traffic flow estimation from cellular network data Master’s thesis presentation Nils Breyer August 27, 2015 LiU/ITN, UC Berkeley (USA)
why are linkflows of interest? Linkflow number of vehicles or people using a link during a certain time Usage • Traffic planning • Traffic control 2
why are linkflows of interest? Linkflow number of vehicles or people using a link during a certain time Usage • Traffic planning • Traffic control 2
datasources Travel surveys Link counts Cellular networks 3
datasources Travel surveys Link counts Cellular networks 3
outline • Overview of the model • Trip extraction and travel demand estimation • Cellpath routing • Network loading • Results & validation 4
Overview of the model 5
cellpath Cellpath a sequence of cells that a user connected to along a trip 6
7 overview of the model Cellpath Network Cellular data routing loading Route link counts Celltower locations
7 overview of the model Cellpath Network Cellular data routing loading Route link counts Celltower locations
problem: data only for a sample 8
problem: data only for a sample 8
od-matrix OD-matrix A matrix containing the traffic demand between pairs of origin and destination zones 9
10 overview of the model OD matrix Cellular data estimation Time-sliced OD matrix Network loading link counts Cellpath Celltower locations routing Route
10 overview of the model OD matrix Cellular data estimation Time-sliced OD matrix Network loading link counts Cellpath Celltower locations routing Route
problem: data contains not only movements 2013-10-01 08:10:00 • Handover data • Call detail records (CDR) Datatypes Table 1: An example of a cellular network dataset 2 2013-10-01 08:20:00 2 3 1 User ID 1 2013-10-01 08:10:00 1 1 2013-10-01 06:50:00 1 Cell ID Timestamp 11
problem: data contains not only movements 2013-10-01 08:10:00 • Handover data • Call detail records (CDR) Datatypes Table 1: An example of a cellular network dataset 2 2013-10-01 08:20:00 2 3 1 User ID 1 2013-10-01 08:10:00 1 1 2013-10-01 06:50:00 1 Cell ID Timestamp 11
overview of the model 12 Trip OD matrix Cellular data extraction estimation Trips with Time-sliced cellpath OD matrix Network loading link counts Cellpath Celltower locations routing Linkpaths
Trip extraction & travel demand estimation 13
overview of the model 14 Trip OD matrix Cellular data extraction estimation Trips with Time-sliced cellpath OD matrix Network loading link counts Cellpath Celltower locations routing Linkpaths
trip extraction Challenges • Depends on the datasource • Cell-switching due to network balancing • Low resolution in time (CDR) Solutions • algorithmic approach (CDR) • movement efficiency metric (handover data) 15
trip extraction Challenges • Depends on the datasource • Cell-switching due to network balancing • Low resolution in time (CDR) Solutions • algorithmic approach (CDR) • movement efficiency metric (handover data) 15
travel demand estimation Challenges • Additional data necessary • Sample might not be representative Solutions • Upscaling using total population • Data fusion with census data 16 • Vehicles ̸ = people
travel demand estimation Challenges • Additional data necessary • Sample might not be representative Solutions • Upscaling using total population • Data fusion with census data 16 • Vehicles ̸ = people
Cellpath routing 17
overview of the model 18 Trip OD matrix Cellular data extraction estimation Trips with Time-sliced cellpath OD matrix Network loading link counts Cellpath Celltower locations routing Linkpaths
cellpath routing problem Input Output A route consisting of connected links on the road network Goal Recover the original route that the user took 19 Cellpath p = ( b 1 , b 2 , ..., b n )
voronoi diagram 20
cellpath routing algorithms Shortest-path Routing shortest path between first and last cell in cellpath Strict Voronoi Routing route must go through every cell in the cellpath Lazy Voronoi Routing route is encouraged to go through cells in the cellpath 21
shortest-path routing shortest path between first and last cell in cellpath 22
strict voronoi routing route must go through every cell in the cellpath 23
waypoint search 24 C 1 3 B 2 A Antenna position Boundary junction (possible waypoint)
lazy voronoi routing route is encouraged to go through cells in the cellpath 25
comparision between algorithms Shortest-path Strict Voronoi Lazy Voronoi 26
Network loading 27
overview of the model 28 Trip OD matrix Cellular data extraction estimation Trips with Time-sliced cellpath OD matrix Network loading link counts Cellpath Celltower locations routing Linkpaths
single od-pair Shortest-path Strict Voronoi Lazy Voronoi 29
single od-pair Shortest-path Strict Voronoi Lazy Voronoi 29
single od-pair Shortest-path Strict Voronoi Lazy Voronoi 29
Results & validation 30
los angeles dataset Coverage I-210 corridor in Los Angeles, USA Input data • Antenna positions • OD-matrix estimated from real cellular data (UC Berkeley) • Simulated cellpaths (UC Berkeley) • OpenStreetMap road network 31
V a E a and r b V b E b have a route similarity of: route validation Definition: Route similarity metric V b V a V b V a s a b Two routes r a 32 Celltower locations MatSim Cellpath OD-matrix simulation routing cellpath estimated route Validation MATSim route
route validation Definition: Route similarity metric 32 Celltower locations MatSim Cellpath OD-matrix simulation routing cellpath estimated route Validation MATSim route Two routes r a = ( V a , E a ) and r b = ( V b , E b ) have a route similarity of: s ( a , b ) := | V a ∩ V b | | V a ∪ V b |
route validation average of 1000 randomly selected routes 33 1.00) 0.80) Shortest) 0.60) Similarity2 Strict) 0.40) Lazy)(ds)=)0.03)) 0.20) 0.31) 0.43) 0.25) 0.00) E0.20)
network loading result using Lazy Voronoi routing 34
network loading validation 35
Conclusions 36
conclusions 1. Lazy Voronoi improves route recoverage over Shortest-path and Strict Voronoi routing 2. Route similarity not enough for a precise network loading 3. Data fusion with traffic counts could improve the result 37
Nils Breyer nilsbreyer.eu 38
4-stage model 39 Trip Trip Mode Route choice & extraction distribution choice network loading link counts
line simplification Ramer-Douglas Peuker algorithm 40 4 1 3 2
lazy voronoi routing details 41 T ¡ W ¡ ¡ ¡ Voronoi-‑routed ¡path ¡ S ¡ Actual ¡path ¡ ¡ ¡ Cellpath ¡ S ¡ Simplified ¡cellpath ¡ ¡ Lazy ¡Voronoi ¡route ¡ T ¡ ¡
la network loading geh 42
Recommend
More recommend
