• Nie Znaleziono Wyników

Optimal control of an isolated two-phase intersection

N/A
N/A
Protected

Academic year: 2022

Share "Optimal control of an isolated two-phase intersection"

Copied!
8
0
0

Pełen tekst

(1)

III IN T E R N A T IO N A L C O N F E R E N C E

T R A N S P O R T S Y S T E M S T E L E M A T I C S T S T '0 3

Z E S Z Y T Y N A U K O W E PO L IT E C H N IK I Ś L Ą S K IE J 2003

T R A N S P O R T z.5 1 , n r kol. 1608

traffic, o p tim a I contro!, B a n g -b a n g c o n tro !

Y o u ssef A H A D A R 1 R achid B O U Y E K H F A bdellah EL M O U D N I

O P T I M A L C O N T R O L O F AN IS O L A T E D T W O -P H A S E I N T E R S E C T IO N

In th is p a p e r w e c o n sid e r an iso lated tw o -p h a se in te rse ctio n streets w ith c o n tro lla b le tra ffic lights on each c o rn er. Iso lated im p lie s th a t w e w ill d e v e lo p sig n al co n tro l at th e in te rse ctio n w ith o u t co n sid e rin g a d ja c e n t in te rse ctio n s. W e c o n stru c t a m odel th a t d e sc rib e s th e ev o lu tio n o f th e q u e u e le n g th s in each lane as a fu n ctio n o f cy cle. W e d iscu ss h o w op tim al co n tro l can m in im ize q u eu es in the o v e rsatu ratio n c o n d itio n s. A n a lg o rith m w ill be p ro p o se d to illu strate the resu lts o b tain ed .

O P T Y M A L N E S T E R O W A N IE

D W U F A Z O W Y M I IZ O L O W A N Y M I S K R Z Y Ż O W A N IA M I

W a rty k u le ro z w a ż a s ię w y iz o lo w a n e d w u fa z o w e sk rz y żo w a n ia ulic w y p o sa ż o n e w sy g n aliza c ję św ietln ą. W y iz o lo w an ie im p lik u je , iż b ę d zie o p ra c o w an e stero w an ie sy g n ała m i n a s k rz y żo w a n iu bez u w zg lęd n ian ia s ą sia d u ją c y c h sk rz y żo w a ń . Z b u d o w a n o m o d e l, k tó ry o p isu je zm ia n ę d łu g o śc i k olejki na k a żd y m p a sie ja k o fu n k c ję cy k lu . R o zw a ż an e je s t o p ty m a ln e stero w an ie , k tó re m o że m in im a liz o w a ć k o le jk ę w w a ru n k a c h p rz e sy c e n ia . Z ap ro p o n o w a n o alg o ry tm do zilu stro w a n ia u z y sk a n y c h w y n ik ó w .

1. IN T R O D U C T IO N

A s the n u m b e r o f veh icles an d the n eed fo r transportation grow , cities around the w o rld face serious ro ad traffic co n g estio n p roblem s. C osts in clu d e lost w ork and leisure tim e, increased fuel co n su m p tio n , air p o llu tio n , health p ro b le m s... In general th ere e x ist different m ethods to tack le th e traffic congestion problem . T he m o st effectiv e m easures in the battle against traffic co n g estio n seem to be a selective construction o f n ew roads and a b etter control o f traffic th ro u g h traffic m an ag em en t. T raffic lig h t control can be used to au g m en t the flow o f traffic in urban en v iro n m en ts b y p ro v id in g a sm ooth circulation o f the traffic o r to regulate the access to hig h w ay s o r m ain roads.

T raffic light co n tro ls have been the subject o f several investigations. In particular, [1]

has show n h ow an optim al traffic light sw itch in g schem e fo r an in tersectio n o f tw o streets can

1 L ab o rato ire S e T -U n iv ersité d e T e c h n o lo g ie B e lfo rt-M o n tb é lia rd - U T B M 9 0 0 1 0 , B elfo rt c ed e x , F ran ce y o u s se f.a h a d a r@ e d u .u n iv -fc o m te .fr.

(2)

14 Y o u sse f A H A D A R , R achid B O U Y E K H F , A b d ellah EL M O U D N I be determ in ed . In g en eral th is leads to a m in im izatio n problem o v er th e so lu tio n set o f an ex ten d ed lin ear co m p le m e n ta rity p ro b le m (E L C P ). H ow ever, it is w ell k now n th a t th e E LC P is N P -h ard pro b lem . H en s, as c laim ed b y [1] this approach is n o t feasib le i f the n u m b e r o f sw itching cy cle s is large. [2] c o n stru cte d m o d els for o versaturation control. B u t th e ir m odels are all co n tin u o u s types an d do n o t address th e p roblem o f op tim izin g cycle length. T he only w o rk th a t m erits o u r atten tio n is [3], In th at paper the authors h ave d e v elo p e d a discrete dynam ic o p tim izatio n m o d els in term o f delay. T he optim al cycle len g th a n d the optim al assigned green tim e are d ete rm in e d fo r th e case o f tw o -p h ase control.

In this p ap er, firstly , b a se d on the w o rk o f [3]-[4]-[5] in w hich the au th o rs calcu late the d elay d u rin g a p e rio d o f o v e rsa tu ra te d co n d itio n s in a cro ssro ad s, w e ad o p t a m odel that describes the ev o lu tio n o f th e q ueue lengths in discrete tim e acco rd in g to th e cycles. This m odel p ro v id es an im p ro v e d level o f control b y using b etter m o d ellin g o f tra ffic variables.

Secondly, th e p ro b lem o f fin d in g th e set o f ad m issib le control to m in im ize the total length q ueue is co n sid ered in th e en tire o v ersatu rated period. In o rd er to ap p ly th e optim al control strateg y w ith re s p e c t to real c o m p o rtm en t o f th e system , an algorithm w ill be p roposed

2. D E F IN IT IO N O F T H E SY STE M

W e c o n sid e r an iso la te d tw o -p h ase in tersectio n as show n in F ig .l. T h e re are tw o lanes and on each c o m e r o f the in tersec tio n th ere is a traffic light. H ence, there are tw o traffic flow s o f veh icles to be serv ed in th e intersection.

Fig. 1. F o u r-le g in te rse ctio n w ith tw o -p h a se sig n al co n tro l

(3)

Optim al control o f an iso lated tw o-phase intersection 15 T o be able to p resen t the m odel, a certain n u m b er o f d efinitions are necessary :

Ef f e c t i v e g r e e n t i me

A s a v eh icle ap p ro ach es an intersection displaying a red signal the d riv er decelerates and stops eith er a t the sto p line o r at th e end o f a queue Fig.2. W hen the signal tu rn s green the driver accelerates until the v eh icle reach es its desired o r m ax im u m p o ssib le speed. It is usually [6], [4] assu m ed th a t after startup lo st tim e the saturation flow rate rem ain s constant until the b e g in n in g o f th e yellow change interval. T he effective green tim e is:

g c = g + y - w = g + y - ( W i + w 2) (1)

w here th e lo st tim e W is the sum o f startup lost tim e W { and clearan ce lo st tim e W 2.

W\ Effective green, & ÏÏ2

H i ■ i

Red Green

F ig.2. E ffectiv e g reen tim e

Yellow Red

R a t e o f s at urati on

R ate o f saturation s is d efin ed as being the m axim um n u m b er o f veh icles b ein g able to use the c o rrid o r w ith o u t in terru p tio n durin g the effective tim e o f the green light g c .

Ov e r s a t u r a t e d condi t i ons

If arrival rate ex ceed s cap acity , the intersection is oversaturated. A p h ase is saturated when a v eh icle at least is co n strain e d to aw ait m o re than one cycle to cro ss the crossroads.

The crossroads is satu rate d w h en at least one o f its phases is saturated. Form ally, the n u m b er o f arrivals A ( k ) du rin g cy cle k e N is d efined b y the equation:

A{ k) =A[ kc] ~A[ ( k- \ ) c\ (2)

A[kc] is defined as the n u m b e r o f th e arrivals at the end o f the cycle. T he dep artu re rate is s during the effectiv e tim e o f the g reen lig h t g e , so that:

D(t + c) - D(t) = s ■ g e (3)

w here D { t ) is the n u m b e r o f dep artu res at the instant t . Fig.3 displays a case w here dem and flow rate in stan tan eo u sly increases above the cap acity at th e b e g in n in g o f a cycle. T he capacity curve C (7 )is n o t the saw -to o th ed departure curve D ( t ) , so th at the area betw een A ( t ) and C ( t ) curves is th e overflow delay [7],

(4)

16 Y o u s s e f A H A D A R , R ach id B O U Y E K H F , A b d ellah EL M O UDN1

F ig .3 . M o d el fo r o v e rflo w d e la y

Let D j( c ) the n u m b e r o f th e dep artu res durin g the cycle k , then th e co n d itio n o f o versaturation is d efin ed by:

A(k)> D k(c) (4)

3. M O D E L IN G

C o n tin u o u s type m o d els are lim ited in th at the sw itch -o v er p o in t d oes n o t n ecessarily o ccu r at th e en d o f a cy cle, n e ith e r does th e term in atio n o f the o v ersatu rated p erio d o ccu r only at the end o f th e final cycle. O n the o th er hand, the sw itch -o v er points d eterm in ed b y a discrete m odel o c c u r e x a c tly at the term in atio n o f a cycle. D iscrete o p eratio n p ro v id es a sm ooth, reg u lar, an d o rd e re d tra n sfe r o f control. C alcu latin g queue is m o re reliab le. Fig.4 illu strates th e situ atio n o f q u eu e d u rin g o v ersatu ratio n and show s th e d u ality b etw een the tw o phases. T o k eep th is d u ality , it is n e cessary th a t the queue rep resen ted w hen th e effective green tim e term in ates at a certain cy cle state k .

X l(k + 1 )

X l( k )

k é m e cycle

F ig .4 . Q u e u e o f a fo u r-le g in te rse ctio n w ith tw o -p h a se c o n tro l

(5)

O ptim al control o f an iso la te d tw o -p h ase intersection 17

Let X2(k) be a q u eu e len g th o f approach 2 w hen the effective green tim e g^(k) term in ates at a certain cycle state k . A cco rd in g to Fig 4, the relation o f the q ueue len g th s betw een state

k and k +1 can b e re p re se n te d b y th e follo w in g equations:

X 2 (k+]) = X 2 (k) + A2 ( k ) - D 2 (k) (5)

w here A2(k) = q r c , rep resen ts the n u m b e r o f arrivals durin g cycle k and D 2(k) = S 2- g c2(k) the n u m b er o f d ep artu res d u rin g cycle k . T he equation (5) becom es:

X 2( k+l ) = X 2(k) +Q2- c - S 2-ge2(k)

(

6

)

Sim ilarly, let x , ( k ) be a q u eu e length o f approach 1 w hen the effectiv e green tim e

= c-g ,(/c) term in ates at a certain cycle state k . T hen

X ^ k + \ ) = X l ( k) +A]( k ) - D i(k) (7 ) w h e r e :

> A](k) = qy g e2( k - l ) + qr ( c - g e2(k)) : represents the n u m b er o f the arrivals at the end o f c - g e2W -

> D\(k) = S v ( c - g e2(k) ) : rep resen ts the n u m b er o f the d epartures at the en d o f c - g ,(& ) ■ The equation (7) becom es:

X\{k + \) = X\{k) + q y g e2{ k - \ ) + q y { c - g c2{ k ) ) ~ S y { c - g e2{k)) (8) L et x (k ) be th e v e c to r d efin ed as Xt®=Q(f® X,(k)J, and U{k) be the v ecto r o f control o f the system d efin es b y U ( k ) = ( g, , ( k ) g r2( k - 1 ) ) '. T hen w e can g ath er the tw o equations (6) an d (8) in the fo llo w in g m atrix form :

X ( k + \ ) = A - X ( k ) + B-U(k) + C (9) W here

' x m

,A = '1 0“

, B=s\ - Q Q

, U(k) = ga lik) - c =

( Q - s J - c

X 2(k) 0 1 -¡2 0 J S e l t t- 0 . Q c

N o tice th a t eq u atio n (9) is th e lin ear form w ith resp ect to the state variab le and control.

Hens, it is easily am en a b le to m athem atical analysis. T herefore, w e can e fficien tly com pute optim al traffic light co n tro l, w hich is the m ean concern in th e follow ing section.

4. O P T IM A L L IG H T S C O N T R O L

C o n sid er the p ro b lem o f fin d in g a control U(k) w hich m in im isin g on o f th e follow ing possible criterions (10) d efin ed as:

N

► J, = ^ x ( k ) - th e sum o f th e queues during oversaturated period.

k=o

(6)

18 Y o u sse f A H A D A R , R achid B O U Y E K H F , A b d ellah E L M O U D N 1

/V

> j 2 = Y ^ \ l 2 X 2(k) ' q u ad ratic form o f the queues length.

k=o N

> y 3 = max ( £ X ¡ ( k ) ) '■ the q u eu e length o v er the w o rst queue.

1 * = o

su b ject to th e d y n am ic eq u atio n :

X ( k + \) = A X ( k ) + B U ( k ) + C (11)

and the in eq u ality c o n stra in ts o f th e form :

n = {(U(k),X(k)) e (R2 x jR2 / Umn < U(k) < t/ _ , X(k) > o} (12) w here N is th e te rm in a tiv e cy cle o f the o v ersatu rated p erio d an d Q is a given set o f constraints.

M in im izin g (1 0 ) su b jected to (9), b ased on the optim al control th eo ry [8], involves eq u iv alen tly to m in im iz in g the H am iltonian fo rm u la (13). T he H am iltonian fo rm u la is d efined as:

H ( k ) = 0(k) + Ar (k + 1 ) [ X ( k ) + B U ( k ) + C ] (13) w here A ( k ) is the co state c o rresp o n d in g to x (k)- N ow , the task is to find an adm issible sequence U ( k ) e Q such th a t (1 3 ) is m inim al subject to (12). A cco rd in g to th e optim al control th eo ry , i f an e x trem u m o f H exists, it m u st satisfy th e follow ing con d itio n s:

= + + (14)

' ’ d X ( k ) d X ( k )

— — -- = X ( k + l ) = X ( k ) + B U ( k ) + C ( 15) 5/t(A + l)

_ d H _ = o (16)

8 U (k)

w here (j){k)\s the function w hich appears in the criterions form s. E quation (13) clearly reveals th at the relatio n b etw een H and the control v ariab le U { k ) is linear. T h is leads to:

- ^ — = At {k + 1 )5 * 0 (12)

d U{ k )

H ens, eq u atio n (17) o b v io u sly in d icates th at on ly the single control v a riab le U { k ) can m i n im iz e / / . In signal co n tro l, the control variable is related to the effectiv e g reen tim e w hich sh ould be taken as the v a lu e b etw een p red eterm in ed u pper and lo w er lim its. T h is p ro p o sitio n is clea rly a b a n g -b a n g control w ith a su ccessiv e tw o -stag e o peration U mirl and (7ni„ :

U ( k ) = U min i f k T (k + \ ) B > 0 V ( k ) = U m ax i f A T (k + ] ) B < 0

(18)

(7)

O ptim al control o f an iso lated tw o -p h ase intersection 19

A t first glance, (18) in d icates th at the im plem entation o f th e control strateg y is note problem atic. H ow ever, th e d iffic u lty arises in the determ ination o f the initial value o f A ( k ) w hich obeys th e re c u rre n t equation (14). T his difficulty can be av o id ed b y tak in g a set o f certain initials v alu es o f A ( k ) th a t respects the m axim al ad m issib le qu eu e len g th w hich is defined b y th e o v ersatu ratio n conditions. W ith this in m ind, the follow ing algorithm is proposed in ord er to tak e this consideration.

Step 1: Let * in itiate A (0 ).

Step 2: W hen Ar (£+1)5>0 D o U (k)=U mii:.

W hen / ( k + \ ) B < 0 D o l_J(k)=Umt>.

Step 3: G ive ( X M, X„, ) T, calcu late X ( k ) .

Step 4: If X i(k) = 0 o r

X i( k ) > X i

back to S te p l. A nd em p lo y in g (14) to define a new value o f A ( k ) to keep X i( k ) > 0 and X l( k ) < X i

Step 5: C alcu late AT (A:) until k = N . End.

A cco rd in g to figure 4, step 4 indicates th at the cum ulative o u tp u t curves do n o t intersect the cu m u lativ e in p u t curves fo r an y o f the approaches. T his fact im plies th at no queue becom es n eg ativ e o r z ero b efo re the end o f the o v ersaturated p eriod. If a q ueue becom es negative w hile th e signal is green, the d esigned green tim e b ecom es in v alid due to the w aste o f control tim e. In Fig.5 w e have plotted the evolution o f queue lengths acco rd in g to the algorithm and the co n strain ts. C learly w e observe that X t(k) decreases ra p id ly w h ile XJ k ) increases slo w ly until 90 th cycle. O nce x , ( k ) becom es zero the control is ch an g ed according to step 4. A fter 90th cy cle x \ k ) increases until x l j m =24 w hich is th e m axim al value that we h ave p erm itted , o f c o u rs e x ,( k ) d ecreases in th is case. A ll th ese o b serv atio n s in d icate th a t the control strateg y forces the q ueue lengths to change according to co n strain ts an d m ak es a flexibility to get o u t the o v ersatu ratio n situation.

F ig.5. E v o lu tio n o f th e q u eu es

(8)

20 Y o u sse f A H A D A R , R achid B O U Y E K H F , A b d ellah EL M O U D N I

5. C O N C L U S IO N

W hen the traffic flow ex ceed s intersection capacity, th is situation cau se q u eu in g o f veh icles th a t can n o t b e e lim in a te d in one signal cycle. T o m in im ise th is c o n g estio n , firstly, w e ex p lain ed a m odel th a t d escrib es the evolution o f the qu eu e lengths at an in tersec tio n o f tw o tw o -w ay streets w ith co n tro llab le traffic lights on each co m er. S eco n d ly , w e h ave p ro ceed ed at the stag e o f th e o p tim isatio n b y d efin in g the appropriate control strategy. T h is resu lts in b an g -b an g like co n tro l, w h ic h is q u ite ap p ro p riate fo r o v ersatu ratio n co n tro l. F inally, an alg o rith m has b een p ro p o se d in o rd er to take into acco u n t th e constraints.

B IB L IO G R A P H Y

[1] S C H U T T E R , B .D ., M O O R , B .D : O p tim a l tra ffic light fo r a sin g le in te rse ctio n .

[2] M IC H A L O P O U L O S , P .G . a n d S T E P H A N O P O U L O S , G ., O v e rsa tu rate d sig n al sy ste m s w ith q u e u e len g th c o n strain ts - 1: S in g le in te rse ctio n . T ra n s p o rta tio n R ese arch 11, 4 1 3 -4 2 1 . (1 9 7 7 a).

[3 ] T a n g -H sie n C h a n g : O p tim a l sig n al tim in g fo r an o v e rsatu rated in te rse ctio n , D e p a rtm e n t o f T ra n s p o rta tio n S c ie n c e, T a m k a n g U n iv e rs ity (1 9 9 8 ).

[4] W E B S T E R , F .V ., 1958. T ra ffic sig n al settin g s, ro ad re search tech n ical p a p er, N o .3 9 , G re a t B rita in R oad R ese arch L a b o ra to ry , L o n d o n .

[5] M A Y , A .D ., J r., 1965. T ra ffic flo w th e o ry -th e tra ffic e n g in e er's c h allen g e . In: P ro c e e d in g s o f th e In stitu te T ra ffic E n g in ee rin g , pp. 2 9 0 ± 3 0 3 .

[6] C L A Y T O N , A .J .H ., 1941. R o ad tra ffic c alcu la tio n s. Jo u rn al o f th e In stitu tio n o f C iv il E n g in ee rs, vo l. 16, no. 7, p. 2 4 7 -2 8 4 . W ith d isc u ssio n and c o rre sp o n d e n c e in n o . 8, pp . 5 8 8 -5 9 4 .

[7 ] G A Z1S, D .C ., P O T T S , R .B ., 1965. T h e o v e rsatu rated in tersectio n . In: P ro c e e d in g s o f th e S eco n d In te rn a tio n a l S y m p o s iu m on th e T h e o ry o f R o ad T r a f c Flow . O rg a n iz atio n fo r E c o n o m ic C o o p e ra tio n and D e v elo p m en t, P a ris, pp. 221 ± 2 3 7 .

[8] S A G E , A n d re w P. an d W H IT E , C h e lse a C.,I1I. O p tim u m S y stem C o n tro l, S e c o n d e d itio n (1 9 7 7 ).

R eview er: P rof. R o m u a ld S zopa

Cytaty

Powiązane dokumenty

1. Let M be an m-dimensional normed complex vector space.. By the above lemma our condition can be carried over—in a classical manner—to the case of manifolds. It suffices to show

Nie można bowiem utrzymywać, że Bóg jest czymś, ponad co niczego miłosierniej- szego nie można pomyśleć, gdzie miłosierdzie polega na poświęceniu się Boga, i

When compared with the relevant past works on optimal control of an aircraft, the proposed controller does not depend on the aircraft dynamics via the design of an

We have introduced a new parameter which is useful, together with moments vector λ M solution of (2.1) and µ CW , or µ CCW , as in (1.14), to characterize attractivity for

We have illustrated the decomposition method by calculating the shortest sequence of operations required to implement the transformation from the standard basis to the Bell basis,

NMR imaging and relaxation time measurements were employed to monitor the mass transfer of TMOS from the oleic to the aqueous phase in a two-phase bulk system.. The

The study of the existence, the structure and properties of (approximate) solu- tions of optimal control problems defined on infinite intervals and on sufficiently large intervals

This paper presents an optimal control problem governed by a quasi- linear parabolic equation with additional constraints.. The optimal control problem is converted to an