CN104809877A - Expressway site traffic state estimation method based on feature parameter weighted GEFCM algorithm - Google Patents

Abstract

The invention belongs to the technical field of a road traffic control system, and discloses an expressway site traffic state estimation method based on a feature parameter weighted GEFCM algorithm. The method comprises the following steps that (1) historical data of three feature parameters including the vehicle flow rate, the average vehicle speed and the average occupation rate are obtained through collection of an expressway microwave vehicle detector, and a sample matrix is formed; (2) the data obtained in the first step is subjected to preprocessing, and the preprocessing includes wrong data recognition and deletion, data restoration and data filtering processing; (3) the weight of the three kinds of feature parameters during clustering analysis is determined; (4) the historical data is subjected to clustering analysis; (5) when the traffic flow parameter of the current cross section is obtained, the traffic state is estimated in real time. The method has the advantages that the imbalance in the historical traffic data sample during the clustering is considered, and the difference of different traffic flow parameters on the clustering is considered, so that a provided feature parameter weighted GEFCM model has a better clustering effect, and further, a better effect and higher reliability are also realized on the traffic state estimation.

Description

The highway place traffic state estimation method of feature based parameter weighting GEFCM algorithm

Technical field

The invention belongs to road traffic control system technical field, concrete is a kind of highway place traffic state estimation method.

Background technology

The importance accounted in China's communications and transportation along with highway is increasing, also more and more serious along with problems such as the traffic congestion occurred, traffic hazard, environmental pollutions.That traffic administration person or the information system management demand of traveler to traffic are all increasing gradually, therefore, how to utilize existing checkout equipment, realize the estimation of traffic status of express way as far as possible effectively and accurately, the traffic holding present road in real time is accurately the prerequisite of efficient management and countermeasure, has important theoretical research and practical application meaning.

Highway is installed the various equipment for traffic data collection, as fixed detector, video detector, Floating Car etc.But, due to the various reason such as coverage rate, cost, making to use in the research estimated for traffic status of express way at present more is the data of fixed detecting device, and the place traffic state estimation method based on fixed detector data is also diversified:

(1) Traffic transport system engineering and information (the 5th volume the 1st phase, in February, 2005) disclose a kind of Classification of Traffic Flow Situation of Urban Freeways method based on fuzzy clustering, it utilizes the classification of the method for fuzzy clustering to traffic flow conditions to be studied, experimental result shows: carrying out traffic flow conditions classification with fuzzy clustering is a kind of feasible method, different traffic flow parameters is different for the impact of classification, very high in speed, directly traffic flow conditions can be judged when speed is very low or occupation rate is very large, need in other situations to carry out comprehensive descision according to traffic flow three variablees,

(2) Transportation science (the 41st volume the 2nd phase, in May, 2007) disclose a kind of traffic status of express way method of estimation based on EKF, it is to be arranged on the data of the detecting device detection on the specific section of highway for input, by the random macroscopic traffic flow of design, and realizing differentiation to road traffic state by means of the method for EKF, experimental result shows that the method can reflect the change of traffic behavior on real road to a certain extent;

(3) Communication and Transportation Engineering and information journal (the 5th volume the 3rd phase, in September, 2007) disclose a kind of road traffic state decision method based on Genetic-Dynamic Fuzzy Clustering Algorithm, it continues to optimize the mapping of the fuzzy comparability between traffic flow parameter and the Euclidean distance between sample by genetic algorithm, achieve dynamic fuzzy clustering, experimental result shows validity and the feasibility of the method;

(4) highway engineering (the 33rd volume the 2nd phase, in April, 2008) disclose a kind of based on fuzzy traffic stream state of city quick road method of discrimination, it is according to the fuzzy characteristics of traffic behavior, in conjunction with knowledge-based fuzzy system, propose the fuzzy set and fuzzy rule that divide for traffic behavior, and traffic behavior is divided into five kinds, the method can show the congested in traffic scope of road network dynamically, for the differentiation and improvement implementing Traffic information demonstration and later stage traffic bottlenecks provides foundation;

(5) systems engineering (the 28th volume the 8th phase, in August, 2010) disclose a kind of urban expressway traffic condition discrimination method based on FCM-rough set, it is for the characteristic of the traffic behavior of city expressway, the differentiation of primary study to the recurrent congestion of through street, experimental result shows that model is feasible under certain condition, effectively can process magnanimity Multiple Source Sensor data, there is higher differentiation rate and lower False Rate.

Make a general survey of the above various method based on fixed detector data, mostly adopt speed, flow, occupation rate three parameter to carry out cluster analysis, and then judge traffic behavior.Cluster analysis is mainly to the analysis of historical sample data, make the correlativity between the data under identical category attribute large, data dependence between different classes of is little, but by finding the analysis of historical sample, the space distribution of sample also exists lack of uniformity, namely the sample size of different conditions classification there are differences, and traditional FCM is responsive to sample size when cluster, can produce erroneous judgement like this when carrying out cluster to this kind of data.In addition, can also find that different traffic flow parameters is different for impact during cluster, therefore, need to consider the otherness of different classes of sample size and the otherness of Different Traffic Flows parameter when adopting cluster analysis to estimate traffic behavior, like this could more scientific and rational estimation traffic behavior.

Summary of the invention

In view of this, the object of the present invention is to provide a kind of highway place traffic state estimation method of feature based parameter weighting GEFCM algorithm, the otherness feature of Data distribution8 lack of uniformity and different characteristic parameters weighting in historical sample can be considered, by adjusting the objective function of cluster, and then optimization Clustering Model, thus reach the object of traffic behavior estimation, improve the reliability of state estimation.

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

The highway place traffic state estimation method of feature based parameter weighting GEFCM algorithm, comprises the following steps:

1) obtain the historical data of these three kinds of characteristic parameters of vehicle flowrate, average speed and average occupancy that highway microwave vehicle checker collects, form sample matrix;

2) to step 1) data that obtain carry out pre-service, and described pre-service comprises the identification of misdata and rejecting, the reparation of data, data filtering process;

3) weight of three kinds of characteristic parameters when cluster analysis is determined;

4) cluster analysis is carried out to historical data;

5) when getting the traffic flow parameter of current section, traffic behavior is estimated in real time.

Further, described step 2) in, specifically adopt the identification and rejecting of carrying out misdata with the following method:

Within a data update cycle, set the threshold range of total vehicle flowrate data as [0, Q _max], the threshold range of average speed is [0, V _max]; If when the data of the total vehicle flowrate data collected or average speed are not in the threshold range of correspondence, then show that these group data are unreliable, and rejected; If when the data of the total vehicle flowrate data collected and average number of vehicles all drop in corresponding threshold range, then show that these group data are reliable, retain this group data; Wherein, Q _max, V _maxbe illustrated respectively in the flow maximum in the data update cycle and speed maximal value;

Set up misdata judgment rule according to traffic flow theory, namely reject rule; Then, judge whether the data sequence gathered meets and reject rule; When satisfied rejecting rule, the data of correspondence are needed reject; When not meeting rejecting rule, retain corresponding data.

Further, described step 2) in, by following formula, data are repaired:

$\hat{x}\left(t\right)=\mathrm{\α}\·x(t-1)+(1-\mathrm{\α})\·x^{\text{'}}\left(t\right)$

Wherein, for the data restore value of t period; X (t-1) is the actual detected value of (t-1) period; X ' (t) is the history average of the image data of n before synchronization days; α is forgetting factor, α ∈ [0,1];

Further, by following formula, filtering process is carried out to data:

S _t＝αX _t+(1-α)S _t-1

In formula, S _tfor the single exponential smoothing value that the t period obtains; S _t-1for the single exponential smoothing value that the t-1 period obtains; X _tfor the observed reading that the t period obtains; α ∈ [0,1] is smoothing factor.

Further, described step 3) specifically comprise the steps:

31) by by step 2) the sample matrix X that forms of these three kinds of characteristic parameters of pretreated vehicle flowrate, average speed and average occupancy carries out Z-score standardization, the matrix Z after obtaining standardization;

32) Calculation of correlation factor is carried out to the matrix Z after Z-score standardization, obtain correlation matrix R;

33) its secular equation being constructed to correlation matrix R | R-λ I|=0, obtains p characteristic root and proper vector, and b _tand c _tt=1,2 ..., p;

Wherein, b _tbe the variance contribution ratio of t major component, c _tfor the accumulative variance contribution ratio of a front t major component;

34) the weight coefficient ω of these three kinds of characteristic parameters of vehicle flowrate, average speed and average occupancy is calculated _o={ ω _o1, ω _o2..., ω _ot.

Further, described step 4) in, by the following method cluster analysis is carried out to historical data:

Iterate and calculate three formulas below, until meet algorithm stop condition, finally determine the cluster centre representing different traffic class;

$J(U,c_1,...,c_c,\mathrm{\ω})=\underset{i=1}{\overset{c}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}\frac{u_{\mathrm{ij}}^m}{\underset{s=1}{\overset{n}{\mathrm{\Σ}}}{u}_{\mathrm{is}}^m}\underset{k=1}{\overset{t}{\mathrm{\Σ}}}{\mathrm{\ω}}_k^2{(x_{\mathrm{jk}}-c_{\mathrm{ik}})}^2;$

$c_i=\frac{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{u}_{\mathrm{ij}}^m{x}_j}{\underset{s=1}{\overset{n}{\mathrm{\Σ}}}{u}_{\mathrm{is}}^m};$

$u_{\mathrm{ij}}=\frac{{\left[\frac{{\left(\underset{s=1}{\overset{n}{\mathrm{\Σ}}}{u}_{\mathrm{is}}^m\right)}^2}{(\underset{s=1}{\overset{n}{\mathrm{\Σ}}}{u}_{\mathrm{is}}^m-u_{\mathrm{ij}}^m)\underset{k=1}{\overset{t}{\mathrm{\Σ}}}{\mathrm{\ω}}_k^2{(x_{\mathrm{jk}}-c_{\mathrm{ik}})}^2}\right]}^{\frac{1}{m-1}}}{\underset{\mathrm{\γ}=1}{\overset{c}{\mathrm{\Σ}}}{\left[\frac{{\left(\underset{s=1}{\overset{n}{\mathrm{\Σ}}}{u}_{\mathrm{\γs}}^m\right)}^2}{(\underset{s=1}{\overset{n}{\mathrm{\Σ}}}{u}_{\mathrm{\γs}}^m-u_{\mathrm{\γj}}^m)\underset{k=1}{\overset{t}{\mathrm{\Σ}}}{\mathrm{\ω}}_k^2{(x_{\mathrm{jk}}-c_{\mathrm{\γk}})}^2}\right]}^{\frac{1}{m-1}}};$

Wherein, x _jfor a jth sample point, x _jka kth characteristic parameter of a jth sample point; U=(u _ik) _{c × n}for subordinated-degree matrix, use u _ijrepresent that a jth sample belongs to the degree of membership of the i-th class, 0≤u _ij≤ 1, m ∈ [1 ,+∞) represent FUZZY WEIGHTED index, characterize the fog-level of subordinated-degree matrix; c _i=(c _i1, c _i2... c _ip) (i=1,2 ..., c) represent different classes of cluster centre; item has showed the capacity attribute of the i-th class; ω={ ω ₁, ω ₂..., ω _t, ω _k∈ [0,1], is a characteristic parameter weight corresponding with the argument sequence of input, represents the importance of a kth characteristic parameter in cluster; T is the number of characteristic parameter; N is sample point quantity.

Further, described step 5) in, by the calculating of following formula, compare the distance at real time data sample and each state class center, based on the principle determination traffic behavior of bee-line;

${\left(d_{\mathrm{ij}}^{\mathrm{\ω}}\right)}^2={\left|\right|\mathrm{\ω}(x_j-c_i)\left|\right|}^2=\underset{k=1}{\overset{t}{\mathrm{\Σ}}}{\mathrm{\ω}}_k^2{(x_{\mathrm{jk}}-c_{\mathrm{ik}})}^2$

Wherein, represent i-th cluster centre c _iwith a jth sample point x _jbetween distance; ω={ ω ₁, ω ₂..., ω _t, ω _k∈ [0,1], is a characteristic parameter weight corresponding with the argument sequence of input, represents the importance of a kth characteristic parameter in cluster; T is the number of characteristic parameter, is 3 for this paper traffic behavior is estimated.

The traffic behavior corresponding based on the principle determination current flows parameter of bee-line is: for sample point x _j, calculate the distance with three class traffic behavior centers respectively according to above formula, obtain with unimpeded, walk or drive slowly, the distance of the three class status center that block up is respectively the relatively size of three's distance, status categories corresponding to minimum distance is then current traffic behavior estimated value.

The present invention has the following advantages relative to prior art tool: the highway place traffic state estimation method of feature based parameter weighting GEFCM algorithm, algorithm is simple, efficiency is high, from the angle that traffic data sample distribution lack of uniformity is different with the impact of Different Traffic Flows parameter on cluster, propose characteristic parameter weighting GEFCM algorithm, weaken the impact of Different categories of samples capacity volume variance on clustering and discriminant, adopt principal component analysis (PCA) to determine the weight of different characteristic parameter simultaneously.While the lack of uniformity that the present invention exists in historical traffic data sample when considering cluster, consider the otherness of Different Traffic Flows parameter for the impact of cluster, thus make the characteristic parameter weighting GEFCM Clustering Model proposed have better Clustering Effect, and then also there are better effect and reliability when the estimation to traffic behavior.

Accompanying drawing explanation

Fig. 1 shows the schematic flow sheet of the highway place traffic state estimation method of feature based parameter weighting GEFCM algorithm.

Embodiment

In order to make the object, technical solutions and advantages of the present invention clearly, will be described in further detail the specific embodiment of the present invention below.

See Fig. 1, the highway place traffic state estimation method of the feature based parameter weighting GEFCM algorithm of the present embodiment, comprises the following steps:

1) vehicle flowrate (q that highway microwave vehicle checker collects is obtained ₁~ q _n), average speed with average occupancy (o ₁~ o _n) historical data of these three kinds of characteristic parameters, form sample matrix $x=\left\{\begin{array}{ccc}\stackrel{\‾}{v_1}& q_1& o_1\\ \stackrel{\‾}{v_2}& q_2& o_2\\ .& .& .\\ .& .& .\\ .& .& .\\ \stackrel{\‾}{v_n}& q_n& o_n\end{array}\right\};$

For cluster analysis, need historical data to have certain richness, the sample size n of the data sequence of the present embodiment chooses the data of current time the last week.

2) data of actual acquisition also exist a series of quality problems, comprise the disappearance of data, invalid, repetition, redundancy, mistake etc., to step 1) historical data that obtains carries out pre-service, and reject misdata and carry out the reparation of data sequence, obtaining the sample matrix X after processing;

Concrete, specifically adopt threshold theory and traffic flow theory identification and reject the misdata not meeting traffic actual conditions:

Read step 1) historical data that obtains; Within a data update cycle, set the threshold range of total vehicle flowrate data as [0, Q _max], the threshold range of average speed is [0, V _max]; If when the data of the total vehicle flowrate data collected or average speed are not in the threshold range of correspondence, then show that these group data are unreliable, and rejected; If when the data of the total vehicle flowrate data collected and average number of vehicles all drop in corresponding threshold range, then show that these group data are reliable, retain this group data; Wherein, Q _max, V _maxbe illustrated respectively in the flow maximum in the data update cycle and speed maximal value, the present embodiment Q _maxvalue is 300, V _maxvalue is 150km/h.

Set up misdata judgment rule according to traffic flow theory, namely reject rule; Then, judge whether the data sequence gathered meets and reject rule; When satisfied rejecting rule, the data of correspondence are needed reject; When not meeting rejecting rule, retain corresponding data; The present embodiment traffic flow theory misdata judgment rule is as shown in table 1.

Table 1: based on the misdata decision rule of traffic flow theory

The value that the present invention adopts the measured data of current road segment and the weighting scheme of historical data to draw is repaired fault data, the method had both considered the real data of current section last period, consider again the trend of the data of same time in historical data, shown in formula specific as follows:

$\hat{x}\left(t\right)=\mathrm{\α}\·x(t-1)+(1-\mathrm{\α})\·x^{\text{'}}\left(t\right);$

Wherein, for the data restore value of t period; X (t-1) is the actual detected value of (t-1) period; X ' (t) is the history average of the image data of n before synchronization days; α is forgetting factor, α ∈ [0,1], and the size of α value determines the data dependence degree for history.

The present invention adopts exponential smoothing to carry out filtering process to data, by eliminating maximum value and the minimal value of data, obtaining the smooth value of data sequence, is a kind of method being distinguished master data pattern and random fluctuation by smooth historical data, the mathematical model of Single Exponential Smoothing as shown in the formula:

S _t＝αX _t+(1-α)S _t-1

In formula, S _tfor the single exponential smoothing value that the t period obtains; S _t-1for the single exponential smoothing value that the t-1 period obtains; X _tfor the observed reading that the t period obtains; α ∈ [0,1] is smoothing factor.

3) for different traffic flow parameters for the different feature of the impact of cluster, the present embodiment adopts the matrix of principal component analysis (PCA) to the traffic flow parameter obtained to carry out main eigen, determines the weight of three kinds of characteristic parameters when cluster analysis; Specifically comprise the steps:

31) by by step 2) the sample matrix X that forms of these three kinds of characteristic parameters of pretreated vehicle flowrate, average speed and average occupancy carries out Z-score standardization, the matrix Z after obtaining standardization;

Because the scope of different its values of traffic flow parameter is different, so before analyzing, under needing to be transformed into identical dimension, namely standardization is carried out to sample data.

Get n traffic flow parameter sample sequence X=(X ₁, X ₂..., X _n) ^t, each sample is made up of p vector, i.e. X _i=(x _i1, x _i2..., x _ip), i=1,2 ..., n, is configured to matrix z-score standardized transformation is carried out to sample:

$Z_{\mathrm{ij}}=\frac{x_{\mathrm{ij}}-\stackrel{\‾}{x_j}}{s_j},i=\mathrm{1,2},...,p;$

Wherein, what represent is the average of sample data sequence, what represent is the variance of sample data sequence.

32) Calculation of correlation factor is carried out to the matrix Z after Z-score standardization, obtains correlation matrix R:

$R={\left(r_{\mathrm{ij}}\right)}_{p\×p}=\frac{Z^{T}Z}{n-1};$

33) its secular equation being constructed to correlation matrix R | R-λ I|=0, obtains p characteristic root λ ₁, λ ₂..., λ _pwith proper vector e ₁, e ₂..., e _p, thus obtain b _t, c _tt=1,2 ..., p:

$b_t=\frac{\mathrm{\λ}}{\underset{j=1}{\overset{p}{\mathrm{\Σ}}}{\mathrm{\λ}}_j},t=\mathrm{1,2},...,p;$

$c_t=\frac{\underset{j=1}{\overset{t}{\mathrm{\Σ}}}{\mathrm{\λ}}_j}{\underset{j=1}{\overset{p}{\mathrm{\Σ}}}{\mathrm{\λ}}_j},t=\mathrm{1,2},...,p;$

Then b _tbe the variance contribution ratio of t major component, c _tfor the accumulative variance contribution ratio of a front t major component, usually in Feature Selection problem, work as c _twhen>=0.85, just can represent whole index by a front t major component, can t=2 be obtained herein by analysis.

And obtain the component matrix of each major component for original index: $\left[\begin{array}{ccc}{r}_{11}& r_{12}& r_{1t}\\ .& .& .\\ .& .& .\\ .& .& .\\ r_{p1}& r_{p2}& r_{\mathrm{pt}}\end{array}\right];$

Wherein, for charge number represents the significance level of each major component for original index.

34) to extract square root the coefficient just obtained in two major components corresponding to each index divided by the characteristic root that major component is corresponding with the charge number in component matrix, two major components obtained are as follows:

$\left\{\begin{array}{c}{F}_1={\mathrm{\α}}_1{X}_1+{\mathrm{\α}}_2{X}_2+{\mathrm{\α}}_3{X}_3\\ F_2={\mathrm{\β}}_1{X}_1+{\mathrm{\β}}_2{X}_2+{\mathrm{\β}}_3{X}_3\end{array}\right.;$

Wherein, F ₁, F ₂represent two major components; X ₁, X ₂, X ₃what represent is different indexs, is speed, flow, occupation rate herein; α ₁~ α ₃, β ₁~ β ₃the coefficient corresponding to each index represented.

Use first principal component F ₁in factor alpha corresponding to each index ₁~ α ₃, be multiplied by first principal component F ₂corresponding contribution rate b ₁, then divided by extract two contribution rate sum c of two major components ₂, then add Second principal component, F ₂in factor beta corresponding to each index ₁~ β ₃, be multiplied by Second principal component, F ₂corresponding contribution rate b ₂, then divided by extract two contribution rate sum c of two major components ₂, integrate score model can be obtained:

Y＝ω ₁X ₁+ω ₂X ₂+ω ₃X ₃；

Wherein, index X ₁~ X ₃corresponding coefficient ω ₁~ ω ₃be expressed as the weight of each index.

Coefficient of colligation is normalized, namely obtains the weight coefficient ω of these three kinds of characteristic parameters of vehicle flowrate, average speed and average occupancy _o={ ω _o1, ω _o2, ω _o3.

4) carry out cluster analysis according to the characteristic parameter weighting GEFCM algorithm improved to historical traffic data to historical data, determine the cluster centre representing different traffic class, concrete cluster analysis step is as follows:

41) the clock rate number c=3 of initialization model, Fuzzy Exponential selects m=2, threshold epsilon=1e-6, maximum iteration time L _max=200, under the prerequisite meeting following formula, adopt the random number initialization subordinated-degree matrix U between [0,1]:

$\underset{i=1}{\overset{c}{\mathrm{\Σ}}}{u}_{\mathrm{ij}}=1,j=\mathrm{1,2},...,n;$

$\underset{i=1}{\overset{c}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{u}_{\mathrm{ij}}=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}\underset{i=1}{\overset{c}{\mathrm{\Σ}}}{u}_{\mathrm{ij}}=n;$

42) the cluster centre c of c=3 class is calculated according to following formula _i, i=1,2,3;

$c_i=\frac{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{u}_{\mathrm{ij}}^m{x}_j}{\underset{s=1}{\overset{n}{\mathrm{\Σ}}}{u}_{\mathrm{is}}^m};$

43) according to the value of following formulae discovery objective function:

$J(U,c_1,...,c_c,\mathrm{\ω})=\underset{i=1}{\overset{c}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}\frac{u_{\mathrm{ij}}^m}{\underset{s=1}{\overset{n}{\mathrm{\Σ}}}{u}_{\mathrm{is}}^m}\underset{k=1}{\overset{t}{\mathrm{\Σ}}}{\mathrm{\ω}}_k^2{(x_{\mathrm{jk}}-c_{\mathrm{ik}})}^2;$

If met:

|| J ^(b+1)-J ^(b)|| < ε or l>=L _max;

Then algorithm stops, and obtains cluster centre c _i, otherwise turn 44);

44) the subordinated-degree matrix U after upgrading is calculated according to following formula:

$u_{\mathrm{ij}}=\frac{{\left[\frac{{\left(\underset{s=1}{\overset{n}{\mathrm{\&Sigma;}}}{u}_{\mathrm{is}}^m\right)}^2}{(\underset{s=1}{\overset{n}{\mathrm{\&Sigma;}}}{u}_{\mathrm{is}}^m-u_{\mathrm{ij}}^m)\underset{k=1}{\overset{t}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_k^2{(x_{\mathrm{jk}}-c_{\mathrm{ik}})}^2}\right]}^{\frac{1}{m-1}}}{\underset{\mathrm{\&gamma;}=1}{\overset{c}{\mathrm{\&Sigma;}}}{\left[\frac{{\left(\underset{s=1}{\overset{n}{\mathrm{\&Sigma;}}}{u}_{\mathrm{\&gamma;s}}^m\right)}^2}{(\underset{s=1}{\overset{n}{\mathrm{\&Sigma;}}}{u}_{\mathrm{\&gamma;s}}^m-u_{\mathrm{\&gamma;j}}^m)\underset{k=1}{\overset{t}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_k^2{(x_{\mathrm{jk}}-c_{\mathrm{\&gamma;k}})}^2}\right]}^{\frac{1}{m-1}}};$

Then step 42 is returned);

Step 41-44) in, x _jfor a jth sample point, x _jkfor a kth characteristic parameter of a jth sample point; U=(u _ik) _{c × n}for subordinated-degree matrix, use u _ijrepresent that a jth sample belongs to the degree of membership of the i-th class, 0≤u _ij≤ 1, m ∈ [1 ,+∞) represent FUZZY WEIGHTED index; c _i=(c _i1, c _i2... c _ip) (i=1,2 ..., c) represent different classes of cluster centre; item has showed the capacity attribute of the i-th class; ω={ ω ₁, ω ₂..., ω _t, ω _k∈ [0,1] is the characteristic parameter weight corresponding with the argument sequence of input; T is the number of characteristic parameter; N is sample point quantity.

5) when getting the traffic flow parameter of current section, according to formula below, the distance belonging to different cluster centre is calculated, according to shortest distance principle, the traffic behavior classification described in judgement.

${\left(d_{\mathrm{ij}}^{\mathrm{\&omega;}}\right)}^2={\left|\right|\mathrm{\&omega;}(x_j-c_i)\left|\right|}^2=\underset{k=1}{\overset{t}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_k^2{(x_{\mathrm{jk}}-c_{\mathrm{ik}})}^2;$

Wherein, represent i-th cluster centre c _iwith a jth sample point x _jbetween distance; ω={ ω ₁, ω ₂..., ω _t, ω _k∈ [0,1], is a characteristic parameter weight corresponding with the argument sequence of input, represents the importance of a kth characteristic parameter in cluster; T is the number of characteristic parameter, is 3 for this paper traffic behavior is estimated.

The traffic behavior corresponding based on the principle determination current flows parameter of bee-line is: for sample point x _j, calculate the distance with three class traffic behavior centers respectively according to above formula, obtain with unimpeded, walk or drive slowly, the distance of the three class status center that block up is respectively the relatively size of three's distance, status categories corresponding to minimum distance is then current traffic behavior estimated value.

6) during the traffic flow parameter Data Update of current section next time, step 5 is performed).

What finally illustrate is, above embodiment is only in order to illustrate technical scheme of the present invention and unrestricted, although with reference to preferred embodiment to invention has been detailed description, those of ordinary skill in the art is to be understood that, can modify to technical scheme of the present invention or equivalent replacement, and not departing from aim and the scope of technical solution of the present invention, it all should be encompassed in the middle of right of the present invention.

Claims (7)

1. the highway place traffic state estimation method of feature based parameter weighting GEFCM algorithm, is characterized in that: comprise the following steps:

1) obtain the historical data of these three kinds of characteristic parameters of vehicle flowrate, average speed and average occupancy that highway microwave vehicle checker collects, form sample matrix;

2) to step 1) data that obtain carry out pre-service, and described pre-service comprises the identification of misdata and rejecting, the reparation of data, data filtering process;

3) weight of three kinds of characteristic parameters when cluster analysis is determined;

4) cluster analysis is carried out to historical data;

5) when getting the traffic flow parameter of current section, traffic behavior is estimated in real time.

2. the highway place traffic state estimation method of feature based parameter weighting GEFCM algorithm as claimed in claim 1, is characterized in that: described step 2) in, specifically adopt the identification and rejecting of carrying out misdata with the following method:

Within a data update cycle, set the threshold range of total vehicle flowrate data as [0, Q _max], the threshold range of average speed is [0, V _max]; If when the data of the total vehicle flowrate data collected or average speed are not in the threshold range of correspondence, then show that these group data are unreliable, and rejected; If when the data of the total vehicle flowrate data collected and average number of vehicles all drop in corresponding threshold range, then show that these group data are reliable, retain this group data; Wherein, Q _max, V _maxbe illustrated respectively in the flow maximum in the data update cycle and speed maximal value;

Set up misdata judgment rule according to traffic flow theory, namely reject rule; Then, judge whether the data sequence gathered meets and reject rule; When satisfied rejecting rule, the data of correspondence are needed reject; When not meeting rejecting rule, retain corresponding data.

3. the highway place traffic state estimation method of feature based parameter weighting GEFCM algorithm as claimed in claim 2, is characterized in that: described step 2) in, by following formula, data are repaired:

Wherein, for the data restore value of t period; X (t-1) is the actual detected value of (t-1) period; X ' (t) is the history average of the image data of n before synchronization days; α is forgetting factor, α ∈ [0,1].

4. the highway place traffic state estimation method of feature based parameter weighting GEFCM algorithm as claimed in claim 3, is characterized in that: carry out filtering process by following formula to data:

S _t＝αX _t+(1-α)S _t-1

In formula, S _tfor the single exponential smoothing value that the t period obtains; S _t-1for the single exponential smoothing value that the t-1 period obtains; X _tfor the observed reading that the t period obtains; α ∈ [0,1] is smoothing factor.

5. the highway place traffic state estimation method of the feature based parameter weighting GEFCM algorithm according to any one of Claims 1-4, is characterized in that: described step 3) specifically comprise the steps:

31) by by step 2) the sample matrix X that forms of these three kinds of characteristic parameters of pretreated vehicle flowrate, average speed and average occupancy carries out Z-score standardization, the matrix Z after obtaining standardization;

32) Calculation of correlation factor is carried out to the matrix Z after Z-score standardization, obtain correlation matrix R;

33) its secular equation being constructed to correlation matrix R | R-λ I|=0, obtains p characteristic root, b _t, c _tt=1,2 ..., p;

Wherein, b _tbe the variance contribution ratio of t major component, c _tfor the accumulative variance contribution ratio of a front t major component;

34) the weight coefficient ω of these three kinds of characteristic parameters of vehicle flowrate, average speed and average occupancy is calculated _o={ ω _o1, ω _o2..., ω _ot.

6. the highway place traffic state estimation method of feature based parameter weighting GEFCM algorithm as claimed in claim 5, is characterized in that: described step 4) in, by the following method cluster analysis is carried out to historical data:

By step 3) principal component analysis (PCA) obtains the weights omega of different characteristic parameter _o={ ω _o1, ω _o2..., ω _ot, bring formula below into, and adopt the traffic flow parameter of three formulas to history below to iterate calculating, until meet algorithm stop condition, finally obtain the cluster centre c representing different traffic class _i;

Wherein, x _jfor a jth sample point, x _jkfor a kth characteristic parameter of a jth sample point; U=(u _ik) _{c × n}for subordinated-degree matrix, use u _ijrepresent that a jth sample belongs to the degree of membership of the i-th class, 0≤u _ij≤ 1, m ∈ [1 ,+∞) represent FUZZY WEIGHTED index; c _i=(c _i1, c _i2... c _ip) (i=1,2 ..., c) represent different classes of cluster centre; item has showed the capacity attribute of the i-th class; ω={ ω ₁, ω ₂..., ω _t, ω _k∈ [0,1] is the characteristic parameter weight corresponding with the argument sequence of input; T is the number of characteristic parameter; N is sample point quantity.

7. the highway place traffic state estimation method of feature based parameter weighting GEFCM algorithm as claimed in claim 6, is characterized in that: described step 5) in, by the calculating of following formula, obtain the distance at real time data sample and each state class center;

Wherein, represent i-th cluster centre c _iwith a jth sample point x _jbetween distance; ω={ ω ₁, ω ₂..., ω _t, ω _k∈ [0,1] is the characteristic parameter weight corresponding with the argument sequence of input; T is the number of characteristic parameter;

For sample point x _j, calculate the distance with three class traffic behavior centers respectively according to above formula, obtain with unimpeded, walk or drive slowly, the distance of the three class status center that block up is respectively the relatively size of three's distance, status categories corresponding to minimum distance is then current traffic behavior estimated value.