ASDA/FOTO based on Kerner s Three-Phase Traffic Theory in North Rhine-Westphalia and its Integration into Vehicles

Transcription

1 ASDA/FOTO based on Kerner s Three-Phase Traffic Theory in North Rhine-Westphalia and its Integration into Vehicles H. Rehborn* and J. Palmer# *Daimler AG and # IT-Designers Abstract Traffic data measured with stationary loop detectors on freeways in North-Rhine Westphalia are used at a major German radio broadcasting station (WDR) for the detection and reconstruction of spatial-temporal congested traffic patterns. This is done with models ASDA (Automatic Jam Recognition) and FOTO (Forecasting Of Traffic Objects) based on Kerner' Three-Phase traffic theory. The detected congested traffic patterns can be suitable as a basis for production and control of RDS/TMC traffic messages which are broadcasted to radio listeners and RDS/TMC receivers in vehicle navigation systems. The article briefly discusses the theoretical background of the models ASDA and FOTO as well as the results achieved until now in the on-line operation in North Rhine-Westphalia without any calibration of model parameters. A quality inde developed for ASDA and FOTO relevant for RDS/TMC messages as well as architectural aspects of the model s integration into vehicles are presented. Inde Terms Cooperative Vehicle Highway Systems, Information Fusion, Telematics, Traffic Information System T I. INTRODUCTION oday more and more applications in vehicles depend on the knowledge of current traffic and a prediction of the further situation. The travel estimation in navigation systems depends substantially on the availability of highquality traffic information. Dynamic route guidance makes it possible to guide drivers around traffic congestions. Other applications like driver assistance systems could benefit from the availability of more up-to-date traffic information. For eample, innovative new automatic cruise control systems react depending on current traffic situation and thus support the dissolution of eisting and prevent the emergence of new traffic congestions. Traffic situation interpretation is one central problem in traffic engineering ([1]-[5] and references there). Traffic data measured on freeways and transmitted to traffic centres serve as a basis for traffic management systems as well as for driver information and assistance in vehicles. Congested traffic situations are aired by broadcasting corporations as Manuscript received January 9, 28. H. Rehborn is with Daimler AG, HPC: 5-G21, D-7169 Sindelfingen, Germany (phone: +49-() ; fa: +49- () ; Hubert.Rehborn@Daimler.com ). J. Palmer is with IT-Designers GmbH, Entennest 2, D-7373 Esslingen, Germany (phone: +49-() ; fa: +49-() ; Jochen.Palmer@IT-Designers.de ). RDS/TMC traffic messages [6]. The WDR as the public broadcasting corporation of North Rhine-Westphalia covers approimately 4% of all German traffic congestions as analyzed in a German message archive [7]. The traffic messages from WDR may originate from different sources, e.g., pilot alarm units ( Staumelder ). Currently, WDR uses automatic processing by models ASDA and FOTO for interpretation of the raw traffic data. The quality of the reconstruction of congested traffic has direct impact on the reliability of traffic messages and therefore influences positive and negative route guidance decisions in vehicles. This article includes some aspects of future development of ASDA and FOTO integration into vehicles to reach a new status of congestion information. II. THREE-PHASE TRAFFIC THEORY Y KERNER An analysis of empirical behaviour of congested freeway traffic has been performed with high amounts of traffic data measured by stationary loop detectors. This has led to the development of Three-Phase traffic theory by oris Kerner [5], which eplains and models empirical traffic phenomena on freeways. This section gives a brief introduction to the theory as well as some statements relevant for traffic congestion analysis and RDS/TMC message generation. The Three-Phase traffic theory differentiates three traffic phases: Free flow (F), Synchronized flow (S) and Wide Moving Jam (J) [5]. In phase F it is possible for vehicles to drive, change lanes and overtake almost freely. In contrast, traffic is no longer free flowing if those manoeuvres are hindered, i.e., movements of vehicles are bound together. The ability to overtake or to change lanes is restricted or not at all possible in phase S. In traffic phase J vehicles are so close together that they often come to a complete stop or drive only with very slow speed. oth congested traffic phases S and J are popular known as "stop-and-go" traffic. Speed (km/h) Synchronized flow (S) Flow rate (Fzg./h) space (km) ottleneck Wide moving jam (J) space (km) Figure 1: Spatial-temporal overview of speed (left) and flow rate of traffic (right) on a selected freeway section [5].

2 Fig. 1 shows the empirical traffic data of vehicle speed and flow over and space on a freeway section. At about location kilometre 18 on the freeway is a bottleneck. The reason can either be a decrease in the number of lanes on the freeway, e.g., from three to two lanes or an intersection with its on- and off-ramps. It can be recognized that directly at the bottleneck an area of traffic phase S is formed. One characteristic criterion of phase S is the fact that the downstream front is fied in most cases at the location of the bottleneck. Within the downstream front vehicles accelerate from lower speeds which they drove in phase S back to higher speeds, which are possible in F. In contrast, the upstream front of the phase S oscillates dependent on the incoming upstream vehicle flow, i.e., the spatial dimension of phase S increases with higher incoming flows and decreases with lower incoming flows. Inside traffic phase S wide moving jams (J) can emerge and propagate further upstream. In areas of phase J neither the upstream nor the downstream front are fied. oth propagate upstream with a certain speed. Characteristic for phase J is that the two fronts are not stopped or slowed down by further bottlenecks while they move upstream (see J in Fig. 1). In both congested traffic phases vehicles have a clearly decreased speed; however, an additional breakdown of flow rates takes place only in areas of phase J. In empirical observations it has been discovered that the emergence of traffic congestion at a freeway bottleneck begins with a phase transition of an area F to an area S, which is called F S transition [5]. Wide moving jams emerge only in areas where already phase S eists, i.e., then S J transitions occur. Direct F J transitions cannot be observed in empirically measured traffic data, but only by a transition over S, i.e., a cascade F S J transition. If a bottleneck is isolated, i.e., far enough away of ones, two congested patterns can be differentiated: 1) Synchronized Pattern (SP): Patterns of this type show only areas of phase S upstream of the bottleneck; no phase J eists. 2) General Pattern (GP): In contrast to patterns of the class SP also areas of phase J emerge. After wide moving jams have emerged in a certain distance upstream from the bottleneck, they propagate further upstream. GP is the most common pattern of congested traffic on freeways. If further bottlenecks are in close neighbourhood, then a third kind of traffic pattern can eist: 3) Epanded Pattern (EP): At each of the bottlenecks separate GP patterns emerge whose regions of phases S and J overlap each other due to the proimity of bottlenecks. EP is more comple both by etent and strength of the patterns in comparison to GP. (a) (c ) trajectory trajectory (b) trajectory Free traffic Synchronized traffic Wide moving jams ottleneck Figure 2: Three major different types of congested traffic patterns: (a) Synchronized Pattern (SP), (b) General Pattern (GP) and (c) Epanded Pattern (EP) [5]. The emergence and development of congested traffic patterns are characteristic for a certain bottleneck and/or a sequence of several bottlenecks on a freeway section. Thus spatial-temporal congested patterns possess predictable and recurring characteristics, e.g., type and etent of the arising spatial-temporal patterns at a certain bottleneck. III. MODELS ASDA AND FOTO FOR TRAFFIC PATTERN RECOGNITION AND PREDICTION For recognition, tracking and prediction of the spatialtemporal congested traffic patterns the models ASDA (Automatische Staudynamikanalyse; automatic congestion analysis) and FOTO (Forecasting of Traffic Objects) ([5], [8]-[12]) have been proposed by Kerner based on his Three-Phase traffic theory. oth models use the above mentioned key elements of the theory. The model FOTO allows the recognition of the current traffic phase at a certain position in space and as well as the detection of the fronts and the etensions of regions of traffic phase S. In model ASDA the fronts of the regions of traffic phase J are recognized and predicted in their movement. oth models work with traffic measurements containing values such as flow rate of traffic, speed of vehicles and proportion of trucks and passenger cars. In addition, the traffic phase recognition based on vehicle measurements (FCD: Floating Car Data, [14]) is intended. The model FOTO determines the position of the fronts (syn) (syn) up, down as functions of t. The model ASDA ( jam) ( jam) updates the fronts up, down of the respective regions of phase J over t (Fig. 3).

3 q 1, v 1 direction of flow Models ASDA and FOTO 1. Recognition of traffic phase: F = free traffic; S = synchronized flow; J = wide moving jams q 2, v 2 2. Tracking and prediction of congested traffic phases F J F S J S F 3 (syn) up down up (syn) up down down Figure 3: Overview ASDA and FOTO [5],[8]-[12] The models ASDA and FOTO determine the positions of the fronts of congested traffic phases also between local detectors. The front positions are determined and updated based on stationary measured traffic data. Thus ASDA and FOTO allow a continuous identification and update of congested traffic patterns even if the distances between individual measuring points become wider, i.e., no complete monitoring of traffic is possible with local detectors. This was proven in etensive investigations with traffic data information of different completeness with an estimation of possible model quality, i.e., the input for the models ASDA and FOTO is reduced during simultaneous evaluation of model results [13]. oth models work without any calibration of the respective model parameters in different environmental and traffic conditions. The quality of traffic phase recognition and prediction could be further improved by data fusion with additional detection systems, e.g., FCD. IV. ASDA/FOTO IN NORTH RHINE-WESTPHALIA AT WDR TRAFFIC CENTRE Freeways in the federal state North Rhine-Westphalia are equipped with stationary detectors conducting on-line traffic measurements by using inductive loops in cyclic intervals. Measurements contain average traffic speed, average traffic flow as well as the proportions for cars and trucks in one minute intervals. The density of the detector net varies on different freeways in North Rhine-Westphalia. Therefore, the quality of traffic data acquisition depends on the detector density at specific freeway locations. Each detector has a unique identification number within the road network. The identification number directly relates to the position of the detector on a specific freeway. In traffic control centres Recklinghausen and Leverkusen all measured data samples of individual detectors are collected. The format used for the description of the measured traffic data is standardized: the so-called TRAILS files contain the measurements of all detectors associated with a stamp, stating when those measurements were conducted. On basis of this data a reconstruction of current traffic conditions as well as prediction of the further development can be computed in the traffic control centres. WDR is the public broadcasting corporation of the federal state North Rhine-Westphalia and member of the ARD, a pool of all German federal state broadcasting corporations [15]. Within this pool the WDR represents the largest federal state broadcasting station. Radio programs aired by the WDR broadcast traffic messages in RDS/TMC format. In the transmission area of WDR about 4% of all German traffic messages are broadcasted [7]. y using the careful assumption of on an average equal data acquisition technique in Germany this also means that about 4% of all German traffic congestions occur in North Rhine- Westphalia. The WDR utilises on-line access to traffic measurements collected with loop detectors. At WDR traffic centre in Cologne the models ASDA and FOTO were installed for on-line monitoring of traffic. The result of raffic data processing with ASDA/FOTO is a reconstruction of current traffic state as spatial-temporal traffic objects. These objects represent the traffic phases S and J and are stored in a traffic object database. The ASDA/FOTO software uses the database in order to determine the travel s which an individual vehicle would eperience when it travels on freeway (TMC-) sections affected by traffic congestions. The travel s determined by ASDA/FOTO can be mapped to RDS/TMC messages [6], when the events stationary traffic and queuing traffic are used together with the respective event length for each TMC section. The conversion of RDS/TMC messages into total travel s is accomplished by: L Lstationary L free queuing T total = + + (1) v free vstationary v queuing with estimation as v free = 12km/ h, v stationary = 1km / h, v queuing = 4km / h as well as L free, Lstationary, Lqueuing representing parts of the entire RDS/TMC road section. Estimation of speeds is based on WDR traffic data analysis. A network of federal state freeways currently controlled by ASDA/FOTO software at WDR site is given in Fig. 4. A freeway section is only shown in Fig. 4 when the stationary loop detection is sufficient for an application of ASDA/FOTO at that specific road section. Hence not every freeway section in North Rhine-Westphalia is shown, because there are quite some sections without or with only insufficient availability of stationary loop detectors. In the current ASDA/FOTO installation the traffic control centre (TCC) Recklinghausen transfers its measurements to WDR traffic centre. Therefore the network is limited to the core area of the Ruhr district of North Rhine-Westphalia. The system is currently not connected to the TCC Leverkusen,

4 hence these measurements are missing. Nevertheless, in the current installation of ASDA/FOTO 516 stationary loop detectors and about 19 km of freeway distance are processed by the on-line system (km) :3 13: 13:3 14: 14:3 15: 15:3 Figure 5: ASDA/FOTO result for A4-East on 31 st Aug : ochum-ruhrstadium, 2: Gelsenkirchen. Figure 4: Freeway sections in North Rhine-Westphalia at the WDR site covered currently by ASDA/FOTO. V. SPATIAL-TEMPORAL CONGESTED PATTERNS ON FREEWAYS IN NORTH RHINE-WESTPHALIA One possible visualization option of ASDA/FOTO results is the representation as a spatial-temporal diagram. Such diagrams present the two congested traffic phases in different colours, yellow for the traffic phase S and red for the traffic phase J. Horizontal lines mark the locations of detectors, vertical lines the elapsed in sections of 15 minutes. The temporal resolution is 1 minute in accordance with the intervals of the detector measurements. The locations 1 and 2 in the diagrams mark the approimate position of the bottlenecks. The bottlenecks are mostly located in the proimity of an intersection where the respective detectors could measure the first phase transition from free to synchronized traffic. In the following two selected spatial-temporal diagrams of traffic congestions recognized and reconstructed with ASDA/FOTO on freeways in North Rhine-Westphalia are presented. The periods during which the analyses were performed are 9 th July 16 th July 27 as well as 29 th Aug. 1 th Sept. 27. The freeway A4-East showed two phase transitions from free to synchronized traffic on 31 st Aug. 27 as outlined in Fig. 5. The first transition occurred shortly before 12:45h at 1, the second one before 13:h at 2. oth of them remained stationary over a longer at the respective bottlenecks. Upstream of these locations areas of synchronized traffic are formed epanding to a maimum of approimately 5 km. In the traffic pattern located at 1 no wide moving jams emerged, hence the pattern remained of type Synchronized Pattern (SP). In the traffic pattern located at 2 no wide moving jams propagating further upstream occurred, too. However, some "narrow" non propagating jams emerged. oth traffic patterns remained independent of each other. The situation of freeway A45-North on the 31 st Aug. 27 with the emergence of a long persisting epanded traffic pattern is shown in Fig. 6: for almost 7 hours and over a distance of up to 17 km several wide moving jams emerged which propagated in some cases over a distance of up to 15 km. The EP emergence is at least influenced by two bottlenecks 1 and 2. Several wide moving jams, which emerged at 1 propagate through the bottleneck 2 without changing their shape and progression. In addition, new wide moving jams emerge at 2. Such an interaction between two bottlenecks is an eample of the complicated interrelations in spatial-temporal congested traffic patterns. They can be eplained by Kerner s Three-Phase traffic theory and ASDA/FOTO allows us to recognize and to predict those patterns in an on-line system environment (km) : 13:3 14: 14:3 15: 15:3 16: 16:3 (km) :3 17: 17:3 18: 18:3 19: 19:3 2: Figure 6: ASDA/FOTO result for A45-North on 31 st Aug : Schwerte-Ergste, 2: Hagen. In a similar way as in equation (1) for RDS/TMC messages it is possible to determine a travel for road sections by using the results of the ASDA/FOTO database: Lfree LJ L T S total = + + (2) v v v free J S 1 with v free = 12km / h (average free flow estimated based 2

5 on WDR data), respective lengths of the events v J = 15km / h and v S as result of FOTO and L free, L S and L J as three parts of the entire road section. The average speed within traffic phase J is estimated to be 15km/h. A distribution of travel based on the traffic patterns shown in Fig. 6 is outlined in Fig. 7. In free traffic the road user needs a of 1 minutes (dashed line in Fig. 7) for a distance of 2 km which increases up three s because of the congested traffic patterns. It would be possible to communicate travel or loss s to road users and navigation systems by using various communication channels Travel Time (min) Free Traffic Congested Traffic 13: 14: 15: 16: 17: 18: 19: 2: Figure 7: ASDA/FOTO travel s for A45-North on 31 st Aug. 27 for the congested pattern shown in Fig. 6. VI. ASDA/FOTO INTEGRATION INTO VEHICLES An additional feature of ASDA and FOTO models is the prediction of the further development of traffic congestion based on assumptions about future data. Those data can either be gained from averaged historical values, regression of historical values or variation curves of local measurements for specific detectors on freeways [5]. A solely traffic pattern based prediction of future traffic congestion which allows to pick the best fitting and thus most likely future congested pattern for a given traffic congestion from a pattern database is another approach [5]. The quality inde takes into account the online-system environment of ASDA/FOTO: therefore, quality inde is a measure which is calculated based on past predictions in comparison to traffic reality and as close as possible to current. In order to define a quality inde for prediction as relative percentage of two series of delay s, a trend of prediction can be defined as 1 prog( n ( ) ref t t ) n loss loss prog ( t) n Loss = 1 (3) ref t loss ased on current real loss tloss ref the n (often set to 3 min) future predictions of loss prog( n t ) loss, usually in prog intervals of one minute, are used to calculate Loss ( t). This can be performed online for each prediction cycle. real( n) After n cycles the real loss s tloss are set in relation to t ref real loss in order to get the real trend Loss ( t) of the loss progression, defined as 1 real( n ( ) ref t t ) n loss loss real ( t) n Loss = 1 ref t loss (4) In both equation (3) and (4) only values with t ref 5 loss > are used for calculation of the trend, because in most cases small travel losses below that value dissolve too soon to justify any trend prediction. The resulting travel trends Trend prog ( t), Trend real ( t) are mapped to three trend classes as outlined below: > 3 Increase (5) Loss Loss < 3 Decrease (6) 3 Loss 3 Const (7) The trend classes associated with each predicted and each real trend for a specific interval t are related to each other as shown in equation (8), in order to determine the quality of the prediction over the last n cycles, whereas each match between a prediction and a current value is counted as a hit. HITS counts cases where Trendprog = Trendreal (8) HITS Quality = 1 n The quality inde is epressed as a percentage and can be used by broadcasting stations to associate each current RDS/TMC message which contains a prediction of future traffic congestions with information on the prediction quality. Vehicles receiving those messages would then be able to rate the quality of the received information. This is of importance when vehicles operate in a setting, where they are able to receive traffic information from various sources. Such a cooperative environment in accordance to the CVIS project [16] is shown in Fig. 8. In addition, to the already eisting I2C communication channel RDS/TMC also other communication channels like C2C and C2I are integrated (Fig. 9). Figure 8: System architecture for vehicles in a cooperative environment

6 Figure 9: Single vehicle system architecture Vehicles may include a local traffic information system, providing information about current traffic condition as well as traffic congestion predictions. The information can be used by other vehicle systems like navigation, information, driver assistance and safety to realize enhanced driver awareness functions and to allow other vehicle systems to react on current traffic conditions. Hence each vehicle generates a local view on the traffic congestions based on its unique position in and space. Each vehicle performs a data fusion of infrastructure data, other vehicles and local vehicle data provided by electronic control units on board. The data fusion is affected by historical data, heuristics and variation curves. In this environment a quality inde associated with incoming messages is an important measure for the overall data fusion process. VII. SUMMARY The on-line operation of models ASDA and FOTO in the freeway network of North Rhine-Westphalia confirms the results previously obtained in Hessen and the USA (see [5], [1]-[12]). Due to the very large freeway net in the Ruhr district with dense populated areas and many bottlenecks, in most cases epanded spatial-temporal traffic patterns (EP) spreading over several bottlenecks arise. The standardized traffic data interface makes it possible for a traffic model such as ASDA/FOTO to realize an online operation in the road network with a dynamic import of large amounts of traffic data. Currently, over 19 freeway kilometres with more than 5 stationary loop detectors are analyzed on-line with ASDA/FOTO at WDR traffic centre. The application is able to compute its results considerable below one minute without any calibration of model parameters. The net steps include completion of freeway network and development of an ASDA/FOTO enhancement able to automatically generate RDS/TMC messages, so that WDR can validate the traffic messages received from other information sources against the source ASDA/FOTO. In addition, traffic prediction - for eample prediction of the future trend of travel for a specific road section - is of great value for traffic message providers. A quality inde associated with each predicted value allows vehicles a better interpretation of traffic messages, especially in a cooperative system environment. Aspects of drive overreactions on predictive information (e.g., [17]) could be possibly integrated in series databases. Further developments of the ASDA/FOTO application in North Rhine-Westphalia are ongoing. eside the current installation at WDR, an additional installation in the traffic control centres of the federal state North Rhine-Westphalia is scheduled for 28 in order to support the interpretation of traffic data at that centre, too. In addition, an integration of ASDA/FOTO in a cooperative vehicle environment is a challenging research topic for the coming years. VIII. ACKNOWLEDGEMENTS We would like to thank WDR for the project in North Rhine-Westphalia. In addition, we thank Andreas Haug and oris Kerner for many fruitful discussions. REFERENCES [1] M. Cremer (1979) Traffic Flow on Freeways (in German). (Springer- Verlag, erlin). [2] R. Wiedemann (1974) Simulation of Traffic Flow (in German). Schriftenreihe Instituts für Verkehrswesen, No. 8, University Karlsruhe. [3] A. D. May (199) Traffic flow fundamentals (Pren. Hall, New Jersey). [4] W. Leutzbach (1988) Introduction to the theory of traffic flow, (Springer Verlag erlin). [5]. S. Kerner (24) The Physics of Traffic (Springer, erlin, N. York). [6] [7] H. Rehborn, A. Haug, M. Aleksic,. S. Kerner, and U. Fastenrath (22) Statistical analysis of traffic messages as decision support (in German). Straßenverkehrstechnik, No. 9, pp [8]. S. Kerner, H. Rehborn, M. Aleksic, and A. Haug (24) Recognition and Tracing of Spatial-Temporal Congested Traffic Patterns on Freeways. Transportation Research C, 12, pp [9]. S. Kerner, H. Rehborn, and H. Kirschfink, German patent DE C2; US-patent US (1998);. S. Kerner and H. Rehborn, German patent publication DE A1 (1998);. S. Kerner, German patent DE C2, USA: US ; Japan: JP (1999);. S. Kerner, M. Aleksic, and U. Denneler, German patent DE C1 (1999). [1]. S. Kerner, H. Rehborn, M. Aleksic, and A. Haug (21) Methods for Tracing and Forecasting of Congested Traffic Patterns on Highways Traffic. Traffic Engineering and Control, 42, No. 9, pp [11]. S. Kerner, H. Rehborn, M. Aleksic, A. Haug, and R. Lange (21) On-line automatic tracing and forecasting of traffic patterns with ASDA and FOTO. Traffic Engineering and Control, 42, No. 11, pp [12]. S. Kerner, H. Rehborn, A. Haug, and I. Maiwald-Hiller (25) Tracing of congested traffic patterns in California. Traffic Engineering and Control, 46, No. 11, pp [13] H. C. Kniss (2) Evaluation of ASDA/FOTO in traffic control centre Hessen (internal report, in German). [14] H. Rehborn, A. Haug,. S. Kerner, M. Aleksic, and U. Fastenrath (23) Floating Car Data and recognition of spatiotemporal traffic patterns (in German). Straßenverkehrstechnik, No. 9, pp [15] [16] [17] M. en-akiva, A. de Palma, and I. Kaysi (1991) Dynamic network models and driver information systems. Transportation Research, Vol. 25A, No. 5, pp