CN103198672A - Method for laying urban road network traffic flow detectors - Google Patents

Abstract

The invention discloses a method for laying urban road network traffic flow detectors. The method includes the following steps: selecting a laying mode for the detectors; carrying out clustering analysis; presetting the positions and the number of the detectors; selecting a forecasting method; forecasting traffic flows of intersections without detectors with the selected forecasting method; and according to forecast results, combined with a clustering pedigree chart, finally confirming the positions and the number of the detectors. Correlation among the traffic flows and then a statistic method are used for determining the laying positions of the urban road network traffic flow detectors, and the method for laying the urban road network traffic flow detectors has the advantage of being wide in coverage. The traffic flow detectors are just laid on a small number of intersections in an urban road network, so that civil engineering cost is low, workload is small, and economical efficiency is good. Only limit detector laying on the intersections and limit infrastructure investments need to be carried out to obtain traffic flow information of the whole urban road network, so that macro-management of the whole urban road network is achieved.

Description

A kind of distribution method of city road network traffic flow amount detector

Technical field

The present invention relates to a kind of detection method of city road network traffic flow amount, particularly a kind of distribution method of city road network traffic flow amount detector.

Background technology

Along with the swift and violent increase of Forecast of Urban Traffic Flow, the difficulty of urban traffic control is progressively increasing.At present, comprise that the world many countries of developed country all generally adopts intelligent traffic control system, as: traffic face control system, intelligent transportation system etc.Along with the progress of science and technology and the development of road traffic cause, at present in some big and medium-sized cities of China (even some world developed countries), the density of its road network is increasing, and the node of road network (referring to the crossing here) number is more and more, for density big road network like this, want to realize its macro-management, just must obtain the telecommunication flow information of overwhelming majority's (or even all) node of road network.And the main crossing in the traffic face control system that generally adopt in these cities has all been buried detecting device underground, the collection of traffic flow data mainly is to rely on the detecting device of burying underground these crossings to obtain, for example: nearly more than 10000 crossings, Tokyo city, but only therein crossing, 7000 left and right sides is embedded with detecting device.Domestic nearly thousands of crossings, Daliang City for another example, but can only bury detecting device underground to adapt to the self-adapting traffic signal control system in about 300 crossings in the city---the basic need of SCOOT system, Nanjing are estimated to bury 350 crossings in the city underground detecting device to adapt to the basic need of the traffic face control system that will adopt.If merely relying on main crossroads buries detecting device underground and obtains telecommunication flow information, will expend a large amount of infrastructure investments, the telecommunication flow information amount of gained can't comprise the telecommunication flow information of whole crossings simultaneously, so just can not satisfy traffic face control system to the macro-management requirement of city road network integral body.

Traditional detecting device distribution method is that vehicle supervision department lays detecting device in main crossing, city as required, and this distribution method does not consider and respectively lay the correlativity between the magnitude of traffic flow between the crossing that laying has certain blindness.

Summary of the invention

Be to solve the problems referred to above that prior art exists, the present invention will design and a kind ofly not only can utilize limited crossing detecting device and reduce infrastructure investment, but also can satisfy traffic face control system to the distribution method of the city road network traffic flow amount detector of city road network integral macroscopic management expectancy.

To achieve these goals, technical scheme of the present invention is as follows: a kind of distribution method of city road network traffic flow amount detector may further comprise the steps:

The laying mode of A, selection detecting device

Select the laying mode of detecting device according to the characteristics of different Area Traffic Control Systems, if the zone centered by each administrative residential quarter or traffic zone control center then selects the distributing detecting device to lay mode; If be to lay the zone or select and determine that detecting device lays the central area by the Competent Authorities of Transport and Communications with the shopping centre, then select center radiation formula detecting device to lay mode;

B, carry out cluster analysis

Utilize clustering methodology to set up the cluster pedigree chart between interior each crossing in the above-mentioned zone;

Position and the quantity of C, default detecting device

At first according to the number of the shown class that goes out in the cluster pedigree chart, in each class, select some crossings of merging for the first time as the target crossing of default detecting device; For su generis crossing, earlier it is assumed to the target crossing of laying detecting device, obtain the target crossing quantity of default detecting device like this;

D, selection Forecasting Methodology

For the class of having only a small amount of number crossing, select the multiple linear regression Forecasting Methodology for use;

For su generis crossing or contain a fairly large number of class in crossing, use a kind of of following two kinds of methods:

D1, elder generation use the principal component analysis (PCA) predicted method, carry out combined prediction with the multiple linear regression Forecasting Methodology again;

D2, progressively Regression Forecast of elder generation are carried out combined prediction with the multiple linear regression Forecasting Methodology again;

The magnitude of traffic flow of the Forecasting Methodology prediction sensorless crossing that E, utilization are selected;

F, predict the outcome and in conjunction with the cluster pedigree chart, finally determine to bury underground quantity and the position of detecting device according to above-mentioned.

Compared with prior art, the present invention has following beneficial effect:

1, because the present invention has utilized the correlativity between the magnitude of traffic flow, determine the installation position of city road network traffic flow detecting device then with statistical method, have the characteristics of broad covered area;

2, because the present invention only lays magnitude of traffic flow detecting device in a spot of crossing in city road network, so civil engineering cost is few, and workload is few, good economy performance;

3, because the adaptive traffic signal control system that the present invention generally adopts for the big and medium-sized cities that comprise China at present can provide real-time and dynamic telecommunication flow information accurately, scientific good, accuracy of detection height, good reliability;

4, because the present invention just utilizes statistical method to solve choosing of city road network traffic flow amount detector installation position, and detector location is clear and definite, and the urban traffic control operation cost is low, detecting device is easy to maintenance.

5, adopt magnitude of traffic flow detecting device distribution method proposed by the invention, only need carry out limited crossing detecting device lays and infrastructure investment, just can obtain the traffic flow information of entire city road network, thereby reach the macro-management to the entire city road network.

Description of drawings

2 in the total accompanying drawing of the present invention, wherein:

Fig. 1 is process flow diagram of the present invention.

Fig. 2 is city road network crossing magnitude of traffic flow cluster synoptic diagram.

The numbering in each crossing of digitized representation or each bar track among the figure.

Embodiment

Below in conjunction with accompanying drawing the present invention is described further.As shown in Figure 1, a kind of distribution method of city road network traffic flow amount detector may further comprise the steps:

The laying mode of A, selection detecting device

Select the laying mode of detecting device according to the characteristics of different Area Traffic Control Systems, if the zone centered by each administrative residential quarter or traffic zone control center, then select the distributing detecting device to lay mode, also can be referred to as " centrifugal " detecting device and lay mode, the detecting device laying is zone centered by each administrative residential quarter or traffic zone control center in this mode, only is suitable for the self-adapting traffic signal control system telecommunication flow information collection of the such hierarchical control of similar SCATS; Lay the zone or select and determine that detecting device lays the central area by the Competent Authorities of Transport and Communications if with the shopping centre be, then select center radiation formula detecting device to lay mode, also can be referred to as " radial inflow " or " gravitative type " detecting device and lay mode, to lay be zone or determine that by the Competent Authorities of Transport and Communications detecting device lays the central area centered by commercial center removes (CBD) to detecting device in this mode, is suitable for the collection of similar SCOOT self-adapting traffic signal control system telecommunication flow information.

B, carry out cluster analysis

Utilize clustering methodology to set up the cluster pedigree chart between interior each crossing in the above-mentioned zone;

Cluster analysis is a kind of method of research " things of a kind come together, people of a mind fall into the same group " in the mathematical statistics.The function of cluster analysis is to set up a kind of sorting technique, it is with a batch sample or variable, classify in qualitative close and distant degree according to them, and describe its close and distant degree two approach are arranged usually: the one, each sample is regarded as the point that m ties up (number of variable is m) space, in the m dimension coordinate, definition distance between points; Another is to describe close and distant degree between the sample spot with certain similarity coefficient.The method of cluster is a lot, as: system's cluster, dynamic clustering, fuzzy clustering, graph theory clustering, cluster forecast etc., the present invention adopts hierarchical clustering method.The method is that individuality one by one is merged into some subclass, until whole totally all in a set till.Similarity coefficient method in the hierarchical clustering method mainly is to utilize the size of the similarity coefficient between each variable to judge and sort out.

The present invention has adopted the similarity coefficient method in system's cluster.The foundation of carrying out cluster analysis in the middle of the present invention is to utilize the size of variable similarity coefficient each other to determine whether variable is classified as a class.Therefore, also must carry out the selection of similarity coefficient here.Being variable with the crossing in this chapter, is sample with the magnitude of traffic flow, utilizes the similarity coefficient method to classify.Basic thought is: the variable that character is more approaching, similarity coefficient between them is more close to 1(or-1), and the variable that has nothing to do each other, similarity coefficient between them is then close to 0, when carrying out clustering processing, be classified as a class comparing similar argument, less similar variable is classified as different classes.

Similarity coefficient is defined as follows:

If C _IjExpression variable y _iWith y _jBetween similarity coefficient, C then _IjShould satisfy following relation:

(1) $C_{\mathrm{ij}}=\±1\⇔y_i={\mathrm{ay}}_j$ (a ≠ 0, a is constant);

(2) | C _Ij|≤1 for all i, and j sets up;

(3) C _Ij=C _JiFor all i, j sets up.

Similarity coefficient has a lot, and the similarity coefficient that the present invention selects is related coefficient.

The present invention adopts maximum similarity coefficient as the cluster standard, and namely two of similarity coefficient maximum being birdsed of the same feather flock together is a class.Specific practice is:

Definition class G _pAnd G _uBetween similarity coefficient R _PuFor:

$R_{\mathrm{pu}}=\underset{q_i\∈G_p,q_j\∈G_u}{{\max r}_{\mathrm{ij}}}---\left(1\right)$

(1) calculates similarity coefficient between each variable, obtain a similarity coefficient matrix R ₍₀₎, this moment, each variable constituted a class by itself, and R is obviously arranged _Pu=r _Pu

(2) seek R ₍₀₎Non-principal diagonal on greatest member, be made as R _Pu, then with G _pAnd G _uBe merged into a new class, be designated as G _s, i.e. G _s={ G _p, G _u.

(3) similarity coefficient of the new class of calculating and other class

$R_{\mathrm{sk}}=\underset{\underset{q_j\∈G_K}{q_s\∈G_s}}{{\max r}_{\mathrm{sj}}}=\max \left\{\underset{\underset{q_j\∈G_k}{q_i\∈G_P}}{{\max r}_{\mathrm{ij}},}\underset{\underset{q_j\∈G_k}{q_i\∈G_u}}{{\max r}_{\mathrm{ij}}}\right\}=\max \{R_{\mathrm{pk}},R_{\mathrm{uk}}\}---\left(2\right)$

Resulting similarity coefficient is designated as R ₍₁₎

(4) to R ₍₁₎Repeat to carry out for R ₍₀₎Step, get R ₍₂₎, by R ₍₂₎Calculate R by same step ₍₃₎Like this up to all variable class and till being a class.

Position and the quantity of C, default detecting device

At first according to the number of the shown class that goes out in the cluster pedigree chart, some crossings of selecting to merge for the first time in each class are as the target crossing of presetting detecting device, as being numbered 16 crossing or being numbered 17 crossing among Fig. 2, other by that analogy; For su generis crossing, can earlier it be assumed to the target crossing of laying detecting device, be respectively crossings such as 10,9,1,7,8,29,32 as the numbering among Fig. 2.Obtain the target crossing quantity of default detecting device like this.As shown in Figure 2, can select 10 crossings to lay detecting device, be respectively: be numbered 16 or 17 detecting device; Be numbered 18 or 19 detecting device; Be numbered 3 or 4 detecting device; Be numbered 24 or 26 or 23 or 25 detecting device; Be numbered 28 or 31 detecting device; Be numbered 9,10,1,7,8 detecting device since under its and its correlativity of the crossing of the laying detecting device in the class relatively poor, so constitute a class by itself, all as target crossing laying detecting device.Its magnitude of traffic flow can profit be predicted in the following method in other crossings of not laying detecting device.

D, selection Forecasting Methodology

For the class of having only a small amount of number crossing, select the multiple linear regression Forecasting Methodology for use.

For su generis crossing or contain a fairly large number of class in crossing, can be with following two kinds of methods a kind of:

D1, elder generation use the principal component analysis (PCA) predicted method, carry out combined prediction with the multiple linear regression Forecasting Methodology again;

D2, progressively Regression Forecast of elder generation are carried out combined prediction with the multiple linear regression Forecasting Methodology again.

Below Forecasting Methodology of the present invention is described in detail:

(1) basic ideas that progressively return are as follows:

In introduction, introduced if predict the magnitude of traffic flow of sensorless crossing and just must utilize the correlativity that has between detecting device crossing and the sensorless crossing, if but there is the magnitude of traffic flow of detecting device crossing to go to predict the magnitude of traffic flow of sensorless crossing with these, then can make accuracy of predicting be subjected to influence to a certain degree.This point can know that if the independent variable that adopts is more many, then regression sum of square is more big from the multiple linear regression analysis method, and residual sum of squares (RSS) is just more little.Yet adopt more variable to come the match regression equation, can make the poor stability of equation, the interval accumulation of error of each independent variable will influence global error, bring irrational explanation can for like this regression coefficient estimated value, and make and do with such equation that forecasting reliability is poor, the precision reduction; If adopted on the other hand the y very little variable of influence is omitted significant variable, can cause estimator to produce bias and inconsistency.In view of the foregoing, in order to obtain sane, a reliable regression model, this just need provide a kind of method, make and from numerous factors that influences y, to pick out the big variable of y contribution, set up the regression equation of " optimum " on the observation data basis of they and y, the progressively regression forecasting method that we will introduce below Here it is.

Set up in the past " Optimal Regression Equation; adopt following several diverse ways usually: the regression analysis of " progressively rejecting ", etc.; this several method has weak point; " progressively rejecting " will calculate the regression equation that comprises all independents variable at the beginning; if more a plurality of remarkable factors are arranged in the original independent variable, and calculated amount will be bigger.The method of " progressively introducing " reckons without because the progressively introducing of new variables makes the variable of original introducing might lose importance.

And the stepwise regression analysis method is a kind of automatically from a large amount of alternative variablees, selection is to setting up the method for the important variable of regression equation, it is a kind of algorithm skill that derives from the multiple linear regression basis, and it has overcome the deficiency of above-mentioned two kinds of methods.The method is similar to the regression analysis of " progressively introducing ", since an independent variable, looks independent variable to the significance degree of y effect, introduces regression equation from big to small one by one.Difference is, when former introducing variable becomes no longer significantly owing to the introducing of back variable, it be rejected.Introduce a variable or from regression equation, reject a variable, be a step that progressively returns.Each step all will be carried out the F check, only comprises significant variable in the regression equation before to guarantee the new conspicuousness variable of each introducing.This process is carried out repeatedly, until both invariably significantly variable from regression equation, reject, when not having remarkable variable again and being selected into regression equation till.

In the present invention, " in the middle of the Optimal Regression Equation, and there are these data of the magnitude of traffic flow of detecting device crossing to be updated to the " data of the magnitude of traffic flow of prediction sensorless crossing in the middle of the Optimal Regression Equation with utilizing the stepwise regression analysis method that the magnitude of traffic flow that the detecting device crossing is arranged that those play the conspicuousness effect to the sensorless crossing is introduced.

Progressively the basic step of Hui Guiing is as follows:

(1) calculates contribution amount

$V_i^{\left(l\right)}=\frac{{\left(b_i^{\left(l\right)}\right)}^2}{c_{\mathrm{ii}}^{\left(l\right)}}---\left(3\right)$

In the formula: b _jBe regression coefficient, c _IiBe the matrix of coefficients (S of normal equations _Ij) inverse matrix (C _Ij) in i element on the principal diagonal.Wherein normal equations is:

$\left\{\begin{array}{c}{S}_{11}{\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">b</figure-callout>}}_1+S_{12}{\mathrm{<figure-callout\; id="2"\; label="b"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">b</figure-callout>}}_2+...+S_{1p}{b}_p=S_{1y}\\ S_{21}{\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">b</figure-callout>}}_1+S_{22}{\mathrm{<figure-callout\; id="2"\; label="b"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">b</figure-callout>}}_2+...+S_{2p}{b}_p=S_{2y}\\ .........\\ S_{p1}{\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">b</figure-callout>}}_1+S_{p2}{\mathrm{<figure-callout\; id="2"\; label="b"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">b</figure-callout>}}_2+...+S_{\mathrm{pp}}{b}_p=S_{\mathrm{py}}\end{array}\right.---\left(4\right)$

Get

Calculate then

If F _Pick≤ F α 2 namely under level of significance α 2 meanings, is less than or equal to critical value F α 2 if test value F picks, and then this variable should be rejected from regression equation, otherwise keeps.

(2) similarly, if variable xi is not for introducing variable, the present establishes it and will introduce regression equation as (l+1) individual variable, and then its contribution is:

$V_i^{(l+1)}=\frac{{\left(b_i^{(l+1)}\right)}^2}{c_{\mathrm{ii}}^{(l+1)}}---\left(6\right)$

The variable of introducing in a certain step that recurrence is calculated should be that all do not introduce one that contributes maximum in the variable at this moment, and might as well establish its sequence number is k ＇, namely

Its corresponding F test value is:

If F _DrawF α 1, namely under level of significance α 1 meaning, F draws test value greater than critical value F α 1, then this variable is introduced regression equation, otherwise will not introduce.

(3) by calculate determine to draw as variable after, set up regression equation.

(4) significance test of regression equation.Here mainly calculate coefficient of multiple correlation R and F test value, if wherein the F test value is greater than its theoretical value F _Reason=0.5, think that then the result meets the demands.

(2) basic ideas of principal component analysis (PCA) are as follows:

Principal component analysis (PCA) is a kind of statistical method that a plurality of indexs is turned to a few overall target.It is in the reduced data structure, and there is important effect choice variable subclass aspect.In multivariable research, often because the variable number is too many, and exist certain correlativity each other, thereby the information that the feasible data of observing reflect to a certain extent has overlapping in a large number, and work as the variable number more for a long time, the regularity of distribution of research sample is cumbersome in higher dimensional space.Principal component analysis (PCA) is carried out abbreviation to this situation just, namely take a kind of method of dimensionality reduction, find out several multi-stresses and represent original numerous variable, make these multi-stresses can reflect the quantity of information of primal variable as much as possible, and uncorrelated mutually each other.

The specific practice of principal component analysis (PCA) is:

(1) raw data is carried out standardization.

$x_{\mathrm{iK}}^{\&prime;}=\frac{x_{\mathrm{iK}}-\stackrel{\&OverBar;}{x_K}}{S_K}$ (i＝1,2,...,nK＝1,2,...,p) (9)

Wherein:

${\stackrel{\&OverBar;}{x}}_K=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{iK}}$ (i＝1,2,...,nK＝1,2,...,p)

${\mathrm{<figure-callout\; id="2"\; label="S\; K"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">S</figure-callout>}}_K^{}=\frac{1}{n-1}\mathrm{\&Sigma;}{(x_{\mathrm{iK}}-{\stackrel{\&OverBar;}{x}}_K)}^2$ (K=1，2，...,p)

We need carry out standardization with raw data in this example.

(2) calculate correlation matrix

$R=\left[r_{\mathrm{ij}}\right]---\left(10\right)$

Wherein:

$r_{\mathrm{ij}}=(\underset{K=1}{\overset{n}{\mathrm{\&Sigma;}}}{x^{\&prime;}}_{\mathrm{KI}}*{x^{\&prime;}}_{\mathrm{Kj}})/(n-1)$ (i，j=1，2，...，p)

(3) corresponding to correlation matrix R, ask secular equation with jacobi method | R-λ I |=0 p non-negative eigenvalue ₁＞λ ₂＞...＞λ _p〉=0, corresponding to eigenvalue _ICorresponding proper vector be:

$C^{\left(i\right)}=(c_1^{\left(i\right)},c_2^{\left(i\right)}...c_p^{\left(i\right)})$ (i＝1,2,...,p) (11)

And satisfy: $C^{\left(i\right)}\&CenterDot;C^{\left(j\right)}=\underset{K=1}{\overset{p}{\mathrm{\&Sigma;}}}{c}_K^{\left(i\right)}\&CenterDot;c_K^{\left(j\right)}=\left\{\begin{array}{cc}1& i=j\\ 0& (i\&NotEqual;j)\end{array}\right.$

(4) select m(m ﹤ p) individual principal component.If preceding m factor z1, z2 ... zm(m ﹤ p) quantity of information of representative accounts for the α % above (α of the present invention gets 95) of gross information content, be that this m principal component has almost reflected the full detail amount, we just choose preceding m factor z1, z2 ... zm is first principal component, second principal component, m principal component, wherein this m principal component is the linear combination of an original p variable.

(3) multiple linear regression is in Application in Prediction

1, basic ideas

In many cases, we will study the correlationship between a plurality of variablees usually.Under real traffic conditions, particularly in many big and medium-sized cities of China, all generally adopted under the situation of traffic face control system, ubiquity correlationship all between most of crossings of city road network, if predict the magnitude of traffic flow of some crossings this moment, we just must study the correlationship between this crossing and other crossings, and set up with this crossing to be dependent variable and be the regression equation of independent variable with other crossings, to utilize this regression equation to go to predict the magnitude of traffic flow of this crossing.

Set up the multiple regression analysis method that the used method of this regression equation will be introduced just here.The multiple regression analysis method is the correlativity of research between a plurality of variablees, and is that the corresponding data that regression equation goes to predict known variables are set up on the basis with this correlativity.Below we will give brief introduction to multiple linear regression method.

2, model determines

Known that by top narration multiple linear regression analysis method is with a kind of method that solves correlationship between a plurality of variablees.Therefore our crossing of getting telecommunication flow information to be predicted is dependent variable here, and getting all crossings associated therewith is that independent variable is set up multiple linear regression equations.

If variable y and variable x ₁, x ₂..., x _pBetween have linear regression relation, its α time test figure is (y _α; x _{α 1}, x _{α 2}..., x _{α p}) α=1,2 ..., N.

So have

$\left\{\begin{array}{c}{\mathrm{<figure-callout\; id="1"\; label="y"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">y</figure-callout>}}_1={\mathrm{\&beta;}}_0+{\mathrm{\&beta;}}_1{x}_{11}+...+{\mathrm{\&beta;}}_p{x}_{1p}+{\mathrm{\&epsiv;}}_1\\ {\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">y</figure-callout>}}_2={\mathrm{\&beta;}}_0+{\mathrm{\&beta;}}_1{x}_{21}+...+{\mathrm{\&beta;}}_p{x}_{2p}+{\mathrm{\&epsiv;}}_2\\ ......\\ y_N={\mathrm{\&beta;}}_0+{\mathrm{\&beta;}}_1{\mathrm{<figure-callout\; id="1"\; label="x\; n"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">x</figure-callout>}}_n+...+{\mathrm{\&beta;}}_p{x}_{\mathrm{np}}+{\mathrm{\&epsiv;}}_n\end{array}\right.---\left(12\right)$

β wherein ₀, β ₁..., β _pBe p+1 parameter to be estimated, x ₁, x ₂..., x _pHere be p the crossing that can accurately measure its telecommunication flow information, ε ₁, ε ₂..., ε _NBe N separate and obey same normal distribution stochastic variable.

If $X=\left[\begin{array}{cccccc}1& x_{11}& & x_{12}& ...& x_{1p}\\ 1& x_{21}& & x_{22}& ...& x_{2p}\\ & & ...& ...& ....& \\ 1& x_{N1}& & {\mathrm{<figure-callout\; id="2"\; label="x\; N"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">x</figure-callout>}}_N& ...& x_{\mathrm{Np}}\end{array}\right]$

Then Shang Mian multiple regression equation can be expressed as:

Y＝Xβ+ε(13)

Y＝(y ₁,y ₂,...,y _N)' β＝[β ₀,β ₁,...,β _p]' ε＝[ε ₁,ε ₂,...,ε _N]'

Wherein parameter beta can utilize least square method to estimate.Its formula is:

b＝A ^-1B＝(X'X) ^-1X'Y (14)

A is the matrix of coefficients of normal equations Q in the formula, and B is the constant term matrix of normal equations right-hand member.The form of normal equations Q is as follows:

$\left\{\begin{array}{c}{\mathrm{Nb}}_0+\left(\underset{\mathrm{\&alpha;}=2}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;}1}\right){\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">b</figure-callout>}}_1+\left(\underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;}2}\right){\mathrm{<figure-callout\; id="2"\; label="b"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">b</figure-callout>}}_2+...+\left(\underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;p}}\right)b_p=\underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{y}_{\mathrm{\&alpha;}}\\ \left(\underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;}1}\right)b_0+\left(\underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;}1}^2\right){\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">b</figure-callout>}}_1+(\underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;}1}\&CenterDot;x_{\mathrm{\&alpha;}2}){\mathrm{<figure-callout\; id="2"\; label="b"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">b</figure-callout>}}_2+...+(\underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;}1}\&CenterDot;x_{\mathrm{\&alpha;p}})b_p=\underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;}1}\&CenterDot;y_{\mathrm{\&alpha;}}\\ .........\\ \left(\underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;p}}\right)b_0+(\underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;p}}\&CenterDot;x_{\mathrm{\&alpha;}1}){\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">b</figure-callout>}}_1+(\underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;p}}\&CenterDot;x_{\mathrm{\&alpha;}2}){\mathrm{<figure-callout\; id="2"\; label="b"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">b</figure-callout>}}_2+...+\left(\underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;p}}^2\right)b_p=\underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;p}}\&CenterDot;y_{\mathrm{\&alpha;}}\end{array}\right.---\left(15\right)$

The form of coefficient matrices A is as follows:

$A=\left[\begin{array}{ccccc}N& \underset{\mathrm{\&alpha;}}{\mathrm{\&Sigma;}}{x}_{\mathrm{\&alpha;}1}& \underset{\mathrm{\&alpha;}}{\mathrm{\&Sigma;}}{x}_{\mathrm{\&alpha;}2}& ...& \underset{\mathrm{\&alpha;}}{\mathrm{\&Sigma;}}{x}_{\mathrm{\&alpha;p}}\\ \underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;}1}& \underset{\mathrm{\&alpha;}}{\mathrm{\&Sigma;}}{x}_{\mathrm{\&alpha;}1}^2& \underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;}1}\&CenterDot;x_{\mathrm{\&alpha;}2}& ...& \underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;}1}\&CenterDot;x_{\mathrm{\&alpha;p}}\\ & ...& ...& ...& \\ \underset{\mathrm{\&alpha;}}{\mathrm{\&Sigma;}}{x}_{\mathrm{\&alpha;p}}& \underset{\mathrm{\&alpha;}=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_{\mathrm{\&alpha;}1}\&CenterDot;x_{\mathrm{\&alpha;p}}& \underset{\mathrm{\&alpha;}}{\mathrm{\&Sigma;}}{x}_{\mathrm{\&alpha;}2}\&CenterDot;x_{\mathrm{\&alpha;p}}& ...& \underset{\mathrm{\&alpha;}}{\mathrm{\&Sigma;}}{x}_{\mathrm{\&alpha;}\mathrm{<figure-callout\; id="2"\; label="p"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">p</figure-callout>}}^2\end{array}\right]$

$=\left[\begin{array}{cccc}1& 1& ...& 1\\ x_{11}& x_{21}& ...& x_{N1}\\ x_{12}& x_{22}& ...& {\mathrm{<figure-callout\; id="2"\; label="x\; N"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">x</figure-callout>}}_N\\ & ...& ...& \\ x_{1p}& x_{2p}& ...& x_{\mathrm{Np}}\end{array}\right]\&CenterDot;\left[\begin{array}{ccccc}1& x_{11}& x_{12}& ...& x_{1p}\\ 1& x_{21}& x_{22}& ...& x_{2p}\\ & ...& ...& ...& \\ 1& x_{N1}& {\mathrm{<figure-callout\; id="2"\; label="x\; N"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">x</figure-callout>}}_N& ...& x_{\mathrm{Np}}\end{array}\right]=X^{\&prime;}X$

The form of constant term matrix B is as follows:

$B=\left[\begin{array}{c}\underset{\mathrm{\&alpha;}}{\mathrm{\&Sigma;}}{y}_{\mathrm{\&alpha;}}\\ \underset{\mathrm{\&alpha;}}{\mathrm{\&Sigma;}}{x}_{\mathrm{\&alpha;}1}\&CenterDot;y_{\mathrm{\&alpha;}}\\ ...\\ \underset{\mathrm{\&alpha;}}{\mathrm{\&Sigma;}}{x}_{\mathrm{\&alpha;p}}\&CenterDot;y_{\mathrm{\&alpha;}}\end{array}\right]=\left[\begin{array}{cccc}1& 1& ...& 1\\ x_{11}& x_{21}& ...& {\mathrm{<figure-callout\; id="1"\; label="x\; N"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">x</figure-callout>}}_N\\ & ...& ...& \\ x_{1p}& x_{2p}& ...& x_{\mathrm{Np}}\end{array}\right]\&CenterDot;\left[\begin{array}{c}{y}_1\\ {\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">y</figure-callout>}}_2\\ ...\\ y_N\end{array}\right]=X^{\&prime;}Y$

Therefore, the form of normal equations is: (X ' X) b=X ' Y or Ab=B.The b that wherein relates to above is parameter beta estimated value in the regression equation.

So obtain the estimated value b of β be:

b＝A ^-1B＝(X′X) ^-1X′Y (16)

So far, we just can utilize the coefficient substitution full scale equation of the regression equation that following formula calculates to go to predict the data of variable to be measured.

The magnitude of traffic flow of the Forecasting Methodology prediction sensorless crossing that E, utilization are selected;

F, predict the outcome and in conjunction with the cluster pedigree chart, finally determine to bury underground quantity and the position of detecting device according to above-mentioned.

With the data instance that 33 detecting devices of different crossings, Nanjing detected on August 6th, 2004 situation of aforementioned two kinds of arranging signals control crossing detecting device is studied respectively.Shown in the data of these 33 detecting devices were seen attached list, its numbering was as follows:

Two-way 5 tracks, Ye Jie north and south are built in two-way 8 tracks of thing, door street, grassland, reed mat battalion two-way 8 tracks of thing, two-way 6 tracks of East Zhongshan Road thing, two-way 6 tracks, north, the central south of road.

The detecting device of sequence number representative is in the table: wherein xc1-xc8 represents two-way 8 tracks of thing, door street, grassland, numbering: 1-8; Two-way 5 tracks, Ye Jie north and south are built in the xj1-xj5 representative, numbering: 9-13; Xlxy1-xlxy8 represents reed mat battalion two-way 8 tracks, north and south, numbering: 14-21; Xzsd1-xzsd6 represents two-way 6 tracks of East Zhongshan Road thing, numbering: 22-27; Xzyl1-xzyl6 represents two-way 6 tracks, north, the central south of road, numbering: 28-33.

At first utilizing clustering methodology to calculate correlation matrix in this example is shown in Table 1.

Table 1 cluster coefficients table

Annotate: the order of the corresponding aforesaid detector cluster process of the sequence number in the table.

Cluster result is as shown in Figure 2:

Annotate: the corresponding aforesaid detector numbering of the sequence number among the figure.

Here the number that it is pointed out that cluster should be determined according to actual needs.But the variable especially little to those related coefficients (crossing) should a single-row class, to guarantee the meaning of classification.As No. 9, No. 10 and No. 22 detecting device in this example, their classification related coefficients only are 0.3441,0.4995 and 0.5974.It is abundant inadequately that this illustrates that the magnitude of traffic flow of other crossings comprises the quantity of information of these three crossing magnitudes of traffic flow.Therefore, these three crossings should a single-row class.

1. track (or highway section) is combined into a class in twos, perhaps track (or highway section) number less (generally being lower than 5 is advisable) in the subclass

For this kind situation, when laying detecting device, can utilize the correlativity between them fully, the method for monobasic or multiple linear regression of using is set up the telecommunication flow information forecast model.In this example, as choose detecting device numbering and be respectively 18,19,20,21 track, if choose track, No. 19 places as laying the detecting device track, then utilize the forecast model in its excess-three track that monobasic or multiple linear regression analysis method set up as shown in table 2.

Table 2 track forecast model

From table 2, all by the F check, precision of prediction all can satisfy request for utilization to the forecast model of 18,19,20,21 lane detectors as can be seen.

2. su generis track (or highway section), perhaps track (or highway section) number is more in the subclass

For this kind situation, when laying detecting device, then need on the basis of above-mentioned cluster analysis result, use the forecast model that principal component analysis (PCA) or stepwise regression analysis method are determined this type of track (or highway section).

(1) principal component analysis (PCA)

Here we are that predict in unknown track to build No. 30 detecting device track of Ye, an ancient place in Henan Province street south orientation, and No. 30 detecting device track and No. 23 detecting device of East Zhongshan Road east orientation, west are classified as a class to the 24th, 25, No. 26 detecting device and the central south of road to No. 29 detecting device, No. 32 detecting device track of north orientation as can be known by above-mentioned cluster pedigree chart.Therefore, our telecommunication flow information that only need get these 6 tracks when carrying out principal component analysis (PCA) gets final product.Concrete steps are as follows:

(1) raw data is carried out standardization.

$x_{\mathrm{iK}}^{\&prime;}=\frac{x_{\mathrm{iK}}-\stackrel{\&OverBar;}{x_K}}{S_K}$ (i＝1,2,?,n;K＝1,2,?p) (17)

In the formula:

${\stackrel{\&OverBar;}{x}}_K=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{iK}}$

${\mathrm{<figure-callout\; id="2"\; label="S\; K"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">S</figure-callout>}}_K^{}=\frac{1}{n-1}\mathrm{\&Sigma;}{(x_{\mathrm{iK}}-{\stackrel{\&OverBar;}{x}}_K)}^2$ (K＝1,2,?,p)

(2) calculate correlation matrix R.

R＝[r _ij] (18)

In the formula:

$r_{\mathrm{ij}}=(\underset{K=1}{\overset{n}{\mathrm{\&Sigma;}}}{x^{\&prime;}}_{\mathrm{Ki}}*{x^{\&prime;}}_{\mathrm{Kj}})/(n-1)$ (i,j＝1,2,?,p)

The correlation matrix that utilizes above-mentioned data computation of passing through standardization to obtain in this example is as shown in table 3.

Table 3 related coefficient

The detecting device numbering	xzsd2	xzsd3	xzsd4	xzsd5	xzyl2	xzyl5
							xzsd2	1.0000	0.9276	0.9718	0.1457	0.7903	0.8052

xzsd3	0.9276	1.0000	0.9215	0.2200	0.8100	0.8304
							xzsd4	0.9718	0.9215	1.0000	0.1083	0.7774	0.8190
xzsd5	0.1457	0.2200	0.1083	1.0000	-0.0413	0.1384
							xzyl2	0.7903	0.8100	0.7774	-0.0413	1.0000	0.8125
xzyl5	0.8052	0.8304	0.8190	0.1384	0.8125	1.0000

As can be seen from Table 3, have and have certain correlationship between the detecting device crossing, and have the correlativity between some detecting devices also more intense, allow my couplet on the door expect having between the track magnitude of traffic flow of the track magnitude of traffic flow of detecting device crossing and sensorless crossing so very naturally and necessarily have correlationship.

(3) corresponding to correlation matrix R, ask secular equation with jacobi method | R-λ I |=0 p non-negative eigenvalue ₁＞λ ₂＞...＞λ _p〉=0, corresponding to eigenvalue _ICorresponding proper vector be:

$C^{\left(i\right)}=(c_1^{\left(i\right)},c_2^{\left(i\right)}...c_p^{\left(i\right)})$ (i＝1,2...p) (19)

And satisfy:

$C^{\left(i\right)}\&CenterDot;C^{\left(j\right)}=\underset{K=1}{\overset{p}{\mathrm{\&Sigma;}}}{c}_K^{\left(i\right)}\&CenterDot;c_K^{\left(j\right)}=\left\{\begin{array}{cc}1& (i=j)\\ 0& (i\&NotEqual;j)\end{array}\right.$

Utilize the method for above-mentioned introduction, the eigenwert that we utilize correlation matrix R to calculate is shown in Table 4.

Table 4 list of feature values

Sequence number	Eigenwert	Each eigenwert accounts for overall number percent (%)	Cumulative percentage (%)
				1	4.4103	73.505	73.505
2	1.0230	17.05	90.555
				3	0.2948	4.913	95.468
4	0.1723	2.8717	98.3397
				5	0.0741	1.235	99.575

6

0.0255

0.425

100

6 new factors being made up of proper vector are

z ₁＝0.3432·x ₁′-0.7011·x ₂′+0.0486·x ₃′+0.4224·x ₄′-0.0056·x ₅′+0.4581·x ₆′

z ₂＝-0.8571·x ₁′-0.0074·x ₂′+0.1424·x ₃′+0.1693·x ₄′-0.0781·x ₅′+0.4586·x ₆′

z ₃＝0.3028·x ₁′+0.7023·x ₂′-0.1170·x ₃′+0.4378·x ₄′+0.0287·x ₅′+0.4570·x ₆′

z ₄＝0.1222·x ₁′+0.0433·x ₂′+0.1230·x ₃′-0.1370·x ₄′-0.9713·x ₅′+0.0761·x ₆′

z ₅＝0.2022·x _′+0.0830·x ₂′+0.6459·x ₃′-0.5550·x ₄′+0.2223·x ₅′+0.4215·x ₆′

z ₆＝0.0067·x ₁′-0.0800·x ₂′-0.7290·x ₃′-0.5239·x ₄′+0.0128·x ₅′+0.4331·x ₆′

In the following formula: x ₁, x ₂, x ₃, x ₄, x ₅, x ₆Be the data of raw data through standardization, new factor z ₁, z ₂, z ₃, z ₄, z ₅, z ₆Between mutually orthogonal, and their variance is successively decreased.

(4) select m(m ﹤ p) individual principal component.If preceding m factor z1, z2 ... zm(m ﹤ p) quantity of information of representative accounts for the α % above (α of the present invention gets 95) of gross information content, be that this m principal component has almost reflected the full detail amount, we just choose preceding m factor z1, z2 ... zm is first principal component, second principal component, m principal component, wherein this m principal component is the linear combination of an original p variable.

The accumulated value that in the middle of this example, utilizes information that eigenwert that previous step calculates and principal component reflect to account for the 95%(of full detail amount that is computation of characteristic values account for the eigenwert summation 95%) this condition chooses principal component, choose preceding two factors as principal component, these two linear combinations that the factor is all 9 variablees through calculating us.The value of these two principal components is shown in Table 5.

Table 5 principal component table

Principal component analysis (PCA) of the present invention is a kind of data processing method, the track data on flows that utilization has detecting device crossing actual detected to arrive is constructed principal component z1, z2...zm by said method, connect by the magnitude of traffic flow and the principal component of multiple linear regression analysis with each track, sensorless crossing, thereby the magnitude of traffic flow in each track, sensorless crossing is predicted.Here the central south of road that detecting device is arranged is assumed to be the sensorless crossing to No. 30 detecting device, utilizes value and this m principal component of track, the sensorless crossing magnitude of traffic flow of standardization to carry out multiple linear regression analysis.

Here the multiple linear regression equations that obtains by regretional analysis is:

y=5.4726E-18+22.1198*z ₁+2.6958*z ₂+15.4557*z ₃ （20）

Adopt F check in this example, calculate the F value be 15.0384, its theoretical value is: 2.13.By check.Multiple correlation coefficient is: R=0.9173.Here need to prove that the predicted value that obtains through standardization, also need be reduced into original data layout here, concrete predicted value is shown in Table 6.

Table 6 predicted value and the detected value table of comparisons

From formula (20) as can be seen, after principal component analysis (PCA) calculating, original 6 variablees comprehensively are 3 principal components finally, can make that like this stability of model is better, are conducive to improve accuracy of predicting.

(2) stepwise regression analysis method

Equally, we also can adopt the method that progressively returns to set up forecast model on the basis of cluster analysis result.

On above-mentioned clustering result basis, we carry out stepwise regression analysis.Here we are that predict in unknown track to build No. 30 detecting device track of Ye, an ancient place in Henan Province street south orientation equally, and No. 30 detecting device track and No. 23 detecting device of East Zhongshan Road east orientation, west are classified as a class to the 24th, 25, No. 26 detecting device and the central south of road to No. 29 detecting device, No. 32 detecting device track of north orientation as can be known by above-mentioned cluster pedigree chart.Therefore, our telecommunication flow information that only need get these 6 tracks when carrying out stepwise regression analysis gets final product.Concrete steps are as follows:

(1) calculates the variable average.

$\stackrel{\&OverBar;}{x_i}=\frac{1}{n}\underset{K=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_{\mathrm{Ki}}$ (i＝1,2,?,p)

$\stackrel{\&OverBar;}{y}=\frac{1}{n}\underset{K=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_K---\left(21\right)$

(2) calculate the deviation matrix.

$\begin{array}{c}{S}_{\mathrm{ij}}=S_{\mathrm{ji}}=\underset{K=1}{\overset{n}{\mathrm{\&Sigma;}}}(x_{\mathrm{Ki}}-\stackrel{\&OverBar;}{x_i})\&CenterDot;(x_{\mathrm{Kj}}-\stackrel{\&OverBar;}{x_j})\\ S_{\mathrm{iy}}=S_{\mathrm{yi}}=\underset{K=1}{\overset{n}{\mathrm{\&Sigma;}}}(x_{\mathrm{Kj}}-\stackrel{\&OverBar;}{x_i})\&CenterDot;(y_k-\stackrel{\&OverBar;}{y})\\ S_{\mathrm{yy}}=\underset{K=1}{\overset{n}{\mathrm{\&Sigma;}}}{(y_K-\stackrel{\&OverBar;}{y})}^2\end{array}$ (i，j=1，2，...，p） (22)

(3) calculate correlation matrix.In this example, the correlation matrix that obtains is shown in Table 7.

$r_{\mathrm{ij}}=r_{\mathrm{ji}}=\frac{S_{\mathrm{ij}}}{\sqrt{S_{\mathrm{ii}}}\&CenterDot;\sqrt{S_{\mathrm{jj}}}}$ (i,j＝1,2,?,p,y) （23）

Table 7 coefficient R

Detecting device n	xzyl3	xzsd3	xzsd4	xzsd5	xzsd6	xzyl2	xzyl5
								zxl3	1.0000	0.9276	0.9718	0.1457	0.7903	0.8052	0.8997
xzsd3	0.9276	1.0000	0.9215	0.2200	0.8100	0.8304	0.8417
								xzsd4	0.9718	0.9215	1.0000	0.1083	0.7774	0.8190	0.8844
xzsd5	0.1457	0.2200	0.1083	1.0000	-0.0413	0.1384	-0.0621
								xzsd6	0.7903	0.8100	0.7774	-0.0413	1.0000	0.8125	0.8498
xzyl2	0.8052	0.8304	0.8190	0.1384	0.8125	1.0000	0.6975
								xzyl5	0.8997	0.8417	0.8844	-0.0621	0.8498	0.6975	1.0000

(4) introducing of regression variable and rejecting

Can prove that in regression process progressively, the variable that m+1 step and m+2 step introduce can not go on foot disallowable at m+3.Therefore, three step is to introduce variable entirely.

Utilize R(k) (k=0,1,2) calculate the contribution that first three goes on foot each variable

$V_i^{\left(m\right)}=\frac{{\left(r_{\mathrm{iy}}^{\left(k\right)}\right)}^2}{r_{\mathrm{ij}}^{\left(k\right)}}$ (i=1,2,, p) (k=0,1,2) (m=1,2,3) (24) are found out and are made Vi(1) and get peaked sequence number k1, namely have

$V_{k1}^{\left(m\right)}=\underset{1\&le;i\&le;p}{\mathrm{<figure-callout\; id="1"\; label="max"\; filenames="HDA00002973014300021.png"\; state="\{\{state\}\}">max</figure-callout>}}\left\{V_i^{\left(m\right)}\right\}$ M=1,2,3 (25) calculate then:

(m＝1,2,3) (k＝0,1,2) （26）

If F draws〉F α 1, then xK1 introducing regression equation and by the correlation matrix transformation for mula compute matrix R(1 that introduces previously)=(rij(1)); If F draws≤F α 1, then stop to calculate.

Result of calculation is as shown in table 8 in this example:

Table 8 is Return Law result of calculation progressively

As can be seen from the table, through the Return Law progressively, introduce the variable number of regression equation and have only 1 on the basis of cluster, the stability that this has increased equation greatly is conducive to improve precision of prediction.

(5) the regression equation calculation result is as follows:

y＝(4.134－105.43*x)*1.0e-007 （27）

The assay of regression equation is as follows: F test value F=14.1307, and greater than its theoretical value F reason=0.5.Its coefficient of multiple correlation R=0.9278 also meets the demands.Here need to prove, is standardized value by the resulting y value of top model, also need be reduced into raw data format, and resulting predicted value result is as shown in table 9:

Table 9 model prediction result

As can be seen from Table 9, prediction result and original value have certain error, and this mainly is owing to introduce the information that the variable of model can't comprise whole road network variablees (lane detector), still is enough to satisfy the requirement on the precision of prediction.

In like manner, lay to adopt to use the same method for the highway section detecting device and set up regressive prediction model.But, it should be noted that because the significantly raising of social economy and people's purchasing power impels the quick growth of vehicle population, is example with the Daliang City, Daliang City's motor vehicle every day increases with 200/day speed, and every month, the Dalian vehicle population will increase about 6000 so through calculating.Therefore, the correlativity between track in the city road network (or highway section) can keep certain stability in a short time, but this stability can for good and all not kept.Based on above-mentioned situation, needed per 3 months or whether just should check its installation site about half a year again reasonable for the detecting device of the signalized crossing that adopts said method to lay? should adjust?

Generally speaking, the present invention at first utilizes statistical correlation technique to set up the cluster pedigree chart between each crossing in institute's favored area, then on the basis of cluster pedigree chart, earlier defaultly choose some crossings in the class as the target crossing of laying detecting device, the telecommunication flow information that contains other crossings in the class of a small amount of crossing then can utilize the correlativity of itself and target crossing to set up the linear regression predictive equation to come the magnitude of traffic flow of this crossing is predicted.Simultaneously, can adopt two kinds of methods to obtain its traffic flow information for the crossing that constitutes a class by itself individually in the road network or contain in a fairly large number of class in crossing:

1, with its target crossing as the laying detecting device, can increase infrastructure investment and workload that the crossing detecting device is laid so greatly, but its telecommunication flow information that obtains accurately and reliably;

2, be according to classification under it, use principal component analysis (PCA) or stepwise regression analysis method and set up its model of traffic flux forecast, may reduce infrastructure investment and workload that the crossing detecting device is laid like this, but because the magnitude of traffic flow of some detecting device also utilizes forecast model to obtain in its affiliated classification, therefore may cause the increase of cumulative errors as a result of its traffic flow forecasting.

By adopting above-mentioned detecting device distribution method, then can utilize limited detecting device to obtain the traffic flow information of most even whole crossings in the whole road network.

Claims (1)

1. the distribution method of a city road network traffic flow amount detector is characterized in that: may further comprise the steps:

The laying mode of A, selection detecting device

Select the laying mode of detecting device according to the characteristics of different Area Traffic Control Systems, if the zone centered by each administrative residential quarter or traffic zone control center then selects the distributing detecting device to lay mode; If be to lay the zone or select and determine that detecting device lays the central area by the Competent Authorities of Transport and Communications with the shopping centre, then select center radiation formula detecting device to lay mode;

B, carry out cluster analysis

Utilize clustering methodology to set up the cluster pedigree chart between interior each crossing in the above-mentioned zone;

Position and the quantity of C, default detecting device

At first according to the number of the shown class that goes out in the cluster pedigree chart, in each class, select some crossings of merging for the first time as the target crossing of default detecting device; For su generis crossing, earlier it is assumed to the target crossing of laying detecting device, obtain the target crossing quantity of default detecting device like this;

D, selection Forecasting Methodology

For the class of having only a small amount of number crossing, select the multiple linear regression Forecasting Methodology for use;

For su generis crossing or contain a fairly large number of class in crossing, use a kind of of following two kinds of methods:

D1, elder generation use the principal component analysis (PCA) predicted method, carry out combined prediction with the multiple linear regression Forecasting Methodology again;

D2, progressively Regression Forecast of elder generation are carried out combined prediction with the multiple linear regression Forecasting Methodology again;

The magnitude of traffic flow of the Forecasting Methodology prediction sensorless crossing that E, utilization are selected;

F, predict the outcome and in conjunction with the cluster pedigree chart, finally determine to bury underground quantity and the position of detecting device according to above-mentioned.

Images