CN106408960A - Signal control time period dividing method based on ordered clustering - Google Patents

Abstract

The invention provides a signal control time period dividing method based on ordered clustering. The signal control time period dividing method utilizes regular fluctuations of actual sequences of a traffic flow, and performs division and clustering on time sequences with high similarities, so as to obtain reasonable division of traffic time periods and define a traffic control time period dividing scheme. The signal control time period dividing method based on ordered clustering realizes automatic optimization of number of intersection signal control time periods, thereby achieving reasonable grouping of the traffic flow.

Description

Signal control time dividing method based on orderly cluster

Technical field

The present invention relates to a kind of orderly cluster method of signal control time segmentation, can be with Automatic Optimal clusters number and defeated Go out splitting scheme, provide for timing controlled at times and support, belong to traffic control research field.

Background technology

Although majority signal control system is provided with adaptive control function, it is limited to the constraint of information integrity, timesharing The fixed signal of section controls one of control mode being still most commonly seen.Adopting currently for the segmentation of control time more Traditional clustering method, according to the attribute of flow itself, mathematically according to certain similitude or otherness index, quantitative description Close and distant relation between sample, and by this close and distant degree, cluster analysis is completed to sample, namely Time segments division.Existing method is deposited Not enough as follows：It is unable to the rational clusters number of Automatic Optimal, preferred plan can only be sought by way of enumerating, its method Time complexity is very big.It is therefore proposed that a kind of time complexity is little and method that can automatically export clusters number and scheme is inevitable The speed of Time segments division and the reliability of result can be greatly promoted, controlling for multi-period fixed signal provides technical support.

Content of the invention

It is an object of the invention to realizing the Automatic Optimal of integrative design intersection fixed number and concrete scheme, it is basic Thought is：Using the regular fluctuation of the actual sequence of the magnitude of traffic flow, time series higher for similitude is carried out segmentation cluster, from And obtain the classifying rationally of traffic slot, define traffic control Time segments division scheme.

The basic step of the present invention is as follows：

C1, set cluster number be k, calculating different value of K in the case of each class diameter；

Error function in the case of c2, design different value of K；

Error recursion value in the case of c3, calculating different value of K；

C4, cluster numbers k are determined by error amount in the case of different value of K_op；

C5, according to k_opObtain optimal classification scheme with error function.

The process of step c1 includes：

C11, the traffic statistics interval of intersection are generally fixing period T, usually 5 minutes, 10 minutes or 15 minutes； Assume that the data on flows of a day has n data sample, then

C12, one day n ordered data sample is made to be followed successively by Z₁,Z₂,…,Z_n, the data sample that class G comprises has { Z_i, Z_i+1,…,Z_j}(j>I), i.e. G={ i, i+1 ..., j }, calculates such diameter D (i, j)

1. calculate such class average

In formula：Average for class G；T numbers for seasonal effect in time series；Z_tFlow value corresponding to the sequence samples t time.

2. calculate the diameter of each class, i.e. sum of squares of deviations in class：

In formula：D (i, j) is the distance between i-th, j section；T is the time；Z_tStream corresponding to the sequence samples t time Value；Average for class G；Weigh the sum of squares of deviations of each class by calculated diameter, for representing data difference DRS in class Degree, numerical value is less to represent that difference is less.

The process of step c2 includes：

C21, calculating cluster sum are error function during k, remember that concrete sorting technique is：

G₁={ i₁,i₁+1,…,i₂-1}

G₂={ i₂,i₂+1,…,i₃-1}

…

G_k={ i_k,i_k+1,…,n}

Wherein branch is 1=i₁<i₂<…<i_k<N=i_k+1- 1 (i.e. n+1=i_k+1).

Its error function expresses formula：

In formula：K is cluster numbers；B (n, k) is a certain sorting technique that n data is divided into k class；L [b (n, k)] is n Data is divided into the error function of k class；T numbers for time series；D(i_t,i_t+1- 1) it is i-th_tWith i_t+1The distance value of -1 data, meter Calculate error function, for the classification of optimal recurrence.

The process of step c3 includes：

C31, calculation error function recursion value, and record the position j of optimal partition point；

In formula：P (n, k) is that n data is divided into k class to make error function reach the sorting technique of minimum；L [p (n, 2)] is n The data on flows of individual section is divided into the error function of 2 classes；L [p (n, k)] is that the data on flows of n section is divided into the error letter of k class Number；D (j, n) is the diameter between jth and n data；Work as n, when k is fixing, L [p (n, k)] is less, represent the deviation of all classes Quadratic sum is minimum, and classification is rationally.Therefore, understand that calculating n sample is divided into the error function of k class to reach minimum by stepping type, then J-1 sample of calculating is needed to be divided into the error function of k-1 class to reach minimum, by that analogy.

The process of step c4 includes：

C41, make the tendency chart that L [p (n, k)] changes with k, as shown in Figure 2；

C42, searched out k value corresponding to maximum flex point by tendency chart as preferable clustering number k_op；

1. calculate the rate of change of error corresponding to any two adjacent k value：

2. the slope difference before and after calculating, obtains front difference diff₁, rear difference diff₂Value:

C43, utilize front difference diff₁With rear difference diff₂Ratio be worth to decision diagram maximum flex point, that is,：

K value when c44, diff value is maximum is optimal cluster number k_op.

The process of step c5 includes：

C51, according to k_opObtain optimal segmentation scheme with error function：

N data sample is divided into k_opThe optimum segmentation of class is divided into k it should set up in j-1 sample_opThe optimal sorting of -1 class On the basis of cutting, by that analogy.Using the error recurrence function calculating, the position j with record optimal partition point, respectively obtain poly- For 2,3 ... k_opSegmentation situation during class.

Beneficial effects of the present invention：The present invention proposes a kind of signal control time dividing method based on orderly cluster, Using the regular fluctuation of the actual sequence of the magnitude of traffic flow, time series higher for similitude is clustered, realize intersecting message The Automatic Optimal of number control time number, thus obtaining the reasonable packet of the magnitude of traffic flow, defines traffic control Time segments division with this Scheme.

Brief description

Fig. 1 algorithm realizes process flow diagram flow chart；

Fig. 2 preferable clustering number number k tendency chart；

Fig. 3 sequence data cluster result figure.

Specific embodiment

, data is split, sees Fig. 1 taking certain 24 hours flow in city intersection sky as a example.

1st, the number setting class is as k, the diameter of each class in the case of calculating different value of K；

The data on flows of (1) one day has 1440 data samples；When the traffic statistics interval of intersection is generally fixing Section T, this fixation period is taken as 5 minutes.

(2) remember that one day 288 Ordered Sample time flow point is followed successively by { Z₁,Z₂,…,Z_n, class G_iComprising sample has { Z_i, Z_i+1,…,Z_j}(j>I), remember G_i=i, i+1 ..., j }；

1. calculate each class average

2. calculate the diameter of each class；

2nd, the error function in the case of calculating different value of K；

(1) calculate cluster sum for error during k；

Remember that concrete sorting technique is：

G₁={ i₁,i₁+1,…,i₂-1}

G₂={ i₂,i₂+1,…,i₃-1}

…

G_k={ i_k,i_k+1,…,n}

Wherein branch is 1=i₁<i₂<…<i_k<N=i_k+1-1.Its error function expresses formula：

3rd, calculate different value of K in the case of error function recurrence formula；

(1) calculation error function recurrence formula, and record taken point position j；

4th, cluster numbers k are determined by error function；

(1) make the tendency chart that L [p (n, 2)] changes with k, as shown in Figure 2；

(2) the k value corresponding to maximum flex point is searched out by tendency chart；

1. calculate the rate of change of error corresponding to any two adjacent k value：

2. the slope difference before and after calculating, obtains front difference diff₁, rear difference diff₂Value:

3. utilize front difference diff₁With rear difference diff₂Ratio be worth to decision diagram maximum flex point, that is,：

4. k value when diff value is maximum is optimal cluster number k_op.

5th, according to k_opObtain optimal cluster segmentation with error function.

(1) according to k_opObtain optimal cluster segmentation with error function：

N data sample is divided into k_opThe optimum segmentation of class is divided into k it should set up in j-1 sample_opThe optimal sorting of -1 class On the basis of cutting, by that analogy.Using the error recurrence function calculating, the position j with record optimal partition point, respectively obtain poly- For 2,3 ... k_opSegmentation situation during class, as shown in Figure 3.

Claims (6)

1. the signal control time dividing method based on orderly cluster is it is characterised in that the method comprises the following steps：

C1, set cluster number be k, calculating different value of K in the case of each class diameter；

Error function in the case of c2, design different value of K；

Error recursion value in the case of c3, calculating different value of K；

C4, cluster numbers k are determined by error amount in the case of different value of K_op；

C5, according to k_opObtain optimal classification scheme with error function.

2. the signal control time dividing method based on orderly cluster according to claim 1 it is characterised in that：Step c1 Specifically：

C11, the traffic statistics interval of intersection, using fixing period T, are 5 minutes, 10 minutes or 15 minutes；Assume one day Data on flows has n data sample, then

$n=\frac{1440}{T}---(1-a)$

C12, one day n ordered data sample is made to be followed successively by Z₁,Z₂,…,Z_n, class G_iThe data sample comprising has { Z_i,Z_i+1,…, Z_j}(j>I), i.e. G_i=i, i+1 ..., and j }, calculate such diameter D (i, j)

1. calculate the class average of each class

$\stackrel{\&OverBar;}{Z_{G_i}}=\frac{1}{j-i+1}\underset{t=i}{\overset{j}{\&Sigma;}}{Z}_t---(1-b)$

In formula：For class G_iAverage；T numbers for seasonal effect in time series；Z_tFlow value corresponding to the sequence samples t time；

2. calculate the diameter of each class, i.e. sum of squares of deviations in class：

$D(i,j)=\underset{t=i}{\overset{j}{\&Sigma;}}|Z_t-\stackrel{\&OverBar;}{Z_{G_i}}|---(1-c)$

In formula：D (i, j) is the distance between i-th, j section；T is the time；Z_tFlow value corresponding to the sequence samples t time；For class G_iAverage；Weigh the sum of squares of deviations of each class by calculated diameter, for representing data difference degree in class, Numerical value is less to represent that difference is less.

3. the signal control time dividing method based on orderly cluster according to claim 2 it is characterised in that：Step c2 Specifically：

C21, calculating cluster sum are error function during k, and note is specifically categorized as：

G₁={ i₁,i₁+1,…,i₂-1}

G₂={ i₂,i₂+1,…,i₃-1}

…

G_k={ i_k,i_k+1,…,n}

Wherein branch is 1=i₁<i₂<…<i_k<N=i_k+1- 1 (i.e. n+1=i_k+1)；

Its error function expresses formula：

$L\&lsqb;b(n,k)\&rsqb;=\underset{t=1}{\overset{k}{\&Sigma;}}D(i_t,i_{t+1}-1)---(1-d)$

In formula：K is cluster numbers；B (n, k) is a certain sorting technique that n data is divided into k class；L [b (n, k)] is n data It is divided into the error function of k class；T numbers for time series；D(i_t,i_t+1- 1) it is i-th_tWith i_t+1The distance value of -1 data, calculates by mistake Difference function, for the classification of optimal recurrence.

4. the signal control time dividing method based on orderly cluster according to claim 3 it is characterised in that：Step c3 Specifically：

C31, calculation error function recursion value, and record the position j of optimal partition point；

$\left\{\begin{array}{c}L\&lsqb;p(n,2)\&rsqb;=\underset{2\&le;j\&le;n}{\min }\{D(1,j-1)+D(j,n)\}\\ L\&lsqb;p(n,k)\&rsqb;=\underset{k\&le;j\&le;n}{\min }\{L\&lsqb;p(j-1,k-1)\&rsqb;+D(j,n)\}\end{array}\right.---(1-e)$

In formula：P (n, k) is that n data is divided into k class to make error function reach the sorting technique of minimum；L [p (n, 2)] is n area The data on flows of section is divided into the error function of 2 classes；L [p (n, k)] is that the data on flows of n section is divided into the error function of k class；D (j, n) is the diameter between jth and n data；Work as n, when k is fixing, L [p (n, k)] is less, represent the deviation square of all classes And minimum, classification is rationally；Therefore, calculating n sample is divided into the error function of k class to reach minimum, then need to calculate j-1 sample The error function being divided into k-1 class reaches minimum, by that analogy.

5. the signal control time dividing method based on orderly cluster according to claim 4 it is characterised in that：Step c4 Specifically：

C41, make the tendency chart that L [p (n, k)] changes with k；

C42, searched out k value corresponding to maximum flex point by tendency chart as preferable clustering number k_op；

1. calculate the rate of change of error corresponding to any two adjacent k value：

$tan\left(t\right)=\frac{L\&lsqb;p(n,i)\&rsqb;-L\&lsqb;p(n,j)\&rsqb;}{i-j}---(1-f)$

2. the slope difference before and after calculating, obtains front difference diff₁, rear difference diff₂Value:

$\left\{\begin{array}{c}{\mathrm{diff}}_1\left(t\right)=|tan\left(t\right)-tan(t+1)|\\ {\mathrm{diff}}_2\left(t\right)=|tan\left(t\right)-tan(t-1)|\end{array}\right.---(1-g)$

C43, utilize front difference diff₁With rear difference diff₂Ratio be worth to decision diagram maximum flex point, that is,：

$diff=\underset{t=k}{max}\frac{{\mathrm{diff}}_1\left(t\right)}{{\mathrm{diff}}_2\left(t\right)}---(1-h)$

K value when c44, diff value is maximum is optimal cluster number k_op.

6. the signal control time dividing method based on orderly cluster according to claim 5 it is characterised in that：Step c5 Specifically：

N data sample is divided into k_opThe optimum segmentation of class is divided into k it should set up in j-1 sample_opThe base of the optimum segmentation of -1 class On plinth, by that analogy；Using the error recurrence function calculating, the position j with record optimal partition point, respectively obtain gather for 2, 3……k_opSegmentation situation during class.