Reader

Abstract

The present invention relates to a trajectory periodic pattern mining method based on k-means. For mobile target trajectory data, the method considers deep mining of the movement pattern of the mobile target trajectory, capturing the periodic pattern of the target trajectory movement, and considering the dynamic non-fixedness factor of the data point contribution. The method adaptively characterizes the current trajectory sequence, extracts deep-level serialized feature representation through trajectory clustering, and realizes the matching of the current trajectory pattern. The method can better mine the target movement pattern based on the target trajectory data, and has the advantages of high interpretability of the result in line with practical significance, fast model convergence, good flexibility, etc.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a k-means-based trajectory periodic pattern mining method, which can be used for trajectory processing of mobile targets such as airplanes and ships, and is particularly suitable for mining periodic regular patterns of activity trajectories of mobile targets such as airplanes and ships. Background Art

[0002] The advancement of current position acquisition technology has generated a large amount of aircraft and ship trajectory data, which contains very rich information. Trajectory data mining refers to obtaining the potential motion information of the target object based on the historical trajectory of the object. By calculating the trajectory data of moving targets such as aircraft and ships, it is helpful to analyze their activities and predict the future activities of the target. Traditional trajectory mining methods have high requirements for the attributes of the original trajectory data, or require experts to manually add labels, etc., and cannot perform more comprehensive mining. Therefore, there is an urgent need for an adaptive trajectory mining method that can meet the growing trajectory needs. Summary of the invention

[0003] The technical problem to be solved by the present invention is to avoid the deficiencies in the above-mentioned background technology and provide a trajectory periodic pattern mining method based on k-means. The method proposed in the present invention can make up for the deficiencies of the label-driven classification method, obtain the periodic regularity information by exploring the regional relationship between the data, which is more in line with the actual physical meaning of the trajectory, avoids the problem of artificially specifying the number of clusters K through rules such as periodic matching and new class generation, and enables the trajectory periodicity to be better mined and analyzed. As an unsupervised clustering method, it has the characteristics of fast convergence speed, excellent clustering effect, and strong interpretability.

[0004] The technical problem to be solved by the present invention is achieved by the following technical solutions:

[0005] A trajectory periodic pattern mining method based on k-means includes the following steps:

[0006] (1) Obtaining all trajectory data of the current target and performing preprocessing; the preprocessing includes removing outliers and converting all trajectory data into a point sequence with unified physical meaning;

[0007] (2) Clustering and matching each segment of the trajectory data of the current target based on the physical centroid, specifically: using the k-means algorithm, first, without specifying the k value, by calculating the distance between the centroid of the current trajectory to be matched and the centroid of the existing clusters, determine whether to add the current trajectory to be matched to the existing cluster, or directly create a new cluster to represent a new target area activity trajectory cluster; and dynamically adjust the centroid of the cluster with trajectory updates until all trajectories are clustered;

[0008] (3) For all target area activity trajectory clusters, first determine the number of clusters and the trajectories they contain, determine the effectiveness of data mining, and then record and output the area visit cycles of the trajectories in the valid clusters.

[0009] Wherein, step (1) specifically comprises the following steps:

[0010] (101) extracting target trajectory data from the original data;

[0011] (102) cleaning the data in the target trajectory data that exceeds the legal range of the target trajectory, wherein the legal range of the target trajectory is determined according to the upper and lower bounds of the position range of the target in the target data set;

[0012] (103) Converting the target’s trajectory data into a point sequence with a unified physical meaning;

[0013] (104) Encapsulate the target’s trajectory data.

[0014] Wherein, step (2) specifically comprises the following steps:

[0015] (201) Calculating the centroid of all trajectories of the current target, wherein the centroid calculation process is as follows: using the GPS longitude and latitude calculation formula to convert the longitude and latitude information of the point series into Euclidean distance, and calculating and solving the centroid of the point series based on the Euclidean distance;

[0016] (202) The k-means algorithm is used to cluster the trajectories. The clustering criterion is the distance between the centroids, that is, the centroid distance between the current trajectory to be matched and the existing cluster is calculated. When the distance is higher than the specified threshold, the current trajectory to be matched is clustered into a new cluster, and the centroid of the current trajectory to be matched is the centroid of the new cluster. When the distance is lower than the specified threshold, the current trajectory to be matched is added to the currently matched cluster, and the centroid of the currently matched cluster is updated, so that the centroid of the cluster as a whole represents the area represented by all the trajectories in the cluster until all the trajectories are clustered.

[0017] Compared with the background technology, the present invention has the following advantages:

[0018] 1. The algorithm used in the present invention has high computational efficiency;

[0019] 2. The present invention is easy to implement and debug;

[0020] 3. The present invention has the characteristics of easy expansion, strong compatibility and flexible parameter adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1A schematic diagram of the workflow of a trajectory periodic pattern mining method based on k-means provided in the embodiment of the present invention;

[0022] Figure 2 This is the running result of the present invention on the test data. The diamond and the square represent the center of the area of ​​two different activity cycles. DETAILED DESCRIPTION

[0023] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0024] like Figure 1 As shown, this embodiment provides a trajectory periodic law mining method based on k-means, comprising the following steps:

[0025] (1) Obtaining all trajectory data of the target and preprocessing the trajectory data of the target, wherein the preprocessing step includes removing outliers and converting the trajectory data into a point sequence with a unified physical meaning, and then encapsulating the trajectory; specifically:

[0026] (101) extracting target trajectory data from the original data;

[0027] (102) cleaning the data in the target trajectory data that exceeds the legal range of the target trajectory, wherein the legal range of the target trajectory is determined according to the upper and lower bounds of the position range of the target in the target data set;

[0028] (103) Converting the target’s trajectory data into a point sequence with a unified physical meaning;

[0029] (104) Encapsulate the target’s trajectory data.

[0030] (2) Clustering and matching of each segment of the current target’s trajectory data based on the physical centroid is performed as follows:

[0031] (201) clustering and initializing the trajectory data of the target. The clustering initialization process is as follows: first, the centroid of all trajectories in the target is calculated. The centroid calculation process is as follows: the latitude and longitude calculation formula of the Global Positioning System (GPS) is used to convert the latitude and longitude information of the original point series into Euclidean distance, and the centroid of the point series is calculated and solved based on the Euclidean distance and encapsulated into the trajectory encapsulation data structure in step 1. The number of clusters will be determined by the model after comparison, so there is no need to manually set it to reduce errors and improve usability and flexibility;

[0032] (202) The k-means algorithm is used to cluster the trajectories. The clustering standard is the distance between the centroids, that is, the centroid distance between the current trajectory to be matched and the existing cluster is calculated. When the distance is higher than the specified threshold, the current trajectory to be matched is clustered into a new cluster, and the centroid of the current trajectory to be matched is the centroid of the new cluster. When the distance is lower than the specified threshold, the current trajectory to be matched is added to the currently matched cluster, and the centroid of the currently matched cluster is updated, so that the centroid of the cluster as a whole represents the area represented by all the trajectories in the cluster until all the trajectories are clustered. Using the clustering method based on physical distance for trajectory planning can make the division results have good practical significance and interpretability, and the adjustable threshold brings more flexibility and usage to the algorithm.

[0033] (3) For the target area activity trajectory clusters generated by k-means clustering, the number of clusters and the trajectories they contain is first determined to determine the effectiveness of data mining, and the visit area cycles of the trajectories in the valid clusters are recorded and output.

[0034] Complete trajectory periodic pattern mining based on k-means.

[0035] like Figure 2 As shown, the running result of a trajectory periodic law mining method based on k-means provided in this embodiment can be expressed as:

[0036] (1) Figure 2 Contains two activity trajectory clusters of an active object and the center of its activity area. The center of mass of the dotted trajectory cluster is a diamond coordinate point; the center of mass of the solid trajectory cluster is a square coordinate point.

[0037] (2) According to the output calculation results, the periodic law of the object activity contained in the graph is described as follows:

[0038] (201) The center of the first activity area is (124.402176, 33.190666), which is the diamond point in the figure. The activity pattern in this area is: Thursday -> (1 day interval) -> Friday -> (5 days interval) -> Wednesday -> (5 days interval) -> Monday -> (2 days interval) -> Wednesday -> (1 day interval) -> Thursday -> (1 day interval) -> Friday. The specific explanation is: the activity of the object near the center of this area has a periodic pattern. It is active near the center of the area on Thursday, and there is an activity track on Friday after 1 day, on Wednesday after 5 days, there is an activity track on Monday after 5 days, there is an activity track on Wednesday after 2 days, there is an activity track on Thursday after 1 day, and there is an activity track on Friday after 1 day;

[0039] (202) The center of the second activity area is (124.759136, 27.426964), which is the square point in the figure. The activity pattern in this area is: Wednesday -> (7 days interval) -> Wednesday -> (3 days interval) -> Saturday -> (3 days interval) -> Tuesday -> (10 days interval) -> Friday. The specific explanation is: the activity of the object near the center of this area has a periodic pattern. It is active near the center of the area on Wednesday, and there is an activity track on Wednesday after 7 days, on Saturday after 3 days, there is an activity track on Tuesday after 3 days, and there is an activity track on Friday after 10 days.

Claims

1. A trajectory periodic pattern mining method based on k-means, characterized by , including the following steps: (1) Obtaining all trajectory data of the current target and performing preprocessing; the preprocessing includes removing outliers and converting all trajectory data into a point sequence with unified physical meaning; (2) Clustering and matching each segment of the current target’s trajectory data based on the physical centroid is performed. Specifically, the k-means algorithm is used. First, without specifying the k value, the distance between the centroid of the current trajectory to be matched and the centroid of the existing clusters is calculated to determine whether the current trajectory to be matched is added to the existing cluster, or a new cluster is directly created to represent a new target area activity trajectory cluster; And dynamically adjust the centroid of the clusters with trajectory updates until all trajectories are clustered; (3) For all target area activity trajectory clusters, first determine the number of clusters and the trajectories they contain, determine the effectiveness of data mining, and record and output the area visit cycles of the trajectories in the valid clusters; Wherein, step (2) specifically includes the following steps: (201) Calculating the centroid of all trajectories of the current target, wherein the centroid calculation process is: using the GPS longitude and latitude calculation formula to convert the longitude and latitude information of the point series into Euclidean distance, and calculating and solving the centroid of the point series based on the Euclidean distance; (202) The k-means algorithm is used to cluster the trajectories. The clustering criterion is the distance between the centroids, that is, the centroid distance between the current trajectory to be matched and the existing cluster is calculated. When the distance is higher than the specified threshold, the current trajectory to be matched is clustered into a new cluster, and the centroid of the current trajectory to be matched is the centroid of the new cluster. When the distance is lower than the specified threshold, the current trajectory to be matched is added to the currently matched cluster, and the centroid of the currently matched cluster is updated so that the centroid of the cluster as a whole represents the area represented by all the trajectories in the cluster until all the trajectories are clustered.

2. The k-means-based trajectory periodic pattern mining method according to claim 1, characterized in that: Step (1) specifically includes the following steps: (101) Extracting target trajectory data from the original data; (102) cleaning the target trajectory data that exceeds the legal range of the target trajectory, wherein the legal range of the target trajectory is determined based on the upper and lower bounds of the target position range in the target data set; (103) Convert the target’s trajectory data into a point sequence with a unified physical meaning; (104) Encapsulate the target’s trajectory data.