Dynamic control and traffic performance in a freeway corridor: a simulation study

Trar, spn R,es -,4. Vol. 25A. No 5, pp 267-276. 1991 Printed i n Great Bream

0191-2607/91

$3 00 + .00

DYNAMIC CONTROL A N D TRAFFIC PERFORMANCE IN A FREEWAY CORRIDOR: A SIMULATION STUDY ROBERT A. REISS Dunn Engineering Associates, Westhampton Beach, NY 11978, U.S.A. NATHAN H. GARTNER Department of Civil Engineering, University of Lowell, Lowell, MA 01854, U.S.A. and STEPHEN L. COHEN Federal Highway Administration, U.S. Department of Transportation, Washington, D.C. 20590, U.S.A. Abstntet-Th~s paper describes simulation studies that were conducted to assess the performance of a freeway corridor control system. The system combines an advanced traffic management system with a motorist information system that provides route guidance to individual drivers. It has a hierarchical structure: The corridor level control acts in a supervisory capacity dynamically allocating traffic among alternative corridor facilities, including freeways, frontage roads, and signalized arterials. The local level control then selects control parameters for the individual facilities based on the predicted usage at the corridor level. A user specified performance function is optimized in the process. Both recurrent and nonrecurrent congestion scenarios were simulated using the SCOT model as a test bed. It is shown that, in most cases, significant benefits in performance can be obtained when the system operates as designed. INTRODUCTION

in light of the recent Intelligent Vehicle/Highway Systems initiatives (USDOT, 1989; F H W A , 1991). In this paper we report on simulation studies on dynamic traffic control in a freeway corridor that were conducted in conjunction with the development and design of the Integrated Motorist Information System (IMIS). IMIS is a freeway and arterial-street traffic management project on 128 miles o f heavily traveled highways in a 35-mile long, east-west corridor. It is located in the densely developed northwestern quadrant of Long Island, New York (Fig. 1). The system's operation is based on routing traffic past traffic jams and lane closures, via existing alternative routes that are not fully used. During incidents, route diversion information provides motorists with knowledge that a better alternative route exists past a congested highway section ahead. Alternate route traffic control, during diversions, based on real-time surveillance, reduces the instabilities of high density traffic flow on that route and improves overall system performance. The simulation studies were conducted to assess the benefits that will accrue from the implementation of IMIS (Reiss et al., 1981; Gartner and Reiss, 1987).

The growing traffic congestion in urban areas is drawing increasing attention from the public, government officials, and transportation professionals. Congestion on U.S. urban freeways, which carry nearly 30~0 o f all traffic in urban areas, is estimated to cause 1.2 billion vehicle-hours of delay, 1.3 billion gallons of wasted fuel, and 9 billion dollars in excess user costs per year. Since urban freeway travel increases at a rate of 1.9°/o per year, and no significant additional physical capacity is contemplated, the problem will continue to increase in severity (ITE, 1986; Lindley, 1987). One of the most effective remedial measures is the installation of a surveillance and control system for the congested segments o f a freeway. Rapid advances in electronics, communications, and information processing technologies make it possible to develop advanced traffic management systems that are coupled with advanced motorist information systems (OECD, 1987). The purpose of the traffic management systems is to help achieve full utilization of the highway network capacity and reduce trip times, congestion, and accidents. Such systems can influence the pattern of route choice by providing early traffic incident detection and manTRAFFIC CONTROL IN A FREEWAY CORRIDOR agement, then redistribute traffic among the facilities of a corridor or a network by utilizing the excess A corridor is a roadway system consisting o f a capacity in some parts of the network. The advanced few primary longitudinal roadways (freeways or mamotorist information systems can provide drivers jor arterials) carrying a major traffic movement with with information on congestion, navigation and lo- interconnecting roads which offer the motorist altercation, traffic conditions, and alternative routes. native paths to his destination. The IMIS corridor in There is an extensive interest in the development of Long Island, New York (Fig. 1) is an example o f such systems both in the U.S. and abroad, especially such a corridor. A corridor which can benefit from 267

268

R. A.

REISS, N. H. GARTNER,and S. L. COHEN

LONG ISLAND EXPflESSWAY

rAY

1 N

NOATHERN STATE PKWY

Fig. 1. The IMIS corridor in Long Island, New York.

a corridor traffic control system is o n e in which o n e or more routes can become congested even though the corridor as a whole has sufficient capacity to provide travellerswith a reasonable level of service. In such corridors, a traffic control system serves to rapidly detect congestion and implement control to minimize trafficdisruption. These controls can take the form of route diversion, ramp metering, and sig-

available to the algorithm user. Selectable by user option are: travel time, speed, throughput, delay, fuel consumption, and pollutant emissions or any linear combination of these. These functions can be computed from conventional sensor outputs. The corridor O - D matrix is synthesized from volume data collected by a real-time surveillancesystem. Flow optimization is accomplished by an iterativetrafficasnal timing. signment procedure based on the Frank-Wolfe In route diversion control, trafficis dynamically method (Gartner, 1977) and is illustratedin Fig. 3. allocated among the various corridor facilitiessuch The output is a set of optimal diversion fractions as freeways, freeway-frontage roads, and signalized which are to be implemented through the motorist arterials.This can be implemented by variable mes- information system during the upcoming control pesage signs, highway advisory radio, or other I V R G riod. Since we directlyminimize system performance (in-vehicle route guidance) systems which convey (the objective function represents a disutility),the route diversion information to motorists. Ramp meresulting assignment is system-optimal. Although tering uses trafficsignals on entrance ramps to con- one could argue that such an assignment does not trol the volume of trafficentering the freeway main- adequately reflect user behavior, there is evidence line.Metering rates are selectedto respond to present that there are only small differences in performance and predicted trafficdemands on the mainline. Sig- between user-optimal and system-optimal flow patnal timing control enables timing plans on corridor terns for large-scalenetworks (LeBlanc et aLE 1975). arterialsto be responsive to changes in demand due For a discussion of the issues involved in the applicato diversion as well as regular time-of-day changes. tion of the two types of assignment in the design of The route diversion control system has a hierar- route guidance systems see Gartner et al.,0980) and chical structureas shown in Fig. 2. The corridor level Boyce (I988). In any case, the system-optimal assignacts in a supervisory capacity, dynamically allocating ment used in the IMIS study represents a lower traffic optimally among the various corridor facili- bound on the improvement in traffic performance ties such as freeways, freeway-frontage roads, and that is attainable by the corridor control algorithm. signalized arterials. Then the local level optimizes The flow optimization procedure also includes a flow over the individual facilities,based on the predicted usage determined at the corridor level. The corridor level control algorithm dynamically SYSTEM OBJECTIVE assigns trafficto corridor routes in such a way as to optimize a selected performance criterion or objecCORR~OO~ I CONTROLDECISION tive function. The corridor-level control process is CO~RmOR CmVEnS~ONSJ LEVEL CONTROLLER I ~r performed at periodic intervals, typically, I0 to 20 minutes, or whenever measured corridor conditions TRAFFIC LOAO$ appear to warrant immediate reoptimization, such as L~:AL when a major incident is detected. The algorithm OnJE~rVE OB.ll~l~tIVE ~lF FUNCTI~S FUNC~I~ generates a diversion policy (consisting of optimal diversion fractions at each control node) which satisLEVEL LEVEL LEVEL fiesthe existingorigin-destinationdesires of all corriCONtROLLeR CONTROLLER CONTROLLER ..... dor users. The major components in corridor level I.OCA~. CONYROL CONTROL CONTROL control are: (i)selectionof an objective function, (ii) DECISIONS DECISIONI OECI~ONS specificationof the corridor O - D matrix, (iii)compuSIGNALiZeD MET{AEO METERED tation of an optimal assignment (effected through ARTER)AL FR((WAy FREEWAYW~TH FRONTAG(:ROAD the corresponding diversion fractions). A wide choice of system objective functions is Fig. 2. Hierarchy of corridor control system.

~

I

1....

.....

I......... 1

1....

I

1 ....

[

1

Dynamic control and traffic performance

i

I

I I I I I l N . 111 11111)11

ISICIII

1ol, LIAITICOST IMlh CALCULAIIOI

I'--"'

.... "-"

. lllllH

LIASI.CoIIr ff~ThS It01 ALL 0..l I ASIIIN ALL IlqmO VlLIIqll ?0 N i l VOU ) A l a ¢LOU P ~ V ~ n n

I~lllS

811.1.111I t l l l l

OF I t l l l A , IIIl$

]

(q,TN$

¢~n

t

,

I . IQ,lUIClq P101155

I IIITI~IlllII • t011 Tills 11111117I I I I

.)

conllnt Oto Iv~ Iltu e t ~

FI(I ) t ~ j l n

PATrtI~

* (l.g)!ll. I

NW CCIPIJT| CIlUlKI01111 1 ~ I I I l l FUCTIrI$

!

I_

IIIIUT[ TIiAP'PlC lqll01CllOll NIImL INI51 0111titS1111,IUIClICINS T0 I l l l l m l l l [ 0LI(L;II~ ,UI?I01q

I . ~

+

N i l TIUIVIL TIH[$

NO

I:l~li

It"

etta

C ~ T ! [NlllSllm011~ OlVtl~lOu PIAc'rlIIS (011PUTt N*IVA?[V($14 r MC COSTS ¢ tll$1-tosv

Mlm ~.CIA.ATll

process by allocating and scheduling computer resources to the various traffic control functions; (2) the corridor level control, which optimizes traffic flow "in the large" by allocating traffic among the various corridor facilities such as freeways and signalized arterials; and (3) the local level controls, which operate to optimize the use of individual facilities independently from one another, but based on the predicted demand as determined at the corridor level. Based on this three-tiered control structure, the control program has been configured into nine distinct modules, or functions, as follows:

llllllle I 1~ III!11{) I I I

IIN~k

:i ::.,.,..,,.-,,:, 41)

269

u

Fig. 3. Iterative flow optimization process.

traffic prediction model which calculates system performance as a function of the predicted traffic states on each link as well as the cost and travel time to be experienced in traversing each link. Traffic prediction on limited access routes is accomplished using a hydrodynamic flow model with provisions for queue generation and spillback onto upstream links. Signalized arterial prediction is done by superimposing changes in fiow as a result of diversions onto "normal" time-of-day traffic changes. The link costs associated with these flows are computed by retrieving previously run TRANSYT program evaluations for various flow levels (Robertson, 1969). After the corridor level control has dynamically assigned the flows, traffic control on individual freeways, freeway-frontage road pairs and signalized arterials must be optimized. To accomplish this, the local level algorithm selects ramp metering rates at freeway on-ramp controllers and controller timing parameters along arterials and frontage roads based on the flows predicted to take place after diversion. The ramp metering strategy is one which adjusts metering rates to prevent flow from exceeding capacity at any point along the freeway. Control along signalized arterials and frontage roads is performed by selecting one of a set of patterns previously generated by the TRANSYT signal optimization program. ROUTE DIVERSION: PROGRAM STRUCTURE

The functional structure of the route diversion algorithm consists of three control levels (I) the corridor executive, which supervises the overall control

I. Corridor Executive Control II. Corridor-Level Control, including: Demand Estimation Corridor Control Message Selection Performance Evaluation III. Local-Level Control, including: Ramp Metering Control Arterial Slgnahzatlon lncident Detection State and Parameter Estimatton The key interactions among these nine functions and between these functions and the corridor surveillance network are shown in Fig. 4. Of interest here are the major and minor control loops by which traffic surveillance information gathered by roadway sensors is fed back to the various control functions. Specific functions are described below (Reiss, 1981).

1. Demand estimatzon A key ingredient of the corridor-level control process is the estimation of the corridor users' timevarying origin-destination demand pattern. The Demand Estimation function calculates estimates of current and projected origin-destination volume flows between every feasible origin-destination node pair in the corridor network. The estimates are based on a combination of (1) historical demand data as obtained from O-D surveys, and (2) synthesized O-D data, generated from estimated on- and off-flows at system entrance and exit ramps and estimated link volumes. Synthesized data are needed when rerouting of existing O-D demands is indicated (due to congestion, incidents, etc.) by the flow optimization algorithm. The Exit Fraction Residual Model is used for the synthesis. This model distributes each entry ramp volume among the downstream exit ramps in accordance with the overall exit fraction observed there (Reiss et al., 1981). A flow diagram o f the Demand Estimation Algorithm is shown in Fig. 5. The future period over which predicted demand data are required is the "look-ahead" interval of the Corridor Control algorithm (typically, 10 to 20 minutes). 2. Corridor control function The Corridor Control algorithm provides outerloop control of traffic flow throughout the corridor network, through the establishment of traffic diver-

R. A. REIss, N. H. GARTNER,and S. L. COHEN

270

CONTROL J REAL-WORLD PROGRAM ENVIRONMENT J ION-LINEVERSION) OR

CORRIDOR EXECUTIVE FUNCTION

SCOT

CORRI DOR CONTROL

JESTIMATED

CORRIDOR. PERFORMANCE bEVEL CON'rROL EVALUATION IIUNCTIONI;

'ON .rEMAN°],

ME.AGE

SELECTION

i m

! m

I

PREDICTED TRAFFIC STATES

LOCALLEVEL CONTROL FUNCTIONS

i~

/

OIVERSION I IoFF'LINE vERsION) MESSAGES .

I

I I I CONTROLLER

I ~l ARTERIAL JPARAMETERS SIGNALIZATIONJ ,

cONTn°L

~ ~m~ ~1

I

I METERING RAMP 1 RATES METERING CONTROL J

TRAFFIC RESPONSE J TRAFFIC STATES

~

INCIDENT NOTIFICATION :TIONJ t A v~

ESTIMATED TRAFFIC STATES

,[

SENSOR I STATEAND I . OUTPUTS PARAMETER I ESTIMATION J"

[SENSORS 1

I

Fig. 4. Functional interactions of control program and simulation model.

sion objectives at the major potential diversion nodes in the system. A controlpolicy is established by specifying a set of desired diversion fractions for each link in the network.

3. Message selection The Message Selection function generates route guidance information to the motorists to attempt to shape the traffic flow into the optimized pattern determined by the Corridor Control function. The type of information that is generated depends on the communication means that are used for its transmission. Possible means include in-vehicular route guidance (IVRG) or external message transmission. Variablemessage signs were designated in the IMIS project.

I OBTAINHISTORICALDE]qANOS eHZST(tk) [ FOR THIS Tiff[ PERIOD

NO

.

:-

UTILIZE HISTORICAL DEHANOS

IrES

-,OR EACN~OR RO~E(

f

EXIT)

4. Performance evaluation

s

JI

SYNTHESIZE O-D D[HANDS eSYl~,,n ALONGTHIS ROIJTE

I-

]

UTILIZE HISTORICAL Drr../'tASDDATA FOR THIS ROUTE

l_

L

I

FOR EACHCONNECTORROUTE 6

SYNTHESIZEO-D OEHANDS tSYNm'n I FOR THIS CONNECTORROUTE

t

COMPUTEESTIHATEOO-D ~TRICES

e(t k)

l

4

J

I

FIg. 5. Schematic of demand estimation algorithm.

I

The Performance Evaluation function provides estimates of the current performance being achieved, both on a link-by-link basis and on a system-wide level. This function utilizes the current link state estimates to compute measures of performance for each link and for the system as a whole. Generally, these performance measures include the objective function which the Corridor Control function is attempting to optimize, as well as the other link and corridor measures which the control function evaluates on a predictive basis.

5. Ramp metering control The Ramp Metering Control function calculates metering rates for all metered ramps along a freeway section. The function may be executed independently for each of several different freeway sections in the network. The metering rates calculated by the function are based on the predicted demands at the on-

Dynamic control and traffic performance

ramp entry points and the downstream reserve capacities. In calculating the ramp r~tering rates, the function explicitly takes into account expected traffic demands resulting from traffic diversions instituted by the Corridor Control function. 6. Arterial signalization control This function establishes cycle length, offsets, and splits for a set of signal controllers along a section of arterial roadway. A TRANSYT-prepared control plan is selected to match most closely the predicted traffic pattern in each section during the upcoming control period. 7. Incident detection The Incident Detection function identifies freeway incidents (e.g. accidents or severe congestion) of a magnitude sufficient to disrupt traffic flow. It also estimates the severity of such incidents and calls the Corridor Control function to reoptimize the network. Incidents are detected by examining occupancy data for adjacent pairs of detector stations and utilizing the California incident detection algorithm to test for incidents. When an incident is identified, its approximate location is determined, together with the remaining (incident) capacity of the affected link for use by the Corridor Control function in reoptimizing the network traffic flow, taking into account the reduced capacity of the incident link. 8. State and parameter estimation The State and Parameter Estimation function provides up-to-date estimates of the current traffic states at each detector station and on each network link. This function processes real-time roadway sensor data to obtain estimates of the traffic states in the network required by the other control ~functions. Specifically, it uses smoothing techniques to calculate volumes, occupancies, and speeds for all detector stations in the network.

iv. Rd. W/B

Willis Rcl.

271

SIMULATION TESTING AND EVALUATION

To evaluate the performance of the freeway corridor control system and the associated motorist information system in IMIS, a series of stimulation tests were conducted using the SCOT model (USDOT, 1985). A set of scenarios was defined to be representative of traffic flow within the IMIS system. All scenarios take place on one of the four networks shown in Figs. 6-9 in simplified form and include the following conditions: typical weekday morning and evening rush hour peaks; typical recreation peak; off peak period; traffic incidents of varying duration and severity, with either one or two lanes blocked, at critical corridor locations during both peak and off peak periods and during recreation peaks; nonincident scenarios. The scenarios are summarized below (Note: LIE = Long Island Expressway, NSP = Northern State Parkway): Scenario 1 (Fig. 6)-Typical westbound weekday AM peak. Single lane blocked by incident on NSP westbound immediately upstream of Willis Avenue off-ramp for 20 minutes. Scenario 2 (Fig. 6)-Typical westbound weekday AM peak. Single lane blocked by incident on LIE westbound, between Willis Avenue offramp and on-ramp, for 15 minutes. Scenario 3 (Fig. 7)-Typical westbound weekday AM peak. Single lane blocked by incident on NSP westbound, approximately 1.6 miles (2.6 kin) west of Route 106/107 for 15 minutes. Scenario 4 (Fig. 7)-Typical westbound weekday AM peak. Single lane blocked by incident on LIE westbound, approximately 0.6 mile (1.0 kin) west of Route 106/107 for 25 minutes. Scenario 5 (Fig. 8 ) - Typical eastbound weekday PM peak. Single lane blocked by incident on NSP eastbound, approximately 0.6 mile (1.0 km) east of Route 135 for 15 minutes. Scenario 6 (Fig. 8 ) - Typical eastbound weekday PM

,v

RoslynRd.

-~Glen Cove Rd.

~.:~"Incident

Fsg. 6.1MIS corridor network for simulation scenarios I and 2.

272

R A. REISS, N. H. GARTNER,and S. L. COHEN

L. I. X-way

~1¢

,\ Nothern State Parkway

Rte 135 ~ ' Incident Fig. 7. IMIS corridor network for simulation scenarios 3 and 4.

peak. Single lane blocked by incident on LIE eastbound approximately 0.3 mile (0.5 km) west of Route 135 for 25 minutes. Scenario 7 (Fig. 9 ) - E a s t b o u n d Friday PM peak prior to summer holiday weekend; traffic volumes 5°7o over normal peak. No incident; length of run: 1:20 hours. Scenario 8 (Fig. 9 ) - E a s t b o u n d Friday PM peak prior to summer holiday weekend; traffic volumes 507oover normal peak. Single lane blocked by incident on LIE eastbound, approximately 3.0 miles (4.8 km) east of Route 110 for 20 minutes. Scenario 9 (Fig. 9)-Typical eastbound weekday midday traffic; traffic volumes 7807o of normal peak. Single lane blocked by incident on LIE eastbound, approximately 3.0 miles (4.8 km) east of Route 110 for 30 minutes. The SCOT simulation model requires a large amount of input data. The specific types of data which were compiled for the simulation study are:

(i) traffic volumes, (ii) monthly and daily traffic conversion factors, by location, (iii) hourly factors, by location, (iv) turning fractions at local intersections, (v) intervening source/sink data for local streets, (vi) incident data, (vii) origin-destination data, (viii) roadway geometric data, both freeway and arterial, (ix) locations of stop and yield signs, (x) locations of intersecting signals, and current timing and phasing data, (xi) lane usage patterns at intersection approaches, (xii) speed-density characteristics for limited access facilities, (xiii) queue discharge statistics for signalized intersections. Most of the input data requirements for the scenarios were satisfied by the above data. However, there were areas where the IMIS data had to be supplemented by field observations. Specific examples are: (i) local intersection volume counts and turning fractions, particularly for intersections which are operating at or near saturation, (ii) freeway volume counts during recreational peak, (iii) number of lanes and lane usage at a few signalized intersections, (iv)

_ "iiii'i t--t . /. Rte. 135 <~ Incident Fig 8 IMIS corridor network for s~mulauon scenarios 5 and 6.

Dynamic control and traffic performance

273

L. I. X-way

Nothern State Parkway

Rte 110 ~'

Inodent

F~g 9. IMIS corridor network for simulatmn scenarios 7, 8, 9. latest signal timing and phasing data at some intersections. After the data had been compiled for a scenario, the next step was to generate the required optimum signal timing plans for the various arterial and frontage road signals within the network by the TRANSYT program. These timing plans had to be generated for both the nominal traffic demands which exist at the time-of-day and day-of-week of the scenario and for the various levels of additional traffic loading which were anticipated due to ramp metering and traffic diversions. In addition to the above data, arterial and freeway vehicle detectors have been located throughout each of the IMIS sections to provide sampled 'realtime' data for the simulation study. These data consist of vehicle count and vehicle presence indications, from which volume, occupancy, speed, density, travel ume, delay, and the various other dependent traffic parameters are derived. The criteria which were used for placing detectors in the simulated networks are similar to those which were used for specifying the actual detector complement in the IMIS network. They are summarized below: 1. All freeway lanes immediately upstream of a metered on-ramp. 2. All freeway lanes every 1/2 mile (0.8 km) if not required by the above. 3. All approach lanes to a signalized intersection. 4. All lanes at off-ramp and at nonmetered onramps. 5. A queue detector and a passage detector at automaucally metered on-ramps. 6. A queue detector only at manually metered onramp. RESULTS

All scenarios have been executed with and without computer control in order to evaluate the expected performance improvements of applying the

algorithm to various networks within the IMIS system. A summary of the simulation results for each scenario is given in Table 1. The details of a specific scenario and its results is given in Table 2. Table 3 summarizes overall performance. The overall effectiveness of the algorithm can be judged by comparing the baseline case (no control) with the control case using the various measures of effectiveness (MOE) computed by SCOT, The first MOE compared is delay, which was selected as the objective function for all the scenarios. The average improvement for the MOE was a respectable 16.2070. As seen in Table 1, reduction in delay is greatest in the case of a "hard" incident (i.e. one in which the excess of demand over capacity at the incident site is at least 750 vehicle/hour) when there is sufficient reserve capacity on the alternative route to accommodate this excess. These conditions are met in Scenarios 2, 3, and 5. Generally, as the excess of demand over capacity increases, the benefits in terms of delay will increase provided reserve capacity on the alternative route is sufficient to accommodate the excess. Expected benefits decrease as reserve capacity on the alternative route decreases below the value of excess demand at the incident site. Expected benefits also decrease as the incident becomes "softer" (less excess of demand over capacity). For example, the no-incident case simulated in Scenario 7 shows only a 3.1°70 improvement. Scenario 9, an incident case where the demand was less than the incident capacity, shows a 3.4% improvement. Scenario 4, a soft incident, actually shows a small (less than 1070)disadvantage with control; while not significant in magnitude, it illustrates the fact that counterintuttive results are possible. Another factor to be considered is that the delay improvement values tend to be diluted, for two reasons. First, the delay computed by SCOT includes that experienced by vehicles at signalized intersections. Even with optimum timing, there is some irreducible minimum value of this delay which occurs in both the no-control and control cases. This tends to

I 2 3 4 5 6 7 8 9

Scenario No.

1038 820 1235 587 1385 403 1576 (132) No excess

Excess Demand (veh/hr)

1112 1394 1530 380 3029 757 627 1587

Reserve Capacity on Alternate Route (veh/hr) 20 15 15 25 15 25 No incid 20 30

Duration o f Incident (minutes) 13414 11391 16527 11398 13928 4074 24550 53847 2093

Total Delay without Control (mm) 12981 9361 12385 11490 4611 3847 23791 46227 2022

Total Delay with Control (min)

Table I. S u m m a r y of performance by scenario

3.2 17.8 25.1 (0.8) 66.9 5.6 3.1 14.2 3.4

Delay Reduction (% Improvement)

16.7 66.7 54.5 50 91.7 0 -50 No backup

Congestion Clearance (% Time Reduction)

12.5 63.6 74.1 57.1 81.3 0 34.1 -

Max. Queue Extension (We Reduction)

Dynamic control and traffic performance

275

Table 2. Typical scenario results SCENARIO 2

Typical westbound weekday AM peak Single land blocked by incident on LIE W/B at Willis Avenue (Between Willis Avenue off-ramp and on-ramp) E VENT

TIME (MINUTES F R O M BEGINNING OF RUN) 0 4

Start of run 1st control period Incident starts Incident detected Incident cleared Clearance detected Congesuon cleared

10 18 25 28 37 W/O Control 29 W/ Control 75

End of run C U M U L A T I V E M E A S U R E S OF EFFECTIVENESS AlOE

Veh trips Veh miles (veh kin) Total delay (rain) Ave delay/veh (sec) Delay (min/veh mi) Travel tlme/veh (min) Congestion clearance time after incident removed (min) Max extension of queue upstream of incident (mi)

W / O CONTROL 18526 49536 11391

W~ CONTROL 18533

(79718)

36.9 0.23 (0.14 min/veh kin) 3.26 12 1.1 (1.8 kin)

reduce the percentage improvement achievable by the algorithm. Second, the time span of the scenarios ordinarily included normal periods both before and after the period of congestion. During these periods, only small delay improvements are possible, but the Irreducible surface street delay just described continues to accumulate. As a result, the overall percentage improvement for the scenario is less than that realized during the incident. Shortening the duration of the scenarios would thus increase the apparent delay Improvement. Measures of effectiveness which apply only during the incident show far greater improvement. Referring to Table 3, which presents a summary of all the simulation runs, we see that the average time required to clear congestion after an incident is removed shows a 56070 improvement with control. The length of queue generated by an incident was reduced on average by about 52°70. These magnitudes indicate the dramatic Improvements potentially available during incident conditions.

CONCLUSIONS

A multitude of route guidance and in-vehicle communication systems are now on the drawing

CHANGE (%) -

49445 (79572) 9361 30.3 0.19 (0.12 mtn/veh kin) 3.15 4 0.4 (0.6 kin)

0.2 17.8 17.9 17.4 3.4 66.7 63.6

board or in various stages of experimentation (OECD, 1988). There are great expectations that advanced technology will enable us to more effectively combat congestion. In this paper we attempt to provide a limited, hopefully realistic assessment of the actual opportunities that such systems can offer. Nine scenarios were studied in a well defined network: a freeway corridor system. It was assumed that route-guidance is provided to all motorists and that the motorists follow this guidance so that a system-optimal flow pattern is effected. In most of the cases that were studied, significant improvements were achieved. Those occurred under incident (i.e. nonrecurrent) conditions. However, there was one case in which the control system produced no benefits, or even disbenefits (Scenario 4). In the only case involving recurring congestion (Scenario 7), only minute improvements were observed. This again raises the intriguing question of the potential benefits to be obtained by system-optimization versus useroptimization, since the implication is that drivers already use their best routes under recurrent conditions. Based on these results, it would appear that the conditions under which appreciable benefits would be derived by in-vehicle guidance systems coupled with optimized routing would involve incident,

Table 3. Overall performance summary PERCENTAGE IMPR 0 V E M E N T

AlOE

W I T H O U T CONTROL

Average total delay per scenario Average congestion clearance time after inczdenIremoved (per scenario) Average extension of queue upstream of incident (per scenario)

16802 veh mm

14080 vehmin

16.2

16.3 mm

7.lmln

56.4

1.74 ml (2.80 kin)

0.84m1(1.35 km)

51.7

TR(^) Z5:5-0

WITH CONTROL

276

R. A. REISS,N. H. GARTNER,and S. L. COHEN

or nonrecurrent situations. Future studies should also investigate whether it is realistic to expect further improvements in performance by designing more responsive control systems.

Acknowledgment-This paper is based on research sponsored by the Offices of Research and Development, Federal Highway Administration, with Sperry Systems Management. REFERENCES Boyce D. E. 0988) Route guidance systems for improving urban travel and location choices. Transpn. Res., 22A, 275-281. Federal Highway Administration (1991) Intelligent Vehicle/Highway Systems, Project Summaries. January 1991. Garmer N. H. (1977) Analysis and control of transportation networks by Frank-Wolfe decomposition. Proc. 7th Int. Symp. on Transpn. and Traffic Theory (T. Sasaki and T. Yamaoka, Eds.). Kyoto, Japan, August, 1977, pp. 591-623. Garmer N. H., et al. (1980) Pilot study of computer-based urban traffic management. Transpn. Res., 14B, 203217. Garmer N. H. and Reiss R. A. (1987) Congestion control in freeway corridors: The IMIS system. In A. Odoni et al. (Eds.), Flow Control of Congested Networks, Springer-Verlag.

Institute of Transportation Engineers (1986) Urban Traffic Congestion: What Does the Future Hold? ITE Publication No. IR-040, 1986. LeBlanc L. J., Morlok E. K. and Pierskaila W. P. 0975) An efficient approach to solving the road network equilibrium traffic assignment problem. Transpn. Res., 9, 309-318. Lindley J. A. (1987) Urban freeway congestion: Quantification of the problem and effectiveness of potential solutions. ITEJourna/, January 1987. Organization for Economic Co-operation and Development (1987) Dynamic Traffic Management m Urban and Suburban Road Systems. Road Transport Research Report, Paris. Organization for Economic Co-operation and Development (1988) Route Guidance and In-Car Communication Systems. Road Transport Research Report, Paris. Reiss, R. A. (1981) Traffic Diversion Software-Applications summary. Report No. FHWA/RD-80/100, Federal Highway Administration, September 198 I. Reiss R. A., et al. 0981) Algorithm development for corridor traffic control. Traffic, Transportation and Urban Planning (Vol. 2). George Goodwin, London. Robertson D. I. (1969). TRANSYT: A Traffic Network Study Tool. RRL Report LR 253. Crowthorne, Berks., U.K. U.S. Dept. of Transportation (1985) Traffic Control Systems Handbook (Revised Edition). Report No. FHWA-IP-85-1 l, Washington, DC. U.S. Dept. of Transportation 0989) Discussion Paper on Intelligent Vehicle-Highway Systems. Office of the Secretary of Transportation, May 1989.