CN100533475C - Traffic signal off-line time distribution optimizing method basedon particle group operation method - Google Patents

Abstract

A offline match time optimum method of traffic signal based on particle group algorithm, using the real number encoded mode denote the green light time of every signal crossings in area, in the solving process of particle group algorithm, different match time project is correspond to different particle, valuating the sufficiency of each particle according to the general delay time simulated by microcosmic traffic, at last ,the optimum match time project is the particle which has the minimal delay time. The invention provides a signal match time intelligent optimum method compared with the existing offline match time technology, its code form is simple, the parameter which needed to adjust is less, the counting rate is quicker too.

Description

Traffic signal off-line timing optimization method based on particle cluster algorithm

Technical field

The present invention relates to the particle group optimizing technology and the flow process of regional traffic signal off-line timing scheme.It is applicable to the timing and the optimization of urban area signalized intersections.

Background technology

Urban traffic signal control mainly refers to the traffic lights control of each intersection of urban road, traffic flow is regulated, is warned and induce to reach and improve transporting safely of people and goods by the timing of signal lamp, improves efficiency of operation.The traffic signals control mode can be divided into 2 big classes: off-line timing and online timing.

The off-line timing belongs to the first generation technology of signal controlling, it is to calculate the signal period at crossing and the length of all directions green time according to traffic flow historical data in a day, signal lamp switches automatically according to the timing scheme of input, or adopt the multi-period timing of TOD (Time Of Day), variation according to flow, being divided into several periods in one day, when signal calculated cycle and each phase place are green respectively in each period.Off-line control is simple, reliable, economical, but can not in time respond the random variation of traffic flow.

Online timing belongs to the second generation technology of signal timing dial, is a kind of real-time technique, by being laid on the underground detecting device in each crossing, gather traffic data in real time, utilize technology such as automatic control, System Discrimination, electronics, online timing parameter is optimized, signal lamp is controlled by controller.Online timing can in time respond the random variation of traffic flow, but control complexity, poor reliability, difficult in maintenance, expense is high.

The online in real time control technology is the research focus of current traffic control.Yet because the imperfection of the complicacy of traffic system, uncertainty, randomness and real-time control technology, the cost performance that timing is in real time at present optimized is not high; In addition, in the process of control in real time, a large amount of traffic flow historical datas of crossing detecting device accumulation, owing to requirement and the technology limitation of timing in real time to the response time, most of historical data all is not utilized.

At present, be accompanied by the continuous development of computer technologies such as data mining, intelligence computation, machine learning and perfect, the optimization effect of utilizing traffic flow historical data and various infotech to improve the off-line timing causes scholar's attention again.Realize based on the off-line timing optimization method of data-driven and heredity calculating is all existing, and confirmed the validity of these technology by test.

Yet genetic algorithm generally adopts scale-of-two to encode, and encoding-decoding process is complicated, needs the parameter adjusted more (comprise duplicate, intersect, the setting of operator such as variation), and iteration time is long and optimization timing scheme that obtain is accurate inadequately.

Summary of the invention

Technology of the present invention is dealt with problems: overcome the deficiencies in the prior art, provide optimize timing accurately, optimizing process is simpler, needs the parameter adjusted still less, the traffic signal off-line timing optimization method based on particle cluster algorithm that timing efficient improves.

Technical solution of the present invention: based on the traffic signal off-line timing optimization method of particle cluster algorithm, its characteristics are mainly to may further comprise the steps:

(1) obtains the correlation parameter such as road, crossing, vehicle, driver of survey region, utilize traffic modeling software (as: TransCAD, TSIS/ITRAF etc.) to carry out the road network modeling.

(2) parameter of setting particle cluster algorithm, described parameter comprises particle number, particle length (be the particle dimension, equal the number of phases sum of each signalized intersections), maximal rate V of each dimension of particle _Max, particle is respectively tieed up the scope [X of position _Min, X _Max] (being the scope of each phase place when green), self-adaptation is adjusted the scope [W of inertia weight _Min, W _Max], study factor c ₁And c ₂, maximum iteration time Maxlter, the fitness minimum change CutOff of adjacent twice iteration.

(3) utilize random number functions to produce the initial value x that each particle is respectively tieed up position and speed _Id ⁽¹⁾And v _Id ⁽¹⁾, require to satisfy condition respectively: X _Min＜x _Id ⁽¹⁾＜X _Max, v _Id ⁽¹⁾＜V _MaxMake iterations t=1, iteration begins.

(4) for each particle:

The signal time distributing conception that it is corresponding is inserted traffic microcosmic Simulation software (as: VisSIM, TSIS/CorSIM, Synchro/SimTraffic etc.), and the road network modeling file that utilizes step (1) to produce carries out traffic simulation by simulation softward.

From the performance index of simulation output, obtain the total delay time d of the t time iteration ^(t), with d ^(t)Fitness evaluation value as this particle.

Compare total delay time and the minimum in history total delay time of this particle, if d ^(t)＜pBest, and particle respectively ties up the position and be in the suitable solution space, i.e. X _Min＜x _Id ^(t)＜X _Max, pBest=d then ^(t)

(5) from all particles, select the particle of the particle of individual extreme value pBest minimum as global extremum gBest correspondence.

(6), calculate the inertia weight w of the t time iteration according to the relation of the linear decrease between following inertia weight scope and the maximum iteration time ^(t)Value.

$w^{\left(t\right)}=\frac{W_{\max }-W_{\min }}{\mathrm{MaxIter}}\·(\mathrm{MaxIter}-t)+W_{\min }$

(7) upgrade each particle according to following formula and respectively tie up position and speed, obtain new position x _Id ^(t+1)And speed v _Id ^(t+1)

$v_{\mathrm{id}}^{(t+1)}=w^{\left(t\right)}\&CenterDot;v_{\mathrm{id}}^{\left(t\right)}+{\mathrm{<figure-callout\; id="1"\; label="c"\; filenames="C200610089781E00101.png"\; state="\{\{state\}\}">c</figure-callout>}}_1\&CenterDot;\mathrm{rand}\left(\mathrm{}\right)\&CenterDot;({\mathrm{pBest}}_{\mathrm{id}}-x_{\mathrm{id}}^{\left(t\right)})+c_2\&CenterDot;\mathrm{Rand}\left(\mathrm{}\right)\&CenterDot;({\mathrm{gBest}}_d-x_{\mathrm{id}}^{\left(t\right)})$

$x_{\mathrm{id}}^{(t+1)}=x_{\mathrm{id}}^{\left(t\right)}+v_{\mathrm{id}}^{\left(t\right)}$

In the following formula, rand () and Rand () are the random numbers of two values between [0,1].

(8) if t=Maxlter, perhaps, the difference of the global extremum of adjacent twice iteration is less than fitness minimum change CutOff, then the particle result of global extremum gBest correspondence promptly is optimum each signalized intersections timing scheme of zone; Otherwise t=t+1 returns step (4).

(9) call traffic Visual Dynamic simulation softward (as: TSIS/TRAFVU, SimTraffic etc.), dynamic demonstration signal optimizing timing result comprises transport condition, formation situation, chocking-up degree, signal lamp variation of vehicle etc.

Principle of the present invention: particle cluster algorithm is to equal a kind of evolutionary computing that nineteen ninety-five proposes by Kennedy and Eberhart.Its core concept is the simulation to biological social behavior.Its initial imagination is the process of simulation flock of birds predation, and in research process, the optimization that is applied to variety of issue has obtained good result.Suppose bevy the predation, found food for wherein one, then some other bird can be followed this bird and be flown to the food place, and makes some can remove to seek better food source.In the whole process of predation, bird can utilize the experience of self and the information of colony to seek food.Particle cluster algorithm takes a hint from this behavior of flock of birds, and uses it for finding the solution of optimization problem.In particle cluster algorithm, each problem separate a bird that all is counted as in the search volume, be called " particle ".The state quality of particle is used by the fitness value of optimised problem decision and is represented.Direction and distance that each particle also has a speed decision particle to circle in the air.Particle is followed current optimal particle and is searched in solution space.When finding the solution a problem, adopt certain coding method to be encoded into particle this problem, carry out interative computation according to the mechanism of particle cluster algorithm then.

For signalized intersections, each phase time mainly comprises green time, yellow time, complete red time, wherein amber light and complete red time are generally definite value, as given 3 seconds and 1 second respectively, are primarily aimed at the green time of each phase place based on the signal timing dial optimization of particle cluster algorithm.

If survey region has n signalized intersections C _i, i=1,2 ..., n, then each particle (Particle) can be encoded in the particle cluster algorithm:

Particle=＜C ₁, C ₂..., C _n, and $C_i=<g_i^1,g_i^2,\&CenterDot;\&CenterDot;\&CenterDot;,g_i^{m_i}>$

In the following formula, m _iBe the number of phases of crossing i,

The green time of representing i crossing j phase, the length of each particle is

In optimizing process, the evaluation utilization of each particle is carried out certain performance index that microcosmic traffic simulation obtains as adaptive value to corresponding timing scheme, as: overall travel time, total delay time, queue length etc.Here the total delay time that obtains with microcosmic Simulation is as the evaluation of estimate of each particle.

In the standard particle colony optimization algorithm, because the enchancement factor that algorithm adds, each particle position and speed that optimization is obtained have the situation that negative or big numerical value occur, and the green time of phase place generally is set in [1,120] in Miao the scope, therefore be necessary in optimizing process, to consider the constraint condition of position and speed value.Utilize particle cluster algorithm to solve the nonlinear optimal problem of belt restraining, it is generally acknowledged to have four class methods to handle constraint: based on keeping suitable method of separating, based on the method for penalty function, suit to separate and non-suitable method of separating based on distinguishing, and other mixed method.Wherein, keeping suitable method of separating is a kind of method of the most direct solution constraint condition, utilize this method, each particle can be searched for (upgrading position and speed) in whole solution space, but keeping aspect historical self cognition (individual extreme value) and social recognition (global extremum) ability, only following the trail of those and be in particle in the feasible solution space; Simultaneously, for quickening the optimization process, when initialization, all particles all adopt the random initial solution of separating within the scope suitable.

In conjunction with the particle cluster algorithm of traffic network modeling, belt restraining condition with utilize the microcosmic traffic simulation to obtain the evaluation result of fitness, the present invention has provided the traffic signals timing method based on particle group optimizing.

The present invention compares with existing traffic signal off-line timing technology: the signal timing dial method coding form, the optimizing process that the present invention is based on population are simpler, the parameter that need adjust still less, speed of convergence is faster, and therefore, it is optimized, and timing is accurate, efficient improves more obvious.

Description of drawings

Fig. 1 is a process flow diagram of the present invention.

Fig. 2 is Beijing's signalized intersections road synoptic diagram.

Fig. 3 adopts the performance of signal optimizing timing method of the present invention and existing timing method for the crossing of Fig. 2 and compares.

Embodiment

Timing scheme optimization with Beijing's signalized intersections is that example specifies.Fig. 2 is the road synoptic diagram of this crossing.When utilizing the method for the invention that the timing scheme optimization is carried out in this crossing, each related parameter values of the particle cluster algorithm of appointment sees Table 1.

Particle cluster algorithm relative parameters setting during table 1 pair signalized intersections timing scheme optimization

Parameter	Value	Parameter	Value
				Particle number	50	The inertia weight scope	[0.9，0.4]
Particle length	4	The study factor	c 1＝c 2＝2
				Maximal rate	110	Maximum iteration time	1000
Position range	[0，120]	Minimum change	0.01

Keeping road parameters, the traffic direction that crossing phase place number and each phase place allow, under the constant situation of vehicle and driver's correlation parameter, four entranceway flows adopt 13:00～13:15 period toroid winding detecting device to detect average discharge, minimum flow (average discharge-3 times standard deviation), the maximum flow (average discharge+3 times standard deviation) of flow in many days respectively, utilize particle cluster algorithm, obtained adopting the period timing prioritization scheme result (table 2) of three kinds of flows, the existing timing scheme of the average discharge in the table is obtained by relevant traffic control department.

This signalized intersections of table 2 13:00～13:15 period particle group optimizing timing scheme result (unit: second)

The existing timing scheme of above-mentioned average discharge is imported microcosmic Simulation software respectively with optimization timing scheme carry out the traffic simulation operation, can obtain the every performance index in crossing of two kinds of schemes when four entrancewaies input average discharges.Table 3 has provided in all vehicle accumulative total total delay times of these two kinds of timing schemes of 13:00～13:15 period, all vehicle accumulative totals and has always expended three index results such as oil mass, the average every kilometer CO of all vehicles (carbon monoxide) discharge capacity, Fig. 3 is that the performance index result with existing timing scheme is a radix, each index result of particle group optimizing scheme reduces ratio, wherein the total delay time of prioritization scheme has reduced 12.7% with respect to existing scheme, always expend oil mass and reduced 1.7%, every kilometer CO discharge capacity has reduced 2.3%.

As can be seen, when adopting particle cluster algorithm to optimize the timing scheme, the time delay of whole crossing, oil plant expend with the exhaust emissions situation all improvement to some extent, wherein being used to determine the principal element of signalized intersections service level---the delay time at stop has reduced 12.7% with respect to existing timing scheme, illustrates to adopt prioritization scheme can improve this Capacity Analysis for Signalized Intersection effectively.

The performance index result of existing scheme of table 3 this intersection signal timing under average discharge and prioritization scheme dry run

	Total delay time (unit: minute)	Total oil plant expends (unit: rise)	Average CO discharges (unit: gram/kilometer)
				Average discharge has scheme now	874.8	254.07	94.43
The average discharge prioritization scheme	763.8	249.72	92.12

Claims (3)

1, a kind of traffic signal off-line timing optimization method based on particle cluster algorithm is characterized in that mainly may further comprise the steps:

(1) establishes survey region n signalized intersections C arranged _i, i=1,2 ..., n, then each particle Particle is encoded in the population:

Particle=＜C ₁, C ₂..., C _n, and $C_i=<g_i^1,g_i^2,\&CenterDot;\&CenterDot;\&CenterDot;,g_i^{m_i}>$

In the following formula, m _iBe the number of phases of crossing i,

The green time of representing i crossing j phase, the length of each particle is

Set the parameter of particle cluster algorithm simultaneously, described parameter comprises particle number, particle length, and promptly the particle dimension equals the number of phases sum of each signalized intersections, the maximal rate V of each dimension of particle _Max, particle respectively ties up the scope [X of position _Min, X _Max], the scope when promptly each phase place is green, self-adaptation are adjusted the scope [W of inertia weight _Min, W _Max], study factor c ₁And c ₂, maximum iteration time Maxlter and adjacent twice iteration fitness minimum change CutOff;

(2) utilize random number functions to produce the initial value x that each particle is respectively tieed up position and speed _Id ⁽¹⁾And v _Id ⁽¹⁾, making iterations t=1, particle group optimizing begins;

(3) for each particle: the signal time distributing conception that it is corresponding carries out traffic simulation, obtains the total delay time d of the t time iteration from the performance index of simulation output ^(t), with d ^(t)Fitness evaluation value as this particle; Compare total delay time and the minimum in history total delay time of this particle, if d ^(t)＜pBest, and particle respectively ties up the position and be in the suitable solution space, i.e. X _Min＜x _Id ^(t)＜X _Max, pBest=d then ^(t), pBest is individual extreme value;

(4) from all particles, select the particle of the particle of individual extreme value pBest minimum as global extremum gBest correspondence;

(5), calculate the inertia weight w of the t time iteration according to the relation of the linear decrease between following inertia weight scope and the maximum iteration time ^(t)Value,

$w^{\left(t\right)}=\frac{W_{\max }-W_{\min }}{\mathrm{MaxItet}}\&CenterDot;(\mathrm{MaxIter}-t)+W_{\min }$

(6) upgrade each particle according to following formula and respectively tie up position and speed, obtain new position x _Id ^(t+1)And speed v _Id ^(t+1),

$v_{\mathrm{id}}^{(t+1)}=w^{\left(t\right)}\&CenterDot;v_{\mathrm{id}}^{\left(t\right)}+c_1\&CenterDot;\mathrm{rand}\left(\right)\&CenterDot;(p{\mathrm{Best}}_{\mathrm{id}}-x_{\mathrm{id}}^{\left(t\right)})+c_2\&CenterDot;\mathrm{Rand}\left(\right)\&CenterDot;({\mathrm{gBest}}_d-x_{\mathrm{id}}^{\left(t\right)})$

$x_{\mathrm{id}}^{(t+1)}=x_{\mathrm{id}}^{\left(t\right)}+v_{\mathrm{id}}^{\left(t\right)}$

In the following formula, rand () and Rand () are the random numbers of two values between [0,1];

(7) if t=Max/ter, perhaps, the difference of the global extremum of adjacent twice iteration is less than fitness minimum change CutOff, then the particle result of global extremum gBest correspondence promptly is optimum each signalized intersections timing scheme of zone; Otherwise t=t+1 returns step (4).

2, the traffic signal off-line timing optimization method based on particle cluster algorithm according to claim 1 is characterized in that: described initial value x _Id ⁽¹⁾And v _Id ⁽¹⁾Satisfy condition respectively: X _Min＜x _Id ⁽¹⁾＜X _Max, v _Id ⁽¹⁾＜V _Max

3, the traffic signal off-line timing optimization method based on particle cluster algorithm according to claim 1, it is characterized in that: described traffic simulation is road, crossing, vehicle, the driver's correlation parameter that obtains survey region, utilizes the traffic modeling software to carry out the road network modeling.

Images