A method for user trajectory aggregation based on base station data

Abstract

This invention discloses a user trajectory aggregation method based on base station data. It includes a signaling acquisition module, a user location point generation module, a user trajectory formation module, a user path-based similarity measurement module, and a user trajectory clustering module. The signaling acquisition module collects base station signaling data. The user location point generation module combines the signaling data with the base station location to convert it into the user's time location. The user trajectory formation module forms trajectory points from the user's location points. The user path-based similarity measurement module obtains a similarity matrix based on path similarity. The user trajectory clustering module clusters the similarity matrix to obtain the user's travel trajectory. The beneficial effects of this invention are: by combining user signaling data from base stations, user location information is obtained; user trajectories are aggregated on a daily basis to form population travel patterns, thereby better supporting population classification and data value mining, and uncovering users' periodic activity patterns.

Description

Technical Field

[0001] This invention relates to the technical field of operator base station data mining, and in particular to a method for user trajectory aggregation based on base station data. Background Technology

[0002] With the widespread adoption of 4G and 5G technologies, the density of operator base stations has significantly increased, especially in urban areas where coverage can reach one base station within a 500-meter radius. Through heartbeat connections between mobile phones and multiple nearby base stations, operators can provide signaling data collected from these stations. Operators have a vast user base, and this signaling data contains various important information about people, such as their travel habits, consumption patterns, and behavioral characteristics. By collecting and analyzing this massive amount of signaling data, the diverse value of users can be further explored, enabling the provision of targeted services. This data can also be shared with governments for purposes such as location tracking of specific individuals and screening of transient populations. Currently, existing technologies do not provide specific analysis and mining of base station signaling data for further value analysis. Summary of the Invention

[0003] The present invention aims to overcome the aforementioned shortcomings in the prior art by providing a user trajectory aggregation method based on base station data that can mine user periodic activity patterns.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A user trajectory aggregation method based on base station data includes a signaling acquisition module, a user location point generation module, a user trajectory formation module, a user path-based similarity measurement module, and a user trajectory clustering module. The signaling acquisition module collects base station signaling data. The user location point generation module combines the signaling data with the base station location to convert it into the user's time location. The user trajectory formation module forms trajectory points from the user's location points. The user path-based similarity measurement module obtains a similarity matrix based on path similarity. The user trajectory clustering module clusters the similarity matrix to obtain the user's travel trajectory. The specific operation steps are as follows:

[0006] (1) The signaling acquisition module collects the signaling data of each base station of the operator. Every 5 minutes, the signaling data is aggregated to the provincial data processing center through the signaling acquisition module.

[0007] (2) The user location point generation module converts the signaling data into the user's real-time latitude and longitude points. By combining the user's signaling data at a certain moment with the associated base station location, the user's latitude and longitude location points are calculated using the base station positioning method.

[0008] (3) The user trajectory formation module generates the path feature space for each user on a daily basis;

[0009] (4) Based on the path sequence similarity measurement module, the similarity of the user's daily trajectory is calculated in pairs on a weekly basis, and finally a weekly similarity matrix of a user is formed.

[0010] (5) The user trajectory clustering module uses a hierarchical clustering algorithm based on matrix transformation to cluster the similarity matrix, selects the cluster with the most paths as the user's frequently used travel trajectories, and those below the threshold are the user's abnormal travel trajectories.

[0011] This invention utilizes signaling data to generate user location points and create user trajectories based on time sequence. For each user, a similarity measure based on path sequence is applied. Finally, based on the similarity measure results, a hierarchical clustering method using matrix transformation is used to aggregate each user's trajectories, thereby forming each user's travel patterns. By combining user signaling data from base stations, user location information is obtained; user trajectories are aggregated on a daily basis to form operator-based population travel patterns, thus better supporting population classification and data value mining. This invention realizes the analysis and calculation of daily user travel trajectories based on user location data (oidd), and aggregates daily user travel trajectories within a certain time period to mine user periodic activity patterns.

[0012] Preferably, in step (2), the base station positioning method calculates the user's latitude and longitude points based on the latitude and longitude of the associated base station and the signaling field strength. The specific steps are as follows:

[0013] (21) Divide the base station signaling data into buckets according to mobile phone numbers, and divide the signaling data of the same user into the same space;

[0014] (22) For all signaling data of each user, select all signaling data of this user at a certain moment to form a signaling set: T = {X t1 X t2 X t3 X t4 , ..., X tn}, where T represents the set of all signaling data for a user at a certain moment, and X ti This represents the voltage level of the user's connection to base station i at time t;

[0015] (23) Select the signaling of a certain station in the signaling set, and calculate the distance of this signaling point from the base station based on the propagation model algorithm, combined with the base station's transmit power, mounting height, antenna and field strength value in the signaling. Similarly, use this method to calculate the distance of all users from the base station where the signaling is located.

[0016] (24) Form a circle with the distance between the user and each base station as the radius, calculate the intersection point of these circles, use these intersection points to form a convex polygon, calculate the centroid of this polygon, and the position of this centroid is the latitude and longitude position of the user at this moment.

[0017] (25) Generate user location data based on operator base stations.

[0018] Preferably, in step (3), the user trajectory formation module processes the user location data to form a path feature space for each user. Each user is merged by day as the smallest unit, and the path feature space of trajectory aggregation is formed by a week. The specific processing flow is as follows:

[0019] (31) Sort the user's location points in chronological order by day:

[0020] Q = q1, q2, ..., q n

[0021] Where Q is the user's n-hop path, which is a path containing n nodes, representing the user's path on a certain day. q1, q2, ..., qn represent the latitude and longitude points that the user passed through on that day, and are arranged in ascending order of time.

[0022] (32) Select one week's worth of data for this user to construct the path feature space required for user trajectory aggregation:

[0023]

[0024] Where f(Q) is the feature space of the aggregation path Q, and r has a maximum value of 7, Q l Let be the set of all possible paths for a given day, i.e.:

[0025] Q l ={q l1 q l2 , ..., q ln ,|l=1,2,...,r。

[0026] Preferably, in step (4), the user calculates the similarity between paths taken on any two days of the week based on the path sequence similarity measurement module.

[0027]

[0028] Among them, Q i Q represents the user's path on day i. j Let sim(Q) represent the user's path on day j. i Q jThe similarity between any two paths is represented by the natural angle between them. i Q i > l It is Q i and Q i The inner product in the feature space, j Q j > l It is Q j and Q j The inner product in the feature space, i Q j > l It is Q i and Q j The inner product in the feature space is calculated as follows:

[0029]

[0030] Where q represents the latitude and longitude points at the same time each day between paths with similarity that need to be compared. T This is the transpose of the matrix.

[0031] As a preferred approach, the feature space of user trajectories over a week is compared pairwise, resulting in a 7x7 similarity matrix:

[0032]

[0033] Where simij represents the similarity between the paths on day i and day j.

[0034] Preferably, in step (5), the user trajectory clustering module uses a hierarchical clustering algorithm based on matrix transformation to diagonally divide the similarity matrix, obtain the optimal division, and obtain the user's travel trajectory by calculating the cohesion of each cluster after division and comparing it with the standard cohesion. The specific processing flow of the clustering algorithm is as follows:

[0035] (51) Perform row and column transformation on the similarity matrix. Select column k where the maximum value of the elements in columns 2 to 7 is located, swap columns 2 and k, and swap rows 2 and k. Then repeat the same data operation on columns 3 to 7 until all swaps are completed, forming the transformed similarity matrix A.

[0036] (52) Divide the transformed similarity matrix A into blocks along the diagonal, and find the partition point d of the cluster with the largest mean similarity within the cluster, where d is the row and column of the best partition.

[0037] (53) By dividing by d, the trajectory similarity matrix is ​​divided into four sub-matrices, labeled A and B respectively. 11 A 12 A​​​21 A 22 Calculate the sum of the elements of each submatrix:

[0038]

[0039] (54) To find the optimal d-partition, we define:

[0040] F d =M (d) (A 1] )*M (d) (A 22 )-M (d) (A 11 M (d) (A 21 )

[0041] When F d When the maximum value is reached, d is the optimal partition;

[0042] (55) After obtaining the optimal d partition, calculate the cluster cohesion T(A). ii The calculation method uses the mean intra-cluster similarity as a metric:

[0043]

[0044] T(A) ii The points in this matrix are compared with a given threshold. If the points are greater than the threshold, then the points in this matrix are the user's frequently used trajectories for the week.

[0045] (56) For the remaining diagonal matrix, continue to cluster according to steps (51) to (55) until it is less than the given threshold. The points less than the given threshold are used as the user's discrete trajectory for the analysis of the user's special preferences.

[0046] (57) After classifying and aggregating the trajectories of all users by week, the commonly used trajectories and abnormal trajectories of all users are obtained and saved as a location feature indicator of the operator's users.

[0047] Preferably, in step (54), in order to find the largest F d Perform the following procedures:

[0048] (541) When d = 1, select the values ​​in the first column and calculate M respectively. (1) (A11), M (1) (A22), M (1) (A12), M (1) (A21) Calculate the value of F1. Similarly, loop through the loop to calculate the values ​​of F2, F3, F4, F5, and F6 respectively.

[0049] (542) Compare all the values ​​of F and select the d corresponding to the largest F as the best partition.

[0050] The beneficial effects of this invention are: by combining user signaling data from base stations, user location information is obtained; user trajectories are aggregated on a daily basis to form travel patterns of the population based on operators, thereby better supporting population classification and data value mining, and thus mining users' periodic activity patterns. Attached Figure Description

[0051] Figure 1 This is a diagram of the method architecture of the present invention;

[0052] Figure 2 This is a flowchart illustrating the generation of the weekly trajectory similarity matrix for users in this embodiment of the invention;

[0053] Figure 3 This is a flowchart of a commonly used trajectory discovery method for user similarity matrices in embodiments of the present invention. Detailed Implementation

[0054] The present invention will now be further described with reference to the accompanying drawings and specific embodiments.

[0055] like Figure 1 In the aforementioned embodiment, a user trajectory aggregation method based on base station data includes a signaling acquisition module, a user location point generation module, a user trajectory formation module, a user path order similarity measurement module, and a user trajectory clustering module. The signaling acquisition module collects base station signaling data. The user location point generation module combines the signaling data with the base station location to convert it into the user's time location point. The user trajectory formation module forms trajectory points from the user's location points. The user path order similarity measurement module obtains a similarity matrix based on path similarity. The user trajectory clustering module clusters the similarity matrix to obtain the user's travel trajectory. The specific operation steps are as follows:

[0056] (1) The signaling acquisition module collects the signaling data of each base station of the operator. Every 5 minutes, the signaling data is aggregated to the provincial data processing center through the signaling acquisition module. The signaling acquisition module collects the signaling data of all base stations in the province and saves it to the provincial data processing center for further processing.

[0057] (2) The user location point generation module converts the signaling data into the user's real-time latitude and longitude points. By combining the user's signaling data at a certain moment with the location of the associated base station, the user's latitude and longitude location points are calculated using the base station positioning method. The base station positioning method calculates the user's latitude and longitude points at this moment based on the latitude and longitude of the associated base station and the signaling field strength. The specific steps are as follows:

[0058] (21) The base station signaling data is divided into buckets according to the mobile phone number, and the signaling data of the same user is divided into the same space; where, bucketing refers to distinguishing according to the mobile phone number, one number corresponds to one bucket, and the signaling data corresponding to this mobile phone will be placed into the space corresponding to this number;

[0059] (22) For all signaling data of each user, select all signaling data of this user at a certain moment to form a signaling set: T = {X t1 X t2 X t3 X t4 , ..., X tn}, where T represents the set of all signaling data for a user at a certain moment, and X ti This represents the voltage level of the user's connection to base station i at time t;

[0060] (23) Select the signaling of a certain station in the signaling set, and calculate the distance of this signaling point from the base station based on the propagation model algorithm, combined with the base station's transmit power, mounting height, antenna and field strength value in the signaling. Similarly, use this method to calculate the distance of all users from the base station where the signaling is located.

[0061] (24) Form a circle with the distance between the user and each base station as the radius, calculate the intersection point of these circles, use these intersection points to form a convex polygon, calculate the centroid of this polygon, and the position of this centroid is the latitude and longitude position of the user at this moment; the multi-base station location positioning method can make the positioning accuracy accurate to 50 meters.

[0062] (25) Generate user location data (oidd) based on operator base stations;

[0063] (3) The user trajectory formation module generates a path feature space for each user on a daily basis. The user trajectory formation module processes the user location data (oidd) to form a path feature space for each user. Since the working days generally cycle on a weekly basis, each user is merged with the daily data as the smallest unit, and the path feature space is formed by merging trajectories on a weekly basis. The specific processing flow is as follows:

[0064] (31) Sort the user's location points in chronological order by day:

[0065] Q = q1, q2, ..., q n

[0066] Where Q is the user's n-hop path, which is a path containing n nodes, i.e., one hop to one point, representing the user's path on a certain day. q1, q2, ..., qn represent the latitude and longitude points that the user passed through on that day, and are arranged in ascending order of time.

[0067] (32) Select one week's worth of data for this user to construct the path feature space required for user trajectory aggregation:

[0068]

[0069] Where f(Q) is the feature space of the aggregation path Q, and r has a maximum value of 7, Q l Let be the set of all possible paths for a given day, i.e.:

[0070] Q l ={q l1 q l2 ,...]q ln ,|l=1,2,...,r];

[0071] (4) Based on the path sequence similarity measurement module, the similarity of each user's daily trajectory is calculated pairwise on a weekly basis, ultimately forming a weekly similarity matrix for a given user. Specifically, the path sequence similarity measurement module calculates the similarity between the paths of any two days within a week.

[0072]

[0073] Among them, Q i Q represents the user's path on day i. j Let sim(Q) represent the user's path on day j. i Q j The similarity between any two paths is represented by the natural angle between them. i Q i > l It is Q i and Q i The inner product in the feature space, j Q j > l It is Q j and Q j The inner product in the feature space, i Q j > l It is Q i and Q j The inner product in the feature space is calculated as follows:

[0074]

[0075] Where q represents the latitude and longitude points at the same time each day between paths with similarity that need to be compared. T It is the transpose matrix;

[0076] ​​​After pairwise similarity comparisons of the user trajectory feature space over a week, the similarity matrix is ​​a 7x7 matrix:

[0077]

[0078] Where simij represents the similarity between the paths on day i and day j;

[0079] (5) The user trajectory clustering module uses a hierarchical clustering algorithm based on matrix transformation to cluster the similarity matrix. The cluster with the most paths is selected as the user's frequently used travel trajectories, while those below a threshold are considered abnormal travel trajectories. Specifically, the user trajectory clustering module uses a hierarchical clustering algorithm based on matrix transformation to diagonally divide the similarity matrix to obtain the optimal division. The cohesion of each cluster after division is calculated and compared with the standard cohesion to obtain the user's travel trajectory. The specific processing flow of the clustering algorithm is as follows:

[0080] (51) Perform row and column transformation on the similarity matrix. Select column k where the maximum value of the elements in columns 2 to 7 is located, swap columns 2 and k, and swap rows 2 and k. Then repeat the same data operation on columns 3 to 7 until all swaps are completed, forming the transformed similarity matrix A.

[0081]

[0082] (52) Divide the transformed similarity matrix A into blocks along the diagonal and find the partition point d of the cluster with the largest mean similarity within the cluster;

[0083] Convert to

[0084] Where d is the row and column of the optimal partition, since d must be located on the diagonal, the row of d is equal to the column number.

[0085] (53) By dividing by d, the trajectory similarity matrix is ​​divided into four sub-matrices, labeled A and B respectively. 11 A 12 A 21 A 22 Calculate the sum of the elements of each submatrix:

[0086]

[0087] Among them, u ij For the corresponding simij, since the original representation is inappropriate after forming the submatrix, the u of the new submatrix is ​​used. ij To represent, i.e., u ij This represents the similarity between the paths on day i and day j in the new submatrix;

[0088] (54) To find the optimal d-partition, we define:

[0089] F d =M (d) (A 11 )*M (d) (A 22 )-M (d) (A 11 M (d) (A 21 )

[0090] When F d When the maximum value is reached, d is the optimal partition; to find the maximum F d Perform the following procedures:

[0091] (541) When d = 1, select the values ​​in the first column and calculate M respectively. (1) (A11), M (1) (A22), M (1) (A12), M (1) (A21) is used to obtain the value of F1. Similarly, the values ​​of F2, F3, F4, F5, and F6 are obtained by looping. The similarity matrix A has a total of 7 rows and 7 columns, and the last one automatically becomes a submatrix.

[0092] (542) Compare all the values ​​of F and select the d corresponding to the largest F as the best partition;

[0093] (55) After obtaining the optimal d partition, calculate the cluster cohesion T(A). ii The calculation method uses the mean intra-cluster similarity as a metric:

[0094]

[0095] T(A) ii The similarity matrix A is compared with a given threshold. If the similarity matrix A is greater than the threshold, the points contained in this matrix are the user's frequently used trajectories this week. The similarity matrix A can only have two submatrices, 1 and 22, and n refers to the number of rows and columns of the submatrices (the number of rows and columns is equal).

[0096] (56) For the remaining diagonal matrix, continue to cluster according to steps (51) to (55) until it is less than the given threshold. The points less than the given threshold are used as the user's discrete trajectory for the analysis of the user's special preferences.

[0097] (57) After classifying and aggregating the trajectories of all users by week, the commonly used trajectories and abnormal trajectories of all users are obtained and saved as a location feature indicator of the operator's users.

[0098] like Figure 2 As shown, the generation of the weekly user trajectory similarity matrix based on base station data mainly involves comparing each user's weekly trajectory with their daily trajectories pairwise to form the user trajectory similarity matrix.

[0099] (1) Using user ID as the aggregation dimension, all location data of the same user are grouped into a set, and a similarity matrix is ​​formed by selecting the location data of the user for one week.

[0100] (2) Select a week's worth of user data. For each day's data, sort and connect the trajectory points according to their time order to form daily user trajectory points.

[0101] (3) For users' weekly trajectories<p1,p2,p3,p4,p5,p6,p7> Compare pairs of items and apply the similarity calculation formula.

[0102]

[0103] Calculate the similarity of the trajectories between each two days and save the results;

[0104] (4) Sort by weekdays to weekdays to form a similarity matrix for each user for a week.

[0105] 1 2 3 4 5 6 7 1 1 0.88 0 0.88 0 0 0 2 0.88 1 0 0.94 0 0 0 3 0 0 1 0 0.96 0 0 4 0.88 0.94 0 1 0 0 0 5 0 0 0.96 0 1 0 0 6 0 0 0 0 0 1 0.99 7 0 0 0 0 0 0.99 1

[0106] like Figure 3 As shown, the user similarity matrix commonly used trajectory discovery methods realize the clustering of users' weekly commonly used trajectories and the discovery of discrete trajectories.

[0107] (a) Transform the similarity matrix by row and column operations to make it compatible with subsequent clustering. The transformation process is as follows: Starting from the first row, find the column where the maximum value of columns 2 to 9 of the first row is located, swap this row with the second row, swap this column with the second column, and so on, until the second to last row is reached, finally forming the transformation matrix:

[0108]

[0109] After conversion:

[0110]

[0111] (b) The transformed matrix is ​​traversed, and the optimal d partition is found by calculating the intra-cluster similarity, resulting in two sub-matrices A11 and A22.

[0112] (c) Calculate the cohesion within clusters A11 and A22 and compare it with the cohesion threshold of 0.96;

[0113] (d) The intra-cluster cohesion of A11 is compared with the cohesion threshold. If the intra-cluster cohesion is greater than the threshold, it means that A11 is a cluster trajectory of the user.

[0114] (e) If A22 is less than the coagulation threshold, then it is further divided;

[0115] (f) After transforming the remaining matrix, divide it into two new submatrices A11 and A22 based on the optimal division d = 3.

[0116] (g) Calculate the intra-cluster cohesion of the re-divided A11 and A22;

[0117] (h)A11 has a cluster cohesion greater than the cohesion threshold, which is the clustering result;

[0118] (i) The cluster cohesion of A22 is greater than the cohesion threshold, which is the clustering result.

[0119] This concludes the detailed process of a user trajectory aggregation method based on base station data.

[0120] Through experimental calculations, the user trajectory aggregation method based on base station data of this invention realizes the extraction of commonly used trajectories based on the base station signaling data of operators, and achieves a quantitative characterization of users' activity habits in the time and space domains. This allows operators to easily grasp users' activity habits and greatly improves the use of user signaling data.

[0121] This invention utilizes signaling data to generate user location points and create user trajectories based on time sequence. For each user, a similarity measure based on path sequence is applied. Finally, based on the similarity measure results, a hierarchical clustering method using matrix transformation is used to aggregate each user's trajectories, thereby forming each user's travel patterns. By combining user signaling data from base stations, user location information is obtained; user trajectories are aggregated on a daily basis to form operator-based population travel patterns, thus better supporting population classification and data value mining. This invention realizes the analysis and calculation of daily user travel trajectories based on user location data (oidd), and aggregates daily user travel trajectories within a certain time period to mine user periodic activity patterns.

Claims

1. A method for user trajectory aggregation based on base station data, characterized in that, The system includes a signaling acquisition module, a user location point generation module, a user trajectory formation module, a user path-based similarity measurement module, and a user trajectory clustering module. The signaling acquisition module collects base station signaling data. The user location point generation module combines the signaling data with the base station location to convert it into the user's real-time location. The user trajectory formation module forms trajectory points from the user's location points. The user path-based similarity measurement module obtains a similarity matrix based on path similarity. The user trajectory clustering module clusters the similarity matrix to obtain the user's travel trajectory. The specific operation steps are as follows: (1) The signaling acquisition module collects the signaling data of each base station of the operator. Every 5 minutes, the signaling data is aggregated to the provincial data processing center through the signaling acquisition module. (2) The user location point generation module converts the signaling data into the user's real-time latitude and longitude points. By combining the user's signaling data at a certain moment with the associated base station location, the user's latitude and longitude location points are calculated using the base station positioning method. (3) The user trajectory formation module generates the path feature space for each user on a daily basis; (4) Based on the path sequence similarity measurement module, the similarity of the user's daily trajectory is calculated in pairs on a weekly basis, and finally a weekly similarity matrix of a user is formed. (5) The user trajectory clustering module uses a hierarchical clustering algorithm based on matrix transformation to cluster the similarity matrix, selects the cluster with the most paths as the user's frequently used travel trajectories, and those below the threshold are the user's abnormal travel trajectories.

2. The user trajectory aggregation method based on base station data according to claim 1, characterized in that, In step (2), the base station positioning method calculates the user's latitude and longitude based on the latitude and longitude of the associated base station and the signaling field strength. The specific steps are as follows: (21) Divide the base station signaling data into buckets according to mobile phone numbers, and divide the signaling data of the same user into the same space; (22) For all signaling data of each user, select all signaling data of this user at a certain moment to form a signaling set: T = {X t1 X t2 X t3 X t4 , ..., X tn }, where T represents the set of all signaling data for a user at a certain moment, and X ti This represents the voltage level of the user's connection to base station i at time t; (23) Select the signaling of a certain station in the signaling set, and calculate the distance of this signaling point from the base station based on the propagation model algorithm, combined with the base station's transmit power, mounting height, antenna and field strength value in the signaling. Similarly, use this method to calculate the distance of all users from the base station where the signaling is located. (24) Form a circle with the distance between the user and each base station as the radius, calculate the intersection point of these circles, use these intersection points to form a convex polygon, calculate the centroid of this polygon, and the position of this centroid is the latitude and longitude position of the user at this moment. (25) Generate user location data based on operator base stations.

3. The user trajectory aggregation method based on base station data according to claim 2, characterized in that, In step (3), the user trajectory formation module processes the user location data to form a path feature space for each user. Each user is merged by day as the smallest unit, and the path feature space of trajectory aggregation is formed by a week. The specific processing flow is as follows: (31) Sort the user's location points in chronological order by day: Q=q1,q2,...,q n Where Q is the user's n-hop path, which is a path containing n nodes, representing the user's path on a certain day. q1, q2, ..., qn represent the latitude and longitude points that the user passed through on that day, and are arranged in ascending order of time. (32) Select one week's worth of data for this user to construct the path feature space required for user trajectory aggregation: Where f(Q) is the feature space of the aggregation path Q, and r has a maximum value of 7, Q l Let be the set of all possible paths for a given day, i.e.: Q l ={q l1 ,q l2 ,...,q ln ,|l=1,2,...,r}。 4. A user trajectory aggregation method based on base station data according to claim 1, 2, or 3, characterized in that, In step (4), specifically, the user calculates the similarity between paths taken on any two days of the week based on the path order similarity measurement module: Among them, Q i Q represents the user's path on day i. j Let sim(Q) represent the user's path on day j. i Q j The similarity between any two paths is represented by the natural angle between them. i Q i > l It is Q i and Q i The inner product in the feature space, j Q j > l It is Q j and Q j The inner product in the feature space, i Q j > l It is Q i and Q j The inner product in the feature space is calculated as follows:​​​ Where q represents the latitude and longitude points at the same time each day between paths with similarity that need to be compared. T This is the transpose of the matrix.

5. The user trajectory aggregation method based on base station data according to claim 4, characterized in that, After pairwise similarity comparisons of the user trajectory feature space over a week, the similarity matrix is ​​a 7x7 matrix: Where simij represents the similarity between the paths on day i and day j.

6. The user trajectory aggregation method based on base station data according to claim 5, characterized in that, In step (5), specifically: the user trajectory clustering module uses a hierarchical clustering algorithm based on matrix transformation to diagonally divide the similarity matrix, obtain the optimal division, and obtain the user's travel trajectory by calculating the cohesion of each cluster after division and comparing it with the standard cohesion; the specific processing flow of the clustering algorithm is as follows: (51) Perform row and column transformation on the similarity matrix. Select column k where the maximum value of the elements in columns 2 to 7 is located, swap columns 2 and k, and swap rows 2 and k. Then repeat the same data operation on columns 3 to 7 until all swaps are completed, forming the transformed similarity matrix A. (52) Divide the transformed similarity matrix A into blocks along the diagonal and find the partition point d of the cluster with the largest mean similarity within the cluster; (53) By dividing by d, the trajectory similarity matrix is ​​divided into four sub-matrices, labeled A and B respectively. 11 A 12 A 21 A 22 Calculate the sum of the elements of each submatrix: (54) To find the optimal d-partition, we define: F d =M (d) (A 11 )*M (d) (A 22 )-M (d) (A 11 ),M (d) (A 21 ) When F d When the maximum value is reached, d is the optimal partition; (55) After obtaining the optimal d partition, calculate the cluster cohesion T(A). ii The calculation method uses the mean intra-cluster similarity as a metric: T(A) ii The points in this matrix are compared with a given threshold. If the points are greater than the threshold, then the points in this matrix are the user's frequently used trajectories for the week. (56) For the remaining diagonal matrix, continue to cluster according to steps (51) to (55) until it is less than the given threshold. The points less than the given threshold are used as the user's discrete trajectory for the analysis of the user's special preferences. (57) After classifying and aggregating the trajectories of all users by week, the commonly used trajectories and abnormal trajectories of all users are obtained and saved as a location feature indicator of the operator's users.

7. The user trajectory aggregation method based on base station data according to claim 6, characterized in that, In step (54), in order to find the largest F d Perform the following procedures: (541) When d = 1, select the values ​​in the first column and calculate M respectively. (1) (A11), M (1) (A22), M (1) (A12), M (1) (A21) Calculate the value of F1. Similarly, loop through the loop to calculate the values ​​of F2, F3, F4, F5, and F6 respectively. (542) Compare all the values ​​of F and select the d corresponding to the largest F as the best partition.

Images