CN109697438B

CN109697438B - Method and system for early detection of special group aggregation behaviors and prediction of aggregation places

Info

Publication number: CN109697438B
Application number: CN201910159609.6A
Authority: CN
Inventors: 高随祥; 王赛楠; 刘敏涛; 谭屯子; 姜志鹏; 桂继宏; 尚康禹; 杨文国
Original assignee: University of Chinese Academy of Sciences
Current assignee: University of Chinese Academy of Sciences
Priority date: 2018-03-08
Filing date: 2019-03-04
Publication date: 2020-10-27
Anticipated expiration: 2039-03-04
Also published as: CN109697438A; CN108416293A

Abstract

The invention belongs to the field of group behavior control, and discloses a method and a system for early detection of special group aggregation behaviors and prediction of aggregation places, wherein effective active members in a group are defined by using an aggregation behavior detection algorithm based on a moving average distance between the effective active members, and the interference of noise members on the detection of the aggregation behaviors is eliminated; for a special group with aggregation behaviors, detecting the aggregation tendency of the group in the early stage of group activity and giving out early warning; and screening out the potential aggregation members by using an aggregation prediction algorithm based on the least square fitting straight line of the movement locus of the potential aggregation members, and performing aggregation prediction. The method does not depend on a video monitoring system, only utilizes the historical movement track data of the group members to realize the rapid judgment of the special group gathering behavior, gives early warning to the group with the gathering behavior in time, and can accurately predict the gathering place of the group.

Description

Method and system for early detection of special group aggregation behaviors and prediction of aggregation places

Technical Field

The invention belongs to the field of group behavior control, and particularly relates to a method and a system for early detection of special group aggregation behaviors and prediction of aggregation places.

Background

Currently, the current state of the art commonly used in the industry is such that:the aggregation behavior of some special groups is likely to threaten public safety and even cause illegal events, so the early detection and prevention of the aggregation behavior of the special groups are particularly important. For some special groups, related departments need to pay close attention to the aggregation behavior of the groups, and once the aggregation tendency is found, precautionary measures should be taken in time. Because the monitoring period of a special population may be long, it is unrealistic to rely on manual monitoring and judgment, and the manual monitoring and judgment has the defects of untimely discovery of the gathering behavior and inaccurate prediction of the gathering place. There is a need to provide automated monitoring techniques to quickly and timely detect aggregation behavior of a particular population and accurately predict the location of the aggregation. The following are the existing research situations related thereto:

(1) and (3) detecting the crowd aggregation: based on a video monitoring system, according to the spatial distribution condition of a foreground image, measuring the density degree of people in a video scene by adopting a distribution entropy, and realizing the detection of people clustering behaviors in the video scene;

(2) and (3) hot spot area prediction: establishing a crowd aggregation threshold calculation model based on the mobile phone access amount data of the urban base station, and constructing a crowd density prediction model by using a Markov chain to realize prediction of a hot spot region;

(3) predicting in aggregate: for a special population with known aggregation behaviors, the position data of population members is utilized to calculate the centroid, and the position of the centroid is used as the aggregation center of the population.

In summary, the problems of the prior art are as follows:

(1) the method depends on a video monitoring system, and only the crowd gathering behavior judgment can be carried out on a specific place presented in a video image. For a special group, firstly, it is unlikely that a perfect video monitoring system can monitor the group all the time, and secondly, the gathering is not known in advance;

(2) the method comprises the steps that hot spot areas are predicted based on mobile phone access amount data of the urban base station, the crowd density degree of a target area can be predicted only in a macroscopic angle, and the position distribution condition, the moving behavior and the gathering behavior of members in a specific group cannot be analyzed and predicted;

(3) when the centroid method is used for solving the aggregation place, interference members (i.e. members not participating in aggregation) cannot be eliminated, and the predicted aggregation place is greatly influenced by factors such as initial positions, departure times, moving directions and moving speeds of the members participating in aggregation, so that the prediction accuracy is poor.

The difficulty and significance for solving the technical problems are as follows:

in practice, it is difficult to continuously monitor a special population by means of video monitoring, and a method independent of a video monitoring system needs to be provided to detect the aggregation behavior of the special population and predict the aggregation place. Under the condition of not depending on a video monitoring system, how to identify target special group members from a huge group of people is a difficult point. Even after the target special population is locked, in the case of aggregation, all members in the population are not always involved in aggregation, and since the aggregation place is not known in advance, the identification and exclusion of interfering members (members not involved in aggregation in the target special population) are difficult. And the interference members will greatly influence the detection of the population aggregation behavior and the prediction of the aggregation place. In addition, in the case of aggregation, the initial positions and the aggregation places of the members of the special population are relatively small in relation, the distances from the members to the aggregation places are different, the departure time, the moving speed and the arrival time are also different, and the possible aggregation of the aggregation places is a large-range continuous two-dimensional plane area (the height is not considered temporarily), so that the advance and accurate prediction of the aggregation places is particularly difficult.

The video monitoring mode is difficult to continuously monitor the special population in practice, and how to continuously identify the target special population members from the large-scale population becomes a difficult point. Even after the target special population is locked, in the case of aggregation, not all members in the population participate in aggregation, and since the aggregation place is not known in advance, the identification and exclusion of interfering members (members not participating in aggregation in the target special population) is difficult. And the interference members will greatly influence the detection of the population aggregation behavior and the prediction of the aggregation place. In addition, for the aggregation situation, in practice, the initial positions of the members of the special population are relatively small in association with the aggregation places, the distances from the members to the aggregation places are different, the departure time, the moving speed and the arrival time are also different, and the possible aggregation of the aggregation places is a large-range continuous two-dimensional plane area, so that the advance and accurate prediction of the aggregation places is particularly difficult, and the problem is difficult to solve only by means of video monitoring.

The invention provides a method and a system for early detection of special group aggregation behaviors and prediction of aggregation places, which are realized by utilizing movement track data of group members. The method overcomes the difficulties, can help related departments to quickly and timely detect the potential gathering behaviors in the special group and accurately predict gathering places, so that the related departments can take countermeasures in time to prevent the diseases in the future.

Disclosure of Invention

Aiming at the problems in the prior art, the invention provides a method and a system for early detection of special group aggregation behaviors and prediction of aggregation. The method is independent of a video monitoring system, only utilizes the historical movement track data of the group members to realize the rapid judgment of the special group gathering behavior, gives early warning to the group with the gathering behavior in time, can accurately predict the gathering place of the group, and assists relevant departments to take corresponding measures in time.

The invention is realized in such a way that a method for early detection and prediction of aggregation behaviors of a special population comprises the following steps:

defining effective active members in a group by using an aggregation behavior detection algorithm based on the moving average distance between the effective active members, and eliminating the interference of noise members on the aggregation behavior detection; for a special group with aggregation behaviors, detecting the aggregation tendency of the group in the early stage of group activity and giving out early warning;

and screening out the potential aggregation members by using an aggregation prediction algorithm based on the least square fitting straight line of the movement locus of the potential aggregation members, and performing aggregation prediction.

Further, the group aggregation information processing method based on the group member historical movement track data comprises the steps of firstly, acquiring real-time position data of a target group member by using a data acquisition unit, and continuously positioning the group member; an aggregation behavior detection algorithm is adopted, an aggregation tendency judgment unit is used for detecting the aggregation tendency of the group in the early stage of the group activity, and an alarm display unit sends out early warning; in the aggregation place prediction stage, potential aggregation members are screened out, and an aggregation place prediction algorithm is adopted to predict an aggregation place by using a prediction processing unit; and presented by the aggregated display unit.

Further, the method for early detection and prediction of aggregation of specific population includes:

A. firstly, carrying out aggregation behavior detection on a population; from the initial moment, finding out active members in the group;

B. if the number of the active members at the current moment does not reach alpha n, jumping to the next moment to continuously ask for the active members in the group; n represents the total number of members in the population, alpha n is the aggregation of the number of members which are supposed to concern the relevant departments only for alpha n and above in the population, and alpha is more than 0 and less than or equal to 1;

C. repeating the step B until the number of the active members at a certain moment is not less than alpha n, calculating the active members in the group at the moment, and jumping to the next moment to calculate the active members in the group if the number of the active members does not reach alpha n;

D. circularly executing the step B and the step C until the number of the effective active members at a certain moment is not less than alphan, and calculating the average distance between the effective active members at the moment;

E. when the average distance between the effective active members at a section of continuous time is obtained, calculating the sliding average distance between the effective active members corresponding to the current time; if the sliding average distance between the effective active members is continuously decreased for a plurality of times and the reduction rate reaches a given threshold value, an early warning is sent out;

F. next, performing aggregate prediction; from the early warning moment, solving a set of effective active members;

G. for each effective active member, solving a least square fitting straight line of a movement track of the active member until the current moment, and solving an intersection point between any two fitting straight lines;

H. calculating all forward intersection points at the current moment, screening effective forward intersection points, and calculating potential aggregation members at the current moment based on the effective forward intersection points;

I. establishing an unconstrained non-linear programming model [ P1]]Solving an optimal position at the current moment, wherein the sum of the distances from the optimal position to least square fitting straight lines corresponding to all potential aggregation members is minimum; introducing new non-negative variables to lead an unconstrained non-linear programming model [ P1]]Conversion to an equivalent constrained linear programming model [ P2](ii) a Solving a Linear programming model [ P2 ]]Obtaining the optimal position P of the current moment_t ^*；

J. After the optimal position of a section of continuous time is obtained, calculating the descending rate of the average distance between the potential aggregation members in a fixed long period; if P is within this fixed long period of time_t ^*The position does not vary beyond a given threshold and the rate of decrease of the average distance between the members of the cluster reaches a given threshold, a series P of the time interval is calculated_t ^*And outputting it as a center of convergence.

Further, in the step a, the definition of the positive members in the population at the time t is as follows:

for a certain member i (i ═ 1.., n) in the population, if the position at time t satisfies:

d(P_i ^t,P_i ¹)≥r

the member is said to be the active member at time t. Wherein P is_i ^tDenotes the position coordinate of the ith member at time t, d (P)_i ^t,P_i ¹) Represents P_i ^t,P_i ¹The distance between two locations, r is a distance threshold; the set of active members at time t is denoted as A_tUsing | A_tI represents the set A_tOf (1) elementCounting;

in the step C, the definition of the effective active members in the population at the time t is as follows:

let | A_t| ≧ 2, for some positive member i (i ∈ A)_t) If the position at the time t meets the following conditions:

i.e. if its average distance from other active members at time t reaches beta compared to the rate of decrease at the initial time₀The member is called as the effective active member at the moment t; wherein, beta₀< 1 represents a threshold for the average distance decrease rate between active members; the set of active members at time t is recorded as EA_t；

In the step D, the calculation formula of the average distance between the effective active members in the group at the time t is as follows:

wherein the content of the first and second substances,

is a combination number;

in the step E, the aggregation of the population is described by using the moving average distance between the effective active members, and the calculation formula of the moving average distance between the effective active members in the population at the time t is as follows:

wherein m is a positive integer representing the number of terms of the moving average;

in the step E, the moving average distance between the effective active members is l₀The calculation formula of the reduction rate in the next successive time is:

wherein l₀Is a positive integer and represents a threshold value of the number of times of the falling of the moving average distance;

in the step H, the method for screening out the effective forward intersection point includes:

by using

A least squares fit straight line representing the trajectory of the movement of the active positive member i before time t,

representing straight lines

And a straight line

The intersection point between them; for a certain

If it satisfies both of the following inequalities:

is the forward intersection point at the time t; where, represents the inner product of the vector; the set of forward intersections at time t is denoted as F_t；

In the step H, the method for determining the potential aggregation members at the current time includes:

for each forward intersection

Calculate the average distance between it and all other forward intersections at time t:

the average distances are ranked from large to small and ranked inFront side

The forward intersection point corresponding to the average distance of the nodes is an effective forward intersection point; wherein the content of the first and second substances,

representing rounding up, gamma is a proportional threshold of the effective forward intersection point, and meets the condition that gamma is more than 0 and less than 1;

in step H, the definition of the potential aggregation members in the population at time t is:

for a certain active positive member i ∈ EA at time t_tIf the fitted straight line corresponding to the member passes through at least one effective forward intersection point at the time t, the member is called a potential aggregation member at the time t; the set of potential aggregation members at time t is denoted as PG_t；

In the step I, the established unconstrained nonlinear programming model [ P1] is:

where K represents the number of potential aggregation members at time t, | PG_tK; accordingly, the equation for the K fitted lines at time t is:

is the coefficient corresponding to the kth straight line

To pair

In the objective function

The distance from the point (x, y) to the k-th line is expressed, and the objective function represents a point P at the time of t_t ^*(x^*,y^*) The sum of the distances to the K straight lines is minimum;

the above-mentionedIn step I, the new non-negative variable introduced is u_kAnd v_kSatisfy the following requirements

Transformed equivalent constrained linear programming model [ P2]Comprises the following steps:

s.t.

u_k≥0,v_k≥0，k＝1,2,...,K；

in step J, the calculation formula of the decreasing rate of the average distance between the potential aggregation members at the time t in q consecutive times before the time t including the time t is as follows:

in the step J, in the step,

the calculation formula of the centroid C gathering center is as follows:

wherein C (x) and C (y) respectively represent the horizontal and vertical coordinates of the point C;

and

respectively representing points

The abscissa and ordinate of (a).

Further, based on active membersIn the algorithm for detecting the aggregation behavior of the inter-moving average distance, order

l: ═ 0, t: ═ 2; wherein l is a counter for recording the number of drops in the running average distance;

the method specifically comprises the following steps:

step 1, solving a set A of active members at the time t_t(ii) a If | A_tIf | is less than alphan, then order

l: ═ 0, t: ═ t +1, return to step 1; otherwise, turning to the step 2;

step 2, solving the set EA of the effective active members at the time t_t(ii) a If EA_tIf | is less than alphan, then order

l is 0, t is t +1, and then step 1 is carried out; otherwise, turning to the step 3;

step 3; calculating the average distance between the active members at the time t

If t is less than m, let

t is t +1, turning to step 1; otherwise, turning to the step 4;

step 4, calculating the moving average distance between the effective active members at the time t

If it is not

Making l: ═ 0, t: ═ t +1, and then turning to step 1; otherwise, enabling the l: (l + 1), and turning to the step 5;

step 5, calculating

If l is greater than or equal to l₀And lambda is more than or equal to lambda₀If yes, sending out early warning and stopping; otherwise, enabling the t to be t +1, and turning to the step 1;

l₀is a positive integer and represents a threshold value of the number of times of the falling of the moving average distance; lambda [ alpha ]₀Is a threshold value of the falling rate of the moving average distance, and satisfies the condition that 0 is more than lambda₀Less than 1; when the moving average distance between active members is continuous l₀A second decrease, and a decrease rate of λ₀And if the group has aggregation tendency, early warning is given out.

Further, in the aggregation prediction algorithm based on the least square fitting straight line of the movement locus of the potential aggregation member, the order

The method specifically comprises the following steps:

step one, solving a set EA of effective active members at the time t_t；

Step two, for each effective active member i belongs to EA_tDetermining a fitted straight line of the movement locus up to the time t

And calculating the intersection point of any two fitting straight lines

Step three, solving all forward intersection points at the time t, and screening out effective forward intersection points at the time t;

step four, solving the potential aggregation members at the time t;

step five, solving a linear programming model [ P2 ]]Get the optimal solution P_t ^*(ii) a If it is not

Turning to the sixth step, otherwise, turning to the first step by making t: (t + 1);

step six, calculating:

if it is not

And is not less than₀If yes, turning to the step seven, otherwise, turning to the step one by making t: ═ t + 1;

step seven, calculating

Centroid C:

outputting the aggregation center C and stopping; c (x) and C (y) respectively represent the horizontal and vertical coordinates of the point C;

and

respectively representing points

The abscissa and ordinate of (a).

Another object of the present invention is to provide a computer program for implementing the method for early detection of specific group aggregation behavior and prediction of aggregation.

Another object of the present invention is to provide an information data processing terminal that implements the method for early detection of specific group aggregation behavior and prediction of aggregation.

It is another object of the present invention to provide a computer-readable storage medium, comprising instructions which, when executed on a computer, cause the computer to perform the method for early detection of specific group aggregation behavior and prediction of aggregation.

Another objective of the present invention is to provide a system for early detection and prediction of specific group aggregation behavior, comprising:

an aggregation behavior detection module based on the moving average distance between the effective active members defines the effective active members in the group by using an aggregation behavior detection algorithm based on the moving average distance between the effective active members, and eliminates the interference of noise members on the aggregation behavior detection; for a special group with aggregation behaviors, detecting the aggregation tendency of the group in the early stage of group activity and giving out early warning;

and the aggregation prediction module screens out the potential aggregation members by utilizing an aggregation prediction algorithm based on a least square fitting straight line of the movement locus of the potential aggregation members to perform aggregation prediction.

The early detection and aggregation prediction system for specific population aggregation behaviors further comprises:

a data acquisition unit: the system is used for acquiring real-time position data of target group members;

an aggregation tendency judgment unit: the system is used for judging whether the aggregation behaviors exist in the group or not based on an aggregation behavior detection algorithm;

an alarm display unit: the system is used for giving out early warning on the aggregation tendency of the population;

a prediction processing unit: for predicting the aggregation based on an aggregation prediction algorithm;

a gathering place display unit: for presenting the predicted aggregate site.

Another object of the present invention is to provide an information data processing terminal equipped with the above-mentioned early detection and aggregation prediction system for specific group aggregation behavior.

The invention provides a method and a system for early detection of special group aggregation behaviors and prediction of aggregation places, which are realized by utilizing movement track data of group members. The technical method overcomes the difficulties, gets rid of the limitation of manual or video monitoring, can automatically and quickly detect the potential gathering behavior in a special group, timely sends out gathering early warning, and further carries out gathering prediction. The system can display the historical movement track and the current position of the group members in real time, and can dynamically display the change of the gathering place for the gathering condition, and the predicted gathering place is more and more accurate along with the increase of time. Based on the method and the system, relevant departments can automatically monitor the movement behaviors of special groups in real time, and once the gathering behaviors are found, corresponding measures can be taken in time to prevent the gathering behaviors.

The invention is based on group member historical movement track data. Generally, group members all hold mobile phone devices, existing research data can position users through mr (measurement report) data generated in a mobile phone communication process, and discrete position points with time marks of the group members, namely moving tracks of the members, can be obtained.

The invention does not assume that the moving tracks of all members in a target group need to be obtained, such as solving all forward intersection points at the current moment, screening effective forward intersection points, then solving the potential aggregation member step at the current moment based on the effective forward intersection points, and the selection of the parameter gamma can possibly exclude individual members which really participate in aggregation, but the detection of aggregation behavior or the prediction of an aggregation place cannot be influenced.

The advantages and positive effects of the invention are also reflected in the following four aspects：

(1) The aggregation and non-aggregation conditions can be accurately distinguished;

(2) the calculation speed is high;

(3) for the aggregation condition, the aggregation trend can be detected and the aggregation place can be predicted at an earlier time;

(4) the method can effectively detect the aggregation trend for various aggregation conditions, has accurate prediction on aggregation places, and has strong robustness.

Compared with the prior art, the invention has the advantages that:

the method for early detection of the special group aggregation behaviors and prediction of the aggregation places provided by the invention is independent of a video monitoring system, and only utilizes the historical movement track data of group members to realize quick judgment of the special group aggregation behaviors, give early warning to the groups with the aggregation behaviors in time, and can accurately predict the aggregation places of the groups. The method provided by the invention has better robustness, can automatically eliminate interference members and increases the reliability of results. For the prediction of the gathering place, the method is not influenced by factors such as initial positions, departure time, moving speed, gathering directions and the like of the members participating in gathering, and can achieve high prediction accuracy.

Through simulation experiments in three aspects, namely tests on aggregation members with different proportions, non-omnidirectional aggregation and aggregation in different population scales, the method disclosed by the invention is proved to be capable of accurately distinguishing aggregation and non-aggregation conditions, and capable of predicting accurate aggregation places for various aggregation conditions, thereby showing stronger robustness of the method disclosed by the invention. In contrast, the centroid method is often poor in prediction result due to lack of processing on the interfering members and no consideration of factors such as differences in distances from the aggregation members to the actual aggregation places. In addition, the experimental results prove that the method can process and finish various cases with higher efficiency, can give early warning and predict the aggregation place for the aggregation case at an earlier time, and completely meets the requirements of practical application.

Drawings

FIG. 1 is a schematic diagram of movement trajectories of group members according to an embodiment of the present invention;

FIG. 2 is a flow chart of an aggregate behavior detection algorithm (Algorithm 1) provided by an embodiment of the present invention;

FIG. 3 is a schematic diagram of a forward intersection provided by an embodiment of the present invention;

fig. 4 is a schematic diagram of forward intersections of an effective active member of interference and interference provided by an embodiment of the present invention;

FIG. 5 is a flow chart of the aggregate prediction algorithm (Algorithm 2) provided by an embodiment of the present invention;

fig. 6 is a flowchart of the overall algorithm provided by the embodiment of the present invention.

Fig. 7 is a comparison diagram of the early warning time and the predicted aggregation place time, the average activity starting time of the group members, and the aggregation final completion time under the five different aggregation proportions provided by the embodiment of the present invention.

FIG. 8 is a graph illustrating the results of an experiment of an example of a 10% aggregation fraction provided by an embodiment of the present invention.

FIG. 9 is a graph showing the prediction results aggregated in the example of aggregation from above provided by an embodiment of the present invention.

FIG. 10 is a graph showing aggregated predicted results for an example aggregated from the top left provided by an embodiment of the present invention.

Fig. 11 is a graph of aggregate predicted results in an aggregation example showing a population membership of 1000 according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The invention comprises two phases of aggregate behavior detection and aggregate prediction. Effective active members in a group are defined in the first stage, interference of noise members on aggregation behavior detection is eliminated, and an aggregation behavior detection algorithm based on the moving average distance between the effective active members is provided. For a special group with aggregation behaviors, the algorithm can detect the aggregation tendency of the group in the early stage of group activities and send out early warning. And then, in an aggregation prediction stage, screening out potential aggregation members, and providing an aggregation prediction algorithm based on least square fitting straight lines of the movement tracks of the potential aggregation members. The invention can effectively help related departments to discover the gathering behavior of special groups as early as possible and accurately lock gathering places so as to assist the related departments to take corresponding measures in time.

The method for early detection and prediction of aggregation behaviors of special populations provided by the embodiment of the invention comprises the following steps:

E. when the average distance between the effective active members at a section of continuous time is obtained, calculating the sliding average distance between the effective active members corresponding to the current time; if a running average distance between active members occurs continuously₀The secondary decrement, and the rate of decrease reaches a given threshold λ₀(0＜λ₀Less than 1), sending out early warning;

F. once the early warning is sent out, the fact that the clustering tendency exists in the population is meant, and then clustering prediction is carried out; from the early warning moment, solving a set of effective active members;

I. establishing an unconstrained non-linear programming model [ P1]]Solving an optimal position at the current moment to ensure that the sum of the distances from the optimal position to least square fitting straight lines corresponding to all potential aggregation members is minimum; introducing new non-negative variables to lead an unconstrained non-linear programming model [ P1]]Conversion to an equivalent constrained linear programming model [ P2](ii) a Solving a Linear programming model [ P2 ]]Obtaining the optimal position P of the current moment_t ^*；

J. After the optimal position of a section of continuous time is obtained, calculating the descending rate of the average distance between the potential aggregation members in a fixed long period; if P is within this fixed long period of time_t ^*The maximum distance between does not exceed d_max(d_max> 0) and the rate of decrease of the average distance between the aggregation members has reached a given threshold₀(0＜₀< 1), a series of P's are calculated over the period_t ^*And outputting it as a center of convergence.

In the step A, the definition of the active members in the population at the time t is as follows:

d(P_i ^t,P_i ¹)≥r

the member is said to be the active member at time t. Wherein P is_i ^tDenotes the position coordinate of the ith member at time t, d (P)_i ^t,P_i ¹) Represents P_i ^t,P_i ¹The distance between two locations, r is a distance threshold. The set of active members at time t is denoted as A_tUsing | A_tI represents the set A_tThe number of the elements in (B).

i.e. if its average distance from other active members at time t reaches beta compared to the rate of decrease at the initial time₀This member is said to be the active member at time t. Wherein, beta₀< 1 represents a threshold for the average distance decrease rate between active members. The set of active members at time t is recorded as EA_t。

wherein the content of the first and second substances,

is a combination number.

In step E, the moving average distance between the active members (rather than directly using the average distance between the active members) is used to characterize the population aggregation, because the average distance sequence may have more frequent fluctuation, and the moving average of the average distance sequence can overcome this disadvantage. the calculation formula of the moving average distance between the effective active members in the population at the time t is as follows:

where m is a positive integer representing the number of terms of the moving average.

6. In the step E, the moving average distance between the effective active members is l₀The calculation formula of the reduction rate in the next successive time is:

wherein l₀Is a positive integer and represents the threshold for the number of falling moving average distances.

In the step H, the definition of the forward intersection at the time t is:

by using

representing straight lines

And a straight line

The intersection between them. For a certain

If it satisfies both of the following inequalities:

it is referred to as the forward crossing point at time t. Where "·" represents the inner product of the vector. the set of forward intersections at time t is denoted as F_t。

In the step H, the definition of the effective forward intersection point is:

for each forward intersection

the average distances are sorted from large to small, and are arranged in the front

And the forward intersection point corresponding to the average distance of the nodes is the effective forward intersection point. Wherein the content of the first and second substances,

indicating rounding up, gamma is the proportional threshold of the effective forward crossing point, and satisfies 0 < gamma < 1.

for a certain active positive member i ∈ EA at time t_tIf the fitted straight line corresponding to the member passes at least one valid forward intersection point at time t, the member is called a potential aggregation member at time t. the set of potential aggregation members at time t is denoted as PG_t。

[P1]

where K represents the number of potential aggregation members at time t, i.e. | PG_tAnd K. Accordingly, the equation for the K fitted lines at time t is:

is the coefficient corresponding to the kth straight line

To pair

In the objective function

The distance from the point (x, y) to the k-th line is expressed, and the objective function represents a point P at the time of t_t ^*(x^*,y^*) So that the sum of its distances to the K lines is minimized.

In the step I, the introduced new non-negative variable is u_kAnd v_kSatisfy the following requirements

[P2]

s.t.

u_k≥0,v_k≥0，k＝1,2,...,K.

in the step J, in the step,

the centroid C (center of aggregation) of (C) is calculated as:

and

respectively representing points

The abscissa and ordinate of (a).

The invention is further described with reference to specific examples.

The method for early detecting the aggregation behaviors of the special population and predicting the aggregation comprises two stages of detection of the aggregation behaviors and prediction of the aggregation.

In the aggregation behavior detection stage, the invention provides an aggregation behavior detection algorithm based on the moving average distance between effective active members. First, a number of necessary definitions for the aggregate behavior detection algorithm are given.

For a particular population with n members, it is assumed that the location data of the group members is continuously obtained from the time t ═ 1. By P_i ^t(i＝1,...,n；t∈N⁺) Indicating the position coordinates of the ith member at time t. Suppose that the relevant department is concerned only with the aggregation of people of α n and above in the population, where 0 < α ≦ 1.

Define 1. active members of the population at time t.

d(P_i ^t,P_i ¹)≥r

the member is said to be the active member at time t. Wherein d (P)_i ^t,P_i ¹) Represents P_i ^t,P_i ¹The distance between two positions, r > 0, is a distance threshold. The set of active members at time t is denoted as A_tUsing | A_tI represents the set A_tThe number of the elements in (B).

Define the effective active members of the population at time t.

As shown in fig. 1, the solid line with an arrow represents the movement trajectory of the group member. For each group member, the point pointed by the arrow represents the position of its current time, and the other end of the arrow is the initial position. The radius of the small circle is r, and the member corresponding to the delta slightly moves but does not exceed the area of the small circle; □, the corresponding member indicates the member that has been in a quiescent state by the current time. According to definition 1, o and the corresponding members are active members at the current moment; according to definition 2, only the members corresponding to O are active members at the current time.

Defining the average distance between active positive members in the population at time t as:

wherein the content of the first and second substances,

is a combination number.

Defining the moving average distance between effective active members in the population at the moment t as:

As shown in fig. 2, the specific implementation steps of the algorithm 1-aggregation behavior detection algorithm are as follows:

first, let

l: ═ 0, t: ═ 2. Where l is a counter to record the number of drops in the running average distance.

Step 1. solving the set A of active members at the time t by using the definition 1_t. If | A_tIf | is less than alphan, then order

And l: ═ 0, t: ═ t +1, and the step 1 is returned. Otherwise, turning to the step 2;

step 2. solving the set EA of the effective active members at the time t by using the definition 2_t. If EA_tIf | is less than alphan, then order

And l: ═ 0, t: ═ t +1, and then step 1 is carried out. Otherwise, turning to the step 3;

step 3, calculating the average between the effective active members at the time tDistance between two adjacent plates

If t is less than m, let

And t ═ t +1, and turning to step 1. Otherwise, turning to the step 4;

If it is not

And (5) enabling l: ═ 0 and t: ═ t +1, and turning to the step 1. Otherwise, enabling the l: (l + 1), and turning to the step 5;

step 5. calculate

If l is greater than or equal to l₀And lambda is more than or equal to lambda₀And if yes, giving out an early warning and stopping. Otherwise, let t: ═ t +1, go to step 1.

In step 5,/₀Is a positive integer and represents a threshold value of the number of times of the falling of the moving average distance; lambda [ alpha ]₀Is a threshold value of the falling rate of the moving average distance, and satisfies the condition that 0 is more than lambda₀Is less than 1. Step 5 illustrates that when the moving average distance between active positive members is continuous l₀A secondary decrease is achieved, and the rate of decrease reaches lambda₀And the group is considered to have aggregation tendency, so that an early warning is given.

The average distance between active members is not used directly, but rather the running average distance between active members is used to characterize the population clustering because the average distance sequence may fluctuate more frequently, and the running average of the average distance sequence overcomes this disadvantage.

Once the algorithm 1 gives an aggregation early warning, it means that some members in the population have aggregation tendency. Then, the aggregation prediction stage is carried out, and the invention provides aggregation of least square fitting straight lines based on the movement tracks of potential aggregation membersA collective prediction algorithm. By using

Indicating the moment when the algorithm 1 gives an early warning.

The effective active members in the population are most likely to be a part of members participating in the aggregation, and the moving tracks of the effective active members have important reference values for searching the aggregation places. For each active member, if the member does participate in the aggregation, then the direction of his trajectory of movement should be pointing to the aggregation site. In order to obtain the direction of the movement locus, least square straight line fitting is carried out on the movement locus of each effective active member, and all intersection points between the fitted straight lines are solved. By using

A fitted straight line representing the trajectory of the movement of the active positive member i before time t,

representing straight lines

And a straight line

The intersection between them.

The forward intersection at time 5.t is defined.

For a certain

If it satisfies both of the following conditions:

and

it is referred to as the forward crossing point at time t. the set of forward intersections at time t is denoted as F_t。

In definition 5, "·" denotes the inner product of two vectors. According to the definition of the inner product,

where | g | represents the length of the vector and θ represents the angle between the two vectors. When the two formulas are simultaneously established, the included angle between the last movement direction of the two members i and j and the direction pointing to the intersection point from the current position of the members i and j does not exceed 90 degrees, so that the intersection point can be considered to be the intersection point

The direction of travel of both members i and j is consistent and can be considered their common destination. All these forward intersections thus represent to some extent the location of the gathering ground.

Fig. 3 shows an example of a forward intersection. In FIG. 3, there are 3 member movement trajectories (broken lines with arrows), and thus 3 fitting straight lines are obtained

And

the 3 straight lines mutually have 3 intersection points

And

for point

In other words, because it satisfies

And

thus, it is possible to provide

Is the forward intersection. Can be easily seen

And

none of the conditions for definition 5 are met and therefore none are forward intersections.

There is some interference in the active member and the forward intersection. For example, in fig. 4, there are 7 active positive members, and the solid lines with arrows represent their movement trajectories and movement directions. Their corresponding fitted straight lines are common

One intersection (● and ≧ wherein ● is the forward intersection. The circular region is a region where the forward intersections are denser and therefore the most likely convergence region.

It is easy to see that the two members corresponding to the two thicker movement trajectories in fig. 4 are interfering members. Straight line

The intersection with the other straight lines is not a forward intersection. It is also easy to see that the trajectory of this member has already crossed the circular area, so there is no reason to think that he is involved in the aggregation. While straight line

Although there are 3 forward intersections with other lines, these three points are significantly further from the circular area, meaning that the direction of movement of this member is not directed to the place of most people's focus, and therefore, it is also believed that there is little likelihood that he will participate in the focus.

These interfering members do not participate in the aggregation and therefore their movement trajectories and their associated forward intersections affect the determination of the true aggregation site. The effective active members of these interferers and the forward intersection of the interferers are removed below.

Defining a valid forward intersection at time 6. t.

For each forward intersection

And the forward intersection point corresponding to the average distance of the nodes is the effective forward intersection point. Here, the first and second liquid crystal display panels are,

Defining potential aggregation members in the population at time t.

The potential aggregation members defined above are effective active members for interference removal. The value of γ should be small enough to exclude as many members as possible that do not participate in the aggregation, and thus may exclude individual members that actually participate in the aggregation, but this does not affect the prediction of the aggregation site, since the prediction of the aggregation site does not require finding all members that actually participate in the aggregation.

The historical movement trajectories of these potential aggregation members are used below to predict the aggregation. For a member participating in the aggregation, the fitted straight line of his historical movement trajectory reflects the direction of his movement to some extent, so the aggregation should be very close to the fitted straight line corresponding to the member. Therefore, a location point at time t is next found so that the sum of its distances to the fitted straight lines for all potential aggregation members is minimized.

The number of potential aggregation members at time t, i.e. | PG, is represented by K_tAnd K. Accordingly, there are K fitted lines, expressed as:

wherein the content of the first and second substances,

is the coefficient corresponding to the kth straight line

To pair

Point P is found_t ^*(x, y) satisfying the minimum sum of the distances from the K straight lines, namely solving the unconstrained nonlinear programming problem [ P1]]：

[P1]

Defining new non-negative variables u_kAnd v_kSatisfy the following requirements

Then the non-linear programming model P1]Conversion to a Linear programming model [ P2 ]]：

[P2]

s.t.

u_k≥0,v_k≥0，k＝1,2,...,K.

By solving a linear programming model [ P2 ]]Can obtain the early warning time

Optimum position P at each moment thereafter_t ^*. If P is_t ^*The position of which varies little over a continuous period of time and the average distance between potentially aggregating members has reached a certain rate of decline over that period of time, and a series of P's obtained over that period of time_t ^*As a center of concentration. The aggregated predictive algorithm is given below.

As shown in FIG. 5, the algorithm 2-the prediction algorithm with aggregation is implemented by the following steps:

first, let

Step 1, solving a set EA of effective active members at the time t by using a definition 2_t；

Step 2, for each effective active member i belongs to EA_tFinding a straight line fitting the movement path of the user up to t

And calculating the intersection point of any two fitting straight lines

Step 3, solving all forward intersection points at the time t by using a definition 5, and screening out effective forward intersection points at the time t by using a definition 6;

step 4, solving potential aggregation members at the time t by using the definition 7;

step 5, solving the linear programming model [ P2 ]]Get the optimal solution P_t ^*. If it is not

And 6, turning to the step 6. Otherwise, enabling the t to be t +1, and turning to the step 1;

and 6, calculating:

if it is not

And is not less than₀If true, go to step 7. Otherwise, enabling the t to be t +1, and turning to the step 1;

step 7, calculate

Centroid C:

the center of aggregation C is output and stopped. Here, C (x) and C (y) respectively represent the abscissa and ordinate of the point C;

and

respectively representing points

The abscissa and ordinate of (a).

In Algorithm 2, q is a positive integer, d_maxIs a positive number, and the number of the positive number,₀is a threshold value of the average distance reduction rate among the potential aggregation members, and satisfies 0<₀Is less than 1. Steps 6 and 7 are described when

The maximum distance between does not exceed d_maxAnd the rate of decrease of the average distance between the members of the cluster reaches the final q consecutive times₀Algorithm 2 gives the center of aggregation and stops.

After the algorithm 1 gives an early warning, the algorithm 2 provides a prediction algorithm of the aggregation, and fig. 6 shows the overall algorithm flow of early detection and prediction of the aggregation of the special group. If no early warning is given by the algorithm 1 until the monitoring time for a special population is over, or no aggregation place is found by the algorithm 2 until the monitoring time is over, the population is not subjected to aggregation behavior. The fact that algorithm 1 gives an early warning does not mean that there must be an aggregate behavior in the population. Due to the severity of the special group aggregation behaviors, in the conservative aspect, the algorithm 1 has higher sensitivity to aggregation detection during parameter setting, and gives early warning to the conditions with aggregation tendency, so that the alertness of related monitoring departments is improved. Furthermore, for the case of aggregation, if algorithm 2 continues to run after it gives an aggregated place, it is possible to obtain an aggregated place that is continuously updated over time. As the historical movement trajectories of population members contain more and more information over time, these continually updated aggregations will get closer to the true aggregations.

The invention is further described below with reference to simulation experiments to verify the accuracy, efficiency, robustness, and prediction accuracy of the method of the invention.

The experiment is mainly divided into three aspects: different proportions of aggregating members (including non-aggregating cases), non-omnidirectionally aggregating, different population size aggregating. All experiments were performed by MATLAB.

The number of members in the group is set to be 50, and the related departments are assumed to focus on the aggregation of no less than 10 persons in the group, namely n is 50 and alpha is 0.2. The aggregation finalization time is set to 100 (for the aggregation case). The threshold values and other parameters involved in the algorithm can be set empirically according to actual problems, or the parameters can be learned by machine learning to obtain the optimal setting. Here, our parameters are set to: r 10, beta₀＝0.2，m＝5，l₀＝5，λ₀＝0.05，γ＝0.35，q＝5，₀＝0.09，d_max＝2.5。

Testing of different proportions of aggregating members: a plurality of random examples were generated for six cases in which the aggregation member ratio was 0 (non-aggregation case), 20%, 40%, 60%, 80%, 100% (corresponding to the numbers of members participating in aggregation being 0, 10, 20, 30, 40, 50, respectively), and tested by the method of the present invention. Experimental results show that the method can accurately distinguish aggregation and non-aggregation conditions, give early warning to the aggregation conditions and further predict the aggregation places. The calculation time for each example is generally between 2 and 7 seconds, and on average about 3.68 seconds, reflecting the faster calculation speed of the method of the invention. In addition, fig. 7 shows the comparison (in average meaning) between the early warning time and the aggregation site prediction time under five different aggregation proportions and the average activity starting time and the aggregation final completion time of the group members, which reflects that the method of the present invention can detect the aggregation trend and predict the aggregation site at an earlier time. To illustrate the effect of aggregation prediction, an example of an aggregation ratio of 10% is illustrated as an experimental result, as shown in fig. 8. The small dots distributed in fig. 8 indicate the positions of the group members at the current time (the time when the aggregation site is predicted), and the broken lines connected thereto indicate the historical movement trajectories thereof. It can be seen that the centroid method predicts the aggregate ground with poor accuracy, while the method of the present invention predicts the aggregate ground with higher accuracy.

Testing of non-omnidirectional aggregation: since the actual aggregation situation is likely not to be as shown in fig. 8, there are members participating in the aggregation from all directions. In practice, the case of omnidirectional focusing tends to be relatively easy to predict, and the centroid method is sometimes more accurate in this case. In order to verify the prediction effect of the method of the present invention on the aggregation situation in non-omnidirectional directions, we performed experimental verification on eight situations aggregated from above, below, left, right, above-left, below-left, above-right, and below-right, respectively. The aggregation member proportion is set to 75% when the instance is generated. The results show that the calculation time, the early warning time, and the time at which the aggregation is predicted for these eight cases are at the same level as the effect in the experiment of the first aspect. For the prediction of the gathering place, the prediction results of the centroid method for the non-omnidirectional gathering condition deviate from the actual gathering place to a large extent, and the algorithm of the invention has better prediction accuracy for the non-omnidirectional gathering condition. Fig. 9 and 10 show aggregated predicted results for two examples of two cases aggregated from above and from the top left, respectively.

Testing of different population-scale aggregation: in both experiments, the population size, i.e., the number of members, was set to 50. In order to verify the prediction effect of the method on the aggregation condition of special populations with different scales, the population member numbers are respectively set to be 20, 50, 100, 200, 500 and 1000, the aggregation final completion time is still set to be 100, and the aggregation member proportion is set to be 75%. Since algorithm 2 needs to update the active positive members, fit straight lines and all intersections between them, etc. for every moment after the pre-warning, the running time of algorithm 2 will increase greatly as the population size increases. Here, a small improvement is made to algorithm 2: and after the effective active member set at the early warning moment is solved, if the number of the effective active members exceeds 50, randomly extracting 50 members from the effective active member set to form a new group. In subsequent iterations, the new population is used to take part in the calculation instead of the original population. This results in a significant reduction in algorithm runtime, and as discussed above, for populations with aggregate behavior, the method of the present invention does not require all members participating in the aggregation to be found to enable aggregate behavior detection and aggregate prediction. The experimental results show that the calculation time, the early warning time and the time for predicting the gathering place of the six population scale conditions have the same effect as the experiment of the first aspect, and the accuracy of the prediction of the gathering place is still high. FIG. 11 shows the aggregate prediction results given in the aggregate example with population membership 1000, where the aggregation centers predicted by the method of the present invention and the actual aggregation centers almost coincide.

Simulation experiments in the three aspects prove the correctness, the high efficiency, the robustness and the prediction accuracy of the method.

In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When used in whole or in part, can be implemented in a computer program product that includes one or more computer instructions. When loaded or executed on a computer, cause the flow or functions according to embodiments of the invention to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions may be transmitted from one website site, computer, server, or data center to another website site, computer, server, or data center via wire (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL), or wireless (e.g., infrared, wireless, microwave, etc.)). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that includes one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims

1. A method for early detection and prediction of aggregation of special population, comprising:

defining effective active members in a group by using an aggregation behavior detection algorithm based on the moving average distance between the effective active members, and eliminating the interference of noise members on the aggregation behavior detection; for a special group with aggregation behaviors, detecting the aggregation tendency of the group in the early stage of group activity and giving out early warning; screening out the potential aggregation members by using an aggregation prediction algorithm based on a least square fitting straight line of the movement locus of the potential aggregation members, and performing aggregation prediction;

the method for early detection and prediction of aggregation behaviors of the special population specifically comprises the following steps:

B. if the number of active members at the current moment does not reach alpha n, jumping to the next moment to continuously ask for the active members in the group; n represents the total number of members in the population, alpha n is the aggregation of no less than n members in the population of interest, and alpha is more than 0 and less than or equal to 1;

I. establishing an unconstrained non-linear programming model [ P1]]Solving an optimal position at the current moment, wherein the sum of the distances from the optimal position to least square fitting straight lines corresponding to all potential aggregation members is minimum; introducing new non-negative variables to lead an unconstrained non-linear programming model [ P1]]Conversion to an equivalent constrained linear programming model [ P2](ii) a Solving a constrained linear programming model [ P2 ]]Obtaining the optimal position P of the current moment_t ^*；

J. When a segment of continuous is obtainedAfter the optimal position of the moment, calculating the descending rate of the average distance between the potential aggregation members in a fixed long period; if P is within this fixed long period of time_t ^*The position does not vary beyond a given threshold and the rate of decrease of the average distance between the members of the cluster reaches a given threshold, a series P of the time interval is calculated_t ^*And outputting it as a center of convergence.

2. The method for early detection and prediction of special group aggregation activities according to claim 1, wherein the group aggregation information processing method based on group member historical movement trajectory data first uses the data acquisition unit to acquire real-time position data of a target group member and continuously locates the group member; an aggregation behavior detection algorithm is adopted, an aggregation tendency judgment unit is used for detecting the aggregation tendency of the group in the early stage of the group activity, and an alarm display unit sends out early warning; in the aggregation place prediction stage, potential aggregation members are screened out, and an aggregation place prediction algorithm is adopted to predict an aggregation place by using a prediction processing unit; and presented by the aggregated display unit.

3. The method for early detection and prediction of aggregation of specific populations according to claim 1, wherein in step a, the positive members of the population at time t are:

for a certain member i in the population, where i ═ 1.

d(P_i ^t,P_i ¹)≥r

Wherein P is_i ^tDenotes the position coordinate of the ith member at time t, d (P)_i ^t,P_i ¹) Represents P_i ^t,P_i ¹The distance between two locations, r is a distance threshold; the set of active members at time t is denoted as A_tUsing | A_tI represents the set A_tThe number of middle elements;

in step C, the effective active members in the population at time t are:

for a certain active member i, where i ∈ A_t，|A_t| ≧ 2; the active positive members at time t are:

wherein, beta₀< 1 represents a threshold for the average distance decrease rate between active members; the set of active members at time t is recorded as EA_t；

wherein the content of the first and second substances,

is a combination number;

in the step E, the calculation formula of the moving average distance between the effective active members in the group at the time t is as follows:

in the step H, the forward intersection point at the time t is

Representing straight lines

And a straight line

The intersection point between them;

a least squares fit straight line representing the trajectory of the movement of the active positive member i before time t, where i, j ∈ EA_t,i≠j；

for each forward intersection

The average distance to all other forward intersections at time t is:

for a certain active positive member i ∈ EA at time t_tThe fitting straight line corresponding to the member passes through at least one effective forward intersection point at the time t and is a potential aggregation member at the time t; the set of potential aggregation members at time t is denoted asPG_t；

where K represents the number of potential aggregation members at time t, | PG_tK; accordingly, the equation for the K fitted lines at time t is: the coefficient corresponding to the k-th straight line satisfies the pair

The distance from the point (x, y) to the k-th line is represented in the objective function, which represents a point P at time t_t ^*(x^*,y^*) The sum of the distances to the K straight lines is minimum;

Transformed equivalent constrained linear programming model [ P2 ]]Comprises the following steps:

u_k≥0,v_k≥0，k＝1,2,...,K；

in the step J, in the step,

the calculation formula of the centroid C gathering center is as follows:

and

respectively representing points

The abscissa and ordinate of (a).

4. The method of claim 1, wherein the algorithm for detecting aggregation behavior based on moving average distance between active and active members comprises

l is 0, t is 2; wherein l is a counter for recording the number of drops in the running average distance;

the method specifically comprises the following steps:

step 1, solving a set A of active members at the time t_t；|A_tIf | is less than alphan, then order

Returning to the step 1 when l is 0 and t is t + 1; otherwise, turning to the step 2;

step 2, solving the set EA of the effective active members at the time t_t；|EA_tIf | is less than alphan, then order

Turning to step 1 when l is 0 and t is t + 1; otherwise, turning to the step 3;

step 3, calculating the average distance between the effective active members at the time t

t is less than m, order

turning to step 1 when t is t + 1; otherwise, turning to the step 4;

Making l equal to 0 and t equal to t +1, and turning to the step 1; otherwise, changing l to l +1, and turning to the step 5;

step 5, calculating

If l is greater than or equal to l₀And lambda is more than or equal to lambda₀If yes, sending out early warning and stopping; otherwise, making t equal to t +1, and turning to the step 1;

5. The method for early detection of aggregation activities of special populations and prediction of aggregation as claimed in claim 1, wherein the algorithm for prediction of aggregation based on least squares fit of lines of movement trajectories of potential aggregation members comprises

The method specifically comprises the following steps:

step one, solving a set EA of effective active members at the time t_t；

And calculating the intersection point of any two fitting straight lines

Wherein i, j ∈ EA_t,i≠j；

step four, solving the potential aggregation members at the time t;

step five, solving a constrained linear programming model [ P2 ]]Get the optimal solution P_t ^*；

Turning to the step six, otherwise, turning to the step one by making t equal to t + 1;

step six, calculating:

and is not less than₀If yes, turning to step seven, otherwise, turning to step one by making t equal to t + 1;

step seven, calculating

Centroid C:

and

respectively representing points

The abscissa and ordinate of (a).

6. An information data processing terminal for implementing the method for early detection of the special group aggregation behavior and prediction of the aggregation place according to any one of claims 1 to 5.

7. A computer-readable storage medium comprising instructions that, when executed on a computer, cause the computer to perform the method for early detection of ad hoc group aggregate behavior and prediction of aggregation as claimed in any one of claims 1 to 6.

8. The early detection of special group aggregation behavior and prediction system of aggregation according to claim 1, wherein the early detection of special group aggregation behavior and prediction system of aggregation comprises: an aggregation behavior detection module based on the moving average distance between the effective active members defines the effective active members in the group by using an aggregation behavior detection algorithm based on the moving average distance between the effective active members, and eliminates the interference of noise members on the aggregation behavior detection; for a special group with aggregation behaviors, detecting the aggregation tendency of the group in the early stage of group activity and giving out early warning; and the aggregation prediction module screens out the potential aggregation members by utilizing an aggregation prediction algorithm based on a least square fitting straight line of the movement locus of the potential aggregation members to perform aggregation prediction.

9. The system for early detection of aggregation and prediction of aggregation by specific group of individuals according to claim 8, wherein the system for early detection of aggregation and prediction of aggregation by specific group of individuals further comprises:

the device comprises a data acquisition unit: the system is used for acquiring real-time position data of target group members;

a gathering place display unit: for presenting the predicted aggregate site.