CN109492788A

CN109492788A - Prediction flow of the people and the method and relevant device for establishing flow of the people prediction model

Info

Publication number: CN109492788A
Application number: CN201710822564.7A
Authority: CN
Inventors: 吴洁璇; 梅铮
Original assignee: China Mobile Communications Group Co Ltd; China Mobile Hangzhou Information Technology Co Ltd
Current assignee: China Mobile Communications Group Co Ltd; China Mobile Hangzhou Information Technology Co Ltd
Priority date: 2017-09-13
Filing date: 2017-09-13
Publication date: 2019-03-19
Anticipated expiration: 2037-09-13
Also published as: CN109492788B

Abstract

The present invention relates to the communications field, discloses prediction flow of the people and establish the method and relevant device of flow of the people prediction model.For realizing the Accurate Prediction of total flow of the people for hot zones.This method are as follows: determine target area at least one on the influential reference zone of the total flow of the people in target area, then, at least one reference zone is obtained in total flow of the people at the first moment based on mobile data, and according to target arma modeling, total flow of the people to target area at the second following moment is predicted；Wherein, which is total flow of the people based at least one reference zone in each historical juncture, the influence to total stream of people of historical juncture of the target area after corresponding and establish.In this way, Accurate Prediction can be carried out to total flow of the people of the future time instance of target area according to the historical data of reference zone, meanwhile, it is high to efficiently solve the problems, such as that source data existing in the prior art covers narrow and data acquisition equipment maintenance cost.

Description

Prediction flow of the people and the method and relevant device for establishing flow of the people prediction model

Technical field

The present invention relates to the communications field more particularly to a kind of prediction flow of the people and the methods for establishing flow of the people prediction model And relevant device.

Background technique

Ultra-large city for large population bases such as Beijing, Shanghai, Shenzhen, full of mobility, comprehensive shopping centre, The hot spot regions such as famous tourist attractions, large-scale activity place often will appear crowd " blowout " phenomenon, this during festivals or holidays Phenomenon is more significant.For various security risks existing for prevention urban human stream aggregation place, scientific and effective information-based prison is needed Survey means carry out flow of the people prediction, find crowd's flow characteristics early, and the coordination for district management resource provides decision-making foundation.

The technical solution of existing stream of people prediction relies primarily on rail and hands over website gate data, wifi hotspot location data, infrared Sensing data etc., using conventional time series prediction algorithm or its modified hydrothermal process, the stream of people for carrying out future time instance is predicted.

Existing method has a degree of effect on the stream of people predicts, but there is problems:

1, the source data that existing stream of people's prediction technique is relied on, as rail hand over website gate data, wifi hotspot location data, Infrared sensing data etc. are all set on fixed position, and detection range is limited, cause source data covering narrow.

2, source data is acquired by fixed point, it is also necessary to which data acquisition device is set on fixed point, and maintenance cost is high.

Summary of the invention

The embodiment of the present application provides a kind of method and relevant device predicted flow of the people and establish flow of the people prediction model. It is narrow for realizing the Accurate Prediction of total flow of the people of city hot spot region, and solution source data covering existing in the prior art Mountain pass and the high problem of data acquisition equipment maintenance cost.

Specific technical solution provided by the embodiments of the present application is as follows:

A method of prediction flow of the people, comprising:

Determine target area and at least one reference zone, wherein total flow of the people of the reference zone is to target area Total flow of the people impact；

At least one described reference zone is obtained in total flow of the people at the first moment based on mobile data；

Total flow of the people according at least one described reference zone at the first moment is slided using preset target autoregression Total flow of the people of the average arma modeling to the target area at the second following moment is predicted；

Wherein, the target arma modeling is total stream of people based at least one described reference zone in each historical juncture Amount, the target area is established in the influence of total flow of the people of the historical juncture accordingly lagged.

Optionally, before determining target area and at least one reference zone, further comprise:

Obtain the sample changed sequence at least two specified regions in designated time period, wherein the sample changed sequence Indicate total flow of the people of each time sampling point of the specified region in the designated time period；

Target area and reference zone are determined according to external command, and the corresponding sample changed sequence in target area is made For target sequence, using the corresponding sample changed sequence of reference zone as reference sequences；

Using Granger Granger Causality Test, at least one ginseng for causing hysteresis quality to influence target sequence is filtered out Examine sequence；

Target arma modeling is established based on the target sequence and at least one described reference sequences.

Optionally, the sample changed sequence for obtaining any one specified region in designated time period, specifically includes:

It obtains in any one specified region described in each time sampling point acquisition in the designated time period Mobile data；

Mobile data based on acquisition determines any one described specified region in the total of each time sampling point respectively Flow of the people, to obtain corresponding sample changed sequence.

Optionally, based on the mobile data of acquisition, determine any one described specified region in any one time sampling Total flow of the people of point, comprising:

The identification information of each base station cell in any one specified region is obtained, and each base station cell is associated with To any one described specified region；

The identification information of each base station cell based on acquisition acquires each base station cell and adopts in any one described time The user data of sampling point；

Duplicate removal processing and denoising are carried out to the user data of acquisition, to determine any one described specified region in institute State the mobile subscriber's quantity of any one time sampling point；

Based on any one described specified region any one time sampling point mobile subscriber's quantity, with reference to pre- If mobile subscriber's quantity and total flow of the people between accounting relationship, any one described specified region it is described any one when Between sampled point total flow of the people.

Optionally, after carrying out duplicate removal processing and denoising to the user data of acquisition, determine it is described any one Specified region further comprises before the mobile subscriber's quantity of any one time sampling point:

From the mobile subscriber's quantity, exclusion residence time in any one described specified region is less than preset time The mobile subscriber of threshold value.

Optionally, which is characterized in that use Granger Causality Test, filter out and cause hysteresis quality to influence target sequence At least one reference sequences, specifically include；

Stationarity detection and adjustment are carried out respectively to the target sequence and the reference sequences；

For the target sequence, each reference sequences is respectively adopted and carries out Granger Causality Test, filters out to institute Stating target sequence causes hysteresis quality to influence, can be as at least one reference sequences of the target sequence dependent variable.

Optionally, stationarity detection and adjustment are carried out respectively to the target sequence and the reference sequences, it is specific to wrap It includes:

Stationarity detection is carried out to the target sequence and each reference sequences respectively, is filtered out without stationarity Target sequence or/and reference sequences；

Whole relationship is assisted to sentence for the target sequence for not having stationarity or/and the reference sequences further progress It is fixed, filter out the target sequence or/and the reference sequences for not having and assisting whole relationship；

For not having stationarity and there is no the target sequences and the reference sequences of assisting whole relationship, using difference Transform method carries out first-order difference or higher order differential transformation until sequence meets stationarity.

Optionally, for the target sequence, Granger Causality Test is carried out using any one reference sequences, comprising:

According to the target sequence and any one described reference sequences construction without constrained regression equation, according to the target Sequence structure has constrained regression equation；

Utilize the residual sum of squares (RSS) RSSu and the residuals squares for having constrained regression equation of the no constrained regression equation F statistic is constructed with RSSr:

Wherein, n indicates that sample size, p indicate that the lag time granularity maximum number of target sequence, q indicate reference sequences Lag time granularity maximum number.

Judge whether F statistic is greater than the response critical value that F is distributed under given level of signifiance α；If so, determining described appoint Reference sequences of anticipating are the dependent variables impacted to the target sequence, otherwise, it determines any one described reference sequences It is not the dependent variable impacted to the target sequence.

Optionally, target arma modeling is established based on the target sequence and at least one described reference sequences, it is specific to wrap It includes:

Based on the target sequence and at least one described reference sequences, initial ARMA model is established；

Parameter Estimation and model order are carried out for the established initial ARMA model, obtains corresponding target ARMA Model.

Optionally, parameter Estimation and model order are carried out to the established initial ARMA model, obtains corresponding mesh Arma modeling is marked, is specifically included:

Based on the target sequence and at least one reference sequences, parameter is carried out using least squares estimate and is estimated Meter, obtaining to enable makes residual sum of squares (RSS) reach the smallest estimates of parameters；

The model order of the initial ARMA model is successively increased since preset initial value to preset upper limit value； Wherein, every to increase once, calculate a minimal information AIC criterion functional value；

It chooses the corresponding model order of the smallest AIC criterion functional value of value and initial ARMA is arranged in the estimates of parameters Model obtains the target arma modeling.

Optionally, after obtaining the target arma modeling, further comprise,

Any one in operating below or any combination are executed for the target arma modeling:

Level of signifiance Student T statistics control is carried out to the target arma modeling；

Steady invertibity inspection is carried out to the target arma modeling；

Residual sequence white noise verification is carried out to target arma modeling.

A method of establishing flow of the people prediction model, comprising:

Optionally, using Granger Causality Test, at least one ginseng for causing hysteresis quality to influence target sequence is filtered out Sequence is examined, is specifically included；

It carries out that whole relationship is assisted to determine for the target sequence or/and the reference sequences for not having stationarity, screening Do not have the target sequence or/and the reference sequences for assisting whole relationship out；

To do not have assist whole relationship and do not have the target sequence of stationarity or/and the reference sequences carry out it is steady Property conversion.

Parameter Estimation and model order are carried out for the established initial ARMA model, obtains corresponding target ARMA Model

Using the specific value of the known target sequence and at least one reference sequences, estimated using least square Meter method carries out parameter Estimation, and obtaining to enable makes residual sum of squares (RSS) reach the smallest estimates of parameters；

Optionally, after obtaining the target arma modeling, further comprise,

Steady invertibity inspection is carried out to the target arma modeling；

A kind of equipment for predicting flow of the people, the equipment include:

Input module, for determining target area and at least one reference zone, wherein total stream of people of the reference zone Amount impacts total flow of the people of target area；

Acquisition module, for obtaining at least one reference zone in total flow of the people at the first moment based on mobile data；

Prediction module, for according to the total flow of the people of at least one reference zone at the first moment, use to be preset Total flow of the people of the target arma modeling to the target area at the second following moment is predicted；Wherein, the target Arma modeling is total flow of the people based at least one described reference zone in each historical juncture, to the target area in phase The influence of the total flow of the people for the historical juncture that should be lagged and establish.

It optionally, further comprise analysis module and processing module,

The acquisition module is further used for:

Before determining target area and at least one reference zone, at least two specified regions in designated time period are obtained Sample changed sequence, wherein the sample changed sequence indicates that the specified region is each in the designated time period Total flow of the people of a time sampling point；

The analysis module is used for:

The processing module is used for:

Using Granger Granger Causality Test, at least one ginseng for causing hysteresis quality to influence target sequence is filtered out Sequence is examined, and target arma modeling is established based on the target sequence and at least one described reference sequences.

Optionally, when obtaining the sample changed sequence in any one specified region in designated time period, the acquisition module It is specifically used for:

Optionally, based on the mobile data of acquisition, determine any one described specified region in any one time sampling When total flow of the people of point, the acquisition module is used for:

Optionally, after carrying out duplicate removal processing and denoising to the user data of acquisition, determine it is described any one Before the mobile subscriber's quantity of any one time sampling point, the acquisition module is further used in specified region:

Optionally, using Granger Causality Test, at least one ginseng for causing hysteresis quality to influence target sequence is filtered out When examining sequence, the processing module is specifically used for；

Optionally, described when carrying out stationarity detection and adjustment respectively to the target sequence and the reference sequences Processing module is specifically used for:

Optionally, for the target sequence, when carrying out Granger Causality Test using any one reference sequences, institute Processing module is stated for including:

Optionally, described when establishing target arma modeling based on the target sequence and at least one described reference sequences Processing module is specifically used for:

Optionally, parameter Estimation and model order are carried out to the established initial ARMA model, obtains corresponding mesh When marking arma modeling, the processing module is specifically used for:

Optionally, after obtaining the target arma modeling, the processing module is further used for,

Steady invertibity inspection is carried out to the target arma modeling；

A kind of equipment for establishing flow of the people prediction model, comprising:

Acquisition module, for obtaining the sample changed sequence at least two specified regions in designated time period, wherein described Sample changed sequence indicates total flow of the people of each time sampling point of the specified region in the designated time period；

Analysis module, for determining target area and reference zone according to external command, and target area is corresponding Sample changed sequence is as target sequence, using the corresponding sample changed sequence of reference zone as reference sequences；

Processing module filters out for using Granger Granger Causality Test and causes hysteresis quality to influence target sequence At least one reference sequences；

Optionally, after carrying out duplicate removal processing and denoising to the user data of acquisition, determine it is described any one When specified region is before the mobile subscriber's quantity of any one time sampling point, the acquisition module is further used for:

Optionally, for the target sequence, when carrying out Granger Causality Test using any one reference sequences, institute Processing module is stated to be used for:

Steady invertibity inspection is carried out to the target arma modeling；

A kind of computer equipment, the computer equipment include:

At least one processor, and

The memory being connect at least one described processor；

Wherein, the memory is stored with the instruction that can be executed by least one described processor, described at least one The instruction that device is stored by executing the memory is managed, the method as described in any one of claim 1-21 is executed.

A kind of computer can storage medium,

The computer-readable recording medium storage has computer instruction, when the computer instruction is run on computers When, so that computer executes the method as described in any one of claim 1-21.

The embodiment of the present application realize the utility model has the advantages that

In conclusion background server determines reference zone at the first moment according to mobile data in the embodiment of the present application Total flow of the people, and use target arma modeling, total flow of the people according to reference zone at the first moment, to target area not Total flow of the people at the second moment come is predicted, wherein target arma modeling is established based on the following information content: at least Total flow of the people of one reference zone in each historical juncture, to target area the historical juncture accordingly lagged total flow of the people Influence.In this way, background server can be based on mobile data, to total flow of the people of each time sampling point in each region Be acquired, handle and update, thus data acquisition when be no longer influenced by regional limitation, can obtain each region more subject to The total flow of the people of true history and background server can use target arma modeling, the total stream of people of history based on reference zone The hysteresis quality of amount is more accurately predicted target area in total flow of the people of future time instance, and then provides timely, accurate and effective Early warning, improve system service quality and efficiency of service.

Detailed description of the invention

Fig. 1 is the flow diagram that target arma modeling is established in the embodiment of the present application；

Fig. 2 is the flow diagram of total flow of the people of predicted city hot spot region in the embodiment of the present application；

Fig. 3 is the illustrative view of functional configuration of the first background server in the embodiment of the present application；

Fig. 4 is the illustrative view of functional configuration of second of background server in the embodiment of the present application.

Specific embodiment

Method in the embodiment of the present application can specifically be used in background server in a mobile communication device In, below to do details description with to the embodiment of the present application in background server.

In order to realize the Accurate Prediction to total flow of the people of city hot spot region, and solution source existing in the prior art The problem that data cover is narrow and data acquisition equipment maintenance cost is high.In the embodiment of the present application, background server determines mesh Region is marked at least one on the influential reference zone of the total flow of the people in target area, then, background server is based on mobile number According to obtaining total flow of the people of at least one reference zone at the first moment, and input preset target autoregressive moving-average model (Auto-Regressive Moving Average Mode, ARMA), to target area the second following moment total stream of people Amount is predicted.Wherein, which is total flow of the people based at least one reference zone in each historical juncture, Influence to total stream of people of historical juncture of the target area after corresponding and establish.

In order to facilitate reading, first the portion of techniques term used in the embodiment of the present application is explained.

Target area sample changed sequence is referred to as target sequence below, target area is contained in target sequence The at the appointed time flow of the people of each time sampling point in section, wherein target area, which refers to, may will form stream of people's congestion Region, e.g., sight spot, amusement park etc.；

Reference zone sample changed sequence is referred to as reference sequences below, a reference is contained in reference sequences The flow of the people sequence of region at the appointed time each time sampling point in section, wherein reference zone is referred to target area The region that the flow of the people in domain impacts, e.g., subway station, bus station, railway station etc..

Under normal conditions, for the prediction of target sequence generally there are two types of conventional method,

The first: predicting target sequence using current time all reference sequences, be suitable for target sequence with Reference sequences the case where there are stronger correlativities.

Both sides relation can specifically be indicated using following formula:

Wherein,For target sequence, e.g., flow of the people of the Disney park at t o'clock, X_iIt (t) is a reference sequences, Such as, flow of the people of the railway station at t o'clock, a_i、ε₁For preset constant.

Second: being predicted using the historical data of target sequence itself, be suitable for target sequence and itself history number According to there is the case where stronger correlativity.It can specifically be indicated using following formula:

Wherein,For target sequence, e.g., flow of the people of the Disney park at t o'clock, Y (t-n) is going through for target sequence History data, e.g., flow of the people of the Disney park at t-n o'clock, a_n、ε₂For constant.

However, the flow of the people of certain target area at a time also will receive reference zone stream of people shunting in practical application The influence of (e.g., bus station, subway station etc.), it is clear that above two method does not all account for going through of including in reference sequences The hysteresis quality of history data.

For example, the flow of the people in Disney park will receive the influence of subway station, e.g., 3 o'clock subway station flow of the people, can be with The flow of the people in 3 thirty Disney parks can be impacted, i.e. shadow of the flow of the people of reference zone to the flow of the people of target area Sound could embody over time, become.

Therefore, in the embodiment of the present application, it is also contemplated that the hysteresis quality of part historical data in reference sequences introduces Granger Causality Test method is excavated on the influential dependent variable of target area flow of the people, and is closed based on the time cause and effect Connection feature improves arma modeling, predicts the flow of the people of target area future time instance.

As shown in fig.1, the detailed process for establishing target arma modeling based on dependent variable is as follows in the embodiment of the present invention:

Step 101: background server obtains at least two fingers of each time sampling point acquisition at the appointed time section Determine the mobile data in region, and based on the mobile data of acquisition determine respectively above-mentioned at least two specified regions each when Between sampled point total flow of the people, to obtain corresponding sample changed sequence respectively.

With any one time sampling point (hereinafter referred to as sampled point x) and any one specified region (hereinafter referred to as area For the x) of domain, in sampled point x using the mobile data in the x of region, and determine region x in sampled point based on the mobile data of acquisition When total flow of the people of x, background server can use but be not limited to following manner:

Firstly, background server can be from each base station cell obtained in the x of region in communication system server cluster Identification information (i.e. CELL-ID), and each base station cell is associated with to region x.

Secondly, background server can be based on the identification information of each base station cell of acquisition, from communication system server Acquired in cluster in each base station cell user data (e.g., user terminal location update message, user terminal and network side it Between interaction signaling message etc.).

Again, background server carries out duplicate removal processing and denoising to the user data of acquisition, to determine in the x of region Mobile subscriber's quantity.

In order to improve the precision of data, background server is additionally provided with preset time threshold, which indicates The maximum time that approach population is stopped in region x, for excluding the mobile subscriber that the residence time is less than the preset time threshold.

For example, around may have other just in Travel vehicle when acquiring the mobile subscriber's quantity of certain bus station's near zone On, the passenger on vehicle is also in pickup area, but these driving vehicles can sail out of the pickup area in a short time, so When determining the mobile subscriber's quantity of the pickup area, it should exclude this partial movement user.Therefore, by the way that preset time is arranged Threshold value to judge whether the residence time of mobile subscriber is less than preset time threshold, and excludes to stop from mobile subscriber's quantity The time is stayed to be less than the mobile subscriber of preset time threshold, to enable the mobile subscriber's quantity of the pickup area of acquisition more smart Really.

Finally, obtaining total flow of the people in the x of region based on the mobile subscriber's quantity in the x of region.

Specifically, it is based on mobile subscriber's quantity, it, can be with the city mobile subscriber when obtaining total flow of the people in the x of region It is referred to the accounting of total number of persons, the permeability (such as 0.7) of a mobile subscriber is preset, by the movement of region x obtained above Permeability of the number of users divided by mobile subscriber, the total flow of the people estimated.

Further, background server can use identical method, when acquiring corresponding on each time sampling point Between total flow of the people, in this way, the flow of the people sequence of each time sampling point of designated time period inner region x can be obtained, that is, sample Change sequence.Wherein, the time interval between time sampling point is referred to as time granularity, and the size of time granularity can be by Administrative staff's rule of thumb flexible setting, details are not described herein.

Such as: assuming that set 30 minutes for time granularity, then within 24 hours one day there are 48 time sampling points, So, the corresponding total flow of the people of 48 different time sampled points can be obtained for region x, it is assumed that the setting sampling period is 30 It, then, in 30 days, total flow of the people of 1440 different time sampled points can be obtained for region x, to constitute Sample changed sequence of the region x in this 30 days.

Step 102: background server chooses the first specified region as target area from above-mentioned at least two specified regions Domain, and using the corresponding sample changed sequence in target area as target area sample changed sequence, i.e. target sequence.

For example, it is assumed that needing total flow of the people for the first specified regional prediction future, then background server refers to first Region is determined as target area, and from the sample changed sequence of the multiple regions of acquisition, by the corresponding sample in the first specified region This change sequence is used as target area sample changed sequence, i.e. target sequence.

Step 103: other from above-mentioned at least two specified regions in addition to the first specified region of background server refer to Determine in region, choose at least one specified region as reference zone, and using the corresponding sample changed sequence of reference zone as Reference zone sample changed sequence, i.e. reference sequences.

For example, server can be from removing first after server determines the above-mentioned first specified region as target area In other specified regions outside secondary specified region, at least one specified region is chosen, as reference zone, and from the multiple of acquisition In the sample changed sequence in region, using the corresponding sample changed sequence of reference zone as reference zone sample changed sequence, i.e., Reference sequences.

Step 104: background server carries out stationarity detection and adjustment to target sequence and each reference sequences respectively.

Specifically, background server can be in the following ways when executing step 104:

A) background server carries out stationarity detection to target sequence and each reference sequences respectively, filters out and does not have The target sequence or/and reference sequences of stationarity.

Stationarity detection is carried out to target sequence and each reference sequences, that is, refers to detection target area and each ginseng Whether total flow of the people of the examination district domain within following a period of time can prolong along " form " of existing target sequence and reference sequences Continue down.

Since the premise that Granger causality is excavated is stable sample changed sequence, so, in order to guarantee Granger it is causal effectively, background server need to using extension Dickey-fowler (Augment Dickey-Fuller, ADF it) examines to the above-mentioned target sequence obtained and each reference sequences (for ease of description, being temporarily referred to as sample herein This change sequence) stationary test is carried out, filter out the sample changed sequence without stationarity.

Optionally, when carrying out ADF inspection, following regression equation can be used:

Wherein, Δ Z_tFor sample changed sequence, t is the serial number of time granularity, Z_t-1For the sample changed sequence at t-1 moment, ΔZ_tFor single order sample changed sequence, Δ Z_t-jFor the first-order difference sample changed sequence for lagging j time granularity, α is preset Constant, β, ρ, λ_jFor preset regression coefficient, P is maximum lag time granularity number, μ_tFor error term.

When carrying out stationarity detection to a sample changed sequence, need to construct corresponding DF (Dickey-Fuller) system Multidigit ADF test value is measured, if ADF test value is less than or equal to preset critical value, then it is assumed that the sample changed sequence has flat Stability, and retain the sample changed sequence, if ADF test value is greater than preset critical value, then it is assumed that the sample changed sequence is not Have stationarity, then needs to carry out co integration test.

B) background server is directed to the target sequence for not having stationarity or/and reference sequences further progress assists whole relationship Determine, assists whole relationship to not having and do not have the target sequence of stationarity or/and reference sequences carry out stationarity conversion.

Wherein, do not have the target sequence of stationarity or/and reference sequences include the following three types situation:

Target sequence has stationarity, and reference sequences do not have stationarity；Target sequence does not have stationarity, reference sequences With stationarity；Target sequence and reference sequences all do not have stationarity.

Specifically, considering that if reference sequences and target sequence, can also further progresss in the presence of whole relationship is assisted Granger Causality Test, thus also need to above-mentioned without stationarity sample changed sequence (including target sequence or/and Reference sequences) it carries out that whole relationship is assisted to determine generally to carry out co integration test using EG (Engle-Granger) two-step method.Two step of EG Method, the first step calculate residual sequence ε_t, second step, the whole property of checklist, if ε_tIt is then that association is whole for stationary sequence.

It is specific as follows:

Step 1: being established using least square method to the reference sequences and target sequence that do not have stationarity and assisting whole recurrence Equation calculates residual sequence ε_tIt is as follows:

Y_t=alpha+beta X_t+ε_t

ε_t=Y_t-α-βX_t

Wherein, X_tTotal flow of the people for reference zone in t moment, Y_tTotal flow of the people for target area in t moment, ε_tFor Residual sequence, α, β are preset constant.

Second step, by assisting whole regression equation calculation to go out residual sequence ε_t, to residual sequence ε_tADF inspection is carried out, is judged residual Difference sequence ε_tWhether stationarity is had, if so, determining reference sequences X_tWith target sequence Y_tWith the whole relationship of association, otherwise, it determines Reference sequences X_tWith target sequence Y_tWithout the whole relationship of association.

C) stationarity conversion is carried out to not having the target sequence for assisting whole relationship and reference sequences.

For not having stationarity and there is no the target sequences and reference sequences of assisting whole relationship, using differential transformation side Method carries out first-order difference or higher order differential transformation until sequence meets stationarity.Step 105: background server is directed to target Sequence is respectively adopted each reference sequences and carries out Granger Causality Test, and filtering out can be as target sequence dependent variable Reference sequences.

In the embodiment of the present invention, a reference sequences can refer to as target sequence dependent variable, in a reference sequences Data variation, the data in target sequence can be impacted, can be real-time influence, e.g., 3 o'clock in reference sequences Data impact the 3 o'clock data in target sequence, are also possible to hysteresis quality influence, e.g., the 3 o'clock number in reference sequences It is impacted according to 3 half datas in target sequence.

Specifically, can verify whether each reference sequences are to cause target sequence respectively by Granger Causality Test Whether the Granger reason for arranging variation, i.e., be target sequence dependent variable.

By taking reference sequences X and target sequence Y as an example, detailed process is as follows for Grannger Causality Test:

A) assume " reference sequences X is not the Granger reason for causing target sequence Y to change ", and default following two is returned Return equation:

Without constrained regression equation (u):

There is constrained regression equation (r):

Wherein, Y_tIndicate target sequence, X_t-iIndicate that reference sequences, i indicate lag time granularity number, p indicates target sequence The lag time granularity maximum number of column, q indicate the lag time granularity maximum number of reference sequences, and p and q can usually be obtained It is somewhat larger, ε_tFor white noise, α_o, β indicate constant term.

B) it utilizes the residual sum of squares (RSS) RSSu of above-mentioned no constrained regression equation (u) and has the residual error of constrained regression equation (r) Quadratic sum RSSr constructs F statistic:

Wherein, n indicate sample size (e.g., carried out daily in 30 days 48 times sampling after obtain 1440 total flows of the people, that N=1440), p indicates that the lag time granularity maximum number of target sequence, q indicate the lag time granularity of reference sequences most Big figure.

Judge whether F statistic is greater than the response critical value F that F is distributed under given level of signifiance α_α(q, n-p-q-1), if It is that then null hypothesis is invalid, that is, determines that reference sequences X is the Granger reason of target sequence Y, that is, think that reference sequences X is pair The dependent variable that target sequence Y is impacted；Otherwise, then null hypothesis is set up, that is, thinking reference sequences X not is target sequence Y Granger reason, that is, thinking reference sequences X not is the dependent variable impacted to target sequence Y.

On the other hand, when the Granger reason that reference sequences X is target sequence Y, and target sequence Y non-reference sequence X In the case where Granger reason, it is believed that there are the individual event causalities of reference sequences X to target sequence Y.

Step 106: background server based on target sequence, can as the reference sequences of target sequence dependent variable, establish Target autoregressive moving-average model (ARMA) model.

Specifically, may include following operation when executing step 106:

Firstly, based on target sequence and initial ARMA model can be established as the reference sequences of target sequence dependent variable.

Optionally, the initial ARMA model for having merged the cause and effect feature between target sequence and each reference sequences can be adopted It is indicated with following formula:

Wherein, Y_tFor total flow of the people of target area t moment, Y_t-iFor target area t-i moment total flow of the people, X_t-i(j) For total flow of the people on the influential reference sequences of target sequence data at the t-i moment, i indicates lag time granularity number, p table Show that the lag time granularity maximum number of target sequence, q indicate that the lag time granularity maximum number of reference sequences, j are reference The label in region, m are the sum of reference zone, and t indicates the moment, and α, β are preset parameter.

Secondly, carrying out parameter Estimation and model order for established initial ARMA model.

Specifically, parameter Estimation is the most critical part of ARMA modeling, directly affects the fitting performance of arma modeling.This Shen Parameter Estimation, the i.e. premise in the specific value of known sample change sequence please be carried out using least squares estimate in embodiment Under the conditions of, regression parameter α is found out using nonlinear least square method_i、β_i, so that residual sum of squares (RSS) E reaches minimum.

Optionally, the residual sum of squares (RSS) that the expression formula based on above-mentioned arma modeling defines is as follows:

By above-mentioned formula, α can be obtained_i、β_iEstimated value, also referred to as estimates of parameters.

On the other hand, optionally, in the embodiment of the present application, minimum information criterion (Akaike information is selected Criterion, AIC) function progress model order, it is main to determine regression order (the i.e. model order used in initial ARMA model Number).

Under normal conditions, the initial value for the model order that initial ARMA model uses is smaller, meanwhile, model can be set Then the upper limit value of order is successively increased the model order of initial ARMA model since initial value to upper limit value, every increase Once, the value of primary red pond information criterion (Akaike information criterion, abbreviation AIC) is calculated；Finally, choosing Initial ARMA model is arranged in the corresponding model order of the smallest AIC criterion functional value of value and above-mentioned parameter estimated value, obtains most Good model of fit, alternatively referred to as target arma modeling.

Wherein, AIC is a kind of standard of measure statistical models fitting Optimality.

AIC=2k-2ln (L)

Wherein, k is model parameter number, and L is likelihood function.Best model is selected from one group of alternative model When, generally select the smallest model of AIC.

So far, target arma modeling, which has been established, finishes, and can input target ARMA mould using reference sequences as input value Type, so that total flow of the people to target sequence in the following specified time point is predicted.

It further, can also one of in the following ways or any combination is to mesh after establishing target arma modeling Mark arma modeling is tested, and is optimized and revised:

Mode one: level of signifiance Student T statistics control is carried out to target arma modeling.

Specifically, estimates of parameters can be set to zero, then tested with Student T statistic.

Mode two: steady invertibity inspection is carried out to target arma modeling.

Specifically, whether 1 can be smaller than with the reciprocal of test-target arma modeling characteristic root.It is examined by steady invertibity It can guarantee the stationarity of target arma modeling.

Mode three: residual sequence white noise verification is carried out to target arma modeling.

Specifically, common residual sequence white noise verification method is auto-relativity function method, i.e., it is by inspection residual sequence It is no that there are autocorrelations to judge in target arma modeling with the presence or absence of white noise.

After target arma modeling uses above-mentioned three kinds of modes, when still cannot pass through inspection, need further to correct The estimates of parameters of target arma modeling, alternatively, the model order of modification target arma modeling, until target arma modeling passes through It examines.

It based on the above embodiment, can be to target sequence when future is specified after completing the optimization of target arma modeling Between total flow of the people for putting predicted.As shown in Fig. 2, being carried out to total stream of people of city hot spot region pre- in the embodiment of the present application The detailed process of survey:

Step 201: background server determine target area and at least one reference zone, wherein reference zone it is total Flow of the people impacts total flow of the people of target area；

For example, it is assumed that total flow of the people to Disney park is predicted, then background server is determined by external command Disney park is target area, and using other multiple specified regions in addition to Disney park as reference zone, wherein ginseng Total flow of the people in examination district domain can cause hysteresis quality to influence total flow of the people of target area, for example, the ground near Disney park Iron station, bus station t moment total flow of the people, can total flow of the people in Disney park to the t+i moment impact.

Step 202: background server obtains total flow of the people of at least one reference zone based on mobile data.

The method that background server can use total flow of the people acquisition in above-mentioned steps 101 to specified region, at least Total flow of the people of one reference zone is acquired, and is not being repeated herein.

Step 203: total flow of the people of the background server according at least one above-mentioned reference zone at the first moment, using pre- If total flow of the people of the target arma modeling to target area at the second following moment is predicted.

For example, having obtained following information: total flow of the people meeting of the reference zone at the t-3 moment by above-mentioned target arma modeling Total flow of the people of target area t moment is impacted, it is possible to which total flow of the people of reference zone t moment is inputted the mesh Mark arma modeling, it is predicted that total flow of the people at target area t+3 moment, wherein t is current time, t-3 be current time t to At the time of 3 time granularities of preceding backtracking, at the time of t+3 current time t is after 3 time granularity of standard.

Based on the above embodiment, refering to Fig. 3, the embodiment of the present application provides a kind of equipment (e.g., backstage clothes for predicting flow of the people Business device), include at least input module 301, acquisition module 302 and prediction module 303, in which:

Input module 301, for determining target area and at least one reference zone, wherein total people of the reference zone Flow impacts total flow of the people of target area.

Acquisition module 302, for acquiring fixed reference zone in total flow of the people at the first moment by mobile data.

Prediction module 303, for total flow of the people of at least one reference zone at the first moment according to, using default Total flow of the people of the target arma modeling to the target area at the second following moment predict；Wherein, the target Arma modeling is total flow of the people based at least one described reference zone in each historical juncture, to the target area in phase The influence of the total flow of the people for the historical juncture that should be lagged and establish.

Optionally, the equipment of above-mentioned prediction flow of the people further comprises analysis module 304 and processing module 305,

The acquisition module 302 is further used for:

Before determining target area and at least one reference zone, acquisition module 302 is obtained in designated time period at least The sample changed sequence in two specified regions, wherein the sample changed sequence indicates the specified region when described specified Between each time sampling point in section total flow of the people；

The analysis module 304 is used for:

The processing module 305 is used for:

Optionally, when obtaining the sample changed sequence in any one specified region in designated time period, the acquisition module 302 are further used for:

Optionally, based on the mobile data of acquisition, determine any one described specified region in any one time sampling When total flow of the people of point, the acquisition module 302 is used for:

Optionally, after carrying out duplicate removal processing and denoising to the user data of acquisition, determine it is described any one Before the mobile subscriber's quantity of any one time sampling point, the acquisition module 302 is further used in specified region:

Optionally, using Granger Causality Test, at least one ginseng for causing hysteresis quality to influence target sequence is filtered out When examining sequence, the processing module 305 is specifically used for；

Optionally, described when carrying out stationarity detection and adjustment respectively to the target sequence and the reference sequences Processing module 305 is specifically used for:

Optionally, for the target sequence, when carrying out Granger Causality Test using any one reference sequences, institute Processing module 305 is stated for including:

Optionally, described when establishing target arma modeling based on the target sequence and at least one described reference sequences Processing module 305 is specifically used for:

Optionally, parameter Estimation and model order are carried out to the established initial ARMA model, obtains corresponding mesh When marking arma modeling, the processing module 305 is specifically used for:

Optionally, after obtaining the target arma modeling, the processing module 305 is further used for,

Steady invertibity inspection is carried out to the target arma modeling；

On the other hand, as shown in figure 4, the embodiment of the present application also provides a kind of equipment for establishing flow of the people prediction model, packet Include acquisition module 302, analysis module 304 and processing module 305, in which:

Acquisition module 302, for obtaining the sample changed sequence at least two specified regions in designated time period, wherein The sample changed sequence indicates total stream of people of each time sampling point of the specified region in the designated time period Amount.

Analysis module 304, it is corresponding for determining target area and reference zone according to external command, and by target area Sample changed sequence as target sequence, using the corresponding sample changed sequence of reference zone as reference sequences.

Processing module 305 filters out for using Granger Causality Test and causes hysteresis quality to influence target sequence At least one reference sequences, and target arma modeling is established based on the target sequence and at least one described reference sequences.

Optionally, when obtaining the sample changed sequence in any one specified region in designated time period, the acquisition module 302 are specifically used for:

Optionally, for the target sequence, when carrying out Granger Causality Test using any one reference sequences, institute Processing module 305 is stated to be used for:

Optionally, after obtaining the target arma modeling, the processing module 305 is further used for:

Steady invertibity inspection is carried out to the target arma modeling；

A kind of computer equipment, the computer equipment include:

At least one processor, and

The memory being connect at least one described processor；

Wherein, the memory is stored with the instruction that can be executed by least one described processor, described at least one The instruction that device is stored by executing the memory is managed, the method as described in any one of the above is executed.

A kind of computer can storage medium,

The computer-readable recording medium storage has computer instruction, when the computer instruction is run on computers When, so that computer executes the method as described in any one of the above.

Further, background server is acquired total flow of the people based on mobile data, hands over website relative to based on rail The prior arts such as gate data, wifi hotspot location data, infrared sensing data, reduce economic cost, realize more efficient People flow rate statistical.

Further, background server also excludes to move using preset time threshold when acquiring mobile subscriber's quantity Mobile subscriber of the residence time less than preset time threshold is excluded in number of users, to enable the movement of the pickup area of acquisition Number of users is more accurate

On the other hand, target arma modeling is that the Granger causality relationship based on target area and reference zone is built Vertical, and the Granger causality of target area and reference zone, it is the historical sample according to target area and reference zone Change sequence is by obtain repetition training, and therefore, target arma modeling has efficiently and accurately been excavated to the total stream of people in target area Influential dependent variable is measured, so as to more accurately determine total flow of the people of reference zone to the total flow of the people in target area It influences.

Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can It realizes by means of software and necessary general hardware platform, naturally it is also possible to pass through hardware.Based on this understanding, on Stating technical solution, substantially the part that contributes to existing technology can be embodied in the form of software products in other words, should Computer software product may be stored in a computer readable storage medium, such as ROM/RAM, magnetic disk, CD, including several fingers It enables and using so that a computer equipment (can be personal computer, server or the network equipment etc.) executes each implementation Method described in certain parts of example or embodiment.

Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although Present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: it still may be used To modify the technical solutions described in the foregoing embodiments or equivalent replacement of some of the technical features； And these are modified or replaceed, technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution spirit and Range.

Claims

1. a kind of method for predicting flow of the people characterized by comprising

Determine target area and at least one reference zone, wherein total flow of the people of the reference zone is to the total of target area Flow of the people impacts；

Total flow of the people according at least one described reference zone at the first moment, using preset target autoregressive moving average Total flow of the people of the arma modeling to the target area at the second following moment is predicted；

Wherein, the target arma modeling is total flow of the people based at least one described reference zone in each historical juncture, The target area is established in the influence of total flow of the people of the historical juncture accordingly lagged.

2. the method as described in claim 1, which is characterized in that before determining target area and at least one reference zone, Further comprise:

Obtain the sample changed sequence at least two specified regions in designated time period, wherein the sample changed sequence indicates Total flow of the people of each time sampling point of the specified region in the designated time period；

Target area and reference zone are determined according to external command, and using the corresponding sample changed sequence in target area as mesh Sequence is marked, using the corresponding sample changed sequence of reference zone as reference sequences；

Using Granger Granger Causality Test, at least one for causing hysteresis quality to influence target sequence is filtered out with reference to sequence Column；

3. method according to claim 2, which is characterized in that obtain the sample in any one specified region in designated time period Change sequence specifically includes:

Obtain the movement in any one specified region described in each time sampling point acquisition in the designated time period Data；

Mobile data based on acquisition determines any one described specified region in total stream of people of each time sampling point respectively Amount, to obtain corresponding sample changed sequence.

4. method as claimed in claim 3, which is characterized in that the mobile data based on acquisition, determine it is described any one refer to Region is determined in total flow of the people of any one time sampling point, comprising:

The identification information of each base station cell in any one specified region is obtained, and each base station cell is associated with to institute State any one specified region；

The identification information of each base station cell based on acquisition acquires each base station cell in any one described time sampling point User data；

Duplicate removal processing and denoising are carried out to the user data of acquisition, to determine any one described specified region at described It anticipates the mobile subscriber's quantity of a time sampling point；

Based on any one described specified region in the mobile subscriber's quantity of any one time sampling point, reference is preset Accounting relationship between mobile subscriber's quantity and total flow of the people, any one described specified region are adopted in any one described time Total flow of the people of sampling point.

5. method as claimed in claim 4, which is characterized in that carry out duplicate removal processing and denoising to the user data of acquisition Later, in any one determining described specified region before the mobile subscriber's quantity of any one time sampling point, into One step includes:

From the mobile subscriber's quantity, exclusion residence time in any one described specified region is less than preset time threshold Mobile subscriber.

6. such as the described in any item methods of claim 2-5, which is characterized in that use Granger Causality Test, filter out pair At least one reference sequences that target sequence causes hysteresis quality to influence, specifically include；

For the target sequence, each reference sequences is respectively adopted and carries out Granger Causality Test, filters out to the mesh Mark sequence causes hysteresis quality to influence, can be as at least one reference sequences of the target sequence dependent variable.

7. method as claimed in claim 6, which is characterized in that carried out respectively to the target sequence and the reference sequences Stationarity detection and adjustment, specifically include:

Stationarity detection is carried out to the target sequence and each reference sequences respectively, filters out the target without stationarity Sequence or/and reference sequences；

Whole relationship is assisted to determine for the target sequence for not having stationarity or/and the reference sequences further progress, sieve Select the target sequence or/and the reference sequences for not having and assisting whole relationship；

For not having stationarity and there is no the target sequences and the reference sequences of assisting whole relationship, using differential transformation Method carries out first-order difference or higher order differential transformation until sequence meets stationarity.

8. method as claimed in claim 6, which is characterized in that the target sequence is directed to, using any one reference sequences Carry out Granger Causality Test, comprising:

According to the target sequence and any one described reference sequences construction without constrained regression equation, according to the target sequence It is configured with constrained regression equation；

Residual sum of squares (RSS) RSSu and the residual sum of squares (RSS) for having constrained regression equation using the no constrained regression equation RSSr constructs F statistic:

Wherein, n indicates that sample size, p indicate that the lag time granularity maximum number of target sequence, q indicate the stagnant of reference sequences Time granularity maximum number afterwards.

Judge whether F statistic is greater than the response critical value that F is distributed under given level of signifiance α；If so, determination is described any one A reference sequences are the dependent variables impacted to the target sequence, otherwise, it determines any one described reference sequences are not The dependent variable that the target sequence is impacted.

9. such as the described in any item methods of claim 2-5, which is characterized in that be based on the target sequence and described at least one A reference sequences establish target arma modeling, specifically include:

Parameter Estimation and model order are carried out for the established initial ARMA model, obtains corresponding target ARMA mould Type.

10. method as claimed in claim 9, which is characterized in that carry out parameter to the established initial ARMA model and estimate Meter and model order, obtain corresponding target arma modeling, specifically include:

Based on the target sequence and at least one reference sequences, parameter Estimation is carried out using least squares estimate, Obtaining to enable makes residual sum of squares (RSS) reach the smallest estimates of parameters；

The model order of the initial ARMA model is successively increased since preset initial value to preset upper limit value；Its In, it is every to increase once, calculate a minimal information AIC criterion functional value；

It chooses the corresponding model order of the smallest AIC criterion functional value of value and initial ARMA mould is arranged in the estimates of parameters Type obtains the target arma modeling.

11. such as the described in any item methods of claim 2-5, which is characterized in that after obtaining the target arma modeling, into One step includes,

Steady invertibity inspection is carried out to the target arma modeling；

12. a kind of method for establishing flow of the people prediction model characterized by comprising

13. method as claimed in claim 12, which is characterized in that obtain the sample in any one specified region in designated time period This change sequence specifically includes:

14. method as claimed in claim 13, which is characterized in that the mobile data based on acquisition, determine it is described any one Total flow of the people of the specified region in any one time sampling point, comprising:

15. method as claimed in claim 14, which is characterized in that carried out at duplicate removal processing and denoising to the user data of acquisition After reason, in any one determining described specified region before the mobile subscriber's quantity of any one time sampling point, Further comprise:

16. such as the described in any item methods of claim 12-15, which is characterized in that use Granger Causality Test, filter out On at least one reference sequences that target sequence causes hysteresis quality to influence, specifically include；

17. the method described in claim 16, which is characterized in that the target sequence and the reference sequences respectively into The detection of row stationarity and adjustment, specifically include:

It carries out that whole relationship is assisted to determine for the target sequence or/and the reference sequences for not having stationarity, filters out not With the target sequence or/and the reference sequences for assisting whole relationship；

It assists whole relationship to not having and does not have the target sequence of stationarity or/and the reference sequences carry out stationarity and turn It changes.

18. the method described in claim 16, which is characterized in that background server is directed to the target sequence, using any One reference sequences carries out Granger Causality Test, comprising:

19. such as the described in any item methods of claim 12-15, which is characterized in that based on the target sequence and it is described at least One reference sequences establishes target arma modeling, specifically includes:

20. method as claimed in claim 19, which is characterized in that carry out parameter to the established initial ARMA model and estimate Meter and model order, obtain corresponding target arma modeling, specifically include:

Using the specific value of the known target sequence and at least one reference sequences, using least squares estimate Parameter Estimation is carried out, obtaining to enable makes residual sum of squares (RSS) reach the smallest estimates of parameters；

21. such as the described in any item methods of claim 12-15, which is characterized in that after obtaining the target arma modeling, Further comprise,

Steady invertibity inspection is carried out to the target arma modeling；

22. a kind of equipment for predicting flow of the people, which is characterized in that the equipment includes:

Input module, for determining target area and at least one reference zone, wherein total flow of the people pair of the reference zone Total flow of the people of target area impacts；

Prediction module, for total flow of the people of at least one reference zone at the first moment according to, using preset target Total flow of the people of the arma modeling to the target area at the second following moment is predicted；Wherein, the target ARMA mould Type is total flow of the people based at least one described reference zone in each historical juncture, is accordingly being lagged to the target area Historical juncture total flow of the people influence and establish.

23. a kind of equipment for establishing flow of the people prediction model, which is characterized in that the equipment includes:

Acquisition module, for obtaining the sample changed sequence at least two specified regions in designated time period, wherein the sample Change sequence indicates total flow of the people of each time sampling point of the specified region in the designated time period；

Analysis module, for determining target area and reference zone according to external command, and by the corresponding sample in target area Change sequence is as target sequence, using the corresponding sample changed sequence of reference zone as reference sequences；

Processing module filters out at least one for causing hysteresis quality to influence target sequence for using Granger Causality Test Reference sequences, and target arma modeling is established based on the target sequence and at least one described reference sequences.

24. a kind of computer equipment, which is characterized in that the computer equipment includes:

At least one processor, and

The memory being connect at least one described processor；

Wherein, the memory is stored with the instruction that can be executed by least one described processor, at least one described processor By executing the instruction of the memory storage, the method as described in any one of claim 1-21 is executed.

25. a kind of computer can storage medium, it is characterised in that:

The computer-readable recording medium storage has computer instruction, when the computer instruction is run on computers, So that computer executes the method as described in any one of claim 1-21.