CN104464291A - Traffic flow predicting method and system - Google Patents

Abstract

The embodiment of the invention provides a traffic predicting method and system. The method comprises the steps that current traffic flow data are acquired, wherein the current traffic flow data comprise current traffic flow data of a road segment to be measured and current traffic flow data road segments related to the road segment to be measured; a first predicting model is determined; the current traffic flow data are input to the first predicting model, and a predicting traffic flow value of the road segment to be measured is obtained, wherein the first predicting model carries out discrete processing on historical traffic flow data by acquiring the historical traffic flow data to obtain discrete historical data, a space-time association rule is obtained by mining the discrete historical data, a set of the road segments related to the road segment to be measured is obtained through the space-time association rule, and the historical traffic flow data of the set of the road segments are generated in a training mode through a preset method.

Description

A kind of traffic flow forecasting method and system

Technical field

The present invention relates to traffic flow forecasting field, particularly relate to a kind of traffic flow forecasting method and system.

Background technology

Along with the development of social economy and transportation, the congested in traffic traffic problems that wait more and more show especially out, have become the problem of global common concern.One of the basis of Urban traffic control and route guidance is not only in prediction for traffic flow, or the gordian technique of solving road congestion problems.If can predict the flow motor in the network of communication lines on each branch road accurately, so we can use planing method reasonably to optimize traffic flow, thus make the utilization factor of road reach maximum, also can solve part congestion problems.

At present, normal employing prediction of short-term traffic volume model carrys out predicting traffic flow amount, have compared with the short-term traffic flow forecast model of getting up early: autoregressive model (AR), moving average model (MA), autoregressive moving-average model (ARMA), history averaging model (HA) and Box-Cox model etc., along with the development in this field, Forecasting Methodology is constantly tending towards accurate, occur many more complicated, the forecast model that precision is higher, generally can be divided into two classes: a class is the forecast model based on the mathematical method that mathematical statistics and infinitesimal analysis etc. are traditional, mainly comprise: time series models, Kalman filter model, Partial Linear Models etc., Equations of The Second Kind is the Short-term Forecasting Model formed for main research means with modern science and technology and method (as analogue technique, neural network, analogue technique), this kind of method does not pursue proper mathematical derivation and clear and definite physical significance, more pay attention to and the closeness of fit of the real magnitude of traffic flow, this kind of method mainly comprises nonparametric Regression Model, ARIMA algorithm, method, analysis of spectrum and the multiple hybrid model for short-term load forecasting etc. relevant to neural network based on wavelet theory.

Predicting traffic flow amount is carried out by using short-term traffic flow forecast model, effectively can predict the concrete numerical value of section to be measured magnitude of traffic flow in certain future time, accurately predicting is carried out to section to be measured magnitude of traffic flow in certain future time, but, non-linear due to traffic system, the features such as complicacy and uncertainty, existing prediction of short-term traffic volume model is easily by the impact of extraneous disturbing factor immediately, because the impact of external world's disturbing factor immediately, existing prediction of short-term traffic volume model is used to carry out predicting traffic flow amount, make the Stability and veracity of the predicting traffic flow value obtained low.

Summary of the invention

In view of this, the embodiment of the present invention provides a kind of traffic flow forecasting method and system, to solve in prior art because the impact of extraneous disturbing factor immediately, use existing prediction of short-term traffic volume model to carry out predicting traffic flow amount, make the problem that the Stability and veracity of the predicting traffic flow value obtained is low.

For achieving the above object, the embodiment of the present invention provides following technical scheme:

A kind of traffic flow forecasting method, comprising:

Obtain current traffic flow data, described current traffic flow data comprise the current traffic flow data in section to be measured and the Current traffic flow data with tool relevance section, section to be measured;

Determine the first forecast model;

By described first forecast model of described current traffic flow data input, obtain the predicting traffic flow value in section to be measured;

Wherein, described first forecast model is by obtaining historical traffic flows data, historical traffic flows data described in discrete processes, obtain discretize historical data, data mining is carried out to described discretize historical data and obtains spacetime correlation rule, obtained gathering with the section of section to be measured tool relevance by described spacetime correlation rule, the historical traffic flow data gathered in described section is generated by preordering method training.

Wherein, describedly determine that the first forecast model comprises:

Current traffic flow data described in discrete processes, obtain discretize current data;

Described discretize current data is inputted the second forecast model, obtains magnitude of traffic flow trend;

According to described section to be measured identification information, current time information and between traffic flow trend and the first forecast model corresponding relation determine the first forecast model matched with the second forecast model.

Wherein, historical traffic flow data after sliding-model control, by obtaining historical traffic flows data, historical traffic flows data described in sliding-model control, is carried out the spacetime correlation rule composition that data mining obtains by described second forecast model.

Wherein, according to current traffic flow data described in following steps discrete processes:

By the current traffic flow data normalization of acquisition in [0,1] interval;

Discretize is carried out to current traffic flow data after described normalized, obtains discretize current data;

Wherein, according to historical traffic flows data described in following steps discrete processes:

The historical traffic circuit-switched data of acquisition is normalized in [0,1] interval;

Discretize is carried out to historical traffic flows data after described normalized, obtains discretize historical data.

Wherein, by evolutionary programming algorithm, data mining is carried out to described discretize historical data and obtain spacetime correlation rule;

Wherein, carry out data mining by evolutionary programming algorithm to described discretize historical data to obtain spacetime correlation rule and comprise:

Determine the number M in section in discretize historical data, generate the individuality with M gene;

Get the individuality composition initial population being not less than predetermined number, and calculate the fitness value of each individuality in described initial population according to discretize historical data;

Fitness value according to individuality each in described initial population carries out Swarm Evolution to described initial population;

Qualified correlation rules all in each individuality after Swarm Evolution are exported, obtains spacetime correlation rule.

Wherein, the individuality that described generation has a M gene comprises:

Generate the determination gene that a span is [-T ,-1], T is the quantity of traffic flow data sampling time interval in historical traffic flows data;

Generate the random gene that M-1 span is [0, T], obtain the individuality with M gene.

Wherein, the described fitness value according to individuality each in described initial population carries out Swarm Evolution to described initial population and comprises:

Two individualities in initial population described in Stochastic choice, select the larger individuality of wherein fitness value and enter the next generation, until select the new colony of predetermined number individuality composition;

From described new colony Stochastic choice go out predetermined number individuality complete chiasma operation;

From described new colony Stochastic choice go out predetermined number individuality complete chromosomal variation operation;

Repeat new colony generating run, chiasma operation and chromosomal variation operation, until colony's fitness value convergence.

Wherein, described current traffic flow data and described historical traffic flows data are and filter out abnormal data, and carry out the data of correction to the abnormal data filtered out.

Wherein, Current traffic stream packets that is described and tool relevance section, section to be measured is drawn together:

Determine section to be measured and spacetime correlation rule;

The section with section to be measured tool relevance is obtained according to described section to be measured and spacetime correlation rule.

A kind of traffic flow forecasting system, comprising: acquisition module, model determination module, prediction module and forecast model generation module; Wherein,

Described acquisition module, for obtaining current traffic flow data, described current traffic flow data comprise the current traffic flow data in section to be measured and the Current traffic flow data with tool relevance section, section to be measured;

Described model determination module, for determining the first forecast model;

Described prediction module, for by described first forecast model of described current traffic flow data input, obtains the predicting traffic flow value in section to be measured;

Described forecast model generation module, for obtaining historical traffic flows data, historical traffic flows data described in sliding-model control, historical traffic flow data after sliding-model control is carried out data mining and obtains spacetime correlation rule, obtained gathering with the section of section to be measured tool relevance by described spacetime correlation rule, the historical traffic flow data gathered in described section is by preordering method training generation first forecast model.

Wherein, described model determination module comprises: forecast model determining unit and forecast model generation unit; Wherein,

Described forecast model determining unit, for current traffic flow data described in sliding-model control, obtains discretize current data; Described discretize current data is inputted the second forecast model, obtains magnitude of traffic flow trend; The first forecast model matched with the second forecast model is determined according to corresponding relation between described traffic flow trend and the first forecast model;

Described forecast model generation unit, for obtaining historical traffic flows data, historical traffic flows data described in sliding-model control, carry out the spacetime correlation rule that data mining obtains, form the second forecast model by historical traffic flow data after sliding-model control.

Based on technique scheme, a kind of traffic flow forecasting method that the embodiment of the present invention provides and system, after acquisition current traffic flow data, by determining the first forecast model, by the first forecast model that the input of the Current traffic flow data of acquisition is determined, then from the first forecast model determined, obtain the predicting traffic flow value in section to be measured, wherein, described current traffic flow data comprise the current traffic flow data in section to be measured and the Current traffic flow data with tool relevance section, section to be measured, wherein, described first forecast model is by obtaining historical traffic flows data, historical traffic flows data described in discrete processes, obtain discretize historical data, data mining is carried out to described discretize historical data and obtains spacetime correlation rule, obtained gathering with the section of section to be measured tool relevance by described spacetime correlation rule, the historical traffic flow data gathered in described section is generated by preordering method training.Because in the traffic flow forecasting method that this transmission embodiment provides, the first forecast model used obtains spacetime correlation rule through data mining for history communication flow data, the history communication flow data message in the section needed for this moment correlation rule obtains is obtained by preordering method training, data mining is carried out to historical traffic data, by carrying out global optimum's search to mass data, obtain frequent item set, improve the accuracy of predicting traffic flow value.

Accompanying drawing explanation

In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, be briefly described to the accompanying drawing used required in embodiment or description of the prior art below, apparently, accompanying drawing in the following describes is only embodiments of the invention, for those of ordinary skill in the art, under the prerequisite not paying creative work, other accompanying drawing can also be obtained according to the accompanying drawing provided.

The process flow diagram of the circulation method for predicting that Fig. 1 provides for the embodiment of the present invention;

The method flow diagram of the first forecast model is determined in the traffic flow forecasting method that Fig. 2 provides for the embodiment of the present invention;

The method flow diagram of discrete processes current traffic flow data in the traffic flow forecasting method that Fig. 3 provides for the embodiment of the present invention;

The method flow diagram of discrete processes historical traffic flows data in the traffic flow forecasting method that Fig. 4 provides for the embodiment of the present invention;

By evolutionary programming algorithm, the method flow diagram that data mining obtains spacetime correlation rule is carried out to discretize historical data in the traffic flow forecasting method that Fig. 5 provides for the embodiment of the present invention;

The method flow diagram with the individuality of M gene is generated in the traffic flow forecasting method that Fig. 6 provides for the embodiment of the present invention;

According to the fitness value of individuality each in initial population, described initial population is carried out to the method flow diagram of Swarm Evolution in the traffic flow forecasting method that Fig. 7 provides for the embodiment of the present invention;

The method flow diagram with tool relevance section, section to be measured is obtained in the traffic flow forecasting method that Fig. 8 provides for the embodiment of the present invention;

The system chart of the traffic flow forecasting system that Fig. 9 provides for the embodiment of the present invention;

The structural representation of model determination module 200 in the traffic flow forecasting system that Figure 10 provides for the embodiment of the present invention.

Embodiment

Below in conjunction with the accompanying drawing in the embodiment of the present invention, be clearly and completely described the technical scheme in the embodiment of the present invention, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art, not making the every other embodiment obtained under creative work prerequisite, belong to the scope of protection of the invention.

The process flow diagram of the circulation method for predicting that Fig. 1 provides for the embodiment of the present invention, by carrying out data mining to historical traffic data, obtain training the first forecast model obtained to carry out predicting traffic flow amount with the historical traffic flow data in tool relevance section, section to be measured, improve the accuracy of predicting traffic flow value, with reference to Fig. 1, this circulation method for predicting can comprise:

Step S100: obtain current traffic flow data, described current traffic flow data comprise the current traffic flow data in section to be measured and the Current traffic flow data with tool relevance section, section to be measured;

Wherein, it should be noted that, when there is no the section with section to be measured tool relevance, then think that the Current traffic flow data obtained with tool relevance section, section to be measured is empty data, also can think, when there is no the section with section to be measured tool relevance, only including the current traffic flow data in section to be measured in the current traffic flow data of acquisition, and not comprising and the Current traffic flow data treating this tool relevance section, survey section.

By obtaining current communication flow data, current time information, section to be measured identification information and current traffic flow value data information can be determined from the current traffic flow data obtained, optionally, section to be measured identification information can be title or the number information in section to be measured.

Optionally, with until this survey section tool relevance section can for be confirm the title in section to be measured or number information and in advance with the spacetime correlation rule learnt after, according to surveying section that is that section and spacetime correlation rule obtain and section to be measured tool relevance.

Step S110: determine the first forecast model, described first forecast model is by obtaining historical traffic flows data, historical traffic flows data described in discrete processes, obtain discretize historical data, data mining is carried out to described discretize historical data and obtains spacetime correlation rule, obtained gathering with the section of section to be measured tool relevance by described spacetime correlation rule, the historical traffic flow data gathered in described section is generated by preordering method training;

Historical traffic flows data comprise the information such as system unit time, current time, time interval quantity, section mark and the magnitude of traffic flow, wherein, the time interval of system unit time and data statistics, can by User Defined, be generally in units of 5 minutes or 10 minutes, minimum value is 1 minute, and the magnitude of traffic flow statistics being less than a minute does not possess statistical significance.

Such as, as shown in table 1, the historical traffic flows data of some day:

table 1the historical traffic flows data form of some day

Wherein, the △ t in form is the system unit time, and T is time interval quantity, and optionally, T can value be 24*60/ △ t, t be current time; Blk_id _ifor section mark, wherein i is expressed as the i-th+1 section, and such as, the section in the 1st section is designated blk_id ₀, the section mark in the 2nd section is blk_id ₁; Tf _ifor the magnitude of traffic flow, i wherein is also expressed as i-th section.

Wherein, it should be noted that, table 1 is only the historical traffic flows data form of some day, and the data included in the historical traffic flows data obtained in the historical traffic flows data form of many days, optionally, be the historical traffic flows data of 30 in the historical traffic flows data obtained, alternatively, obtain 30 historical traffic flows data forms.

Optionally, can by certain moment t, certain section blk_id _ithe magnitude of traffic flow be designated as tf _i,t, tf _i,tthe historical traffic flows data obtained shown in being, with reference to table 1, can find out, during the 1st moment, i.e. and t ₁during=△ t, the 1st section, i.e. blk_id _ifor blk_id ₀time the magnitude of traffic flow be 123, the tf namely in table 1 _0,1=123, namely can get historical traffic flows data tf in Table 1 _0,1=123, section to be measured identification information, historical time information and historical traffic flows value data information can be determined from the historical traffic flows data obtained.

Optionally, can from flow collection system or the road history traffic flow data obtaining appointed area from the data on flows storehouse of business platform.

Optionally, discrete processes historical traffic flows data can be, first the historical traffic circuit-switched data of acquisition are normalized in [0,1] interval, then carry out discretize to historical traffic flows data after normalized, obtain discretize historical data.

Optionally, can be L, M and H tri-numerical value by historical traffic flows data discrete, represent respectively magnitude of traffic flow trend low, neutralize high.

Optionally, the historical traffic flows data of acquisition for filter out abnormal data, and can carry out the data of correction to the abnormal data filtered out.Use and filter out abnormal data, and the historical traffic flows data of correction were carried out to the abnormal data filtered out, the situation follow-up result being produced to noise can be avoided; There is the situation of zero or negative value in speed, the magnitude of traffic flow, occupation rate data; The magnitude of traffic flow exceedes more than design discharge certain limit, or speed is greater than the situation of sensing range; The judgement relation of speed, occupation rate and vehicle commander's scope, namely by speed, occupation rate, the product in traffic flow collection cycle, the Vehicle length that can extrapolate, for the off-limits situation of vehicle commander; The generation of the situation of Missing data etc., increases the Stability and veracity of traffic flow forecasting further.

Optionally, because traffic flow data has the advantages that can not undergo mutation within the system unit time, therefore, arest neighbors method of interpolation can be adopted to be revised by the abnormal data filtered out.

Table 2 gives through data correction and historical traffic flows data form after discrete processes:

The historical traffic flows data form of table 2 through data correction and after discrete processes

Table 1 is contrasted mutually with table 2, can find, in table 2 data of grey shading be revise after historical traffic flows data.

Optionally, data mining can be carried out by evolutionary programming algorithm to described discretize historical data and obtain spacetime correlation rule.Generally, each row is as an independently attribute using in tables of data for data mining algorithm, and data line, as a data tuple, completes the correlation rule between excavation two community sets by statistical study.Two main differences are had: one, traffic data has spatiotemporal for the association rule mining of traffic flow forecasting and traditional data mining, namely correlation rule does not exist only between two property sets of synchronization, can be present in not between two property sets in the same time yet; This space-time characterisation can cause the growth of data-mining search space exponentially rank, can cause great challenge for prognoses system computing power; Two, the follow-up item of the correlation rule of forecasting traffic flow is fixing, such as will predict blk_id ₂₀at t=t _jthe magnitude of traffic flow in moment, the follow-up item of its correlation rule is with

Therefore, adopt evolutionary programming algorithm to complete the search of correlation rule, due to high concurrency and the probabilistic search feature of evolutionary programming algorithm, effectively can reduce the complicacy of correlation rule; Meanwhile, because the space-time characterisation of traffic parameter can become the complicacy of the increase data mining of the order of magnitude, adopt the high concurrency based on evolutionary programming algorithm effectively can improve the efficiency of data mining; Integrating parallel computing environment can reach real-time update forecast model.

Optionally, carry out data mining by evolutionary programming algorithm to described discretize historical data to obtain spacetime correlation rule and can be, first determine the number M in section in discretize historical data, generate the individuality with M gene, wherein, each gene and section one_to_one corresponding in individuality, get the individuality composition initial population being not less than predetermined number again, and the fitness value of each individuality in described initial population is calculated according to discretize historical data, then according to the fitness value of individuality each in described initial population, Swarm Evolution is carried out to described initial population, finally qualified correlation rules all in each individuality after Swarm Evolution are exported, obtain spacetime correlation rule.

Optionally, determine the number M in section in discretize historical data, generating has in the individuality of M gene, M gene in the individuality generated, comprises the determination gene that a span is [-T ,-1], the section that this gene is corresponding, for section to be measured, wherein, T is the quantity of traffic flow data sampling time interval in historical traffic flows data; With the random gene that M-1 span is [0, T], wherein, value be 0 the gene representation section corresponding with this gene and section to be measured have nothing to do, do not participate in the calculating of spacetime correlation rule.

Optionally, uniform random number generation method can be adopted generate M-1 span to be the random gene of [0, T].

Such as, as shown in table 3, an individuality with 635 genes:

Table 3, an individuality with 635 genes

0	7	…	-1	0	…	0	9
								Gene 1	Gene 2		Gene 78	Gene 79		Gene 634	Gene 635

In table 3, can find out that this individuality has 635 genes altogether, individuality altogether has in the discretize historical data of 635 gene representations acquisitions and has 635 sections;-1 represents that the 78th section that gene is corresponding is section to be measured, and what will predict is the traffic flow data in the 1st moment; Other numerals, as 7, represent the 2nd and there is spacetime correlation relevance between the section that gene is corresponding and section to be measured, and the association moment was the 7th moment.

According to table 3, can obtain this individuality representative correlation rule forerunner attribute set, namely section subset be blk_id2, blk_id635}, following 12 correlation rules that can wherein be contained:

R1：

R2：

……

R7：

R8：

……

R12：

Optionally, can select η NT individual composition initial population, represent the traffic flow modes of each section in each moment, as the follow-up item of spacetime correlation rule, wherein, η is population size controling parameters, and optionally, the numerical value of η can be taken between 10-100.

In genetic algorithm, evaluate the good and bad degree of each individuality with the size of ideal adaptation degree, thus determine the size of its hereditary chance, generally fitness function gets nonnegative value, and with the fitness value maximizing colony for optimization aim.

Optionally, the quality weighing each correlation rule based on support support and degree of confidence confidence can be adopted.Wherein, emphasis utilizes degree of confidence to characterize the separating capacity of rule, takes into account the ability of aggregation of correlation rule simultaneously.Suppose whole historical traffic flow data table always total K bar record, for R1, calculate support and degree of confidence by all data recorders in convenience data storehouse:

When R1 meets Sup (R1) > θ Sup and Conf (R1) > θ Conf simultaneously, R1 just can think one and qualified correlation rule.Optionally, the span of θ Sup can be 0.25-0.6; The span of θ Conf can be 0.6-0.9.

Optionally, a fitness function can be adopted to calculate the weighted sum of all correlation rule supports of body one by one and degree of confidence, as follows:

$\mathrm{fitness}(g,{\mathrm{ind}}_j)={\mathrm{\ω}}_1\underset{i}{\mathrm{\Σ}}\mathrm{Sup}\left(\mathrm{Ri}\right)+{\mathrm{\ω}}_2\underset{i}{\mathrm{\Σ}}\mathrm{Conf}\left(\mathrm{Ri}\right)$

Wherein, ω 1+ ω 2=1,0< ω 1<1,0< ω 2<1, ω 2< ω 1; It is that g is for population that g represents current.

Optionally, can by two individualities in Stochastic choice initial population, select the individuality that in these two individualities, fitness numerical value is larger and enter the next generation, until select the new colony of predetermined number individuality composition, then from the new colony obtained Stochastic choice go out predetermined number individuality complete chiasma operation, from the new colony obtained Stochastic choice go out predetermined number individuality complete chromosomal variation operation, repeat new colony generating run, chiasma operation and chromosomal variation operation, until colony's fitness value convergence, carry out the Swarm Evolution of initial population.

Optionally, the number of the new individual in population selected can be η NT.

Optionally, Stochastic choice goes out total individual number order 40%-70% from the new colony obtained individual amount chiasma operation can be completed.

Optionally, Stochastic choice goes out total individual number order 0.1%-10% from the new colony obtained individual amount chromosomal variation operation can be completed.

Optionally, colony's fitness value convergence formula can be:

|∑fitness(g，ind _j)-∑fitness(g-1，ind _j)|＜ε；

After the spacetime correlation rule obtained by data mining, utilize spacetime correlation rule to find out the section with front section to be measured with relevance to gather, the i.e. preposition community set of all correlation rule subsets, the historical traffic flows data gathered in this section set up corresponding first forecast model as training data.

Optionally, because traffic flow data belongs to nonstationary time series, therefore, the first forecast model of training ARIMA (p, d, q) model that can be.

Optionally, with { Ri|tf _78,1=L} is example, and training obtains parameter p to be estimated when obtaining ARIMA (p, d, q) model, the method for d and q can comprise:

Steps A): obtain historical traffic flows data tf _78,1, tf _78,0..., tf _78,1-p; Tf _2,7, tf _2,6..., tf _2,7-p, and tf _635,9, tf _635,8..., tf _635,9-p;

Step B): the traffic flow data of acquisition is converted to magnitude of traffic flow differential data;

Step C): use single method of inspection determination parameter d;

Step D): according to training data and parameter d determination parameter p;

Step e): by AR (∞), parameter d and p determine parameter q.

Optionally, with { Ri|tf _78,1=L} is example, and the traffic flow data of acquisition is converted to magnitude of traffic flow differential data can be comprised:

${\mathrm{tf}}_{\mathrm{78,1}}=\underset{i=1}{\overset{p}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_i(w_{\mathrm{78,1}-i}^d+w_{\mathrm{2,7}-i}^d+w_{\mathrm{635,9}-i}^d)+\underset{j=1}{\overset{q}{\mathrm{\&Sigma;}}}{\mathrm{\&delta;}}_j{\mathrm{\&epsiv;}}_{1-j};$

Δtf _i，t＝tf _i，t-tf _i，t-1；

Δ ²tf _i，t＝Δtf _i，t-Δtf _i，t-1；

……

$w_{i,t}^d={\mathrm{\&Delta;}}^{d}t{f}_{i,t};$

Wherein, Δ is first difference operator; Δ ²position 2 rank difference operators; be check the mark operator in d rank with Δ d; α i is auto-regressive parameter; δ j is moving average parameter; P, d and q are model parameter to be estimated; ε is white noise.

Different sections, when different when, the first different forecast models is all correspond to according to different magnitude of traffic flow trend, therefore, the historical traffic flow data gathered by the section with section to be measured tool relevance trains the first forecast model generated to be multiple first forecast models by preordering method, needed before use first forecast model predicting traffic flow value, select needing the first forecast model used, determine the first forecast model that the current traffic flow data that the determination in section to be measured obtains use, just can after the Current traffic flow data of acquisition be inputted the first forecast model, obtain the predicting traffic flow value in section to be measured.

Step S120: described Current traffic flow data is inputted the first forecast model, obtains the predicting traffic flow value in section to be measured.

Based on technique scheme, a kind of traffic flow forecasting method that the embodiment of the present invention provides and system, after acquisition current traffic flow data, by determining the first forecast model, by the first forecast model that the input of the Current traffic flow data of acquisition is determined, then from the first forecast model determined, obtain the predicting traffic flow value in section to be measured, wherein, described current traffic flow data comprise the current traffic flow data in section to be measured and the Current traffic flow data with tool relevance section, section to be measured, wherein, described first forecast model is by obtaining historical traffic flows data, historical traffic flows data described in discrete processes, obtain discretize historical data, data mining is carried out to described discretize historical data and obtains spacetime correlation rule, obtained gathering with the section of section to be measured tool relevance by described spacetime correlation rule, the historical traffic flow data gathered in described section is generated by preordering method training.Because in the traffic flow forecasting method that this transmission embodiment provides, the first forecast model used obtains spacetime correlation rule through data mining for history communication flow data, the history communication flow data message in the section needed for this moment correlation rule obtains is obtained by preordering method training, data mining is carried out to historical traffic data, by carrying out global optimum's search to mass data, obtain frequent item set, improve the accuracy of predicting traffic flow value.

Optionally, Fig. 2 shows the method flow diagram determining the first forecast model in the traffic flow forecasting method that point inventive embodiments provides, and with reference to Fig. 2, this determines that the method for the first forecast model can comprise:

Step S200: current traffic flow data described in discrete processes, obtain discretize current data;

Optionally, discrete processes historical traffic flows data can be, first the historical traffic circuit-switched data of acquisition are normalized in [0,1] interval, then carry out discretize to historical traffic flows data after normalized, obtain discretize historical data.

Step S210: described discretize current data is inputted the second forecast model, obtain magnitude of traffic flow trend, described second forecast model is by obtaining historical traffic flows data, historical traffic flows data described in sliding-model control, carry out the spacetime correlation rule composition that data mining obtains by historical traffic flow data after sliding-model control;

Second forecast model carries out by historical traffic flow data after sliding-model control the spacetime correlation regular collection that data mining obtains and forms, each section has several independently spacetime correlation rules subset at moment t, if by historical traffic flows data discrete be L, M and H tri-numerical value, then each section has 3 independently spacetime correlation rules subset at moment t.

For in table 3, section blk_id78, its forecast model is by { Ri , { Ri and { Ri three correlation rule subset compositions, wherein,

…；

…；

…；

Traffic flow data with correlation rule subset { Ri|tf _78,1in=L}, the matching degree of certain correlation rule Rir is as follows:

Traffic flow data tf _twith correlation rule subset the matching degree of Ri} is as follows:

Wherein, | Ri| represents correlation rule quantity in correlation rule subset.

Model prediction result realize by maximizing correlation rule matching degree:

Optionally, if when the historical traffic flows data of acquisition and current traffic flow data are carried out discrete processes, be L, M and H tri-numerical value by data discrete, then by the second forecast model can obtain magnitude of traffic flow trend can for low, neutralize high.

Step S220: determine the first forecast model matched with the second forecast model according to corresponding relation between described traffic flow trend and the first forecast model.

Because different sections, when different when, the first different forecast models is all correspond to according to different magnitude of traffic flow trend, when obtaining current traffic flow data, just mark and the current time in section to be measured can be learnt, therefore, after passing through the second forecast model, after learning magnitude of traffic flow trend, just can determine the first required forecast model.

Optionally, Fig. 3 shows the method flow diagram of discrete processes current traffic flow data in the traffic flow forecasting method that the embodiment of the present invention provides, and with reference to Fig. 3, the method for these discrete processes current traffic flow data can comprise:

Step S300: by the current traffic flow data normalization of acquisition in [0,1] interval;

Optionally, the historical traffic circuit-switched data tf will obtained _{i, t}the formula being normalized to [0,1] interval can be:

minTF＝min{tf _i}；

maxTF＝max{tf _i}；

${{\mathrm{tf}}_{i,t}}^{,}=\frac{{\mathrm{tf}}_{i,t}-\min \mathrm{TF}}{\max \mathrm{TF}-\min \mathrm{TF}};$

Wherein, tf _{i, t}' for being normalized to the historical traffic flows data behind [0,1] interval.

Step S310: discretize is carried out to current traffic flow data after described normalized, obtains discretize current data.

Optionally, carry out discretize to historical traffic flows data after normalized, the formula obtaining discretize historical data can be:

${\mathrm{\&mu;}}_i=\frac{1}{M}{\mathrm{\&Sigma;}}_{t=t_1}^{t_M}{{\mathrm{tf}}_{i,t}}^{,};$

${\mathrm{\&sigma;}}_i=\frac{1}{M-1}\sqrt{{\mathrm{\&Sigma;}}_{t=t_1}^{t_M}{({{\mathrm{tf}}_{i,t}}^{,}-{\mathrm{\&mu;}}_i)}^2};$

${\mathrm{\&theta;}}_l=\frac{1}{2}({\mathrm{\&mu;}}_i+({\mathrm{\&mu;}}_i-{\mathrm{\&alpha;}}_i{\mathrm{\&sigma;}}_i));$

${\mathrm{\&theta;}}_h=\frac{1}{2}({\mathrm{\&mu;}}_i+({\mathrm{\&mu;}}_i+{\mathrm{\&alpha;}}_i{\mathrm{\&sigma;}}_i));$

Wherein, for the traffic flow data of discretize; μ _ifor label is blk_id _ithe history average traffic flow in section; σ _ifor label is blk_id _ithe historical traffic flows standard deviation in section; θ _land θ _hbe discretize threshold value; α _ifor regulating weighted value, its numerical value can adjust according to actual conditions, optionally, can say α _inumerical value be set between 0.25-0.5.

Optionally, Fig. 4 shows the method flow diagram of discrete processes historical traffic flows data in the traffic flow forecasting method that the embodiment of the present invention provides, and with reference to Fig. 4, the method for these discrete processes historical traffic flows data can comprise:

Step S400: the historical traffic circuit-switched data of acquisition is normalized in [0,1] interval;

Step S410: discretize is carried out to historical traffic flows data after described normalized, obtains discretize historical data.

Wherein, carrying out discrete processes with carrying out discrete processes to current traffic flow data to historical traffic flows data is use identical data formula to process.

Optionally, Fig. 5 shows in the traffic flow forecasting method that the embodiment of the present invention provides and carries out to discretize historical data the method flow diagram that data mining obtains spacetime correlation rule by evolutionary programming algorithm, with reference to Fig. 5, carrying out to discretize historical data the method that data mining obtains spacetime correlation rule by evolutionary programming algorithm can comprise:

Step S500: the number M determining section in discretize historical data, generate the individuality with M gene;

Wherein, in M gene in the individuality generated, comprise the determination gene that a span is [-T ,-1], the corresponding section of this gene is section to be measured, be [0 with M-1 span, T] random gene, wherein, T is the quantity of traffic flow data sampling time interval in historical traffic flows data, value be 0 the gene representation section corresponding with this gene and section to be measured have nothing to do, do not participate in the calculating of spacetime correlation rule.

Step S510: get the individuality composition initial population being not less than predetermined number, and calculate the fitness value of each individuality in described initial population according to discretize historical data;

In genetic algorithm, evaluate the good and bad degree of each individuality with the size of ideal adaptation degree, thus determine the size of its hereditary chance, generally fitness function gets nonnegative value, and with the fitness value maximizing colony for optimization aim.

Optionally, can select η NT individual composition initial population, represent the traffic flow modes of each section in each moment, as the follow-up item of spacetime correlation rule, wherein, η is population size controling parameters, and optionally, the numerical value of η can be taken between 10-100.

Step S520: the fitness value according to individuality each in described initial population carries out Swarm Evolution to described initial population;

Optionally, can by two individualities in Stochastic choice initial population, select the individuality that in these two individualities, fitness numerical value is larger and enter the next generation, until select the new colony of predetermined number individuality composition, then from the new colony obtained Stochastic choice go out predetermined number individuality complete chiasma operation, from the new colony obtained Stochastic choice go out predetermined number individuality complete chromosomal variation operation, repeat new colony generating run, chiasma operation and chromosomal variation operation, until colony's fitness value convergence, carry out the Swarm Evolution of initial population.

Step S530: qualified correlation rules all in each individuality after Swarm Evolution are exported, obtains spacetime correlation rule.

Optionally, the quality weighing each correlation rule based on support support and degree of confidence confidence can be adopted.Wherein, emphasis utilizes degree of confidence to characterize the separating capacity of rule, takes into account the ability of aggregation of correlation rule simultaneously.

For the R1 obtained in table 3, only have when to meet Sup (R1) > θ Sup and Conf (R1) > θ Conf simultaneously, just can think that R1 is and qualified correlation rule.

Optionally, the span of θ Sup can be 0.25-0.6; The span of θ Conf can be 0.6-0.9.

Optionally, Fig. 6 shows the method flow diagram generating in the traffic flow forecasting method that the embodiment of the present invention provides and have the individuality of M gene, reference Fig. 6, and the method that this generation has the individuality of M gene can comprise:

Step S600: generate the determination gene that a span is [-T ,-1];

Wherein, T is the quantity of traffic flow data sampling time interval in historical traffic flows data.

The span generated is the corresponding section of the determination gene of [-T ,-1] is section to be measured.

Step S610: generate the random gene that M-1 span is [0, T].

Optionally, uniform random number generation method can be adopted generate M-1 span to be the random gene of [0, T].

Optionally, Fig. 7 shows the method flow diagram according to the fitness value of individuality each in initial population, described initial population being carried out to Swarm Evolution in the traffic flow forecasting method that the embodiment of the present invention provides, with reference to Fig. 7, this can comprise the method that described initial population carries out Swarm Evolution according to the fitness value of individuality each in initial population:

Step S700: two individualities in initial population described in Stochastic choice, selects the larger individuality of wherein fitness value and enters the next generation, until select the new colony of predetermined number individuality composition;

Optionally, the number of the new individual in population selected can be η NT.

Step S710: from described new colony Stochastic choice go out predetermined number individuality complete chiasma operation;

Optionally, Stochastic choice goes out total individual number order 40%-70% from the new colony obtained individual amount chiasma operation can be completed.

Step S720: from described new colony Stochastic choice go out predetermined number individuality complete chromosomal variation operation;

Optionally, Stochastic choice goes out total individual number order 0.1%-10% from the new colony obtained individual amount chromosomal variation operation can be completed.

Step S730: repeat new colony generating run, chiasma operation and chromosomal variation operation, until colony's fitness value convergence.

Optionally, colony's fitness value convergence formula can be:

|∑fitness(g，ind _j)-∑fitness(g-1，ind _j)|＜ε

Optionally, Fig. 8 shows in the traffic flow forecasting method that the embodiment of the present invention provides and obtains the method flow diagram with tool relevance section, section to be measured, and with reference to Fig. 8, this obtains can comprising with the method in tool relevance section, section to be measured:

Step S800: determine section to be measured and spacetime correlation rule;

Step S810: obtain the section with section to be measured tool relevance according to described section to be measured and spacetime correlation rule.

The traffic flow forecasting method that the embodiment of the present invention provides, the first forecast model used obtains spacetime correlation rule through data mining for history communication flow data, the history communication flow data message in the section needed for this moment correlation rule obtains is obtained by preordering method training, data mining is carried out to historical traffic data, by carrying out global optimum's search to mass data, obtain frequent item set, improve the accuracy of predicting traffic flow value.

Be introduced the traffic flow forecasting system that the embodiment of the present invention provides below, traffic flow forecasting system described below can mutual corresponding reference with above-described traffic flow forecasting method.

The system chart of the traffic flow forecasting system that Fig. 9 provides for the embodiment of the present invention, with reference to Fig. 9, this traffic flow forecasting system can comprise: acquisition module 100, model determination module 200, prediction module 300 and forecast model generation module 400; Wherein,

Acquisition module 100, for obtaining current traffic flow data, described current traffic flow data comprise the current traffic flow data in section to be measured and the Current traffic flow data with tool relevance section, section to be measured;

Model determination module 200, for determining the first forecast model;

Prediction module 300, for by described first forecast model of described current traffic flow data input, obtains the predicting traffic flow value in section to be measured;

Forecast model generation module 400, for obtaining historical traffic flows data, historical traffic flows data described in sliding-model control, historical traffic flow data after sliding-model control is carried out data mining and obtains spacetime correlation rule, obtained gathering with the section of section to be measured tool relevance by described spacetime correlation rule, the historical traffic flow data gathered in described section is by preordering method training generation first forecast model.

Optionally, Figure 10 shows the structural representation of model determination module 200 in the traffic flow forecasting system that the embodiment of the present invention provides, and with reference to Figure 10, this model determination module 200 can comprise: forecast model determining unit 210 and forecast model generation unit 220; Wherein,

Forecast model determining unit 210, for current traffic flow data described in sliding-model control, obtains discretize current data; Described discretize current data is inputted the second forecast model, obtains magnitude of traffic flow trend; The first forecast model matched with the second forecast model is determined according to corresponding relation between described traffic flow trend and the first forecast model;

Forecast model generation unit 220, for obtaining historical traffic flows data, historical traffic flows data described in sliding-model control, carry out the spacetime correlation rule that data mining obtains, form the second forecast model by historical traffic flow data after sliding-model control.

In this instructions, each embodiment adopts the mode of going forward one by one to describe, and what each embodiment stressed is the difference with other embodiments, between each embodiment identical similar portion mutually see.For device disclosed in embodiment, because it corresponds to the method disclosed in Example, so description is fairly simple, relevant part illustrates see method part.

To the above-mentioned explanation of the disclosed embodiments, professional and technical personnel in the field are realized or uses the present invention.To be apparent for those skilled in the art to the multiple amendment of these embodiments, General Principle as defined herein can without departing from the spirit or scope of the present invention, realize in other embodiments.Therefore, the present invention can not be restricted to these embodiments shown in this article, but will meet the widest scope consistent with principle disclosed herein and features of novelty.

Claims (10)

1. a traffic flow forecasting method, is characterized in that, comprising:

Obtain current traffic flow data, described current traffic flow data comprise the current traffic flow data in section to be measured and the Current traffic flow data with tool relevance section, section to be measured;

Determine the first forecast model;

By described first forecast model of described current traffic flow data input, obtain the predicting traffic flow value in section to be measured;

Wherein, described first forecast model is by obtaining historical traffic flows data, historical traffic flows data described in discrete processes, obtain discretize historical data, data mining is carried out to described discretize historical data and obtains spacetime correlation rule, obtained gathering with the section of section to be measured tool relevance by described spacetime correlation rule, the historical traffic flow data gathered in described section is generated by preordering method training.

2. traffic flow forecasting method according to claim 1, is characterized in that, describedly determines that the first forecast model comprises:

Current traffic flow data described in discrete processes, obtain discretize current data;

Described discretize current data is inputted the second forecast model, obtains magnitude of traffic flow trend;

According to described section to be measured identification information, current time information and between traffic flow trend and the first forecast model corresponding relation determine the first forecast model matched with the second forecast model.

Wherein, historical traffic flow data after sliding-model control, by obtaining historical traffic flows data, historical traffic flows data described in sliding-model control, is carried out the spacetime correlation rule composition that data mining obtains by described second forecast model.

3. traffic flow forecasting method according to claim 2, is characterized in that,

According to current traffic flow data described in following steps discrete processes:

By the current traffic flow data normalization of acquisition in [0,1] interval;

Discretize is carried out to current traffic flow data after described normalized, obtains discretize current data;

According to historical traffic flows data described in following steps discrete processes:

The historical traffic circuit-switched data of acquisition is normalized in [0,1] interval;

Discretize is carried out to historical traffic flows data after described normalized, obtains discretize historical data.

4. traffic flow forecasting method according to claim 1, is characterized in that, carries out data mining obtain spacetime correlation rule by evolutionary programming algorithm to described discretize historical data;

Carry out data mining by evolutionary programming algorithm to described discretize historical data to obtain spacetime correlation rule and comprise:

Determine the number M in section in discretize historical data, generate the individuality with M gene;

Get the individuality composition initial population being not less than predetermined number, and calculate the fitness value of each individuality in described initial population according to discretize historical data;

Fitness value according to individuality each in described initial population carries out Swarm Evolution to described initial population;

Qualified correlation rules all in each individuality after Swarm Evolution are exported, obtains spacetime correlation rule.

5. traffic flow forecasting method according to claim 4, is characterized in that, the individuality that described generation has M gene comprises:

Generate the determination gene that a span is [-T ,-1], T is the quantity of traffic flow data sampling time interval in historical traffic flows data;

Generate the random gene that M-1 span is [0, T], obtain the individuality with M gene.

6. traffic flow forecasting method according to claim 4, is characterized in that, the described fitness value according to individuality each in described initial population carries out Swarm Evolution to described initial population and comprises:

Two individualities in initial population described in Stochastic choice, select the larger individuality of wherein fitness value and enter the next generation, until select the new colony of predetermined number individuality composition;

From described new colony Stochastic choice go out predetermined number individuality complete chiasma operation;

From described new colony Stochastic choice go out predetermined number individuality complete chromosomal variation operation;

Repeat new colony generating run, chiasma operation and chromosomal variation operation, until colony's fitness value convergence.

7. traffic flow forecasting method according to claim 1, is characterized in that, described current traffic flow data and described historical traffic flows data are and filter out abnormal data, and carries out the data of correction to the abnormal data filtered out.

8. traffic flow forecasting method according to claim 1, is characterized in that, the Current traffic stream packets in tool relevance section, section described and to be measured is drawn together:

Determine section to be measured and spacetime correlation rule;

The section with section to be measured tool relevance is obtained according to described section to be measured and spacetime correlation rule.

9. a traffic flow forecasting system, is characterized in that, comprising: acquisition module, model determination module, prediction module and forecast model generation module; Wherein,

Described acquisition module, for obtaining current traffic flow data, described current traffic flow data comprise the current traffic flow data in section to be measured and the Current traffic flow data with tool relevance section, section to be measured;

Described model determination module, for determining the first forecast model;

Described prediction module, for by described first forecast model of described current traffic flow data input, obtains the predicting traffic flow value in section to be measured;

Described forecast model generation module, for obtaining historical traffic flows data, historical traffic flows data described in sliding-model control, historical traffic flow data after sliding-model control is carried out data mining and obtains spacetime correlation rule, obtained gathering with the section of section to be measured tool relevance by described spacetime correlation rule, the historical traffic flow data gathered in described section is by preordering method training generation first forecast model.

10. traffic flow forecasting system according to claim 9, is characterized in that, described model determination module comprises: forecast model determining unit and forecast model generation unit; Wherein,

Described forecast model determining unit, for current traffic flow data described in sliding-model control, obtains discretize current data; Described discretize current data is inputted the second forecast model, obtains magnitude of traffic flow trend; The first forecast model matched with the second forecast model is determined according to corresponding relation between described traffic flow trend and the first forecast model;

Described forecast model generation unit, for obtaining historical traffic flows data, historical traffic flows data described in sliding-model control, carry out the spacetime correlation rule that data mining obtains, form the second forecast model by historical traffic flow data after sliding-model control.