The application of microscopic activity based travel demand modelling in large scale simulations Georg Hertkorn, Peter Wagner georg.hertkorn@dlr.de, peter.wagner@dlr.de German Aerospace Centre Deutsches Zentrum für Luft- und Raumfahrt Institute of Transport Research Institut für Verkehrsforschung Rutherfordstrasse 2 Rutherfordstraße 2 12489 Berlin-Adlershof 12489 Berlin-Adlershof Germany Deutschland Institut für Verkehrsforschung/Institute of Transport Research 1
Outline Approach for traffic flow simulation Approach for travel demand estimation Results of a case study: Cologne Results of a brigde blockade scenario Conclusions Institut für Verkehrsforschung/Institute of Transport Research 2
Model traffic systems Evaluate...... traffic system performance,... infrastructure projects, age pyramid (german population)... travel management measures (tolls, congestion pricing, parking restrictions),... changes in society (life styles, demographic structure). Institut für Verkehrsforschung/Institute of Transport Research 3
Background Dynamic features of traffic need to be considered traffic flow itself Flow depends on the temporal variation of traffic load (e.g. spill back, upstream propagation of jams). Demand varies with time (time of day, day, season,...). x [km] t Institut für Verkehrsforschung/Institute of Transport Research 4
Background Traffic demand depends on the travel times experienced by the travellers. (Close this feedback loop.) network load travel demand network performance travel times Traffic is caused by the desire of people to perform out-of-home activities (activity-based approach). Activities are not planned independently from each other. Microscopic scale Activity-based travel demand Efficient procedures Institut für Verkehrsforschung/Institute of Transport Research 5
Traffic flow simulation Use a (fast) queueing model for traffic flow simulation Each lane of the network is a FIFO queue, with limited storage capacity. A vehicle has to stay at least T = L v in the queue. The time to enter the next queue is given by a minimal (service) time or depends on the traffic states in the queues. L Institut für Verkehrsforschung/Institute of Transport Research 6
Travel demand Derive travel demand from observed activity patterns. No "behavioural theory of time allocation" included. Schedules are consistent. Model has to handle temporal shifts due to local conditions compared to original data Establish classification of activity patterns other other work travel at at home Full-time 1 other work travel at at home Active Leisure 2 0 3 6 9 12 15 18 21 24 0 3 6 9 12 15 18 21 24 Full-time 2 Active Leisure 4 other work travel at at home 0 3 6 9 12 15 18 21 24 other work travel at at home 0 3 6 9 12 15 18 21 24 Institut für Verkehrsforschung/Institute of Transport Research 7
Handle time shifts Derive measure for the temporal flexibility of episodes: determine variation in starting time and duration of similar episodes in the same class of diaries. variation in starting time and duration α β 0 3 6 9 12 15 18 21 24 time of day Alternative: Determine flexibility according to activity characteristics, e.g. starting time and duration of a film. Institut für Verkehrsforschung/Institute of Transport Research 8
Arrive at feasible diaries Equilibrate stress for the schedule as a whole. Compare total stress to a given threshold value. Choose new locations and modes if stress is too high. 0:00 activities sleeping breakfast trip work trip theatre trip travel times sleeping breakfast trip work trip theatre trip balance stress sleeping breakfast trip work trip theatre trip stress per episode: s i = α i ( t i ) 2 + β i ( d i ) 2 t i : difference of starting times (new - original) :difference of duration sleeping sleeping sleeping 24:00 Institut für Verkehrsforschung/Institute of Transport Research 9 time of day d i αβ, :episode specific parameters minimize total stress: i S = s i
Locations and modes first step second step third step Institut für Verkehrsforschung/Institute of Transport Research 10 level level Establish a hierarchical ordering among the episodes of a tour, determine location and mode for the episode on the highest level, determine locations and modes for the episodes on the following levels according to locations and modes already set. level level Locations: Model of intervening opportunities. Respect capacities for certain activities (payed work, school) Modes: Decision tree based on empirical data (CHAID-algorithm), Check how many cars are still available in the household. hierarchy of episodes time of day episodes at home
The City of Cologne population [1/ha] 1-7.5 7.5-15 15 30 30 75 75-150 Automobile [1/ha] 0-2.5 2.5-5 5-10 10 20 20-40 car density [1/ha] 5 0 5 10 15 Kilometers population density [1/ha] cars per capita Institut für Verkehrsforschung/Institute of Transport Research 11
City of Cologne: Locations locations [1/ha] 0-20 20 60 60-200 200 600 600-1800 workplaces [1/ha] 0 5 5-15 15 45 45-150 150-300 density of locations [1/ha] density of work places [1/ha] Institut für Verkehrsforschung/Institute of Transport Research 12
Trip length distribution work shopping 0.16 0.14 0.12 0.3 0.25 0.2 0.15 0.1 0.05 0 MOP MiD TAPAS 0 5 10 15 20 25 30 distance [km] Institut für Verkehrsforschung/Institute of Transport Research 13 rel. frequency 0.1 0.08 0.06 0.04 rel. frequency 0.02 0 MOP MiD TAPAS 0 5 10 15 20 25 30 distance [km] ZBE: Time budget survey MiD: Mobility in Germany 2002 TAPAS: Simulation
Destinations by quarters Altstadt Süd 0-0.1% 0.1 0.4% 0.4 2.0% 2.0 8.0% 8.0-40.0% Chorweiler 0-0.1% 0.1 0.4% 0.4 2.0% 2.0 8.0% 8.0-40% Altstadt-Süd (35% internal) Chorweiler (17% internal) Destination of trips for different home locations: share by quarters Institut für Verkehrsforschung/Institute of Transport Research 14
Average distance per trip work shopping Institut für Verkehrsforschung/Institute of Transport Research 15
Average travel time per trip work shopping Institut für Verkehrsforschung/Institute of Transport Research 16
Compare the situation on the different sides of River Rhine home left trips per person 3.84 3.80 share of car trips (driver) 35% 38% distance per person and day 15.5 km 16.9 km travel time per person and day 65 min 66 min avg. trip length (car) 5.3 km 5.8 km home right home right trips per person 3.73 3.72 share of car trips (driver) 40% 40% distance per person and day 20.5 km 21.0 km travel time per person and day 70 min 70 min avg. trip length (car) 6.9 km 7.0 km without city centre without city centre Institut für Verkehrsforschung/Institute of Transport Research 17
Scenario: Deutzer Brücke open/blocked open [veh/day] 0-3000 3000 6000 6000-12000 12000 24000 24000-60000 Institut für Verkehrsforschung/Institute of Transport Research 18
Differences in simulated traffic flow less flow more flow dneg [veh/day] 0 1-1000 1000 2000 2000 4000 4000 8000 8000-16000 dpos [veh/day] 0 1-1000 1000 2000 2000 4000 4000 8000 8000-16000 Institut für Verkehrsforschung/Institute of Transport Research 19
Number of trips crossing the Rhine home left bridge open home right direction trips [10 3 ] share [%] direction trips [10 3 ] share [%] left side 1987 94.6 left to right 54 2.6 right to left 53 2.5 right side 6 0.3 left side 37 3.2 left to right 151 13.0 right to left 153 13.2 right side 818 70.6 bridge blocked direction trips [10 3 ] share [%] direction trips [10 3 ] share [%] left side 1941 94.9 left to right 50 2.4 right to left 48 2.3 right side 6 0.3 left side 29 2.5 left to right 126 11.2 right to left 127 11.3 right side 853 75.1 Institut für Verkehrsforschung/Institute of Transport Research 20
Conclusions Travel demand can be estimated from consistent activity patterns in an efficient way. dynamic travel demand for a working day Simulation results are sensitive to local traffic conditions. Travel demand characteristics vary on a small spatial scale. Outlook Better empirical data of tour formation and changes in activity patterns under various (spatial) conditions needed. Include surrounding districts for incoming/outgoing traffic. Institut für Verkehrsforschung/Institute of Transport Research 21
Thank you! Institut für Verkehrsforschung/Institute of Transport Research 22
Diary data Time use survey of the Federal Statistical Office in Germany (1991/1992): Sample: 7,200 households with a German head of the household. Each member was asked to fill in two diaries for consecutive days. Time interval: 5 minutes. Activity catalogue: free description, coded with a set of 231 activity types. Data element (diary): sequence of 288 activity codes Additional variables: location, parallel activities, presence of other persons, socio-demographic variables of the individuals, regional data. Repetition: 2001/2002 Restriction to (Tuesday, Wednesday, Thursday), elimination of incosistent patterns: 14 000 patterns Institut für Verkehrsforschung/Institute of Transport Research 23
Activity sequencing in the diary classes 24 other work travel at home Full-time 1 0 3 6 9 12 15 18 21 24 A 21 18 15 12 other work travel at home Full-time 2 0 3 6 9 12 15 18 21 24 time of day D A D 9 6 3 0 Institut für Verkehrsforschung/Institute of Transport Research 24
Determine the rigidity of starting times Classify diaries according to their structure (hierachical clustering algorithm). Compare episodes to corresponding episodes of diaries in the same class. The paramters 0 3 6 9 12 15 18 21 24 time of day television hobby leisure at home care leisure not at home school work shopping trips housework eating at home sleeping Institut für Verkehrsforschung/Institute of Transport Research 25......
Comparison of episodes Weighting functions dependent on differences in the starting time and duration. 0 3 6 9 12 15 18 21 24 time of day television hobby leisure at home care leisure not at home school work shopping trips housework eating at home sleeping Institut für Verkehrsforschung/Institute of Transport Research 26
Evaluation of the schedules Set the duration of trips according to time dependent travel times. Adjust starting times: minimize costs for the whole schedule: costs per episode: ux ( 1, x 2 ) = α 1 x 1 s 1 ( ) 2 + β 1 x 2 x 1 ( d 1 ) 2 Time shifts propagate in both directions. Compare the total costs with some threshold value and eventually reject the schedule. Institut für Verkehrsforschung/Institute of Transport Research 27
Location choice (intervening opportunities) The set of alternatives is ordered by travel times (requires preliminary mode choice). A location is selected according to f ( k, q) = q k 1 ( 1 q) ; F( i, q) = Pk ( iq, ) = f ( k, q ) = 1 q i i k = 1 q = q( A, a, s) activity age gender (activity catalogue) Institut für Verkehrsforschung/Institute of Transport Research 28
Conclusions Diary data offer the opportunity to estimate travel demand for status quo scenarios in a reliable and efficient way. Restrictions: People react by using known time use patterns. The model can be coupled with synthetic pattern modelling and help to discern the effect of each modelling step. Moderate computing time facilitates the integration in feed back loops (e.g. traffic flow modelling). Consistent patterns are required: Some effort is needed for consistency checks or modifications. The adaptation to the local situation is a crucial step: A quadratic cost function was proposed. Institut für Verkehrsforschung/Institute of Transport Research 29
Further validation and investigation of the interplay of parameters is planned. Institut für Verkehrsforschung/Institute of Transport Research 30
Institut für Verkehrsforschung/Institute of Transport Research 31
Institut für Verkehrsforschung/Institute of Transport Research 32