CN101739823A - Road-section average travel time measuring method suitable for low-frequency sampling - Google Patents

Abstract

The invention relates to a road-section average travel time measuring method suitable for low-frequency sampling, which comprises the following steps: (1) sampling taxi GPS data at low frequency to obtain the estimated time of each vehicle at the crossroads; (2) according to the geographical position information at each crossroad and the estimated time of each vehicle at the crossroads, obtaining the vehicle travel time of a certain road section as a sample *i; and (3) calculating the average travel time according to the following reference formula (8), wherein in the clustering center, tf stands for quick type, tm stands for medium type, and tl stands for slow type. The road-section average travel time measuring method suitable for low-frequency sampling can effectively measure the average travel time of the road section and meet the requirements for traffic real-time performance and accuracy.

Description

A kind of road-section average travel time measuring method of suitable low-frequency sampling

Technical field

The present invention relates to a kind of road-section average travel time measuring method.

Background technology

Road section traffic volume parameter estimation model: multivariate regression model, neural network, Fuzzy Inference Model, rate integrating model, end points time difference method etc.The discrete instantaneous velocity time series that the rate integrating model has utilized the GPS probe vehicles to detect is used the distance of travelling that numerical integration method calculates probe vehicles.The end points time difference method need be determined the moment of vehicle through two end points in investigation highway section, and difference constantly is the journey time in highway section.Relatively short (under the situation of t＜10s), rate integrating model and end points time difference model result are more approaching, and to compare error ratio less with actual result in sampling interval.Along with the increase in sampling period, and because of urban road situation complexity, and the influence of intersection signal lamp, the contact between each probe vehicles instantaneous velocity will be interrupted, and has formed discrete relatively data collection point, and the rate integrating model is with this situation of incompatibility.Though end points time difference model not with speed as parameter, can when working as sampling interval and becoming big, vehicle is at last sampled point instantaneous velocity and the condition that no longer satisfies traditional end points time difference model of rolling a nearest sampled point in this highway section away from certain highway section.

Summary of the invention

For the not energy measurement road-section average journey time that overcomes existing road section traffic volume parameter estimation model, the deficiency of real-time difference, the invention provides and a kind ofly can effectively measure the road-section average journey time, satisfy the road-section average travel time measuring method of the suitable low-frequency sampling of traffic real-time and accuracy requirement.

The technical solution adopted for the present invention to solve the technical problems is:

A kind of road-section average travel time measuring method of suitable low-frequency sampling, described road-section average travel time measuring method comprises:

1), with low-frequency sampling taxi gps data, the estimation that obtains each vehicle way crossing is constantly;

2), according to the geographical location information of each crossing, and the estimation of each vehicle way crossing of obtaining of step 1) is constantly, the journey time that obtains the vehicle in certain highway section is sample x _i,

If U=is (u _Ik) _N*cBe the fuzzy classification matrix, wherein, n represents number of samples, c presentation class number, and this locates c=3, promptly is divided into quick, medium, three classes of going slowly, u _IkRepresent that i sample belongs to the degree of membership of k classification, establishes X={x ₁, x ₂... x _nFor being classified sample set, define an objective function J _FCM(U T) is:

$J_{\mathrm{FCM}}(U,T)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{k=1}{\overset{c}{\mathrm{\Σ}}}{u_{\mathrm{ik}}}^m{\left|\right|x_i-t_k\left|\right|}^2---\left(4\right)$

Wherein, v _kBe cluster centre, m ∈ [0 ,+∞) be a FUZZY WEIGHTED index, m be for change flexibly to x _iDegree of membership, objective function represent all kinds of in unique points to the distances of clustering centers quadratic sum, clustering problem promptly requires to satisfy the U and the T of formula (4), thereby makes objective function reach minimum value:

The fuzzy C mean algorithm is by the objective function J to formula (4) _FCM(U, iteration optimization T) is obtained the fuzzy classification to data set, i.e. iteration:

$u_{\mathrm{ik}}={\left[\underset{j=1}{\overset{c}{\mathrm{\Σ}}}{\left(\frac{d_{\mathrm{ik}}}{d_{\mathrm{jk}}}\right)}^{\frac{2}{m-1}}\right]}^{-1},$ d _ik＝0，1≤i≤n

(5)

u _ik＝1，d _ik＝0，k＝1

u _ik＝0，d _ik＝0，k≠i

$t_k=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{u_{\mathrm{ik}}}^m{x}_i/\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{u_{\mathrm{ik}}}^m,1\≤k\≤c---\left(6\right)$

Make objective function J _FCM(U T) converges to a local minimum point or saddle point, and an optimum fuzzy c that obtains X is divided U=[u _Ik ^*];

3), according to the t of cluster centre _f, t _m, t _l, t _fRepresent quick type, t _mRepresent medium type, represent to go slowly type, calculate the average stroke time according to following formula:

$\stackrel{\‾}{t}=\frac{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{T}_f+\underset{j=1}{\overset{m}{\mathrm{\Σ}}}(\frac{m}{m+n}{T}_{m+l}+\frac{n}{m+n}{t}_f)}{N}---\left(8\right)$

T wherein _fExpression is with t _fBe the fast travel time set of cluster centre, T _M+lExpression T _mWith T _lTwo union of sets collection, t _iExpression T _fMaximal value in the set, n represents T _fThe number of element in the class, m represents remaining element number.

Further, in described step 1), the sampled data of crossing vehicle mainly is divided into three kinds of situations:

＜1〉intersection entrance and outlet scope all have the bicycle data

＜2〉only the bicycle sampled data is arranged in the intersection exit scope

＜3〉only there is bicycle to adopt data in the intersection entrance scope

Adopt the end points time difference method to adapt to low-frequency sampling, specifically have:

For the＜1〉the kind situation, if v _pNon-vanishing, then have

t _a＝t _p-(L ₂/v _p+ε) (1)

T wherein _aVehicle arrives the estimation moment of crossing, t in the presentation graphs 1 _pVehicle is at the moment that P is ordered, v in the presentation graphs 1 _pThe expression vehicle is at the instantaneous velocity of P, and ε represents that the crossing is to delay that vehicle caused;

If v _pBe zero, v _P-1Non-vanishing, then working as the p point is the empty wagons data, is crossing passenger loading or following car data; Then have

t _a＝t _p-1+L ₁/v _p-1+ε (2)

If v _P-1Be zero, then vehicle is in the crossing wait situation, then has

t _a＝t _p-1+L/TL ₁ (3)

Wherein L is average queue length, and T is this phase average red time;

For the＜2〉the kind situation, if v _pNon-vanishing, then adopt formula (1) to calculate; If v _pBe zero, looking this sampled point is the empty wagons data, with its rejecting;

For the＜3〉the kind situation, if v _P-1Non-vanishing, then adopt formula (2) to calculate, if v _P-1Be zero, then adopt formula (3) to calculate.

Technical conceive of the present invention is: the FCD original data processing: (1) improved map-matching algorithm

In traditional map-matching algorithm, introduce the travel direction in the vehicle location data, and the lane width of road finds preliminary match point as constraint condition, then introduces the continuity of vehicle ', utilize historical data that the initial matching point is revised, obtain final match point.

(2) vehicle route is followed the tracks of: by improving final match point and the normal driving behavior that map-matching algorithm obtains, determine that by the space search algorithm vehicle may be through the crossing, and with these crossings as the zonule road network, by the dijkstra's algorithm shortest route searched as this vehicle driving trace.

Beneficial effect of the present invention mainly shows: 1, can effectively measure the road-section average journey time, satisfy traffic real-time and accuracy requirement; 2, by improving the end points time difference method, overcome the only defective of the situation of suitable high frequency sampling of classic method, under the lower situation of sample frequency, also can guarantee certain precision, thereby well obtain bicycle road-section average journey time; The section bicycle is classified and the element of sorted cluster centre and each class is weighted smoothly when 3, utilizing FCM to the highway section, can get rid of abnormal data preferably, also can better remedy shortcoming, make the road-section average journey time of calculating approaching with actual travel time as far as possible as Floating Car general data source (GPS taxi) itself.

Description of drawings

Fig. 1 is the synoptic diagram of end points time difference method.

Embodiment

Below in conjunction with accompanying drawing the present invention is further described.

With reference to Fig. 1, a kind of road-section average travel time measuring method of suitable low-frequency sampling, described road-section average travel time measuring method comprises:

1), with low-frequency sampling taxi gps data, the estimation that obtains each vehicle way crossing is constantly;

2), according to the geographical location information of each crossing, and the estimation of each vehicle way crossing of obtaining of step 1) is constantly, the journey time that obtains the vehicle in certain highway section is sample x _i,

If U=is (u _Ik) _N*cBe the fuzzy classification matrix, wherein, n represents number of samples, c presentation class number, and this locates c=3, promptly is divided into quick, medium, three classes of going slowly, u _IkRepresent that i sample belongs to the degree of membership of k classification, establishes X={x ₁, x ₂... x _nFor being classified sample set, define an objective function J _FCM(U T) is:

$J_{\mathrm{FCM}}(U,T)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{k=1}{\overset{c}{\mathrm{\Σ}}}{u_{\mathrm{ik}}}^m{\left|\right|x_i-t_k\left|\right|}^2---\left(4\right)$

Wherein, v _kBe cluster centre, m ∈ [0 ,+∞) be a FUZZY WEIGHTED index, m be for change flexibly to x _iDegree of membership, objective function represent all kinds of in unique points to the distances of clustering centers quadratic sum, clustering problem promptly requires to satisfy the U and the T of formula (4), thereby makes objective function reach minimum value:

The fuzzy C mean algorithm is by the objective function J to formula (4) _FCM(U, iteration optimization T) is obtained the fuzzy classification to data set, i.e. iteration:

$u_{\mathrm{ik}}={\left[\underset{j=1}{\overset{c}{\mathrm{\Σ}}}{\left(\frac{d_{\mathrm{ik}}}{d_{\mathrm{jk}}}\right)}^{\frac{2}{m-1}}\right]}^{-1}$ ，d _ik＝0，1≤i≤n

(5)

u _ik＝1，d _ik＝0，k＝1

u _ik＝0，d _ik＝0，k≠i

$t_k=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{u_{\mathrm{ik}}}^m{x}_i/\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{u_{\mathrm{ik}}}^m,1\≤k\≤c---\left(6\right)$

Make objective function J _FCM(U T) converges to a local minimum point or saddle point, and an optimum fuzzy c that obtains X is divided U=[u _Ik ^*];

3), according to the t of cluster centre _f, t _m, t _l, t _fRepresent quick type, t _mRepresent medium type, represent to go slowly type, calculate the average stroke time according to following formula:

$\stackrel{\‾}{t}=\frac{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{T}_f+\underset{j=1}{\overset{m}{\mathrm{\Σ}}}(\frac{m}{m+n}{T}_{m+l}+\frac{n}{m+n}{t}_f)}{N}---\left(8\right)$

T wherein _fExpression is with t _fBe the fast travel time set of cluster centre, T _M+lExpression T _mWith T _lTwo union of sets collection, t _iExpression T _fMaximal value in the set, n represents T _fThe number of element in the class, m represents remaining element number.

In the present embodiment, at first carry out the road-section average journey time and estimate:

(1) improves the end points time difference method

The sampled data of crossing vehicle mainly is divided into three kinds of situations:

＜1〉intersection entrance and outlet scope all have the bicycle data

＜2〉only the bicycle sampled data is arranged in the intersection exit scope

＜3〉only there is bicycle to adopt data in the intersection entrance scope

These three kinds of situations are considered as three kinds of calculative strategies, and the information of input sample point selects own required strategy to calculate the moment of vehicle through this crossing, makes the end points time difference method adapt to low-frequency sampling.

＜1〉for first kind of situation, if v _pNon-vanishing, then have

t _a＝t _p-(L ₂/v _p+ε) (1)

T wherein _aVehicle arrives the estimation moment of crossing, t in the presentation graphs 1 _pVehicle is at the moment that P is ordered, v in the presentation graphs 1 _pThe expression vehicle is at the instantaneous velocity of P, and ε represents that the crossing is to delay that vehicle caused (down with).

If v _pBe zero, v _P-1Non-vanishing, then working as the p point is the empty wagons data, is crossing passenger loading or following car data.Then have

t _a＝t _p-1+L ₁/v _p-1+ε (2)

If v _P-1Be zero, then vehicle is in the crossing wait situation, then has

t _a＝t _p-1+L/TL ₁ (3)

Wherein L is average queue length, and T is this phase average red time.

＜2〉for second kind of situation, if v _pNon-vanishing, then adopt formula 1 to calculate.If v _pBe zero, looking this sampled point is the empty wagons data, with its rejecting.

＜3〉for the third situation, if v _P-1Non-vanishing, then adopt formula 2 to calculate, if v _P-1Be zero, then adopt formula 3 to calculate.

(2) FCM road-section average journey time is estimated

The FCM sorting technique thinks that each sample standard deviation in the classification samples set belongs to a certain class with different degrees of membership. therefore a certain class is just thought a fuzzy subset on the sufficient sample set, so the pairing classification matrix of each such classification results is exactly a fuzzy classification matrix.

If U=is (u _Ik) _N*c(wherein, n represents number of samples, c presentation class number, u for the fuzzy classification matrix _IkRepresent that i sample belongs to the degree of membership of k classification), establish X={x ₁, x ₂... x _nFor being classified sample set, each sample x wherein _iM characteristic index, i.e. xi={x are all arranged _I1, x _I2... x _Im.Sample set X is divided into the c class, and (2＜c＜n), establish c cluster centre vector for T in order to obtain optimum fuzzy classification, has defined an objective function J _FCM(U T) is

$J_{\mathrm{FCM}}(U,T)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{k=1}{\overset{c}{\mathrm{\Σ}}}{u_{\mathrm{ik}}}^m{\left|\right|x_i-t_k\left|\right|}^2---\left(4\right)$

Wherein, v _kBe cluster centre, m ∈ [0 ,+∞) be a FUZZY WEIGHTED index, m be for change flexibly to x _iDegree of membership. objective function represent all kinds of in unique points to the distances of clustering centers quadratic sum, clustering problem promptly requires to satisfy the U and the T of formula (4), thereby makes objective function reach minimum value.

The fuzzy C mean algorithm is by the objective function J to formula (4) _FCM(U, iteration optimization T) is obtained the fuzzy classification to data set, i.e. iteration

$u_{\mathrm{ik}}={\left[\underset{j=1}{\overset{c}{\mathrm{\Σ}}}{\left(\frac{d_{\mathrm{ik}}}{d_{\mathrm{jk}}}\right)}^{\frac{2}{m-1}}\right]}^{-1}$ ，d _ik＝0，1≤i≤n

(5)

u _ik＝1，d _ik＝0，k＝1

u _ik＝0，d _ik＝0，k≠i

$t_k=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{u_{\mathrm{ik}}}^m{x}_i/\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{u_{\mathrm{ik}}}^m,1\≤k\≤c---\left(6\right)$

Make objective function J _FCM(U T) converges to a local minimum point or saddle point, and an optimum fuzzy c that obtains X is divided U=[u _Ik ^*].

(2) the road-section average running time calculates

Utilize FCM, according to the t of cluster centre _f, t _m, t _l, data are divided into three classes, be respectively quick, medium, go slowly.Central value by three classes is calculated the road-section average running time:

$\stackrel{\‾}{t}=\underset{k=1}{\overset{c}{\mathrm{\Σ}}}{t}_k/c---\left(7\right)$

Formula (7) account form is partial to intermediate value, has ignored the frequent characteristic of stopping of taxi data, does not conform to the actual traffic situation, therefore needs the new account form of definition, so computing formula below introducing:

$\stackrel{\‾}{t}=\frac{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{T}_f+\underset{j=1}{\overset{m}{\mathrm{\Σ}}}(\frac{m}{m+n}{T}_{m+l}+\frac{n}{m+n}{t}_f)}{N}---\left(8\right)$

T wherein _fExpression is with t _fBe the fast travel time set of cluster centre, T _M+lExpression T _mWith T _lTwo union of sets collection, t _iExpression T _fMaximal value in the set, n represents the number of element in the Tf class, m represents remaining element number.

The floating car data in many cities derives from the GPS taxi, and the raw data great majority are partial to the data of going slowly, and therefore by formula (8), to the data processing of going slowly, the data of going slowly is raised speed by certain weight, makes result of calculation more accurate.

Example: selecting and purchasing is got north orientation south, highway section, west part of the ring road, Hangzhou to as experiment highway section and carry out traffic study, this highway section total length is about 430m, base closed, the crossing is the pedestrian's street crossing crossing that signal controlling upstream and downstream crossing is signal controlling up and down.Select such highway section to analyze, avoid vehicle to lose on the one hand, can obtain more complete data, upstream and downstream crossing in highway section all has signal controlling on the other hand, and is similar with most highway section, has dissemination.Select hire a car gps data 13:30～14:30 period (flat peak period) and 16:30～17:30 period (peak period on and off duty) as experimental data.The computing method that adopt us to study are calculated, and concrete outcome sees Table 1,2, and table 1 is flat peak period road-section average journey time; Table 2 is road-section average journey time peak period:

Period is estimated the accurate average stroke time (s) of average stroke time (s)

13:30-13:35 37.45 48.3

13:35-13:40 51.69 49.6

13:40-13:45 50.3 53.2

13:45-14:50 53.94 4903

13:50-13:55 47.91 45.7

13:55-14:00 42.16 52.3

14:00-14:05 43.73 46.3

14:05-14:10 47.32 50.1

14:10-14:15 51.91 48.2

14:15-14:20 44.13 49.5

14:20-14:25 50.48 52.3

14:25-14:30 55 57.5

Table 1

Period is estimated the accurate average stroke time (s) of average stroke time (s)

17:00-17:05 55.04 58.2

17:05-17:10 57.43 59.3

17:10-17:15 57.45 62.2

17:15-17:20 64.44 78.9

17:20-17:25 87.18 89.8

17:25-17:30 92.51 100.7

17:30-17:35 84.29 96.2

17:35-17:40 80.74 106.8

17:40-17:45 128.68 115.6

17:45-17:50 123.32 124.5

17:50-17:55 1860.2 118.5

17:55-17:60 112.16 107.2

Table 2

Table 1,2 result show that the method estimated value on the whole of utilizing this paper to propose is more approaching with actual value.Simultaneously, find that also the gap on indivedual time points is still very big, main cause has two kinds, source of error is in when asking the bicycle Link Travel Time on the one hand, vehicle condition is not listed in three kinds of situations of aforementioned analysis, very few relevant with sample size in addition, even to weights optimization, the result is deviation to some extent also.See that on the whole estimated value and exact value are more approaching.

Claims (2)

1. the road-section average travel time measuring method of a suitable low-frequency sampling, it is characterized in that: described road-section average travel time measuring method comprises:

1), with low-frequency sampling taxi gps data, the estimation that obtains each vehicle way crossing is constantly;

2), according to the geographical location information of each crossing, and the estimation of each vehicle way crossing of obtaining of step 1) is constantly, the journey time that obtains the vehicle in certain highway section is sample xi,

If U=is (u _Ik) _N*cBe the fuzzy classification matrix, wherein, n represents number of samples, c presentation class number, and this locates c=3, promptly is divided into quick, medium, three classes of going slowly, u _IkRepresent that i sample belongs to the degree of membership of k classification, establishes X={x ₁, x ₂... x _nFor being classified sample set, define an objective function J _FCM(U T) is:

$J_{\mathrm{FCM}}(U,T)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{k=1}{\overset{c}{\mathrm{\Σ}}}{u_{\mathrm{ik}}}^m{\left|\right|x_i-t_k\left|\right|}^2---\left(4\right)$

Wherein, v _kBe cluster centre, m ∈ [0 ,+∞) be a FUZZY WEIGHTED index, m be for change flexibly to x _iDegree of membership, objective function represent all kinds of in unique points to the distances of clustering centers quadratic sum, clustering problem promptly requires to satisfy the U and the T of formula (4), thereby makes objective function reach minimum value:

The fuzzy C mean algorithm is by the objective function J to formula (4) _FCM(U, iteration optimization T) is obtained the fuzzy classification to data set, i.e. iteration:

$u_{\mathrm{ik}}={\left[\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\left(\frac{d_{\mathrm{ik}}}{d_{\mathrm{ik}}}\right)}^{\frac{2}{m-1}}\right]}^{-1},d_{\mathrm{ik}}=\mathrm{0,1}\≤i\≤n$ (5)

u _ik＝1，d _ik＝0，k＝1

u _ik＝0，d _ik＝0，k≠i

$t_k=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{u_{\mathrm{ik}}}^m{x}_i/\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{u_{\mathrm{ik}}}^m,1\≤k\≤c---\left(6\right)$

Make objective function J _FCM(U T) converges to a local minimum point or saddle point, and an optimum fuzzy c that obtains X is divided U=[u _Ik ^*];

3), according to the t of cluster centre _f, t _m, t _l, t _fRepresent quick type, t _mRepresent medium type, represent to go slowly type, calculate the average stroke time according to following formula:

$\stackrel{-}{t}=\frac{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{T}_f+\underset{j=1}{\overset{m}{\mathrm{\Σ}}}(\frac{m}{m+n}{T}_{m+l}+\frac{n}{m+n}{t}_f)}{N}---\left(8\right)$

T wherein _fExpression is with t _fBe the fast travel time set of cluster centre, T _M+lExpression T _mWith T _lTwo union of sets collection, t _iExpression T _fMaximal value in the set, n represents T _fThe number of element in the class, m represents remaining element number.

2. the road-section average travel time measuring method of a kind of suitable low-frequency sampling as claimed in claim 1, it is characterized in that: in described step 1), the sampled data of crossing vehicle mainly is divided into three kinds of situations:

＜1〉intersection entrance and outlet scope all have the bicycle data

＜2〉only the bicycle sampled data is arranged in the intersection exit scope

＜3〉only there is bicycle to adopt data in the intersection entrance scope

Adopt the end points time difference method to adapt to low-frequency sampling, specifically have:

For the＜1〉the kind situation, if v _pNon-vanishing, then have

t _a＝t _p-(L ₂/v _p+ε) (1)

T wherein _aVehicle arrives the estimation moment of crossing, t in the presentation graphs 1 _pVehicle is at the moment that P is ordered, v in the presentation graphs 1 _pThe expression vehicle is at the instantaneous velocity of P, and ε represents that the crossing is to delay that vehicle caused;

If v _pBe zero, v _P-1Non-vanishing, then working as the p point is the empty wagons data, is crossing passenger loading or following car data; Then have

t _a＝t _p-1+L ₁/v _p-1+ε (2)

If v _P-1Be zero, then vehicle is in the crossing wait situation, then has

t _a＝t _p-1+L/TL ₁ (3)

Wherein L is average queue length, and T is this phase average red time;

For the＜2〉the kind situation, if v _pNon-vanishing, then adopt formula (1) to calculate; If v _pBe zero, looking this sampled point is the empty wagons data, with its rejecting;

For the＜3〉the kind situation, if v _P-1Non-vanishing, then adopt formula (2) to calculate, if v _P-1Be zero, then adopt formula (3) to calculate.

Images