1

VMS INFORMATION AND GUIDANCE CONTROLSTRATEGIESIN AALBORG

S. Mammar1, H. Haj-Salem2, A. Messmer3, M. Papageorgiou4, L. Jensen5

1 Université d’Évry Val d’Essonne, CEMIF, Boulevard des Coquibus 91025 Évry, France2 I.N.R.E.T.S : Institut National de Recherche sur les Transports et leur Sécurité

2 avenue du Général Malleret-Joinville, 94114 Arcueil, France3 Ingenieurbüro A. Messmer, Sommerstr. 12, D-81543 Munchen, Germany

4 Technical University of Crete, GR-73100 Chania, Greece5 Danish Road Directorate, Niels Juels Gade 13, P.O. Box 1569, DK 1020 Copenhagen, Denmark

Tel : 33 (1) 47 40 72 91 Fax : 33 (1) 45 47 56 06 Email : said@inrets.fr

ABSTRACTThis paper describes some of the work conducted within the QUO-VADIS project for the development of a VariableMessage Sign (VMS) information and guidance system at the Aalborg site. The employed control strategy is basedon simple automatic control concepts with decentralized feedback loops. Description of the approach used anddevelopment of the automatic control strategy are presented along with off-line simulation results. The simulationstudies demonstrate the potential improvements achievable with this kind of control measures and control strategies.

INTRODUCTION

The urban and freeway networks around major metropolitan areas in the industrial nations have reached a high levelof geometrical complexity and in fact they offer to the network users several possible routes connecting each origin-destination pair in the network. It has been assessed that, in the case of recurrent congestions, the experienced usersusually optimise their proper path. However with respect to non-recurrent congestions which are due to exceptionalevents such as holidays, sport events, incidents or maintenance works, drivers daily create queues of several tenskilometres in spite of the capacity reserve available elsewhere in the network. This situation motivated the roadauthorities to plan and to install extended nets of traffic detectors and communication links, in order to providecomplete real time information to central stations. After the processing, the most convinient solution to returninformation, in a comprehensive manner, to the drivers, is to use the Variable Message Signs (VMS).

The main objective of the European Advanced Transport Telematics (ATT) project QUO-VADIS was to establishhow VMS are best managed to give high quality information and guidance to drivers and to improve the capacity ofa road network. Three alternative methods (optimal selection from a library plan, automatic control and expertsystem) for the real time VMS control were investigated. These methods are based on loop detectors measurements.These methods were developed and applied to the Scottish motorway network and to the Danish Aalborg citynetwork.

This paper describes some of the work conducted within the QUO-VADIS project at the Aalborg site. In particular,description of the approach used and development of the automatic control strategy are presented along with off-linesimulation results.

BASIC APPROACH

The conducted investigations can be subdivided into 4 main steps :- Calibration and validation of the utilized simulation model

2

- Control strategy development and travel time algorithm calibration and validation- Off-line control strategy testing using the validated simulation model- On -line testing and evaluation.

3

With respect to the control strategy development, two main control policies have been considered. The first one isbased on the delay display. By delay we mean the difference between the actual travel time and free flow travel time.The second policy attempts user guidance and the corresponding strategy development is based on the application ofautomatic control theory. Concerning the off-line simulation studies, the first steps have been the calibration andvalidation of the METACOR model that was developed in the frame of the EUROCOR project (1).

The last step was conducted by the site owners (Danish Road Directorate) in cross collaboration with the involvedpartners (INRETS, TUC).

THE AALBORG NETWORK

The network consists of two subnetworks on both sides of the inlet Limfjorden and two connecting links. The twoLimfjorden connections are : the bridge which connects the urban subnetworks at both sides of Limfjorden andwhich leads traffic through the centre of the town; the tunnel which leads traffic East of the city and is part of thenational road network (E45) through Outland and with connections to Norway and Sweden. The overall networkincludes some 85 urban junctions and 20 km of motorway network (figure 1). In general there is enough sparecapacity in the network to absorb fluctuations in traffic level. In connection with roadworks or incidents, however,queues may occur on a daily basis. Both the bridge and the tunnel face major repair works over the coming yearsand fairly heavy traffic problems are foreseen especially in connection with tunnel closures.

Figure 1. Aalborg network with VMS positions Figure 2. Two VMS types : Delay information (a)and route guidance (b)

Loop detectors are placed along the motorway and in the critical area of the urban network. 14 VMS have beeninstalled on the site covering the north and the south of the city. The locations of the VMS are depicted in figure 1.Two types of messages can be diplayed :

- delay information

- guidance.

In the delay display mode, two main destinations are considered : the bridge and the tunnel. The information givento the users corresponds to the estimated delay, from the VMS location to the bridge or the tunnel. As far as theguidance is concerned, three main destinations have been considered at each VMS and one out of two directions

180n

180n

180n

Fr.havn

Lokken

Nr.sundby

(a)

(b)

E45n

180n

Ventetid via

Tunnel 10 min

Bro 5 min

4

may be

5

recommended for each destination at each VMS bifurcation : These destinations are hard coded in the prism of theVMS and activated along with the respective recommended direction by the developed strategy. The two VMS typecan be seen on figure 2.

THE SIMULATION MODEL

METACOR (2, 3) is a simulation program developed in the framework of the EUROCOR Project. Based on amacroscopic modelling approach, it can be used for mixed urban motorway network traffic flow simulation. Themodel may be applied to existing or hypothetical, multiorigin, multidestination networks with arbitrary topologyand geometric characteristics, including motorway bifurcations, junctions, on ramps and off-ramps and urbanjunctions with or without signal control. That modelling approach allows simulations of all kinds of trafficconditions (free, dense and congested) and of capacity reducing events (incidents) with prescribed characteristics(location, intensity and duration). METACOR also considers traffic control measures such as ramp metering and/orcollective route guidance. Several options are offered for describing or prescribing the average route choicebehaviour of drivers groups with particular destinations.

The model parameters were calibrated and validated on basis of the available measurements detectors (5, 6). It wasfound that the model is able to reproduce the complicated traffic phenomena that occur in the network withacceptable accuracy. Complete results are reported in (6). This step is a prerequisite to start the simulation testing ofthe control strategies under different scenarios.

CONTROL STRATEGY DEVELOPMENT

Figure 3 gives the overall architecture of the automatic control strategy either for the information display or the userguidance. In the following, the main blocks included in the overall architecture are described.

R e a l M e a s u r e m e n t D a t a

D a t a f i lt e r in g

L i n k s t r a v e l t im e c a lc u la t io n

I n f o D e l a y i n f o r m a t i o n

g u id a n c eo rO p e r a t o r

F i l t e r in g

Q u a n t iz a t io n(5 minutes)

R u l e s

D e c e n t r a l i z e d c o n t r o lf o r e a c h V M S

L o c a l r es t r ic t io ns

G lo b a l c o m p a t ib il it y c o n s t ra in ts

D e la y d is p l a y

G u id a n c e d i s p l a y

R o u te d e la y c a l c u la t io n

R o u t e s t r a v e l t i m e c a l c u l a t io n

Figure 3. Control software architecture

6

In order to avoid overreaction of the control strategy, the real data measurements are filtered. The selected filter is anexponential smoothing filter. A travel time is calculated for each network link based on the filtered data. At thisstage, the operator is able to choose between the two VMS display modes.

Delay Display Mode

For each link the delay time is calculated by subtracting the normal "free floating" travel time from the computedtravel time. The whole delay times from each VMS until the bridge and the tunnel are then obtained by summing thedifferent links delay times on each corresponding route.

Before displaying, all the delay times are filtered using an exponential smoothing filter. A quantization module putsthe delay time into steps of 5 minutes. This is done in order to make the VMS display more robust, and increasecredibility of the message.

Route Guidance Mode

The chosen automatic control approach is based on decentralized feedback control that does not require on-lineoptimization (4). The main objective of the developed control strategy is the travel time equalization betweenalternative routes leading to the same destination. In fact at every VMS equipped motorway node or urban junctionwhere a route choice toward a given destination is available, drivers can be routed via one route or the other,according to the current travel time of each route. The control acts like a simple proportional regulator which is fedby an error signal corresponding to the travel time difference. Following this concept the synthesis procedure issimplified because the control can be designed separately for each VMS. The applied method (7, 8) first calculatesfor each destination the route travel times based on real time measurements coming from the loop detectors.According to the travel time difference one of the four following displays may be chosen

- no display- guidance through the tunnel or the bridge for one destination d1- guidance through the tunnel or the bridge for two destinations d1 and d2

- guidance through the tunnel or the bridge for three destinations d1, d2, and d3.

When the traffic is equally congested both near the bridge and the tunnel, the previous Proportional controller willterminate with no display. Mixed displays are then chosen according to special rules (ex: one destination is selectedto guide the users through the tunnel and two through the bridge).

The control law is applied for each VMS. However according to the location of the VMS, the messages to bedisplayed are checked with a compatibility constraints matrix in order to avoid contradictory consecutive messagesalong a route. In fact, the drivers should never see contradictory guidance on their way through the network. Forexample if the first guidance given by one VMS is the bridge, the following VMS should not display guidancethrough the tunnel.

The determined VMS display is compared to the actual display. If there is more than one step between them (ex : theactual display is 1 destination guidance and regulator proposal display is 3 destinations guidance), the proposedregulator display is changed to the nearest display to the actual one with respect to the regulator direction (previousex : 2 destinations guidance). The result is that no step is skipped during the guidance.

The maximum frequency of VMS switching is limited by the measurements frequency which is set to 1 minute.However the switching frequency is a software parameter and can be changed by the operator.

Travel time algorithm description

As explained above the two strategies developed are based on the travel time calculation from the VMS location tothe tunnel and the bridge on one hand, on the other hand from the VMS location to each main hard codeddestination. An accurate on-line travel time algorithm is essential for good automatic control strategy behaviour.

The travel time can be calculated from the occupancy or the speed measurements. In the urban area, the speed isusually not available and the occupancy measurements depend on several factors. In particular, the measured

7

occupancies are very sensitive to the location of the detectors (distance between the detector location and the signallight). If this distance is short, the measurements are influenced by the queuing spillback over the detector due to thesignal light. On the contrary, if the detectors are located far from the signal light, we can assume that the occupancymeasurements are less influenced by the queue. The second factor is the cycle time duration and the ratio betweenthe red and the green phase. In order to take into account these factors and to limit the corresponding deviations,weighting factors are introduced for each partial travel time (single link travel time).

The travel time algorithm was validated using the travel time surveys which were organized by the Danish RoadDirectorate during 2 months. The results of the calibration and validation procedure are reported in (6). Thecalibration procedure of the travel time algorithm can be splitted into two steps.

- Fundamental diagram calibration : parameters are optimised for each network link using the available detectormeasurements

- Weighting factors are calculated by minimising the difference between the calculated travel time and the real traveltimes from travel time surveys.

SIMULATION STUDIES

A number of scenarios were tested using the METACOR mode. The model is fed by several realistic demandprofiles at each origin. The scenarios are chosen such that they reproduce different traffic situations (congestion,incidents). The impact of the automatic control strategy is evaluated by the improvement of four quantitative criteria: Total Time Spent (T.T.S.) in (Veh Hour× ), Total Travel Distance (T.T.D.) (Veh Km× ), Mean Speed (M.S.) (Km Hour/ )and the Travel-Time trajectories (7, 8). The simulations studies are organized in 6 scenarios.− Scenario 1: Normal conditions− Scenario 2: Incident on the bridge− Scenario 3: Incident in the tunnel− Scenario 4: Exceptional demand− Scenario 5: Exceptional demand combined with an incident on the bridge− Scenario 6: Exceptional demand combined with an incident in the tunnel.

For each scenario, simulations are run without VMS and with VMS.

Due to the particularity of the test site, the demand flow for the normal conditions (scenario 1) is such that thecontrol strategy doesn't improve the traffic situation. In fact the VMS displays are always blank, so no user isaddressed. This situation constitutes the reference to which the incident and congested situations are compared. Inthe incident cases the number of lanes of the bridge and of the tunnel was respectively reduced by 1 lane and 2 lanesduring respectively 15 and 30 minutes.

The analysis of the level of the demand in the real data collection shows that this level is low and consequently noparticular traffic problem can be generated, except around the bridge. Several trials have been done in order toidentify a suitable profile of the demand at the main origins which generates congestion (without junction blocking)in the network. The profile was applied at each main origin. This profile generates a realistic congestion around thebridge and in particular in the urban southern part without blocking the bridge junction which can be considered as acritical junction of the network. Compared to reality this profile overestimates the demand but in a fairly realisticway. The maximum demand level was applied during 15 minutes corresponding to the peak morning (evening)period. For the motorway origins the same profile has been applied, but with a higher minimum and maximum. Thisprofile is also chosen for the stability analysis of the control strategy. The stability of the control strategies can beevaluated by the analysis in time of the destinations displayed.

During the normal condition tests, the following assumptions have been made : for most Origin/Destination couplessome 70 % of the traffic is served by the motorway path and 30 % by the urban network. In closed loop mode (withVMS), we assume that the splitting rate for a recommended direction is increased (decreased) by 10%. In otherwords, if the bridge (tunnel) direction is recommended for some destination at a particular VMS, then the driverssplit towards this destination becomes 60%-40% (80%-20%) for tunnel-bridge respectively.

Travel Time Impact

8

The first assessment criterion used were the travel time trajectories by VMS for the different demand profiles appliedat each origin. Travel times is a major aspect of the impact of the control strategy.

As expected, in the case of normal conditions (scenario 1), the travel time trajectories via the bridge and the tunnelare not modified by the automatic control strategy. In fact this situation is chosen such as it constitutes the referencecase. The VMS do not display any guidance message. In the case of incident on the bridge (scenario 2), the controlstrategy decreases the travel time through the bridge with no control case. However the travel time through thetunnel is not significantly modified because the motorway part of the network has sufficient reserve capacity. Forscenario 3, the incident occurs in the tunnel. The control strategy diverts users from the motorway to the urban part.This diversion reduces the travel time via the tunnel for the VMS's which are located on the motorway part. At thesame time it increases the travel time via the bridge for the VMS's which are located on the route borrowed by thediverted vehicles. Thus the global gain over the whole network is not very high. Theses remarks are verified in thequantitative results below.

In the case of exceptional demand ( scenario 4), the automatic control strategy is found to improve the travel time onthe urban routes. The motorway travel time is not modified because the motorway has sufficient reserve capacity.When an incident occurs near the bridge just after a peak of congestion (scenario 5), in the case without VMS, thenetwork is still congested at the end of the simulation while the network with guidance is already in free flowconditions. In summary, concerning the travel time trajectories it can be stated that the mean travel time by VMSand destination has been reduced by 20 %.

Quantitative results

The above described criteria have been calculated for each simulation run. The horizon of the simulation has beenfixed according to the duration of the congested situation. However the simulations start and finish at fluidconditions. The calculation of the criteria by scenario is performed both for the whole corridor network and for theurban part and the motorway part separately. The results obtained are reported in Tables 1 to 6. The reportedchanges per criterion and scenario are calculated as follows :

Change = (Criterion - reference criterion) / ( reference criterion)

where the reference criterion is the one calculated for the scenario in the same situation but without VMS.

With respect to these indicated results the following general remarks can be made ::

− In the case of normal traffic conditions which corresponds to an uncongested network, there is no improvementof the different criteria. In fact the VMS do not display any guidance.

− In the case of normal traffic conditions associated to an incident on the bridge, the guidance control strategyimproves the different criteria by some 5 % for the overall network. However the gain is greater on the urbanpart of the network, especially for the total time spent criteria.

− In the case of normal traffic conditions associated to an incident on the tunnel, the guidance strategy doesn'timprove the overall network situation. In fact the motorway part is still less congested than the urban part forthis scenario (tables 1, 2 and 3).

SCENARIOS

T.T.D. change (%) T.T.S. change (%) M.S. change (%)

1 0 0 02 5 -5 63 0 1 0

Table 1. VMS assessment for the overall network (urban + motorway) under normal demand

SCENARIOS T.T.D. change (%) T.T.S. change (%) M.S. change (%)1 0 0 02 1.7 2 03 -0.8 -3 2.2

9

Table 2. VMS assessment for the motorway part under normal demand

SCENARIOS T.T.D. change (%) T.T.S. change (%) M.S. change (%)1 0 0 02 0 -7 73 0.4 2 -2

Table 3. VMS assessment for the urban part under normal demand

10

− In the case of exceptional traffic demand conditions, the guidance control strategy improves respectively theT.T.D, S. and M.S. by 1%, 10 % and 10% for the overall considered network (table 4).

− The analysis of the calculated criteria by component of the network indicated that the improvement of the trafficcondition in the urban part is clearly higher than in the motorway part.

− Also, in the case of incident either on the bridge or the tunnel, the benefits are in advantage more for the urbanpart than the motorway part. In particular, the guidance with an incident on the bridge is more efficient than foran incident in the tunnel. This is due to the free flow condition on the motorway part.

SCENARIOS T.T.D. change (%) T.T.S. change (%) M.S. change (%)4 1 -10 105 1 -10 136 1 -6 7

Table 4. VMS assessment for the overall network (urban + motorway) with exceptional demand

SCENARIOS T.T.D. change (%) T.T.S. change (%) M.S. change (%)4 5 4 15 5 -4 06 4 -1 6

Table 5. VMS assessment for the motorway part with exceptional demand

SCENARIOS T.T.D. change (%) T.T.S. change (%) M.S. change (%)4 2 -12 135 2 -14 146 2 -7 5

Table 6. VMS assessment for the urban part with exceptional demand

CONCLUSION

The described control strategy can be applied to any kind of mixed network (urban and motorway) with a similarstructure. The developed approach is based on decentralized control concept. The size and the complexity of theconsidered network doesn't affect the computation effort. During three months the two strategies have been appliedin real life. At the same time questionary surveys have been organized by the Danish Road Directorate (9).

ACKNOWLEDGEMENTS

This work was conducted with support of the program Advanced Transport Telematic of the European Commission.Project QUO-VADIS (V2042). The authors acknowledge the support of the Danish Road Directorate.

ENDNOTES

1. DRIVE II Project V2017, EUROCOR. "Deliverable N°4: Integrated model development : Application to theCorridor Périphérique test site", 1993.

2. Haj-Salem, H., Elloumi, N., Mammar, S., Papageorgiou, M., Chrisoulakis, J., Middelham, F., "METACOR :A Macroscopic Modelling Tool for Urban Corridor", First World Congress on Application of TransportTelematics and Intelligent Vehicle-Highway Systems, Paris, pp. 1333-1340, 1994.

11

3. Haj-Salem, H., Mammar, S., "Urban and Freeway Model Calibration and Validation", 13th World CongressIFAC'96, San Francisco, June 30 - July 5, 1996.

4. Papageorgiou, M., "Dynamic modelling, assignment, and route guidance in traffic networks". TransportationResearch 24B, No., pp. 471-495, 1990.

5. Elloumi, N., "Modélisation et Commande du Trafic sur un Corridor, Application de la méthode LP", Thèsede Doctorat Université Paris XI, 1995.

6. DRIVE II Project V2042, QUO-VADIS, "Deliverable N°6B: Validated Network Models for use in NetworkControl", 1994.

7. Mammar, S., Messmer, A., Papageorgiou, M., Jensen, P., Haj-Salem, H., Jensen, L., "Automatic Control ofVMS in Aalborg", To appear in Transportation Research Part C, 1996.

8. DRIVE II Project V2042, QUO-VADIS. "Deliverable N°8: Off-line Control Strategy Evaluation", 1994.

9. DRIVE II Project V2042, QUO-VADIS. "Deliverable N°11: Evaluation of Control Implementation IncludingDriver Response", 1995.