EVALUATION AND FIELD IMPLEMENTATION OF THE FUZZY LOGIC BASED RAMP METERING ALGORITHM ACCEZZ

Key topic: traffic control Keywords: ramp metering, fuzzy logic, field installation Dr.-Ing. Svetlana Vukanovic Dipl.-Ing. Oliver Ernhofer TRANSVER GmbH Maximilianstrasse 45 D-80538 Munich vukanovic@transver.de, ernhofer@transver.de Abstract: Ramp metering has proven to be one of the most direct and effective control measures to overcome temporary disturbances on motorway traffic flow and to help to reach an improved utilization of 1 motorway infrastructure. The ACCEZZ ramp metering algorithm is an adaptive control approach based on Fuzzy Logic. Results from its calibration and validation as well as its evaluation show how different control strategies can improve traffic conditions especially during the peak hours. The ramp metering software ACCEZZ including a GUI has been implemented at the demonstration site A94 in st Munich since December 1 , 2005 and controls traffic at two consecutive on ramps successfully. Data for a cross-algorithm (with ALINEA) and a cross-site evaluation will be collected at A94 until April 2006 2 within the European research project EURAMP .

INTRODUCTION
Ramp metering has proven to be one of the most direct and effective control measures to overcome temporary disturbances on motorway traffic flow and to help to reach an improved utilization of motorway infrastructure. Within the EURAMP project, a European RTD project, forming part of the European Commissions FP6 Action Line e-safety of road and air transports ramp metering algorithms have been modified and re-developed to aim to EURAMPs major objectives to advance, promote and harmonize ramp metering control measures in European motorways. These major objectives are pursued within EURAMP via a number of multifaceted actions and subobjectives:
-

Advancement of methodological issues, with particular focus on traffic flow safety, to secure a European technological leadership in the area. Consolidation, harmonization and advancement of ramp metering practice in Europe. Demonstration of new developments in European sites and paving the way for a new generation of extended (network-wide) ramp metering installations. Co-operation of ramp metering with signal control and further heterogeneous control measures for maximum synergy in terms of traffic flow efficiency and safety.

The first stage in the project is the (re-) design and simulation testing of ramp metering algorithms, in particular ALINEA and ACCEZZ in several demonstration and virtual sites in France (A6 Paris), the Netherlands (A28 and A2 Utrecht), Germany (A9 and A94 Munich) and Israel (Ayalon Highway in Tel Aviv). This will be followed by the field implementation and verification of the ramp metering strategies and the field demonstrations, allowing assessment of the strategy and a comparative evaluation. At the German demonstration site the genetic fuzzy algorithm of ACCEZZ has been implemented, tested (by using the microscopic simulation tool AIMSUN) and verified. In the following a summary of the calibration and evaluation of the algorithm and its fuzzy sets is given as well as the main aspects of the software development, the graphical user interface GUI, the field implementation at the motorway A94 and the next steps within the demonstration phase.

1 2

ACCEZZ: Adaptive Control Approach at Entrance Ramps based on Fuzzy Logic EURAMP: European Ramp Metering Project; www.euramp.org

ACCEZZ RAMP METERING ALGORITHM

The ACCEZZ ramp metering algorithm is an adaptive control approach based on Fuzzy Logic. Fuzzy logic seems to be well suited for ramp metering for several reasons. The rule base, defined as the set of rules in the fuzzy logic algorithm, incorporates human expertise. Since rules are easy to define, alter or eliminate fuzzy logic allows simple development and modification. Because a fuzzy controller can handle nonlinear systems with unknown models, it has a distinct advantage over traditional controllers for the ramp metering problem. Implementations of fairly simple fuzzy logic ramp controllers in Seattle and Amsterdam have proven these capabilities. A detailed description of ACCEZZ can be found in Bogenberger, et al. (2001).

SIMULATION SCENARIOS AND RESULTS

The simulation-based evaluation of ramp control with ACCEZZ is demonstrated for a particular 33 km motorway segment of the motorway A9 Mnchen-Nrnberg in northbound direction (see Figure 1). The simulated network is one of the motorways with the highest traffic load in Germany and recurrent daily congestion occurs especially in the afternoon peak hours. The simulation shows ten on-ramps of which two are motorway-to-motorway interchange ramps and eight off-ramps. At present, traffic is controlled with a link control system and a re-routing system while ramp metering is not installed. The calibration of the simulation model (AIMSUN) has been an extensive process and involved more than 100 calibration/simulation runs with real traffic data other than that used in the simulation later, until a good approximation of reality was achieved. The implementation of ACCEZZ in the microscopic simulation tool AIMSUN took part via its GETRAM extension kit. The approach taken in the microscopic simulation tool allows to consider advanced telematic applications to be tested as an external application that can communicate with AIMSUN. Using specified functions from a DLL library, detector, VMS and traffic lights can be modeled and their attributes defined. With this approach AIMSUN emulates the detection process providing ACCEZZ with the required simulation detection data. This allows ACCEZZ to decide which control actions have to be applied on the road network. The corresponding information is then sent back to the simulation model which then emulates their operation through the corresponding traffic lights. In total eight different scenarios have been tested under different traffic conditions leading up to 16 comparisons. For two different demand patterns a reference case (no control) is compared with different control scenarios: whereas the typical weekday afternoon congestion lasted from 4 to 7pm, the typical Friday congestion period runs from 12 to 2pm and from 4 to 7pm. Taking into the consideration the possibility and needs at the field and as well as scientific interests the following scenarios have been compared to the reference case: 1. Metering of 1 ramp - AS Frankfurter Ring - with the ACCEZZ algorithm and capacity information from the global bottleneck. This scenario represents the most basic solution at A9 but can create major success. AS Frankfurter Ring is the ramp with the highest demand and the longest storage lanes for the ramp-queue. Therefore numerous vehicles can be stored on the on-ramp without negatively influencing the urban street network. This ramp is from great interest for road authorities. The bottleneck threshold value is set to 5500 veh/h. 2. Metering of 1 ramp - AS Freimann - with the ACCEZZ algorithm and capacity information from the global bottleneck. AS Freimann is another on-ramp with very high inflow and would be alternative to the first scenario (AS Frankfurter Ring). A comparison between these strategies should give insight under which circumstances which approach would bring more benefits, as well as how these strategies influences the performances of the entire network. The bottleneck threshold value is set to 5500 veh/h. 3. Metering of 3 on-ramps located at the beginning of the simulated stretch with the ACCEZZ algorithm and global bottleneck information. The controlled on-ramps are AS Frankfurter Ring, AS Freimann and AS Frttmaning. These ramps are located shortly behind each other. High traffic flow coming from these ramps

together with mainline flow merges with the demand coming from A9/A99 motorway-tomotorway interchange. High flows from the mainline and the A9/A99 interchange are the main cause for the congestion at A9 northbound direction. Since controlling the motorway-tomotorway interchange is a questionable issue in Germany, the potential of ramp metering to reduce such congestion by metering the upstream demand will be considered. The bottleneck capacity value is set to 5700 veh/h as determined from the fundamental diagram. 4. Metering of 3 on-ramps located at the beginning of the simulated stretch with the ACCEZZ algorithm and global utilized bottleneck information. The controlled on-ramps are AS Frankfurter Ring, AS Freimann and AS Frttmaning as before. To investigate how the bottleneck threshold value influences the overall performances in this scenario, values of 5000 veh/h will be used instead of the bottleneck capacity value of 5700 veh/h determined from the fundamental diagram. 5. Metering of 4 on-ramps at the beginning of the simulated stretch with the ACCEZZ algorithm and global bottleneck information. The controlled on-ramps are AS Frankfurter Ring, AS Freimann and AS Frttmaning as before but extended by an additional ramp metering at the freeway-to-freeway interchange (AK MNord) which represents one of the most critical ramps at the stretch. This scenario is not likely to be applied in reality but to show further potentials of network wide ramp metering by controlling motorway-to-motorway interchanges. The bottleneck capacity value is set to 5700 veh/h as determined from the fundamental diagram. 6. Metering of 4 on-ramps at the beginning of the stretch with the ACCEZZ algorithm and global utilized bottleneck information. The controlled on-ramps are AS Frankfurter Ring, AS Freimann and AS Frttmaning and the freeway-to-freeway interchange (AK M-Nord). To investigate how the bottleneck threshold value influences the overall performances in this scenario, bottleneck capacity values of 5000 veh/h will be used instead of 5700 veh/h determined from the fundamental diagram. In addition there is also a 7 and 8 scenario by applying the 3 and 4 ramp strategies with only local information. This corresponds to controlling several local on-ramps along the stretch without their coordination or interaction. Using the data from a typical weekday traffic pattern, different ramp metering control strategies are tested, compared and explained in the following. The performances of a 3 on-ramp scenario (AS Frankfurter ring, AS Freimann, AS Froettmaning), a 4 on-ramp scenario (like the previous plus AK MNord) and two 1 on-ramp scenarios (AS Frankfurter Ring and AS Freimann) have been analysed and compared with the no control case. The control scenarios involving several on-ramps are local control strategies since each controlled on-ramp uses only the local data without considering the situation at the bottleneck. As input into the simulation measured traffic demand was used with deactivated dynamic speed limits. Contour plots of the no-control scenario and simulation outputs produced with different ramp metering strategies are given in Figure 2. According to the average journey times (see Figure 3 and Figure 4), all strategies manage to improve the situation at the mainline during the entire congestion period. Travel time for the considered stretch is reduced up to 8 min/veh (more than 35%). However as expected, an increased travel time can be measured at the on-ramps. By observing the journey time at the entire motorway, all ramp metering strategies managed to reduce the congestion at its beginning but during the course of the congestion the high demand at the on-ramps protected the ramp metering from further improvements. The cumulative travel time for the motorway section of interest for all scenarios as well as the improvements in percentage of each control strategy in comparison with the no-control case is given in Table 1. Positive percent means that the control improved the situation at the road, while the negative values mean that motorway performances were decreased due to the ramp metering control. The best improvement at the mainline is achieved by the 4 on-ramp scenario but due to the significant increase of travel times at on-ramps (mainly at the AK M-Nord) overall benefits are reduced. The 3 onramp scenario not only improved the situation at the mainline, but in addition due to the clearing of the on-ramp areas at the beginning of the motorway stretch lower travel times at the on-ramps are achieved also as on-ramps are not severely blocked by the congestion at the mainline. The 1 on-ramp scenarios improved traffic flow performances by the on-ramp but could not significantly reduce congestion since the bottleneck is further downstream by AK M-Nord. The 1 on-ramp Freimann control
th th

scenario increased traffic related performances in comparison with the scenario where only Frankfurter Ring is controlled. In overall, cumulative travel time is reduced, with the lowest improvement achieved with 1 on-ramp Frankfurter Ring scenario of 3.5% and highest reduction by the 3 on-ramp scenario of 10.8%. As expected most changes happen in those areas where the different scenarios are applied, i.e. the first 4 on-ramps respectively the first 5 kilometers. Whereas in the reference case speed at the onramps are constantly 20 km/h or below, conditions recover especially in the 3 and 4 on-ramp scenarios with speeds around 80 km/h. Due to the high demands, the positive effects of the 4 on-ramp scenario on the mainline traffic were strongly reduced by induced queues at the last on-ramp (AK MNord). The throughput after the bottleneck is increased by each ramp metering scenario and the best environmental impacts are measured with the 3 on-ramp scenario. All scenarios significantly improved the situation at the mainline for the price of larger queues at on-ramps.

SOFTWARE DEVELOPMENT AND FIELD INSTALLATION

On the high level of abstraction the ACCEZZ software to be implemented in a microscopic simulation or in the field (here the EURAMP demonstration site A 94) consists of several logical groups: algorithms, data models, data builders (responsible for forming the data models), sequences (responsible for the correct sequence of actions in vase of a model state change e.g. new data are received), interfaces, etc.. Algorithms include ramp metering algorithms, algorithms for transferring the metering rate received from the algorithm into appropriate control programs or cycle times, but also data plausibility checks, data correction algorithms, LOS algorithms, genetic algorithms and traffic flow models. Currently the software integrates interfaces to the AIMSUN DLL, an infrastructure server (ISS) responsible for sending control actions further to the controller at the motorway and a general socket server implementation and corresponding protocols allowing to connect a new client. The context diagram in Figure 5 shows the general structure including all components and their connections at the EURAMP demonstration site A94. It can be seen, that the ramp metering algorithms (here ALINEA and ACCEZZ) are directly installed at the Data Server (DS) in the traffic control centre (in German: Verkehrsrechnerzentrale, VRZ) Freimann. As the DS is connected via the Infrastructure Server (ISS) and the Communication Computer (in German: Kommunikationsrechner, KR) to all sub control centers (in German: Unterzentralen, UZ) and further to all road side stations (in German: Streckenstationen, SST), the algorithm can receive traffic data from all detectors (here loop, radar and video) and send control data to all control interfaces (i.e. ramp meters) at the motorways around Munich. At VRZ the algorithms continually calculate the metering rates which then are sent to the single local controllers. The controllers themselves then choose the according signal programs and cycle times. The local parameter settings of each metered ramp can also be changed from VRZ. A graphic user interface (GUI) has been specially designed for the visualisation of the ramp metering related values and is also used for the user interaction with the ramp metering control software (see Figure 6). The GUI allows to choose between test sites, working and connection modes, and also specifies the input values for testing and running ramp metering algorithms. Most of all it allows a brief overview on almost all results and parameters, shown as graphs or bars for a quicker and easier apprehension. All of them can be examined closely as single graphs.

CONCLUSION AND FURTHER DEVELOPMENTS

The ACCEZZ ramp metering algorithm is an adaptive control approach based on Fuzzy Logic. The verification of the genetic fuzzy algorithm incorporated in the ramp metering model of ACCEZZ shows that utilising bottleneck information in terms of bottleneck capacity has advantages over utilising only local detector information. By applying a lower bottleneck capacity value than the critical capacity value as identified at the fundamental diagram is even more advantageous. Having tested different control strategies in terms of numbers and locations of metered ramps it can be said that each ramp metering strategy managed to improve the situation at the mainline, especially at the location of the ramp metering and its vicinity (~2km distance). By observing the entire motorway

stretch a distinct advantage of the 3 on-ramp strategy at AS Frankfurter Ring, AS Freimann and AS Frttmaning and the 1 on-ramp strategy at AS Freimann is observed. Controlling only the ramp at AS Frankfurter Ring leads to the release of the bottleneck which again results in much higher flows arriving to the major bottleneck downstream at AK M-Nord. Moreover, very high demand at the onramp leads to a fast queue forming and reduces the ability of ramp metering to control incoming traffic. In terms of the cumulative travel times the maximal improvement that can be achieved hereby is up to 10% during weekdays and 13% during a typical Friday. The ramp metering software ACCEZZ including a GUI has been implemented at the EURAMP st demonstration site A94 in Munich since December 1 , 2005 and controls traffic at two consecutive on ramps successfully. Data for a cross-algorithm (with ALINEA) and a cross-site evaluation within EURAMP will be collected at A94 until April 2006. Full demonstration results will be available by summer 2006 and will have an influence on the recommendations given in the Handbook of Ramp Metering that will we published by the end of the EURAMP project in March 2007.

FIGURES AND TABLES

AQ9-110 9031599 AQ9-350 9031012 AQ9-135 9031550 AQ9-155 9031517 AQ9-166 9031470 AQ9-170 9031423 AQ9-210 9031342 AQ9-240 9031287 9031257 AQ9-260 9031233 AQ9-280 9031150 AQ9-310 9031076 AQ9-360 9030975 AQ9-370 9030910 AQ9-380 9030872

9031701

9031635

9031701

9031589

9031376

9031054

Frankfurter Ring

Freimann

Frttmaning

Mnchen-Nord

Garching-Sd

Garching-Nord

Eching

Neufahrn

Allershausen

Figure 1 Demonstration Site Munich: A9 northbound direction: road geometry, metered ramps (yellow), detectors (red) and VMS signs (blue)

Speed Contour Plots for A9 Mnchen-Nrnberg, 26.02.2004 a) reference case

916070

b) 1 on-ramp (Frankfurter Ring)

c) 1 on-ramp (Freimann)

d) 3 on-ramps

e) 4 on-ramps

Figure 2 Speed contour plots for the no-control scenario (a) and different control scenarios b) e)

Figure 3: Average journey time at the mainline; red reference case

Figure 4: Average journey time for the entire motorway including on-ramps; red reference case

<< component >> ACCEZZ << component >> ALINEA

Figure 5: Context diagram Test Site A 94

Figure 6 Graphical user interface (GUI) of ACCEZZ showing the considered part of a network and its traffic conditions (LOS) as well as the main control parameters

Measured cumulative travel time [h] Difference in hours [h] Improvement [%] mainline on-ramp total mainline on-ramp total mainline on-ramp total Reference 4194 630 4824 Frankfurter Ring 3974 680 4654 220 -50 170 5,25 -7,94 3,52 Freimann 3905 539 4444 288 91 380 6,88 14,49 7,87 3 on-ramps 3743 559 4303 450 71 521 10,74 11,24 10,81 4 on-ramps 3699 827 4526 495 -197 298 11,80 -31,24 6,18 Scenario

Table 1: Cumulative travel time and ramp metering impacts

REFERENCES
K. Bogenberger, (2001), Adaptive Fuzzy Systems for Traffic Responsive and Coordinated Ramp Metering. Ph.D. Thesis at the Fachgebiet Verkehrstechnik und Verkehrsplanung, Munich, 2001. K. Bogenberger, S. Vukanovic, H. Keller: A Neuro-Fuzzy Algorithm for Coordinated, Traffic Responsive Ramp Metering, IEEE 4th International Conference on Intelligent Transportation Systems, Oakland, California, August 25-29, 2001 H. Haj-Salem, A motorway-to-motorway control: Overview of simulation and field trial results in France. Preprints Triennial Symp. On Transportation Analysis IV, 2001, pp. 415-420. M. Papageorgiou and A. Kotsialos, Freeway ramp metering: An overview, IEEE Trans. on Intelligent Transportation Systems, vol. 3, pp. 271-281, 2002 EURAMP (EUropean RAmp Metering Project), contract number 507645, is part of the European Commission's FP6 Action Line "e-safety of road and air transports". More information: www.euramp.org