CN104933858A - Space traffic characteristic Kernel-KNN matching road traffic state obtain method - Google Patents

Abstract

A space traffic characteristic Kernel-KNN matching road traffic state obtain method comprises the following steps: firstly, extracting representative road traffic data of a target road section and space related road sections, and setting up a road traffic running characteristic reference sequence after data pre-treatment; then selecting a space road traffic data sequence, and forming a space road traffic data sequence Mercer core; finally selecting the space road traffic characteristic reference data sequence and a present space road traffic characteristic sequence, obtaining an Euclidean distance in characteristic space between the space road traffic characteristic reference data sequence and the present space road traffic characteristic sequence, using a KNN method to select traffic states of k roads of the target road section, and finally using weight average to obtain the road traffic state of the target road section. The method can carry out real time effective dynamic evaluation of road traffic running state, and can effectively obtain road traffic state with good accuracy.

Description

A kind of road traffic state acquisition methods mated based on space communication characteristic Kernel-KNN

Technical field

The invention belongs to traffic behavior and obtain field, relate to a kind of road traffic state acquisition methods.

Background technology

It is the important prerequisite of carrying out the traffic administration such as Traffic flux detection and induction that road traffic state obtains, be the necessary basis formulating the driving safe measures such as traffic safety management strategy, traffic hazard detection, traffic accident causation analysis, be traffic infrastructure management, monitor and safeguard the indispensable firsthand information.Therefore traffic behavior obtains is that traffic administration, traffic insurance and traffic infrastructure monitor the basic major issue safeguarded.

In existing road traffic state acquisition methods, on the one hand, due to the property complicated and changeable of road traffic system, when the road traffic state data in section self are unavailable, the road traffic state acquisition methods of effective popularity is lacked; On the other hand, very wide to the scope of the general assumed condition of population distribution in nonparametric model (KNN), the model set up based on nonparametric statistics has good robustness and adaptability, but existing method lacks taking into full account the many particle properties of road traffic state multidimensional.The technological deficiency existed is: accuracy is poor.

Summary of the invention

In order to the deficiency that the accuracy overcoming existing road traffic state acquisition methods is poor, the invention provides a kind of effective acquisition road traffic state, accuracy preferably based on the road traffic state acquisition methods that space communication characteristic Kernel-KNN mates.

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

Based on the road traffic state acquisition methods that space communication characteristic Kernel-KNN mates, described acquisition methods comprises the following steps:

(1) road traffic features reference sequences is set up:

Design road traffic features reference sequences, extracts representative highway traffic data, carries out data prediction, obtain target road section and the road traffic operation characteristic information with its space correlation section, and stored in road traffic operation characteristic reference sequences;

(2) kernel function of space highway traffic data sequence is built:

Choose the highway traffic data sequence with the many granularities of multidimensional in target road section space correlation section, utilize kernel function by this highway traffic data sequence mapping to feature space, build the kernel function of space highway traffic data sequence;

(3) road traffic state is obtained based on Kernel-KNN:

Extract space road traffic features reference data sequence and current spatial highway traffic data sequence, acquisition space road traffic features reference data sequence and current spatial highway traffic data sequence, at the Euclidean distance of high-dimensional feature space, choose k arest neighbors road traffic features reference data sequence by KNN method; From road traffic features reference data sequence, choose the road traffic state of target road section corresponding to this k arest neighbors road traffic features reference data sequence, obtain road traffic state finally by this k road traffic state weighted mean.

Further, in described step (1), suppose that the time interval of Traffic flow detecting device collection traffic state information is Δ t, the quantity of the traffic state information of single Traffic flow detecting device collection every day is N _um, setting the time dimension chosen is c × Δ t, and the road traffic parameter granularity chosen is d, choose with target road section space correlation and data can section number be r, then space highway traffic data sequence X (t) that t is chosen is:

X(t)＝[S ₁(t-(c-1)Δt) … S ₁(t-Δt) S ₁(t)

S ₂(t-(c-1)Δt) … S ₂(t-Δt) S ₂(t)

…

S _r(t-(c-1)Δt) … S _r(t-Δt) S _r(t)] ^T

S _j(t)＝[S _j1(t) S _j2(t) … S _jd(t)] ^T

Wherein, S ₁(t-(c-1) Δ t) is the road traffic state parameter set in the 1st article of section in the time period [t-(c-1) Δ t, t-(c-1) Δ t+ Δ t], S ₁(t-Δ t) is the road traffic state parameter set in the 1st article of section in the time period [t-Δ t, t], S ₂(t-(c-1) Δ t) is the road traffic state parameter set in time period [t-(c-1) Δ t, t-(c-1) Δ t+ Δ t] the 2nd article of section, S ₂(t-Δ t) is the road traffic state parameter set in the 2nd article of section in the time period [t-Δ t, t], S _r(t-(c-1) Δ t) is the road traffic state parameter set in r article of section in the time period [t-(c-1) Δ t, t-(c-1) Δ t+ Δ t], S _r(t-Δ t) is the road traffic state parameter set in time period [t-Δ t, t] r article section, S _jt () is the road traffic state parameter set in time period [t, t+ Δ t] interior jth bar section, j=1,2 ... r, S _ji(t) for interior jth bar section is at time period [t, t+ Δ t] interior i-th kind of road traffic state parameter value, i=1,2 ... d;

Space highway traffic data sequence X (t) is mapped to high-dimensional feature space by Nonlinear Mapping φ and obtains φ (X (t)) by described step, then t in feature space ₁moment and t ₂the dot product Mercer kernel representation of the space highway traffic data sequence high dimensional feature in moment is:

K(X(t ₁)，X(t ₂))＝<φ(X(t ₁))，φ(X(t ₂))>。

Further again, in described step (3), the process obtaining road traffic state based on space communication characteristic Kernel-KNN is as follows:

Step 3.1: choosing of different highway traffic data sequence

Set the road traffic state data choosing a days altogether and set up road traffic features reference sequences, then in road traffic features reference sequences, the quantity of the traffic state information in every bar section is a × N _um.In definition road traffic features reference sequences, the moment of h data point is (h Δ t), c-1≤h≤(a × N _um-1), the then space road traffic features reference data sequence X in this moment _s(h Δ t) is:

X _S(h·Δt)＝[S ₁(h·Δt-(c-1)Δt) … S ₁(h·Δt-Δt) S ₁(h·Δt)

S ₂(h·Δt-(c-1)Δt) … S ₂(h·Δt-Δt) S ₂(h·Δt)

…

S _r(h·Δt-(c-1)Δt) … S _r(h·Δt-Δt) S _r(h·Δt)] ^T

S _j(h·Δt)＝[S _j1(h·Δt) S _j2(h·Δt) … S _jd(h·Δt)] ^T

Wherein, S ₁(h Δ t-(c-1) Δ t) is the road traffic features reference value in the 1st article of section in the time period [h Δ t-(c-1) Δ t, h Δ t-(c-1) Δ t+ Δ t], S ₁(h Δ t-Δ t) is the road traffic features reference value in the 1st article of section in the time period [h Δ t-Δ t, h Δ t]; S ₂(h Δ t-(c-1) Δ t) is the road traffic features reference value in the 2nd article of section in the time period [h Δ t-(c-1) Δ t, h Δ t-(c-1) Δ t+ Δ t], S ₂(h Δ t-Δ t) is the road traffic features reference value in the 2nd article of section in the time period [h Δ t-Δ t, h Δ t]; S _r(h Δ t-(c-1) Δ t) is the road traffic features reference value in r article of section in the time period [h Δ t-(c-1) Δ t, h Δ t-(c-1) Δ t+ Δ t], S _r(h Δ t-Δ t) is the road traffic features reference value in r article of section in the time period [h Δ t-Δ t, h Δ t]; S _j(h Δ t) is the road traffic features reference value in time period [h Δ t, h Δ t+ Δ t] interior jth bar section, j=1,2 ... r, S _ji(h Δ t) is the i-th kind road traffic state parameter reference values of jth bar section within the time period [h Δ t, (h+1) Δ t], i=1,2 ... d; C-1≤h≤(a × N _um-1);

T _nfor current time scale, then the space highway traffic data sequence X (t that chooses of current time _n) be:

X(t _N)＝[S ₁(t _N-(c-1)Δt) … S ₁(t _N-Δt) S ₁(t _N)

S ₂(t _N-(c-1)Δt) … S ₂(t _N-Δt) S ₂(t _N)

…

S _r(t _N-(c-1)Δt) … S _r(t _N-Δt) S _r(t _N)] ^T

S _j(t _N)＝[S _j1(t _N) S _j2(t _N) … S _jd(t _N)] ^T

Wherein, S ₁(t _n-(c-1) Δ t) be time period [t _n-(c-1) Δ t, t _n-(c-1) Δ t+ Δ t] in the road traffic state parameter set in the 1st article of section, S ₁(t _n-Δ t) be time period [t _n-Δ t, t _n] in the road traffic state parameter set in the 1st article of section, S ₂(t _n-(c-1) Δ t) be time period [t _n-(c-1) Δ t, t _n-(c-1) Δ t+ Δ t] in the road traffic state parameter set in the 2nd article of section, S ₂(t _n-Δ t) be time period [t _n-Δ t, t _n] in the road traffic state parameter set in the 2nd article of section, S _r(t _n-(c-1) Δ t) be time period [t _n-(c-1) Δ t, t _n-(c-1) Δ t+ Δ t] in the road traffic state parameter set in r article of section, S _r(t _n-Δ t) be time period [t _n-Δ t, t _n] in the road traffic state parameter set in r article of section, S _j(t _n) be time period [t _n, t _n+ Δ t] the road traffic state parameter set in interior jth bar section, j=1,2 ... r, S _ji(t _n) for jth bar section is at time period [t _n, t _n+ Δ t] in i-th kind of road traffic state parameter value, i=1,2 ... d;

Step 3.2: the acquisition of Euclidean distance in highway traffic data sequence signature space

By Mercer core by the space highway traffic data sequence mapping of many for multidimensional granularities to feature space, based on the definition of kernel function, the space highway traffic data of the many granularities of multidimensional has been mapped to φ (X (t)), then space road traffic features reference data sequence X _s(h Δ t) and current spatial highway traffic data sequence X (t _n) Euclidean distance in feature space is expressed as:

$\begin{array}{c}d(X_S(h\&CenterDot;\mathrm{\&Delta;t}),X\left(t_N\right))=\sqrt{{\left|\right|\mathrm{\&phi;}\left(X_S(h\&CenterDot;\mathrm{\&Delta;t})\right)-\mathrm{\&phi;}(X\left(t_N\right)\left|\right|}^2}=\\ \sqrt{<\mathrm{\&phi;}\left(X_S(h\&CenterDot;\mathrm{\&Delta;t})\right),\mathrm{\&phi;}\left(X_S(h\&CenterDot;\mathrm{\&Delta;t})\right)>-2<\mathrm{\&phi;}\left(X_S(h\&CenterDot;\mathrm{\&Delta;t})\right),\mathrm{\&phi;}(X\left(t_N\right)>+<\mathrm{\&phi;}(X\left(t_N\right),\mathrm{\&phi;}(X\left(t_N\right)>}\\ =\sqrt{K(X_S(h\&CenterDot;\mathrm{\&Delta;t}),X_S(h\&CenterDot;\mathrm{\&Delta;t}))-2K(X_S(h\&CenterDot;\mathrm{\&Delta;t}),X\left(t_N\right))+K(X\left(t_N\right),X\left(t_N\right))}\end{array}$

Step 3.3: the acquisition of road traffic state

(3.3.1) range formula is utilized, calculate the core distance between current spatial highway traffic data sequence and space road traffic features reference data sequence, k arest neighbors space road traffic features reference data sequence X of selected distance current spatial highway traffic data sequence _s(g _iΔ t), 1≤i≤k, c-1≤g _i≤ (a × N _um-1);

(3.3.2) from road traffic features reference data sequence, X is chosen _s(g _iΔ t) road traffic state of corresponding target road section, be designated as S (g _iΔ t), 1≤i≤k, c-1≤i≤(a × N _um-1);

(3.3.3) t _nthe road traffic state in moment obtained by following formula:

$\stackrel{\&OverBar;}{S\left(t_N\right)}=\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_i\&CenterDot;S(g_i\&CenterDot;\mathrm{\&Delta;t})$

Wherein, μ _ibe the weighted value of i-th road traffic state, itself and current spatial highway traffic data sequence and space road traffic features reference data sequence are inversely proportional at the Euclidean distance of feature space.

Beneficial effect of the present invention is mainly manifested in: by introducing the kernel function of space highway traffic data sequence, by space highway traffic data sequence mapping to higher dimensional space, method finally by KNN coupling finally realizes effective acquisition of road traffic state, and its result can be applied in traffic state analysis, traffic guidance and control system.

Accompanying drawing explanation

Fig. 1 is the schematic diagram of the time format of road traffic features reference sequences.

Embodiment

Below in conjunction with accompanying drawing, the invention will be further described.

With reference to Fig. 1, a kind of road traffic state acquisition methods mated based on space communication characteristic Kernel-KNN, comprises the following steps:

(1) road traffic features reference sequences is set up:

Design road traffic features reference sequences, extracts representative highway traffic data, carries out data prediction, obtain target road section and the road traffic operation characteristic information with its space correlation section, and stored in road traffic operation characteristic reference sequences;

(2) space highway traffic data Sequence kernel function is built:

Choose the highway traffic data sequence with the many granularities of multidimensional in target road section space correlation section, utilize kernel function by this highway traffic data sequence mapping to feature space, build the kernel function of space highway traffic data sequence;

(3) road traffic state is obtained based on Kernel-KNN:

Extract space road traffic features reference data sequence and current spatial highway traffic data sequence, acquisition space road traffic features reference data sequence and current spatial highway traffic data sequence, at the Euclidean distance of high-dimensional feature space, choose k arest neighbors space road traffic features reference data sequence by KNN method; From road traffic features reference data sequence, choose the road traffic state of target road section corresponding to this k arest neighbors space road traffic features reference data sequence, obtain road traffic state finally by this k road traffic state weighted mean.

The road traffic state acquisition methods based on Kernel-KNN coupling of the present embodiment, containing following steps:

(1) step of road traffic features reference sequences is set up

Step 1.1: the structure of design road traffic features reference sequences

The collection period of setting road traffic state data is Δ t, then the time format of Traffic Information template as shown in Figure 1.

The sheet format of road traffic features reference sequences as shown in Table 1 and Table 2.

Table 1 road traffic features reference sequences information table:

Table 1

Table 2 road traffic features reference sequences description list:

Table 2

Step 1.2: by data prediction, sets up road traffic operation characteristic reference sequences

Obtain target road section and the representative road traffic state historical data with its space correlation section, line number of going forward side by side Data preprocess (mainly for reparation that is extremely indivedual and missing data), by in the road traffic state data input road traffic features reference sequences after data prediction, thus set up road traffic features reference sequences.

(2) step of space highway traffic data Sequence kernel function is built

The space road traffic state data choosing the many granularities of multidimensional are mated.Supposing that Traffic flow detecting device gathers time interval of traffic state information is Δ t, and the quantity of the traffic state information that single Traffic flow detecting device gathers every day is N _um.Setting the time dimension chosen is c × Δ t, and the road traffic parameter granularity chosen is d, with target road section space correlation and data can section number be r.Space highway traffic data sequence X (t) that then t is chosen is:

X(t)＝[S ₁(t-(c-1)Δt) … S ₁(t-Δt) S ₁(t)

S ₂(t-(c-1)Δt) … S ₂(t-Δt) S ₂(t)

…

S _r(t-(c-1)Δt) … S _r(t-Δt) S _r(t)] ^T

S _j(t)＝[S _j1(t) S _j2(t) … S _jd(t)] ^T

Wherein, S ₁(t-(c-1) Δ t) is the road traffic state parameter set in the 1st article of section in the time period [t-(c-1) Δ t, t-(c-1) Δ t+ Δ t], S ₁(t-Δ t) is the road traffic state parameter set in the 1st article of section in the time period [t-Δ t, t], S ₂(t-(c-1) Δ t) is the road traffic state parameter set in the 2nd article of section in the time period [t-(c-1) Δ t, t-(c-1) Δ t+ Δ t], S ₂(t-Δ t) is the road traffic state parameter set in the 2nd article of section in the time period [t-Δ t, t], S _r(t-(c-1) Δ t) is the road traffic state parameter set in r article of section in the time period [t-(c-1) Δ t, t-(c-1) Δ t+ Δ t], S _r(t-Δ t) is the road traffic state parameter set in r article of section in the time period [t-Δ t, t], S _jt () is the road traffic state parameter set in time period [t, t+ Δ t] interior jth bar section, j=1,2 ... r, S _jit () is the i-th kind road traffic state parameter value of jth bar section within the time period [t, t+ Δ t], i=1,2 ... d;

Space highway traffic data sequence X (t) is mapped to high-dimensional feature space by Nonlinear Mapping φ and obtains φ (X (t)).Then t in feature space ₁moment and t ₂the dot product Mercer kernel representation of the space highway traffic data sequence high dimensional feature in moment is:

K(X(t ₁)，X(t ₂))＝<φ(X(t ₁))，φ(X(t ₂))>

Nonlinear Mapping φ is determined by the expression-form of kernel function.

(3) step of road traffic state is obtained based on space communication characteristic Kernel-KNN

Step 3.1: choosing of different highway traffic data sequence

Mating based on space communication characteristic Kernel-KNN in the process obtaining road traffic state, the data acquisition related to mainly contains two: space road traffic features reference data sequence and current spatial highway traffic data sequence.

Set the road traffic state data choosing a days altogether and set up road traffic features reference sequences, then in road traffic features reference sequences, the quantity of every bar road section traffic volume status information is a × N _um.In definition road traffic features reference sequences, the moment of h data point is (h Δ t), c-1≤h≤(a × N _um-1), the space road traffic features reference data sequence X in this moment _s(h Δ t) is:

X _S(h·Δt)＝[S ₁(h·Δt-(c-1)Δt) … S ₁(h·Δt-Δt) S ₁(h·Δt)

S ₂(h·Δt-(c-1)Δt) … S ₂(h·Δt-Δt) S ₂(h·Δt)

…

S _r(h·Δt-(c-1)Δt) … S _r(h·Δt-Δt) S _r(h·Δt)] ^T

S _j(h·Δt)＝[S _j1(h·Δt) S _j2(h·Δt) … S _jd(h·Δt)] ^T

Wherein, S ₁(h Δ t-(c-1) Δ t) is the road traffic features reference value in the 1st article of section in the time period [h Δ t-(c-1) Δ t, h Δ t-(c-1) Δ t+ Δ t], S ₁(h Δ t-Δ t) is the road traffic features reference value in the 1st article of section in the time period [h Δ t-Δ t, h Δ t]; S ₂(h Δ t-(c-1) Δ t) is the road traffic features reference value in the 2nd article of section in the time period [h Δ t-(c-1) Δ t, h Δ t-(c-1) Δ t+ Δ t], S ₂(h Δ t-Δ t) is the road traffic features reference value in the 2nd article of section in the time period [h Δ t-Δ t, h Δ t]; S _r(h Δ t-(c-1) Δ t) is the road traffic features reference value in r article of section in the time period [h Δ t-(c-1) Δ t, h Δ t-(c-1) Δ t+ Δ t], S _r(h Δ t-Δ t) is the road traffic features reference value in r article of section in the time period [h Δ t-Δ t, h Δ t]; S _j(h Δ t) is the road traffic features reference value in time period [h Δ t, h Δ t+ Δ t] interior jth bar section, j=1,2 ... r, S _ji(h Δ t) is the i-th kind road traffic state parameter reference values of jth bar section within the time period [h Δ t, (h+1) Δ t], i=1,2 ... d; C-1≤h≤(a × N _um-1);

T _nfor current time scale, then the space highway traffic data sequence X (t that chooses of current time _n) be:

X(t _N)＝[S ₁(t _N-(c-1)Δt) … S ₁(t _N-Δt) S ₁(t _N)

S ₂(t _N-(c-1)Δt) … S ₂(t _N-Δt) S ₂(t _N)

…

S _r(t _N-(c-1)Δt) … S _r(t _N-Δt) S _r(t _N)] ^T

S _j(t _N)＝[S _j1(t _N) S _j2(t _N) … S _jd(t _N)] ^T

Wherein, S ₁(t _n-(c-1) Δ t) be time period [t _n-(c-1) Δ t, t _n-(c-1) Δ t+ Δ t] in the road traffic state parameter set in the 1st article of section, S ₁(t _n-Δ t) be time period [t _n-Δ t, t _n] in the road traffic state parameter set in the 1st article of section, S ₂(t _n-(c-1) Δ t) be time period [t _n-(c-1) Δ t, t _n-(c-1) Δ t+ Δ t] in the road traffic state parameter set in the 2nd article of section, S ₂(t _n-Δ t) be time period [t _n-Δ t, t _n] in the road traffic state parameter set in the 2nd article of section, S _r(t _n-(c-1) Δ t) be time period [t _n-(c-1) Δ t, t _n-(c-1) Δ t+ Δ t] in the road traffic state parameter set in r article of section, S _r(t _n-Δ t) be time period [t _n-Δ t, t _n] in the road traffic state parameter set in r article of section, S _j(t _n) be time period [t _n, t _n+ Δ t] the road traffic state parameter set in interior jth bar section, j=1,2 ... r, S _ji(t _n) for jth bar section is at time period [t _n, t _n+ Δ t] in i-th kind of road traffic state parameter value, i=1,2 ... d;

Step 3.2: the acquisition of Euclidean distance in highway traffic data sequence signature space, space

By Mercer core by the space highway traffic data sequence mapping of many for multidimensional granularities to feature space, after making to map, similar sample is close, and foreign peoples's sample becomes far away.Based on the definition of kernel function, the space highway traffic data of the many granularities of multidimensional has been mapped to φ (X (t)), then space road traffic features reference data sequence X _s(h Δ t) and current spatial highway traffic data sequence X (t _n) Euclidean distance in feature space is expressed as:

$\begin{array}{c}d(X_S(h\&CenterDot;\mathrm{\&Delta;t}),X\left(t_N\right))=\sqrt{{\left|\right|\mathrm{\&phi;}\left(X_S(h\&CenterDot;\mathrm{\&Delta;t})\right)-\mathrm{\&phi;}(X\left(t_N\right)\left|\right|}^2}=\\ \sqrt{<\mathrm{\&phi;}\left(X_S(h\&CenterDot;\mathrm{\&Delta;t})\right),\mathrm{\&phi;}\left(X_S(h\&CenterDot;\mathrm{\&Delta;t})\right)>-2<\mathrm{\&phi;}\left(X_S(h\&CenterDot;\mathrm{\&Delta;t})\right),\mathrm{\&phi;}(X\left(t_N\right)>+<\mathrm{\&phi;}(X\left(t_N\right),\mathrm{\&phi;}(X\left(t_N\right)>}\\ =\sqrt{K(X_S(h\&CenterDot;\mathrm{\&Delta;t}),X_S(h\&CenterDot;\mathrm{\&Delta;t}))-2K(X_S(h\&CenterDot;\mathrm{\&Delta;t}),X\left(t_N\right))+K(X\left(t_N\right),X\left(t_N\right))}\end{array}$

Step 3.3: the acquisition of road traffic state

(3.3.1) range formula is utilized, calculate the core distance between current spatial highway traffic data sequence and space road traffic features reference data sequence, k arest neighbors space road traffic features reference data sequence X of selected distance current spatial highway traffic data sequence _s(g _iΔ t), 1≤i≤k, c-1≤g _i≤ (a × N _um-1).

(3.3.2) from the road traffic features reference data sequence of space, X is chosen _s(g _iΔ t) road traffic state of corresponding target road section, be designated as S (g _iΔ t), 1≤i≤k, c-1≤i≤(a × N _um-1);

(3.3.3) t _nthe road traffic state in moment obtained by following formula:

$\stackrel{\&OverBar;}{S\left(t_N\right)}=\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_i\&CenterDot;S(g_i\&CenterDot;\mathrm{\&Delta;t})$

Wherein, μ _ibe the weighted value of i-th road traffic state, itself and current spatial highway traffic data sequence and space road traffic features reference data sequence are inversely proportional at the Euclidean distance of feature space.

Example: a kind of road traffic state acquisition methods mated based on space communication characteristic Kernel-KNN,

Comprise the steps:

(1) experimental data is chosen

Consider availability and the validity of real road traffic state data, the highway traffic data choosing typical section on Beijing's Second Ring Road carries out algorithm application and checking.The specifying information in section is as shown in table 3, and table 3 is experimental road segment information table:

Table 3

HI7036b (Capital Library-> stabilizes the raft of pontoons) is carried out experimental verification as target road section.

The road traffic historical data extracting two weeks (2011.06.01-2011.06.14) sets up road traffic features reference sequences.The acquisition interval of delta t of road traffic state data is 2min.

Using the highway traffic data of 2011.06.18 as training dataset, carry out algorithm parameter setting.Using the highway traffic data of 2011.06.19,2011.06.25,2011.06.26 as experimental data collection, carry out proof of algorithm.

(2) kernel function is chosen

In view of gaussian kernel function has separability and locality, gaussian kernel function is selected to carry out algorithm application and checking.The citation form of gaussian kernel function is as follows:

$K(x,y)=e^{-\frac{{\left|\right|x-y\left|\right|}^2}{2{\mathrm{\&sigma;}}^2}}$

(3) parameter is determined

Obtaining in the process of road traffic state based on gaussian kernel function, the parameter related to mainly comprises: σ value, time dimension c Δ t, k value.For different road traffic state data sets, parameters corresponding when obtaining optimum road traffic state parameter is different.Here done setting parameter is just to the general impact analysis of parameter to the road traffic state acquisition algorithm mated based on space communication characteristic Kernel-KNN.

Because the precision of these parameters on algorithm respectively has impact, analyzing separately each parameter can not guarantee algorithm optimum on the impact of arithmetic accuracy, therefore should consider the impact of all parameters on this road traffic state acquisition algorithm when carrying out Algorithm Analysis simultaneously.

Introduce normalization mean absolute error to analyze the impact of parameter on arithmetic accuracy:

$\mathrm{NMAE}=\frac{\left|\stackrel{\&OverBar;}{S(t_N+\mathrm{\&Delta;t})}\right|-S(t_N+\mathrm{\&Delta;t})}{S(t_N+\mathrm{\&Delta;t})}$

Namely for different (σ, c, k), there is NMAE corresponding with it.Therefore there is following equation:

NMAE＝ω(σ,c,k)

Namely there is certain distribution relation ω in (σ, c, k) and NMAE, finds time corresponding (σ, c, the k) that NMAE is minimum, be optimized parameter assignment procedure.Therefore can obtain as drag:

Min ω(σ,c,k)

Where $\mathrm{NMAE}=\frac{\left|\stackrel{\&OverBar;}{S(t_N+\mathrm{\&Delta;t})}\right|-S(t_N+\mathrm{\&Delta;t})}{S(t_N+\mathrm{\&Delta;t})}$

Finally the value of (σ, c, k) can be determined by the statistical study of road traffic state historical data.

(4) experimental result

Based on the data extraction in experiment section and the determination of each parameter, the road traffic state of target road section is estimated.Speed and flow are as the parameter that the most effectively can reflect road traffic state, and this experimental result is estimated mainly for the travelling speed value in section and statistic fluid value.Comparative for making experimental result have, by experimental result with obtain road traffic state data based on simple historical data and contrast based on the result that KNN obtains road traffic state data.

The standard variance σ of absolute error e, percentage error PE and absolute error is utilized to carry out check algorithm precision,

e＝|S*-S|，PE＝|S ^*-S|/S，

Wherein, S* is the traffic behavior parameter value calculated; V is actual traffic behavior parameter value; N is experiment numbers.

The statistical study of the traffic behavior acquisition result of all experiment section 2011.06.16-2011.06.29 is as shown in the table.Wherein, e _ker, PE _kerthe absolute error and the percentage error that obtain traffic behavior based on Kernel-KNN, σ _kerit is the standard variance of its absolute error; e _kNN, PE _kNNthe absolute error and the percentage error that obtain traffic behavior based on KNN, σ _kNNit is the standard variance of its absolute error; e _sim, PE _simthe absolute error and the percentage error that obtain traffic behavior based on simple historical data, σ _simit is the standard variance of its absolute error.

Table 4 is added up for speed obtains result:

Table 4

Table 5 is added up for flow obtains result:

Table 5.

Claims (3)

1., based on the road traffic state acquisition methods that space communication characteristic Kernel-KNN mates, it is characterized in that: described acquisition methods comprises the following steps:

(1) road traffic features reference sequences is set up:

Design road traffic features reference sequences, extracts representative highway traffic data, carries out data prediction, obtain target road section and the road traffic operation characteristic information with its space correlation section, and stored in road traffic operation characteristic reference sequences;

(2) kernel function of space highway traffic data sequence is built:

Choose the highway traffic data sequence with the many granularities of multidimensional in target road section space correlation section, utilize kernel function by this highway traffic data sequence mapping to feature space, build the kernel function of space highway traffic data sequence;

(3) road traffic state is obtained based on Kernel-KNN:

Extract space road traffic features reference data sequence and current spatial highway traffic data sequence, acquisition space road traffic features reference data sequence and current spatial highway traffic data sequence, at the Euclidean distance of high-dimensional feature space, choose k arest neighbors road traffic features reference data sequence by KNN method; From road traffic features reference data sequence, choose the road traffic state of target road section corresponding to this k arest neighbors road traffic features reference data sequence, obtain road traffic state finally by this k road traffic state weighted mean.

2. as claimed in claim 1 based on the road traffic state acquisition methods that space communication characteristic Kernel-KNN mates, it is characterized in that: in described step (1), supposing that Traffic flow detecting device gathers time interval of traffic state information is Δ t, and the quantity of the traffic state information that single Traffic flow detecting device gathers every day is N _um, setting the time dimension chosen is c × Δ t, and the road traffic parameter granularity chosen is d, choose with target road section space correlation and data can section number be r, then space highway traffic data sequence X (t) that t is chosen is:

X(t)＝[S ₁(t-(c-1)Δt) … S ₁(t-Δt) S ₁(t)

S ₂(t-(c-1)Δt) … S ₂(t-Δt) S ₂(t)

…

S _r(t-(c-1)Δt) … S _r(t-Δt) S _r(t)] ^T

S _j(t)＝[S _j1(t) S _j2(t) … S _jd(t)] ^T

Wherein, S ₁(t-(c-1) Δ t) is the road traffic state parameter set in the 1st article of section in the time period [t-(c-1) Δ t, t-(c-1) Δ t+ Δ t], S ₁(t-Δ t) is the road traffic state parameter set in the 1st article of section in the time period [t-Δ t, t], S ₂(t-(c-1) Δ t) is the road traffic state parameter set in time period [t-(c-1) Δ t, t-(c-1) Δ t+ Δ t] the 2nd article of section, S ₂(t-Δ t) is the road traffic state parameter set in the 2nd article of section in the time period [t-Δ t, t], S _r(t-(c-1) Δ t) is the road traffic state parameter set in r article of section in the time period [t-(c-1) Δ t, t-(c-1) Δ t+ Δ t], S _r(t-Δ t) is the road traffic state parameter set in time period [t-Δ t, t] r article section, S _jt () is the road traffic state parameter set in time period [t, t+ Δ t] interior jth bar section, j=1,2 ... r, S _ji(t) for interior jth bar section is at time period [t, t+ Δ t] interior i-th kind of road traffic state parameter value, i=1,2 ... d;

Space highway traffic data sequence X (t) is mapped to high-dimensional feature space by Nonlinear Mapping φ and obtains φ (X (t)) by described step, then t in feature space ₁moment and t ₂the dot product Mercer kernel representation of the space highway traffic data sequence high dimensional feature in moment is:

K(X(t ₁)，X(t ₂))＝<φ(X(t ₁))，φ(X(t ₂))>。

3. as claimed in claim 2 based on the road traffic state acquisition methods that space communication characteristic Kernel-KNN mates, it is characterized in that: in described step (3), the process obtaining road traffic state based on space communication characteristic Kernel-KNN is as follows:

Step 3.1: choosing of different highway traffic data sequence

Set the road traffic state data choosing a days altogether and set up road traffic features reference sequences, then in road traffic features reference sequences, the quantity of the traffic state information in every bar section is a × N _um, in definition road traffic features reference sequences, the moment of h data point is (h Δ t), c-1≤h≤(a × N _um-1), the then space road traffic features reference data sequence X in this moment _s(h Δ t) is:

X _S(h·Δt)＝[S ₁(h·Δt-(c-1)Δt) … S ₁(h·Δt-Δt) S ₁(h·Δt)

S ₂(h·Δt-(c-1)Δt) … S ₂(h·Δt-Δt) S ₂(h·Δt)

…

S _r(h·Δt-(c-1)Δt) … S _r(h·Δt-Δt) S _r(h·Δt)] ^T

S _j(h·Δt)＝[S _j1(h·Δt) S _j2(h·Δt) … S _jd(h·Δt)] ^T

Wherein, S ₁(h Δ t-(c-1) Δ t) is the road traffic features reference value in the 1st article of section in the time period [h Δ t-(c-1) Δ t, h Δ t-(c-1) Δ t+ Δ t], S ₁(h Δ t-Δ t) is the road traffic features reference value in the 1st article of section in the time period [h Δ t-Δ t, h Δ t]; S ₂(h Δ t-(c-1) Δ t) is the road traffic features reference value in the 2nd article of section in the time period [h Δ t-(c-1) Δ t, h Δ t-(c-1) Δ t+ Δ t], S ₂(h Δ t-Δ t) is the road traffic features reference value in the 2nd article of section in the time period [h Δ t-Δ t, h Δ t]; S _r(h Δ t-(c-1) Δ t) is the road traffic features reference value in r article of section in the time period [h Δ t-(c-1) Δ t, h Δ t-(c-1) Δ t+ Δ t], S _r(h Δ t-Δ t) is the road traffic features reference value in r article of section in the time period [h Δ t-Δ t, h Δ t]; S _j(h Δ t) is the road traffic features reference value in time period [h Δ t, h Δ t+ Δ t] interior jth bar section, j=1,2 ... r, S _ji(h Δ t) is the i-th kind road traffic state parameter reference values of jth bar section within the time period [h Δ t, (h+1) Δ t], i=1,2 ... d; C-1≤h≤(a × N _um-1);

T _nfor current time scale, then the space highway traffic data sequence X (t that chooses of current time _n) be:

X(t _N)＝[S ₁(t _N-(c-1)Δt)…S ₁(t _N-Δt)S ₁(t _N)

S ₂(t _N-(c-1)Δt)…S ₂(t _N-Δt)S ₂(t _N)

…

S _r(t _N-(c-1)Δt)…S _r(t _N-Δt)S _r(t _N)] ^T

S _j(t _N)＝[S _j1(t _N)S _j2(t _N)…S _jd(t _N)] ^T

Wherein, S ₁(t _n-(c-1) Δ t) be time period [t _n-(c-1) Δ t, t _n-(c-1) Δ t+ Δ t] in the road traffic state parameter set in the 1st article of section, S ₁(t _n-Δ t) be time period [t _n-Δ t, t _n] in the road traffic state parameter set in the 1st article of section, S ₂(t _n-(c-1) Δ t) be time period [t _n-(c-1) Δ t, t _n-(c-1) Δ t+ Δ t] in the road traffic state parameter set in the 2nd article of section, S ₂(t _n-Δ t) be time period [t _n-Δ t, t _n] in the road traffic state parameter set in the 2nd article of section, S _r(t _n-(c-1) Δ t) be time period [t _n-(c-1) Δ t, t _n-(c-1) Δ t+ Δ t] in the road traffic state parameter set in r article of section, S _r(t _n-Δ t) be time period [t _n-Δ t, t _n] in the road traffic state parameter set in r article of section, S _j(t _n) be time period [t _n, t _n+ Δ t] the road traffic state parameter set in interior jth bar section, j=1,2 ... r, S _ji(t _n) for jth bar section is at time period [t _n, t _n+ Δ t] in i-th kind of road traffic state parameter value, i=1,2 ... d;

Step 3.2: the acquisition of Euclidean distance in highway traffic data sequence signature space

By Mercer core by the space highway traffic data sequence mapping of many for multidimensional granularities to feature space, based on the definition of kernel function, the space highway traffic data of the many granularities of multidimensional has been mapped to φ (X (t)), then space road traffic features reference data sequence X _s(h Δ t) and current spatial highway traffic data sequence X (t _n) Euclidean distance in feature space is expressed as:

$\begin{array}{c}d(X_S(h\&CenterDot;\mathrm{\&Delta;t}),X\left(t_N\right))=\sqrt{{\left|\right|\mathrm{\&phi;}\left(X_S(h\&CenterDot;\mathrm{\&Delta;t})\right)-\mathrm{\&phi;}\left(X\left(t_N\right)\right)\left|\right|}^2}=\\ \sqrt{\&lang;\mathrm{\&phi;}\left(X_S(h\&CenterDot;\mathrm{\&Delta;t})\right),\mathrm{\&phi;}\left(X_S(h\&CenterDot;\mathrm{\&Delta;t})\right)\&rang;-2\&lang;\mathrm{\&phi;}\left(X_S(h\&CenterDot;\mathrm{\&Delta;t})\right),\mathrm{\&phi;}\left(X\left(t_N\right)\right)\&rang;+\&lang;\mathrm{\&phi;}(X\left(t_N\right),\mathrm{\&phi;}(X\left(t_N\right)\&rang;}\\ =\sqrt{K(X_S(h\&CenterDot;\mathrm{\&Delta;t}),X_S(h\&CenterDot;\mathrm{\&Delta;t}))-2K(X_S(h\&CenterDot;\mathrm{\&Delta;t}),X\left(t_N\right))+K(X\left(t_N\right),X\left(t_N\right))}\end{array}$

Step 3.3: the acquisition of road traffic state

(3.3.1) range formula is utilized, calculate the core distance between current spatial highway traffic data sequence and space road traffic features reference data sequence, k arest neighbors space road traffic features reference data sequence X of selected distance current spatial highway traffic data sequence _s(g _iΔ t), 1≤i≤k, c-1≤g _i≤ (a × N _um-1);

(3.3.2) from road traffic features reference data sequence, X is chosen _s(g _iΔ t) road traffic state of corresponding target road section, be designated as S (g _iΔ t), 1≤i≤k, c-1≤i≤(a × N _um-1);

(3.3.3) t _nthe road traffic state in moment obtained by following formula:

$\stackrel{\&OverBar;}{S\left(t_N\right)}=\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_i\&CenterDot;S(g_i\&CenterDot;\mathrm{\&Delta;t})$

Wherein, μ _ibe the weighted value of i-th road traffic state, itself and current spatial highway traffic data sequence and space road traffic features reference data sequence are inversely proportional at the Euclidean distance of feature space.