Travel Time Estimation from Multiple Data Sources

Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 1 Travel Time Estimation Travel Time Estimation from Multiple Data Sources F. Soriguera, D. Abeijon, D. Rosas, L. Thorson and F. Robuste Submission date: February 1st, 2007 4,259 Words 1 Table 10 Figures F. Soriguera (corresponding author), CENIT – Center for Innovation in Transport, Technical University of Catalonia (UPC), Jordi Girona 29, 2-A, 08034 Barcelona (Spain), Tel:+34934137667, Fax:+34934137675, [email protected]. D. Abeijon, CENIT – Center for Innovation in Transport, Technical University of Catalonia (UPC), Jordi Girona 29, 2-A, 08034 Barcelona (Spain), Tel:+34934137667, Fax:+34934137675, [email protected] D. Rosas, CENIT – Center for Innovation in Transport, Technical University of Catalonia (UPC), Jordi Girona 29, 2-A, 08034 Barcelona (Spain), Tel:+34934137667, Fax:+34934137675, [email protected]. Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 2 L. Thorson, CENIT – Center for Innovation in Transport, Technical University of Catalonia (UPC), Jordi Girona 29, 2-A, 08034 Barcelona (Spain), Tel:+34934137667, Fax:+34934137675, [email protected]. F. Robuste, School of Civil Engineering, Technical University of Catalonia (UPC), Jordi Girona 1-3, B1, 08034 Barcelona (Spain), Tel:+34934017104, Fax:+34934017264, [email protected] Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 3 ABSTRACT Travel time is the best indicator of the level of service in a road link, and perhaps the most important variable for measuring congestion. This paper presents a method for estimating accurate travel times in toll highways using data from multiple sources, as loop detectors and toll tickets. The proposed methodology consists of a data fusion technique using different travel time estimations in order to obtain a more accurate fused value with less error than individual estimations by itself. Finally results obtained in the application of the methodology to the AP-7 highway, near Barcelona in Spain, are presented. ACKNOWLEDGMENTS This research was performed within a contract managed by Mr. Josep A. Grau and in cooperation with ABERTIS, the Spanish leading private transportation and communications infrastructures management corporation. The contents of this report reflect the views of the authors who are responsible for the facts and accuracy of the data presented herein. ARTICLE TITLE Abbreviated title: Travel Time Estimation Full Title: Travel Time Estimation from Multiple Data Sources Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 4 1. INTRODUCTION Europe is involved in a restrain process in the construction of new road infrastructures. This limitation is due to environmental, land occupancy and budget restrictions. On the other hand, mobility demand is still increasing in all countries. This situation leads to congestion becoming an important problem. The road network management is then essential to optimize the usage of available infrastructures. Travel time and travel time reliability are key factors in road management systems, as they are the best indicators of the level of service in a road link, and perhaps the most important parameter for measuring congestion [1]. Travel time estimation is necessary to assess the operational management and planning of a road network. Moreover, travel time information is the best and most appreciated traffic information for road users. Most European countries (Spain, France, Denmark, Italy, Finland, United Kingdom, Sweden, the Netherlands, Norway and Germany) consider travel time as an emerging issue in Europe and acknowledge a growing demand for real time travel time information. Most of the Trans-European Road Network (TERN) in these countries is now covered by a travel time project [2]. From these experiences it can be concluded that the deployment of a travel time measurement system in a road network is a complex issue. There is not a unique methodology to be applied in all cases, due to different road characteristics, different traffic flow patterns, different surveillance equipments, budget and information limitations, etc. The key factor is to obtain the maximum accuracy from the available data. In this situation is where data fusion techniques play an important role. Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 5 Basically there are two methodologies to measure travel time in a road link: the direct measure and the indirect estimation. The direct travel time measure is based on measuring the time interval that a particular vehicle takes to travel from one point to another. The data collection techniques used in this case are the floating car data, the license plate matching or the AVI (Automated Vehicle Identification) from toll infrastructure. Travel time data is directly obtained from these measures. The alternative is the indirect travel time estimation from traffic flow characteristics (density, flow and speed), obtained from magnetic loop detectors. To obtain travel time data from these measurements some type of algorithm must be applied. Because of the growing interest in measuring travel time, there have been several studies attempting to determine link travel times in a road segment. Previous research on indirect travel time estimation focuses on loop detector data, because loop detectors are currently the most widely used detection technology. Some of these methods are based on the speed estimation in detection points. These algorithms have been proven to be quite effective in road segments with high density of detection points and free flow traffic conditions, but they do not perform well in roads with a lower surveillance level or under heavily congested situations. Several authors have tried to develop better algorithms to overcome these limitations and problems in the spot speed estimation. [2-6]. A different approach to determine the link travel time from loop detector data is presented in [7], where a cumulative flow balance algorithm at successive detector sites is proposed. The counter detector drift is the problem in this case. Other studies, [8-10] focus their efforts on obtaining the detailed signature of a particular vehicle when crossing an inductive loop, so that this vehicle can be reidentified downstream. Another strategy in the identification of vehicles using loop detectors is the platoon recognition [11, 12]. These methods propose the Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 6 estimation of average travel time by matching unique features of vehicle platoons such as the position and/or distribution of vehicle gaps or unique vehicles. Regarding direct measure of travel time, many advanced techniques have been applied [13]. For example [14-17] analyze the direct measure of travel time using AVI systems or [18] using video license plate matching. Other applied techniques are the electronic distancemeasuring, automatic vehicle location, cellular phone tracking, video imaging, and so on. This paper proposes an accurate travel time estimation method using multiple data sources. A spot speed algorithm and a cumulative flow algorithm are used to estimate travel times from loop detector data. Moreover a simple algorithm for estimating travel times in toll highways is developed: the algorithm uses the travel time data included in the toll tickets for different routes in order to estimate the section travel time between consecutive entry and exit ramps. The paper is organized as follows: Section 2 describes the methodologies to obtain travel times from loops and section 3 presents the direct measurement travel time algorithm using toll ticket data. The main ideas of the data fusion method are outlined in section 4, and finally in section 5 a review of travel time dissemination technologies and some results of the application of the model to the AP-7 highway in Spain are presented. The paper finishes with some general conclusions. 2. TRAVEL TIME ESTIMATION FROM LOOP DETECTOR DATA Indirect travel time estimation is based in the measurement of fundamental traffic flow variables (flow, speed and density) in a particular spot of a highway and the extrapolation of these point measurements to a stretch of the highway. These fundamental variables capture the whole Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 7 physical traffic process, and so it should be possible to determine any other variable from them. Loop detectors are the most widely spread technology to collect flow, speed and density of traffic. Take into account that single detectors only collect flow, and sped and density must be approximated. On the other hand dual loop detectors are capable of measure all traffic flow variables. Travel time estimation from loop detector data is based in two basic methodologies. First and most widely used algorithm is the spot speed algorithm. The alternative consists on a cumulative flow algorithm. 2.1. Spot Speed Algorithm As stated earlier this method is based in the speed measurement in a highway section by means of single or dual loop detectors. Travel time could then be obtained by simply applying the following equation: ),( ),( ti i ti v l t= (1) Where: li is the length of the highway stretch considered to be associated with loop detector “i”. v(i,t) is the average 5 minutes speed measured in loop detector “i” and time interval “t”. t (i,t) is the average 5 minutes travel time in the highway stretch “i” and time interval “t”. The hypothesis considered in the application of this algorithm is that traffic flow characteristics maintain constant in the whole stretch and in the whole time period. This means Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 8 that for the algorithm to be effective a high surveillance density (loop detectors every 500m) and a frequent actualization of parameters (every 5 minutes) are needed. Moreover, in highly congested highways with frequent stop and go situations, travel time estimation using this algorithm can be very different from reality. To smooth this problem a calibration method is proposed. This calibration is based in averaging the measured speed in loop detector “i” and time interval “t” with previous measured speeds, in time and space. Then the spot speed travel time algorithm is finally formulated as follows: ),( ),( ti i ti V l t= (2) ),(),1()1,(),( titititi vvVV ⋅⋅= −− (3) Where: V(i,t) is the calibrated spot speed in loop detector “i” and time interval “t”. # FIGURE 1 Spot speed algorithm required surveillance configuration # 2.2. Cumulative Flow Balance Algorithm Due to the required high surveillance density and the lack of accuracy of the spot speed algorithm in congested situations, an alternative travel time estimation methodology is proposed. The cumulative flow balance algorithm estimates travel time directly from loop detector flow measurement, without the previous imprecise calculation of speed. The algorithm uses the entrance and exit flows in the highway stretch to calculate the travel time using a simple flow balance method. The algorithm responds to the following equation: () () dtqqSS t ttiotiititi ∫− −−+= 1,,)1,(),( )( (4) Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 9 Where: S(i,t) is the total cumulated vehicles in the highway stretch “i” and in the time interval “t”. qi(i,t) is the entering flow in the highway stretch “i” and in the time interval “t”. qo(i,t) is the exiting flow in the highway stretch “i” and in the time interval “t”. Finally, travel time is calculated using equation (5): ),( ),( ),( tio ti ti Q S t= (5) Where: Qo(i,t) is the output flow of the main highway trunk in stretch “i” and in the time interval “t”. Obviously, to apply this algorithm, all the highway ramps must be equipped with loop detector units. The surveillance scheme required is displayed in figure 2. # FIGURE 2 Cumulative flow balance algorithm required surveillance configuration. # This algorithm considers traffic as a continuous flow. Therefore the algorithm will be more efficient under heavy flow situations, and should not be applied when traffic flow is lower than 1500 veh/h/l (i.e. Qo(i,t) > 125 veh/5min/l). Note that to evaluate equation (4) it is necessary to establish an initial value for “S0”. This initial value is obtained from the average measured traffic density in the 5 minutes previous to the first 5 minutes time interval with a traffic flow “q” higher than 125 veh/5min/l. iii lKS ⋅ =−)1,()0,( (6) )1,()1,()1,( −−− ⋅= ioiii kkK (7) Where: K(i,-1) is the calibrated traffic density of highway stretch “i” in the 5 minutes time Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 16 3-m ,...,0 2 )( 3 2 ,1, )1,( =∀ − = ∑ + += + +i tt t i ij jiji iis (19) mmmmmms ttt ,1,2)1,2( −−−− − = for the last but one section (20) Moreover it must be taken into account that the last section travel time “ mm t,1−” cannot be split into the main highway trunk travel time and the exit time with the proposed algorithm. Nevertheless, this lack of information is not so important, because the last toll plaza is usually located in the main highway trunk and all the vehicles traveling along the last stretch must go through this toll plaza. In such a way, the interesting information for the driver in this last stretch of the highway is the total aggregated travel time, including both the main trunk travel time and the exit time. On the other hand, the exit time in the last toll plaza would be useful information for the highway operator in order to determine the level of service of this last toll plaza. 4. DATA FUSION FROM MULTIPLE DATA SOURCES To improve the performance of the three previously presented algorithms, a data fusion system is proposed. Each one of the travel time estimation methods imply a margin of error, related to the sensor technologies and to the implemented algorithms. The performance of the sensors depends on the variable that is being monitored and on the traffic flow. The algorithm error depends also on the traffic flow and on the type of travel time estimation used (RTT - Reconstructed Travel Time or ITT -Instantaneous Travel Time). Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 17 4.1. Data Fusion techniques Data fusion is a compendium of different mathematical techniques to improve the robustness of a travel time estimation system, improving also its reliability and achieving a uniform temporal and spatial coverage [20]. The fusion operators used are Context Independent [20] and use probabilistic, evidential or fuzzy logic [21]. Two different fusion procedures will be performed. The first one fuses the ITT values obtained from both algorithms related to loop detector data. The second, fuses the PTTs (Predicted Travel Time) obtained from the ITT (loop detectors) and the RTT (toll tickets). # FIGURE 7 Data fusion flow chart. # 4.2. First Fusion This first data fusion process uses a fuzzy logic technique to determine the congestion state, using the speed and it variance as references. The temporal mean speed decreases in congestion conditions while its variance increases. The use of fuzzy logic represents the probability functions of these statements to be true (see Fig 8) # FIGURE 8 Probability functions for the binomial flow state characterization. # The fuzzy logic operator is independent of context, and it increases the accuracy of the determination of the traffic flow state in this binomial characterization (congestion or not). Once the congestion state is determined, the results (ITT1 and ITT2) from the spot speed algorithm and the cumulative flow balance algorithm are fused using a weighted average Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 18 operator. The weight applied to each travel time estimation depends on its associated error. This error is related to the traffic flow condition: congestion or free flow. The result of this first fusion is a more consistent and reliable ITT. This fusion could be replaced by another one fusing each PTT obtained from every single ITT, but this would require more computational effort to obtain a similar fused PTT. 4.3. Second Fusion In this section the performance of some different data fusion operators are analyzed. These are probabilistic, evidential and fuzzy logic operators. All the algorithms are independent of the context. The probabilistic approach uses the Bayesian Theory: () ( ) ( ) ( ) ()() 21 12 21 || ,| PTTpPTTp PTTpPTTPTTpPTTPTTp PTTPTTPTTp iii i= (21) zai ,...,= Where: i PTT represents the result of the Data Fusion 1 PTT represents the PTT obtained from the Loop Data i PTT represents the PTT obtained from the Ticket Data The a priori probability () i PTTp is supposed to be equal to unit. This hypothesis does not give any extra information, but it will improve with the historical data of the road. The probability functions ( ) ij PTTPTTp | are determined during the training of the algorithm depending on the validation results. The evidential approach uses the generalization of the Bayes Theory, implementing the Dempster Rule. Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 19 Finally the fuzzy logic uses a Context Independent Operator to fuse the probability of every independent PTT to belong to any of the fused PTT sets (Fig 9). The higher probability set is chosen as the right one. # FIGURE 9 Probability functions of every independent PTT to belong to any fused sets. # 5. MODEL RESULTS AND TRAVEL TIME DISSEMINATION The proposed travel time estimation method from multiple data sources is currently being applied in the AP-7 toll highway in Spain. The AP-7 highway runs along the Mediterranean cost corridor, from the French border to the Gibraltar strait. Nevertheless, the pilot test is restricted to the north east stretch of the highway from “La Roca del Vallès” toll plaza, near Barcelona, to the French border at “La Jonquera”. This stretch is approximately 120 km long. Figure 10 provides a graphical summary of the average speeds resulting from the calculated single section travel times. The figure provides average speeds for both directions and for three different time periods. It has to be taken into account that the speed limit on highways in Spain is 120 km/h, and the off peak time speed has been limited according to this legal speed. In this figure it can be observed that the most problematic period was the pm period in the southbound direction. This situation is a result of the massive return to Barcelona city on Sunday afternoon or evening, after spending a day or a weekend at the beach. Note that the average speeds in the single sections before a main highway toll plaza (in the two edges of the stretch), are very low. These low speeds are due to the inclusion of the exit time (i.e. the time required to pay the fee) in these final sections travel time. # FIGURE 10 Graphic summary of average speeds in AP-7 highway (July 10th, 2005 data). # Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 20 Once probed the accuracy of the proposed travel time estimation method, the dissemination of this traffic information to the highway users is the key factor. Traffic information dissemination techniques can be clearly divided in two main blocks: pre-trip information and on-trip information. Pre-trip information allows trip planning while on-trip information is useful in order to modify the initial planning according to current traffic conditions. For traffic information to be effective it must be short, concise, quantified and specifically addressed to the receptor. Travel time information itself fulfills the three first conditions, and the dissemination technology must fulfill the last one. Taking into account these criteria, a multicriterion analysis of different traffic information dissemination technologies has been performed (Table 1). # TABLE 1 Multicriterion Analysis of Different Traffic Information Dissemination Technologies # From table 1 results it is stated that car navigators and VMS are the technologies with higher dissemination potentialities. These results are in accordance to the current practices of Spanish operators who are installing VMS panels in most of the primary network. Moreover results also agree with user perceptions, as car navigation devices are currently best selling car items. 6. CONCLUSIONS AND FURTHER RESEARCH Link travel time is the most appreciated information for road users. The proposed approach for calculating travel times in toll highways using multiple data sources is a simple one and can be Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 21 easily put into practice with the existing infrastructure. The scheme uses data obtained from loop detectors and toll tickets on highways with a closed tolling system. The proposed data fusion method is capable of increasing the robustness and accuracy of each particular travel time estimation algorithm. The final travel time estimation relies on both loop detector data and toll ticket data, so that if one source of information is unavailable or the associated error is too high, the final travel time estimation is still available relying on the best available data. The results of the pilot test carried out on the AP-7 highway in Spain indicate the suitability of the method for the link travel time estimation in a closed toll highway system. Moreover, this accuracy in the travel time estimation should make the development of a robust incident detection system possible, by comparing the real time estimations to the recurrent travel times. Travel time prediction on the basis of the present scheme is also a key factor for future research. Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 22 REFERENCES 1. Palen, J. The Need for Surveillance in Intelligent Transportation Systems. Intellimotion, 6:1, pp. 1-3, 10. University of California PATH, Berkeley, CA, 1997. 2. Hopkin, J., D. Crawford, and I. Catling. Travel Time Estimation. Summary of the European Workshop organized by the SERTI project, Avignon, November 2001. 3. Dailey, D. Travel Time Estimation Using Cross Correlation Techniques. Transportation Research, 27B:2, 97-107,1993. 4. Petty, K. F., P. Bickel, M. Ostland, J. Rice, F. Schoenberg, J. Jiang, and Y Rotov. Accurate Estimation of Travel Times from Single Loop Detectors. Transportation Research, 32A, 1- 17, 1998. 5. Nam, D. H., and D. R. Drew. Automatic Measure of Traffic Variables for Intelligent Transportation Systems Applications. Transportation Research, 33B, 437-457, 1999. 6. Cortés, C. E., R. Lavanya, J. Oh, and R. Jayakrishnan. General-Purpose Methodology for Estimating Link Travel Time with Multiple-Point Detection of Traffic. Transportation Research Record, 1802, 181-189, TRB, National Research Council, Washington, D.C. 2002. 7. Coifman, B. Estimating Travel Times and Vehicle Trajectories on Freeways Using Dual Loop Detectors. Transportation Research, 36A:4, 351-364, 2002. 8. Westerman, M., and L. Immers. A Method for Determining Real-Time Travel Times on Motorways. Road Transport Informatics/Intelligent Vehicle Highways Systems, ISATA, 221- 228, 1992. 9. Coifman, B., and M. Cassidy. Vehicle Reidentification and Travel Time Measurement on Congested Freeways. Transportation Research, 36A:10, 899-917, 2002. Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 23 10. Kwon, J., B. Coifman, and P. Bickel. Day-to-Day Travel Time Trends and Travel Time Prediction from Loop-Detector Data. Transportation Research Record, 1717, 120-129. TRB, National Research Council, Washington, D.C., 2000. 11. Sun, C., G. Arr, and R. P. Ramachandran. Vehicle Reidentification as Method for Deriving Travel Time and Travel Time Distributions. Transportation Research Record, 1826, 25-31, TRB, National Research Council, Washington, D.C., 2003. 12. Gaur, A., and P. B. Mirchandani. Method for Real-Time Recognition of Vehicle Platoons. Transportation Research Record, 1748, 8-17. TRB, National Research Council, Washington, D.C., 2001. 13. Lucas, D. E., P. B. Mirchandani, and N. Verma. Online Travel Time Estimation Without Vehicle Identification. Transportation Research Record, 1867, 193-201. TRB, National Research Council, Washington, D.C., 2004. 14. Turner, S. M., W. L. Eisele, R. J. Benz, and D.J. Holdener. Travel Time Data Collection HandBook. Research Report FHWA-PL-98-035. Texas Transportation Institute, Texas A&M University System, College Park, 1998. 15. Dion, F., and H. Rakha. Estimating Spatial Travel Times Using Automatic Vehicle Identification Data. Presented at 82nd Annual Meeting of the Transportation Research Board, Washington, D.C., 2003. 16. Levine, S., and W. McCasland. Monitoring Freeway Traffic Conditions with Automatic Vehicle Identification Systems. ITE Journal, 64:3, 23-28, 1994. 17. Christiansen, I., and L. Hauer. Probing for Travel Time: Norway Applies AVI and WIM Technologies for Section Probe Data. Traffic Technology International, pp 41-44, UK & International Press, Surrey, Uk, Aug/Sep 1996. Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 24 18. Seki, S. Travel-Time Measurement and Provision System Using AVI Units. Proc. of 2nd World Congress on Intelligent Transport Systems, VERTIS, pp 50-55, 1995. 19. Cui, Y. and Q. Huang. Character Extraction of License Plates from Video. Proc.1997 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp. 351-356. IEEE, 1997. 20. Hall, D.L. and S.A.H. McMullen. Mathematical Techniques in Multisensor Data Fusion. Artech House Publishers, 2004. 21. Bloch, I. Information Combination Operators for Data Comparative Review with Classification. IEEE Transactions on Systems, Man and Cybernetics – Part A: Systems and Humans Vol. 26 nº 1, 1996. Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 25 TABLE 1 Multicriterion Analysis of Different Traffic Information Dissemination Technologies Mark Technology Description Pre- Trip On- Trip Specifically addressed On demand User Friendly 10 Car Navigator On vehicle device RDS-TMC √ √ √ X √ 10 VMS Variable Message Signs √ √ √ X √ 8 Radio broadcasts Radio bulletin √ √ X X √ 7,5 Cellular Phone Text services √ √ √ √ = 7 Phone Traffic Call Center √ √ √ √ X 2,5 TV Broadcasts TV bulletin √ X X X √ 2,5 Press Conflictive days announcements √ X X X √ 2,5 Internet Online services √ X √ √ √ 2 Information Points Service Area information √ √ √ √ XX Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 32 E=0 E=1 E=2 01 t 12 t 02 t Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 33 Loop Data Ticket Data Spot Speed Algorithm Cumulative Flow Balance Algorithm Travel time estimation for toll Ticket Data ITT1 ITT2 RTT Fusion 1 ITT PTT1 PTT2 Fusion 2 PTT Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 34 Prob. Speed Con g estion Free flow Prob. Variance Free flow Con g estion 1 1 Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 35 PTT 1 1 PTT a PTT b PTT c PTT d Prob. PTT 2 1 PTT a PTT b PTT c PTT d Prob. Soriguera, Abeijon, Rosas, Thorson & Robuste CENIT – Barcelona – Spain 36 BARCELONA 3 5 13120 120 120 Southbound LEGEND >50% delay in relation to free flow travel time >25% delay in relation to free flow travel time Peak Hour Speed (Km/h) 87 92 11586 69 78 120 120 120120 120 120 97 110 119120 120 120 103 107 112119 104 98 98 111 112120 93 80 89 96 120114 79 90 100 100 100117 44 102 120 120 120115 42 105 120 120 120120 49 114 119 108 119102 70 102 98 93 98110 85 110 35 44 11233 22 26 OFF PEAK PM AM N o r thbound AM PM OFF PEAK FRANCE