Skip to content

Traffic assignment (MATSim)

At a glance

Engine MATSim 2024.0, queue simulation, SwissRailRaptor transit router
Code assignment/
Road network OpenStreetMap via pt2matsim; 214,716 nodes, 474,037 links
Transit 18 GTFS feeds, one weekday, mapped with pt2matsim
Demand MAST weekday plans, 10 % sample, modes fixed
Choices Route, departure time
Runs 100 iterations; three independent chains, seeds 1–3

Each iteration, every agent executes one plan in a queue simulation; the plan is scored, and a share of agents changes its route or departure time. After 100 iterations the link volumes are read for validation.

1. Demand

Table 1. Demand passed to MATSim (2024, 10 % sample).

Item Value
Persons 262,842
— residents working in the study area 209,401
— residents working outside 25,290
— in-commuters 28,151
Legs 1,032,562
Modes car, ride, transit, walk, bike (fixed)
Not modelled trucks, commercial vehicles, visitors, through traffic

Table 2. Activity locations.

Activity Location Activities
Home OSM residential building in the home zonea all
Other, inside the study area OSM point of interest of the matching category in the chosen zoneb 590,509
Outside the study area Coordinate from MAST 82,891

a One building per household; a residential street point where the zone has no building.

b Drawn at random. Categories: work → work; school, escort → education; shopping → shopping; recreation → recreation; other → other. Fallback: other, then any POI of the zone.

MATSim attaches each activity to its nearest car link. The 1,036 activities (0.08 %) nearest to one of seven unreachable dead-end links are attached to the nearest connected link (pipeline/snap_to_network.py).

2. Road network

2.1 Conversion

Source: Geofabrik OpenStreetMap extract of Maryland (2026-06-11), clipped to the network area and converted with pt2matsim 24.4 to EPSG:26985.

curl -O https://download.geofabrik.de/north-america/us/maryland-latest.osm.pbf
java -cp pt2matsim-24.4-shaded.jar org.matsim.pt2matsim.run.CreateDefaultOsmConfig osm_convert_config.xml
java -Xmx8g -cp pt2matsim-24.4-shaded.jar org.matsim.pt2matsim.run.Osm2MultimodalNetwork osm_convert_config.xml

Table 3. Converter settings (osm_convert_config.xml).

Parameter Value
outputCoordinateSystem EPSG:26985
keepPaths false
maxLinkLength 500 m
keepTagsAsAttributes true
keepWaysWithPublicTransit true
parseTurnRestrictions false
scaleMaxSpeed false
routableSubnetwork car; car + bus

Table 4. Converter defaults for untagged ways.

OSM class Free speed (km/h) Capacity (veh/h/lane) Lanes One-way
motorway 120 2,000 2 yes
motorway_link 80 1,500 1 yes
trunk 80 2,000 2 no
trunk_link 50 1,500 1 no
primary 80 1,500 1 no
primary_link 60 1,500 1 no
secondary (_link) 30 1,000 1 no
tertiary (_link) 25 600 1 no

Tagged values take precedence; most motorways carry a 65 mph tag (105 km/h). Capacities are replaced by Tables 5 and 6.

2.2 Cleaning

Table 5. Corrections to the converted network.

Step Rule Source
Road class Upgraded to the MDOT class where MDOT rates a road highera MDOT SHA functional classification
Ramp lanes MDOT value where higher (18 ramps) MDOT SHA count inventory
I-695 capacity 2,197 veh/h/lane, basic; 2,043 at merges, diverges, lane drops; HCM formula at weavesb HCM
Other motorways 2,000 veh/h/lane converter
Trunk roads 1,500 veh/h/lane network audit
Ramp capacity 1,800–2,200 veh/h/lane by free speedc HCM
Exit storage + 300 m of storage on the first exit-ramp link HCM, ch. 38
Transit-only links 754 removed before transit mapping clean_network.py
Key Bridge removed (closed March 2024) MDTA
I-95 ETL extension removed (opened December 2024; 2026 network only) MDTA

a Nearest MDOT segment within 20 m and 30°; links are never downgraded.

b 2,350 pc/h/lane × heavy-vehicle factor 0.935; 0.93 queue-discharge factor at bottlenecks; all 604 mainline links.

c 2,200 (> 50 mph), 2,100 (40–50), 2,000 (30–40), 1,900 (20–30), 1,800 (< 20 mph).

2.3 Surface streets

Table 6. Free speed and capacity of primary and lower roads.

Item Rule
Posted speed OSM maxspeed, else MDOT sign, else local mediana
Free speed posted × 0.9 below 51 km/h; posted otherwise
Capacity (veh/h/lane) primary 1,500; secondary 800; tertiary, residential 600; living street 300
Link capacity per-lane capacity × max(lanes, 1)

a Median of the same road name and class in the same 5 km cell, then of the class in the cell, then of the class. Free speed and capacity rules follow the MATSim OSM reader.

No traffic count was used to set any speed, capacity or lane value.

2.4 Tolls

Baltimore Harbor and Fort McHenry tunnels: $3.00 per passage, both directions, six links (roadpricing_HARB_2024.xml).

3. Transit network

3.1 Feeds

Table 7. GTFS feeds.

Operator group Feeds
MTA local bus, Light RailLink, Metro SubwayLink, MARC, commuter bus
Baltimore region Charm City Circulator, Harbor Connector, Towson Loop, RTA, Anne Arundel County Transit, Annapolis Transit, Harford Transit LINK, Carroll Transit
Washington region WMATA Metrorail, WMATA Metrobus, Ride On, TheBus
Intercity Amtrak

Service day: Wednesday 21 October 2026 (last published Wednesday for expired feeds). Stops more than 250 m from the network are dropped.

3.2 Mapping

python3 assignment/transit/merge_gtfs.py <raw GTFS folder> network_2024.xml.gz gtfs_merged/
java -Xmx8g -cp pt2matsim-24.4-shaded.jar org.matsim.pt2matsim.run.Gtfs2TransitSchedule \
     gtfs_merged/ 20261021 EPSG:26985 schedule_unmapped.xml.gz transitVehicles_gtfs.xml.gz
java -Xmx9g -cp classes:pt2matsim-24.4-shaded.jar de.umd.transit.MapWithShapes \
     ptmapper_config_2024.xml gtfs_merged/shapes.txt EPSG:26985 300 100
java -Xmx8g -cp pt2matsim-24.4-shaded.jar org.matsim.pt2matsim.run.CheckMappedSchedulePlausibility \
     transitSchedule_2024.xml.gz network_2024_pt.xml.gz EPSG:26985 plausibility_2024/

Table 8. PTMapper settings (ptmapper_config_2024.xml).

Parameter Value
travelCostType travelTime
maxLinkCandidateDistance 90 m
nLinkThreshold 6
candidateDistanceMultiplier 1.2
routingWithCandidateDistance true
maxTravelCostFactor 5.0
scheduleFreespeedModes artificial, rail, subway
Shape corridor / weight distancea 300 m / 100 m

a Routes with a GTFS shape use only links within 300 m of it; link cost rises with distance from the shape.

Buses run on the car network, light rail and Baltimore Metro on their own links, MARC and Amtrak on rail links; Metrorail and ferries on artificial links.

Table 9. Mapped schedule.

Item Value
Lines 354
Routes 4,319
Departures 25,099
Stop facilities 16,687
Stops within 25 m of their link 91.7 %
Stops within 50 m of their link 96.2 %
Route length on artificial links 4.6 %
Car links opened to buses 38,040

3.3 Vehicles

Table 10. Transit vehicle types.

Type Operators Seats Standing Length (m) PCE
Local bus MTA, WMATA, county services 38 32 12.2 2.0
Commuter coach MTA 57 0 13.7 2.5
Light rail, 2 cars MTA 170 200 58 5.2
Metro, 4 cars MTA 304 360 92 4.4
Metrorail, 8 cars WMATA 512 880 183 4.4
Commuter rail MARC 780 0 180 27.1
Intercity rail Amtrak 550 0 200 27.1
Ferry Harbor Connector 50 0 15 1.0

Capacity and PCE are multiplied by the sample rate (0.1) in each run.

4. Simulation set-up

Values of the final 2024 runs.

Table 11. Simulation.

Parameter Value
Iterations 100
Random seeds 1–3 (one run per land-use seed)
Flow capacity factor 0.1
Storage capacity factor 0.1778 (= 0.10.75)
Traffic dynamics queue, FIFO
Simulated mode car (transit vehicles also simulated)
Stuck time 120 s
End time 36:00
Link volumes mean of the last 5 iterations

Table 12. Routing.

Parameter Value
Network modes car, ridea
Walk speed 1.23 m/s (beeline × 1.3)
Bike speed 3.1 m/s (beeline × 1.3)
Routing randomness 3.0
Transit router SwissRailRaptor
Stop search radius 4,000 m
Extension radius 2,000 m
Transfer walk, max. 300 m
Transfer penalty −0.13 utils (about 10 in-vehicle minutes)
Capacity constraints off

a Ride is routed on congested car times but does not occupy road space.

Table 13. Replanning.

Strategy Weight Effect
ReRoute 0.15 new route on last-iteration times
TimeAllocationMutator 0.10 end times shifted ± 30 min
ChangeExpBeta 0.75 logit choice among stored plans

Plans kept per agent: 5. Innovation off after iteration 90. No mode choice.

Table 14. Scoring parameters.

Parameter Value Unit
Performing 0.78 utils/h
Late arrival −1.872 utils/h
Waiting for transit 0 utils/h
Transfer (line switch) −0.13 utils
Marginal utility of money 0.78 / VOTa utils/$

a VOT = 0.5 × max(household income, $5,000) / 2,080 h, per person (USDOT rule); $23.91/h at the median income, used in route search.

Table 15. Mode parameters.

Mode Travel time (utils/h) Distance cost ($/km)
Car 0 0.1112
Ride 0 0
Transit 0 0
Walk 0 —
Bike 0 —

Every travel minute loses only the performing rate (0.78 utils/h), so all minutes are valued alike, as in the generalised cost of the MAST work tour-mode model. Car operating cost: AAA 2017; the car passenger does not pay; transit fares are flat per ride and do not affect the route. Mode constants are 0 (no mode choice).

Table 16. Activity types.

Activity Typical duration
Home 12 h
Worka 8 h
Education 6 h
Shoppingb 2 h
Other 2 h
Social and recreation 1.5 h
Eat 45 min
Errand 30 min
Escort 15 min

Home, work, shopping and other are split into 162 duration classes of 10 min.

a Open 06:00–21:00; 5,184 types with a planned arrival time, used by the late-arrival penalty.

b Open 08:00–20:00.

5. Outputs

Table 17. Outputs.

Output Use
Link volumes, hourly and daily counts, screenlines, continuous counters
Trips and legs with routes vehicle-miles, trip lengths, travel times
Transit boardings by line National Transit Database
Congested travel times analysis; not fed back (skims are fixed)

Validation: Traffic, Interstates by station.

Running

miles assignment network      # road networks 2024 and 2026
miles assignment transit      # GTFS merge, conversion, mapping
miles assignment plans        # MAST plans -> MATSim population
miles assignment run          # one run
miles assignment seeds        # several seeds
miles assignment validate     # counts, screenlines, counters