CN103500362B - A kind of urban road speed predicting method based on analysis of spectrum - Google Patents

Abstract

The present invention relates to a kind of urban road speed predicting method based on analysis of spectrum, including: a. collects the vehicle speed data that GPS collects；B. given data is utilized to select road speeds model parameter based on analysis of spectrum；C. according to the unknown road speeds of model formation prediction.Needed for the present invention has training pattern, historical data is relatively fewer, operation relative ease, parameter in model quickly can dynamically adjust with the change in section, period, can be suitably used for the section that traffic flow fluctuation is bigger, and the advantages such as preferable are mated with real road speed condition, improve accuracy and the reliability of prediction.

Description

A kind of urban road speed predicting method based on analysis of spectrum

Technical field

The present invention relates to a kind of urban road speed predicting method based on analysis of spectrum, belong to intelligent transportation system and grind Study carefully category.

Background technology

Urban highway traffic Forecasting Methodology is by the road speeds of respective stretch in prediction short time range, it is thus achieved that phase Answering road traffic state, and utilize multiple channel issue to predict the outcome, induction driver and crew select rationally to go out walking along the street Line, can play the effect alleviating traffic congestion, the most increasingly receive publicity.

Analysis of spectrum is a kind of mathematical method that spatial data carries out general layout, dimensional analysis, it is possible to present powerful Expressive ability, even a single point on road, it is also possible to strong earth's surface its real traffic existing, thus Reduce the impact that fluctuation brings, it is adaptable to analyze the time series that fluctuation is bigger.But directly utilize analysis of spectrum skill The method of art predicted city road speeds have not been reported.

At present, the main method of road speeds prediction includes ARIMA(difference ARMA model), Kalman filter model etc..ARIMA is a kind of Time Series Forecasting Methods being based entirely on empirical statistics, its The certain mathematical model of random sequence elapsed in time by prediction object and formed carrys out approximate description, and utilizes Past value and the present value of this sequence predict future value.Kalman filter model uses least mean-square error conduct The optimum criterion of prediction, utilizes the predictive value of previous moment and the observation of current time to update state variable Prediction, obtain the predictive value of current time.Although the data of the unknown can comparatively fast be predicted by both models, but It is not to be suitable for the bigger time series that fluctuates.In city road network, road traffic state existence change is fast, ripple The features such as dynamic scope is big, said method cannot reflect that these change in time, causes forecast error bigger.

Summary of the invention

Present invention the deficiencies in the prior art to be overcome, propose a kind of urban road prediction of speed side based on analysis of spectrum Method, alleviates the impact that road speeds is predicted by urban road traffic state rapid fluctuations, improves forecasting accuracy.

For achieving the above object, first the present invention collects road history data and prediction given data on the same day, then profit Selecting prediction model parameters by both corresponding interval relations, last bind profile is analyzed model prediction and is obtained the unknown Data.

The method of the present invention is realized by following steps:

Step 1. collects the vehicle speed data that GPS collects

GPS collects data such as comprising acquisition time, car number, car speed, wherein chooses car speed Data are as source data；

Step 2. preference pattern parameter

Represent that data to be predicted are right with it by the relation predicting the corresponding historical data of given data on the same day Answer the relation of historical data, i.e. refer to the parameter matrix C in forecast model；

Wherein road speeds forecast model formula based on analysis of spectrum is as follows:

$\hat{X}=C\·{\mathrm{\φ}}^T---\left(1\right)$

In formula:Represent certain continuum road speeds matrix that prediction obtains；C representation parameter matrix；φ is The eigenvectors matrix of respective bins road speeds variation tendency is described；

Parameter matrix C is asked to specifically comprise the following steps that

First GPS car speed is converted into road speeds by map-matching method by 2.1；

Then 2.2 will be divided equally into N number of interval every day, it is known that N number of interval, history continuous N sky and J interval road speeds data before M+1 days；

The matrix that data are M*N rank in 2.3 definition N number of intervals, history continuous N sky, is denoted as X.According to R=X^TX tries to achieve the covariance matrix R (N*N) of matrix X；

2.4 use QR decomposition method (matrix decomposition becomes an orthonomal matrix Q and upper triangular matrix R) to ask Obtain the eigenvectors matrix of covariance matrix R, be designated as φ (N*N).N*K is chosen in eigenvectors matrix φ Part is designated as φ ', and wherein, K (1≤K < N) represents exponent number, the quantity of the characteristic vector i.e. used.Choose square In battle array φ ', the 1st walks to jth row part, is designated as φ₁(j*K)；In like manner, the middle jth of φ '+1 walks to Nth row part, It is designated as φ₂((N-j)*K)；

The matrix of 2.5 the M+1 days known interval censored datas compositions is denoted asThe φ tried to achieve with previous step₁Together substitute into Model formation (1) converts the formula obtainedIn, the M+1 days known districts of expression can be tried to achieve Between parameter matrix between (interval 1 to interval j) data and corresponding M days history interval censored datas, be designated as C₁(1*K), it is possible to (, do not include interval j) data and go through after interval j as the unknown in M+1 days is interval The parameter matrix of history M days interval censored data；

Step 3. is according to model formation predicted link speed

By parameter matrix C obtained in the previous step₁And φ₂Substitute in formula (1) simultaneously, the most measurable the M+1 days Interval road speeds later for j.

The remarkable result of the present invention is: needed for urban road speed prediction model based on analysis of spectrum, historical data is relatively Few, model parameter can quickly make dynamically adjustment with section, the change of period, with real road speed condition Coupling preferably, and can alleviate the impact that road speeds is predicted by urban road traffic state rapid fluctuations, improves Forecasting accuracy.

Accompanying drawing explanation

Fig. 1 is urban road speed predicting method flow chart based on analysis of spectrum.

Detailed description of the invention

Below in conjunction with drawings and Examples, technical scheme is described in further detail.Following example Implementing under premise the technical scheme is that, giving detailed embodiment and process, but this The protection domain of invention is not limited to following embodiment.

It is as follows that the present embodiment is embodied as step:

Step 1. utilizes GPS to collect road speeds data

1.1GPS collects data such as comprising acquisition time, car number, car speed, wherein chooses vehicle Speed data is as source data.

1.2 with west part of the ring road, stadium road, city of Hangzhou, white sand road, west part of the ring road, protect hill path, the north of a road, 40 sections such as Bao Lufengqi road are as data acquisition region, and 00:00-23:59 is as data acquisition time Section, the vehicle speed data that GPS collected every 15 minutes is as source data.

It is internal memory 2G that 1.3 data are stored in hardware environment, on the PC of hard disk 300G.

Step 2. preference pattern parameter

Represent that data to be predicted are right with it by the relation predicting the corresponding historical data of given data on the same day Answer the relation of historical data, i.e. refer to the parameter matrix C in forecast model.

Wherein road speeds forecast model formula based on analysis of spectrum is as follows:

$\hat{X}=C\&CenterDot;{\mathrm{\&phi;}}^T---\left(1\right)$

In formula:Represent certain continuum road speeds matrix that prediction obtains；C representation parameter matrix；φ is The eigenvectors matrix of respective bins road speeds variation tendency is described.

Parameter matrix C is asked to specifically comprise the following steps that

First GPS car speed is converted into road speeds by map-matching method by 2.1.

Then 2.2 will be divided equally into 96 intervals (every 15 minutes being an interval) every day, and this example is known JIUYUE in 2012 10 and JIUYUE 11 192 intervals of continuous 2 days and No. 12 the 1st to the 3rd districts of JIUYUE Between road speeds data.Choose 6 the interval censored data declarative procedures comprised in initial 00:00 01:30, Wherein exponent number K takes 2.

2.3 6 interval censored datas composition matrix X ' (2*6) of continuous 2 days of definition history.

$\left\{\begin{array}{cccccc}41& 38& 32& 19& 30& 22\\ 24& 44& 36& 39& 38& 31\end{array}\right\}$

According to formula R=X^TX tries to achieve covariance matrix R (6*6), as follows:

$\left\{\begin{array}{cccccc}2257& 2614& 2176& 1715& 2142& 1646\\ 2614& 3380& 2800& 2438& 2812& 2200\\ 2176& 2800& 2320& 2012& 2328& 1820\\ 1715& 2438& 2012& 1882& 2052& 1627\\ 2142& 2812& 2328& 2052& 2344& 1838\\ 1646& 2200& 1820& 1627& 1838& 1445\end{array}\right\}$

2.4 use QR decomposition method to try to achieve the eigenvectors matrix φ (6*6) of R, as follows:

$\left\{\begin{array}{cccccc}-0.435& 0.764& -0.078& 0.056& 0.466& 0.033\\ -0.504& -0.033& 0.766& 0.141& -0.319& 0.190\\ -0.420& 0.008& -0.248& 0.010& -0.393& -0.780\\ -0.331& -0.595& 0.096& 0.012& 0.695& -0.209\\ -0.413& -0.137& -0.294& -0.748& -0.146& 0.378\\ -0.317& -0.207& -0.500& 0.646& -0.147& 0.410\end{array}\right\}$

2.5 from the eigenvectors matrix φ of corresponding window selected part eigenvectors matrix φ ' (6*2), as follows:

$\left\{\begin{array}{cc}-0.435& 0.764\\ -0.504& -0.033\\ -0.420& 0.008\\ -0.331& -0.595\\ -0.413& -0.137\\ -0.317& -0.207\end{array}\right\}$

2.6 choose front 3 intervals and rear 3 interval road speeds change in this matrix of expression from φ ' respectively The eigenvectors matrix φ of trend₁(3*2) and φ₂(3*2).The most as follows:

${\mathrm{\&phi;}}_1=\left\{\begin{array}{cc}-0.435& 0.764\\ -0.504& -0.033\\ -0.420& 0.008\end{array}\right\}$ ${\mathrm{\&phi;}}_2=\left\{\begin{array}{cc}-0.331& -0.595\\ -0.413& -0.137\\ -0.317& -0.207\end{array}\right\}$

2.7 the following is the matrix of 3 corresponding interval data compositions front with this window in the 3rd day

{50 42 41}

WillAnd φ₁Substituting into formula (1), calculate C (1*2), result is as follows:

{-89.52 14.66}

Step 3. is according to model formation predicted link speed

By φ₂Substitute in model formation (1) with the parameter matrix C tried to achieve in step b, the most measurable JIUYUE 12 Number rear 3 unknown data corresponding with this window, result is as follows:

{21 35 25}

The present invention, on 40 sections of Hangzhou major trunk roads, acquires the gps data checking of 30 days.Experiment knot Fruit shows, compared with using ARIMA and kalman filter method predicted link speed, based on analysis of spectrum pre- Survey method makes forecast error (RMSE: root-mean-square error) reduce by more than 40%.

Claims (1)

1. a urban road speed predicting method based on analysis of spectrum, it is characterised in that comprise the following steps:

Step 1. collects the vehicle speed data that GPS collects

GPS collects the data comprising acquisition time, car number, car speed, wherein chooses car speed Data are as source data；

Step 2. preference pattern parameter

By the relation of given data corresponding history interval censored data on prediction same day represent data to be predicted with The relation of its corresponding historical data, i.e. refers to the parameter matrix C in forecast model；

Wherein road speeds forecast model formula based on analysis of spectrum is as follows:

$\hat{X}=C\&CenterDot;{\mathrm{\&phi;}}^T---\left(1\right)$

In formula:Represent certain continuum road speeds matrix that prediction obtains；C representation parameter matrix；φ is The eigenvectors matrix of respective bins road speeds variation tendency is described；

Parameter matrix C is asked to specifically comprise the following steps that

First GPS car speed is converted into road speeds by map-matching method by 2.1；

Then 2.2 will be divided equally into N number of interval every day, it is known that N number of interval, history continuous N sky and J interval road speeds data before M+1 days；

The matrix that data are M*N rank in 2.3 definition N number of intervals, history continuous N sky, is denoted as X；According to R=X^TX tries to achieve the covariance matrix R (N*N) of matrix X；

2.4 use QR decomposition method that matrix decomposition becomes an orthonomal matrix Q and upper triangular matrix R, ask Obtain the eigenvectors matrix of covariance matrix R, be designated as φ (N*N), in eigenvectors matrix φ, choose N*K Part is designated as φ ', and wherein, K represents exponent number, 1≤K < N, the quantity of the characteristic vector i.e. used；Choose square In battle array φ ', the 1st walks to jth row part, is designated as φ₁(j*K)；In like manner, the middle jth of φ '+1 walks to Nth row part, It is designated as φ₂((N-j)*K)；

The matrix of 2.5 the M+1 days known interval censored datas compositions is denoted asThe φ tried to achieve with previous step₁Together substitute into Model formation (1) converts the formula obtainedIn, the M+1 days known districts of expression can be tried to achieve Between parameter matrix between 1 to j data and corresponding M days history interval censored datas, be designated as C₁(1*K), it is possible to make Being the M+1 days the unknown interval censored datas and the parameter matrix of history M days interval censored data, wherein unknown interval is district Between after j, do not include interval j；

Step 3. is according to model formation predicted link speed

By parameter matrix C obtained in the previous step₁And φ₂Substitute in formula (1) simultaneously, the most measurable the M+1 days Interval road speeds later for j.