CN104615881A - User normal track analysis method based on movable position application - Google Patents

Abstract

The invention discloses a user normal track analysis method based on a movable position application. The user normal track analysis method based on the movable position application includes the following steps that firstly, a user track set P is input; secondly, an orderly convex polygonal track set Q comprising all Pn track points is obtained from the track set P; thirdly, the area of the convex polygon Q is calculated; fourthly, the cluster radius R and cluster density T are determined; fifthly, all core clusters are recorded; sixthly, the core clusters are combined; seventhly, recursion is executed until all the core clusters can not be combined, and then it is judged as convergence; eighthly, the points not contained in the core clusters are judged as noise points, the cluster radius R serves as the threshold value, and the noise points are gathered into a plurality of noise clusters through a proximity clustering principle; ninthly, intra-cluster track point election is carried out on the core clusters and the noise clusters, and it is guaranteed that one point is elected from one cluster.

Description

A kind of user's normality trajectory analysis method based on shift position application

Technical field

The present invention relates to Mobile data trajectory analysis field, be that one carries out mining analysis for User Activity track data, and then obtain the method for user's normality track.

Background technology

Along with the progress of science and technology, Mobile solution changes daily life dearly, for application developer, in order to provide better more humane service to user, usually need to carry out analysis mining to various user behavior data, and User Activity track data this wherein important one just.

At present, being directed to track data, to carry out the algorithm of mining analysis more, and clustering algorithm is also first-selected.But developer is mostly based on respective development requirement, all specific settings are done to the enter factor of clustering algorithm, caused the result applicability of clustering algorithm not to be very wide.

Summary of the invention

Goal of the invention: technical matters to be solved by this invention is for the deficiencies in the prior art, provides a kind of user's normality trajectory analysis method based on shift position application.

In order to solve the problems of the technologies described above, the invention discloses a kind of user's normality trajectory analysis method based on shift position application, comprising the following steps:

Step 1, input user trajectory set P, and object clustering cluster number K, track set P comprises the raw position data of tracing point;

Step 2, for Pn tracing point in track set P, obtains the convex polygon track set Q comprising whole Pn tracing point that in track set P one group is orderly;

Step 3, calculates the area of convex polygon Q;

Step 4, determines bunch radius R and bunch density T;

Step 5, with any one tracing point M in track set P for the center of circle, the some number comprised within bunch radius R is greater than T, then thinking that M is core point, take M as the center of circle, and institute's pointed set that R comprises for radius is combined into core bunch C; If the some number comprised within bunch radius R is less than a bunch density T, then think that M is general point; Points all in traversal P, records all cores bunch;

Step 6, suppose with M1 be core point formed core bunch C1 in comprise T1 tracing point, be comprise T2 tracing point in the core bunch C2 of core point formation with M2, if C1 and C2 two bunches of public tracks are counted, minimum tracing point in more than two bunches bunch counts 50%, then C1 and C2 is merged, form new bunch C12;

Step 7, recurrence performs step 7, until cannot merge between all cores bunch, then judges convergence;

Step 8, by not being included in the point in core bunch, is judged to be noise point; With bunch radius R for threshold value, take the principle of cluster nearby, noise point is gathered into several noises bunch;

Step 9, tracing point election in core bunch and noise bunch being carried out bunch, ensures that cluster elects a bit.

In the present invention, described raw position data comprises: user account, longitude, latitude and acquisition time.

In the present invention, in step 2, the algorithm of convex hull ConvexHull that increases income is used to obtain the orderly convex polygon track set Q comprising whole Pn tracing point of in track set P one group.

In the present invention, in step 3, computing method are: convex polygon Q is divided into several triangles, calculate triangle area and are added.

In the present invention, in step 9, longitude and latitude weighted mean is carried out for all tracing points in each bunch, generate bunch virtual center point, then the distance of this virtual center point is arrived a little in compute cluster, normality tracing point after election distance smallest point merges as this bunch, elects the normality tracing point after all bunches of merging and exports.

The present invention adopts BA-DBScan (Base area DBscan) algorithm when doing mining analysis, according to existing subscriber's track data, the method of dynamic calculation bunch density and bunch radius is carried out by footprint, and then the extraction realized user's normality track, improve the applicability of arithmetic result greatly.

Accompanying drawing explanation

To do the present invention below in conjunction with the drawings and specific embodiments and further illustrate, above-mentioned and/or otherwise advantage of the present invention will become apparent.

Fig. 1 is invention overview flow chart.

Fig. 2 is user A initial trace figure.

Fig. 3 is user A scope of activities figure.

Fig. 4 is user A normality trajectory diagram.

Embodiment

The present invention, from existing subscriber's track data, excavates and extracts user's normality event trace.The User Activity track data collected with Mobile solution, for sample, adopts BA-DBScan as user trajectory Data Clustering Algorithm, then to bunch carrying out a bunch inner track point election after cluster success, is exported by the point elected as normality tracing point;

The input of algorithm: trace information set P, object clustering cluster number K;

The output of algorithm: normality track set;

Introduce concrete steps below:

1> carries out pre-service to raw position data, as inputting data after removing abnormal longitude and latitude data; Ensure to comprise the information such as user account, longitude, latitude, acquisition time in input data.

2>, again for the track set P of wherein user A, supposes to count as Pn containing track in P.Use the algorithm of convex hull ConvexHull that increases income to obtain the orderly convex polygon track set Q of in P one group, ensure that Pn tracing point is all comprised by Q;

3> calculates the area of convex polygon Q.Convex polygon is divided into several triangles, calculates triangle area and be added;

P is triangle semi-perimeter:

p＝(ab+bc+ac)/2；

Wherein a, b, c are an Atria summit respectively, and ab, bc, ac are the Atria limit length of side respectively, be converted into the mode that Mercator's coordinate calculates distance again can calculate the length of side by 2 longitudes and latitudes;

Triangle Sqabc areal calculation formula (Heron's formula) be made up of any 3 a, b, c:

$\mathrm{Sqabc}=\sqrt{p(p-\mathrm{ab})(p-\mathrm{bc})(p-\mathrm{ac})},$

The area Sqn computing formula of convex polygon area Q:

$\mathrm{Sqn}=\underset{n=3}{\overset{n}{\mathrm{\Σ}}}\mathrm{Sq}1(n-1)n,$

Wherein n refers to the number of vertex of convex polygon Q, and the area of Q equals n-2 triangle area cumulative sum;

4> determines bunch radius and bunch density of clustering algorithm:

Bunch radius:

$R=\sqrt{\mathrm{Sqn}/K/\mathrm{\Π}},$

Bunch density:

T＝Pn/K，

5> is with any point M in the track set P of user A for the center of circle, and the some number comprised within radius R is greater than T, then thinking that M is core point, take M as the center of circle, and institute's pointed set that R comprises for radius is combined into core bunch C; If the some number comprised within radius R is less than T, then think that R is general point.Points all in traversal P, records all cores bunch information;

If comprise T1 tracing point in the core bunch C1 that 6> take M1 as core point to be formed, be comprise T2 tracing point in the core bunch C2 of core point formation with M2, what simultaneously C1 and C2 two bunches of public tracks counted minimum tracing point in more than two bunches bunch counts 50%, then C1 and C2 is merged, form new bunch C12;

7> recurrence performs step 6, until all bunches all relatively independent, cannot merge, then think algorithm convergence success, the core bunch information after record merges;

8>, for the point be not included in core bunch, is referred to as noise point.Cluster principle is nearby: mark all noise points for " non-cluster ", calculate the distance of any two " non-cluster " noise points, if distance is less than a bunch radius R, then these two noise points being gathered into noise bunch, is " cluster " with these two noise points of tense marker; If distance is greater than a bunch radius R, then continue to calculate other " non-cluster " noise point distances, cycle calculations, until all " non-cluster " noise points are all greater than a bunch radius R with the distance of all " cluster " noise points, cycle calculations terminates.Remaining " non-cluster " noise point is independently become noise bunch separately, records all noises bunch information;

Tracing point election in the core produced above bunch and noise bunch are carried out bunch by 9>, ensures that cluster is selected a bit.Electoral machinery is as follows: carry out longitude and latitude weighted mean for all tracing points in each bunch, generate bunch virtual center point, then arrive a little the distance of this virtual center point in compute cluster, the normality tracing point after election distance smallest point merges as this bunch also exports.

By above several step, the normality track of user A can be obtained.

Embodiment 1

Fig. 1 is overview flow chart of the present invention, and concrete implementing procedure is described below:

1> prepares a collection of existing subscriber's track sample data, after cleaning, ensures must comprise the information such as user account, longitude, latitude, acquisition time in data;

Extract the initial trace data of user A in the sample data of 2> from step 1 as input trajectory data, the original input trajectory information of concrete user A as shown in Figure 2;

After 3> gets the initial trace figure of user A, utilizing the algorithm of convex hull ConvexHull that increases income to calculate the scope of activities of user A, as shown in Figure 3, is a convex pentagon surrounded by A.B.C.D.E coordinate points.

4> calculates the scope of activities area of user A, i.e. convex pentagon area.Can be three triangles by convex pentagon cutting.Wherein triangle length of side ab, bc, ac can be converted into Mercator's coordinate by two summit latitude and longitude coordinates and calculate.

Triangle semi-perimeter:

P＝(ab+bc+ac)/2

Triangle area:

$\mathrm{Sabc}=\sqrt{p(p-\mathrm{ab})(p-\mathrm{bc})(p-\mathrm{ac})}$

Convex pentagon area:

Sabcdef＝Sabc+Sacd+Sade

5> goes out the enter factor of BA-DBScan algorithm according to scope of activities areal calculation, i.e. bunch radius and bunch density;

User A initial trace data acquisition and object cluster number K input as algorithm by 4>, execution algorithm, until algorithm convergence, Output rusults bunch, and election in carrying out bunch result bunch, obtain final user A normality track, algorithm terminates.The normality trajectory diagram of user A, as Fig. 4, can find, the present invention remains the normality track of user A preferably.

The invention provides a kind of user's normality trajectory analysis method based on shift position application; the method and access of this technical scheme of specific implementation is a lot; the above is only the preferred embodiment of the present invention; should be understood that; for those skilled in the art; under the premise without departing from the principles of the invention, can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.The all available prior art of each ingredient not clear and definite in the present embodiment is realized.

Claims (5)

1., based on user's normality trajectory analysis method of shift position application, it is characterized in that, comprise the following steps:

Step 1, input user trajectory set P, and object clustering cluster number K, track set P comprises the raw position data of tracing point;

Step 2, for Pn tracing point in track set P, obtains the convex polygon track set Q comprising whole Pn tracing point that in track set P one group is orderly;

Step 3, calculates the area of convex polygon Q;

Step 4, determines bunch radius R and bunch density T;

Step 5, with any one tracing point M in track set P for the center of circle, the some number comprised within bunch radius R is greater than T, then thinking that M is core point, take M as the center of circle, and institute's pointed set that R comprises for radius is combined into core bunch C; If the some number comprised within bunch radius R is less than a bunch density T, then think that M is general point; Points all in traversal P, records all cores bunch;

Step 6, suppose with M1 be core point formed core bunch C1 in comprise T1 tracing point, be comprise T2 tracing point in the core bunch C2 of core point formation with M2, if C1 and C2 two bunches of public tracks are counted, the track of minimum tracing point in more than two bunches bunch is counted 50%, then C1 and C2 is merged, form new bunch C12;

Step 7, recurrence performs step 7, until cannot merge between all cores bunch, then judges convergence;

Step 8, by not being included in the point in core bunch, is judged to be noise point; With bunch radius R for threshold value, take the principle of cluster nearby, noise point is gathered into several noises bunch;

Step 9, tracing point election in core bunch and noise bunch being carried out bunch, ensures that cluster elects a bit.

2. a kind of user's normality trajectory analysis method based on shift position application according to claim 1, it is characterized in that, described raw position data comprises: user account, longitude, latitude and acquisition time.

3. a kind of user's normality trajectory analysis method based on shift position application according to claim 1, it is characterized in that, in step 2, the algorithm of convex hull ConvexHull that increases income is used to obtain the orderly convex polygon track set Q comprising whole Pn tracing point of in track set P one group.

4. a kind of user's normality trajectory analysis method based on shift position application according to claim 1, it is characterized in that, in step 3, computing method are: convex polygon Q is divided into several triangles, calculate triangle area and are added.

5. a kind of user's normality trajectory analysis method based on shift position application according to claim 1, it is characterized in that, in step 9, longitude and latitude weighted mean is carried out for all tracing points in each bunch, generate bunch virtual center point, then arrive a little the distance of this virtual center point in compute cluster, the normality tracing point after election distance smallest point merges as this bunch, elects the normality tracing point after all bunches of merging and exports.