2
Topics: Destination Choice in MATSim …2
now: future: filling the gaps between small-scale choice models and large-scale microsimulations
→ how fancy can the models be?→ which data are required?
• fixed initial random seed• freezing the generating order of ij
• storing all ij
destinations
persons
00
nn
10
iji
j
personi alternativej
store seed ki store seed kj
regenerate ij on the fly with random seed f(ki,kj)
one additional random number can destroy «quench»
i,j ~ O(106) -> 4x1012Byte (4TByte)
Repeated Draws: Quenched vs. Annealed Randomness4
tdeparture tarrival
Dijkstra forwards 1-n Dijkstra backwards 1-n
approximation
probabilistic choice
search space
work homeshopping
Search Space Optimum5
Filling the Gaps6
• future study (IATBR, STRC)
• crux of matter in practice → study will be exemplified for specific microsimulation and data (MATSim and Swiss data)
• triumvirate of sources for gap:• I. lack of necessary data• II: computational issueso III: lack of recent progress in implementation/application
State-of-Art Theoretical Destination Choice Models7
• broad range of disciplines such as transport and urban planning, marketing and retailing science, economics, geography, psychology.
• attributes/choice determinants: • examples:
• standard attributes: • prices & incomes
• no comprehensive literature review yet
7
State-of-Art Microsimulation Destination Choice Models8
person•age, gender, mobility tools, occupancy, home loc (ha), work loc (municipality)•act chain randomly assigned from microcensus
8
destination•location (ha), open times, rough type (h, w, s, l, e)
validation•counts
→ relatively limited set of attributesother microsims → similar but no comprehensive literature review yet available
I. Lack of Data: Available Data Switzerland
9
main data sets, complete Switzerland:•census of population (full survey) -> population / person attributes•microcensus (person sample) -> demand•business census (hectare) -> supply (infrastructure)•miv counts (lane) -> validation
only locally available (canton, municipality, city)•pt schedules and lines (*)•parking supply (*)•green times (*)•open times (*)
(*) ZH scenario only
9
I. Lack of Data:“Missing”
1010
“missing” attributes in MATSim•incomes •vot•activity / shop subtypes•size categories (shop)•education•number of employees
•store price level•store hours (complete CH)•parking prices
I. Lack of Data:“Missing”
11
• Should a manual data collection effort be undertaken?
11
transferability /flexibility
collection costs
local sim quality
detail level of data
II. Computational Issues:Choice Sets … Variability
12
• huge destination choice sets, in particular routing very expensive (for assessing an alternative)
• microsimulation stochasticity:• no data → sophisticated rolling of a dice + correlations
→ model is not restrained→ large variability → many runs
required
12
Research Avenues13
• further heterogeneity of agents and alternatives, in particular, prices, income, vot (household budget survey)
• finer activity classification• available in microcensus and business census
• spatial correlations• agglomeration terms + correlated error terms
• choice dimensions interrelation• e.g., mode – destination etc.
Closing Gaps: Destination Choice Research Avenues13
Research Avenues14
• interaction effects at destinations (e.g., at parking lots)• similar to space-time competition on roads
Closing Gaps: Destination Choice Research Avenues14
0
5000
10000
15000
20000
25000
1 2 3 4
Load category
Vis
ito
rs it_0_config2/3
it500_config2
it500_config3
Load category1: 0 – 33 %2: 33 - 66 %3: 66 - 100 %4: > 100%
10 % ZH Scenario: 60K agents
reduces number of implausibly overloaded facilities
• destination choice equilibratione.g., approximate calculation of travel times
• errtot = w err + w err + wtt errtt • if (wtt = small) -> tt can be approximated• equilibrium concept is approximate itself