Simulation of a Bike Sharing System with MATSim

Simulation of a Bike Sharing System with
MATSim
Thibaut Dubernet, Kay W. Axhausen
Institute for Transport Planning and Systems (IVT)
ETH Zurich
ABM Workshop 2015
Introduction
MATSim Implementation
Results
Conclusions
Introduction
MATSim Implementation
Results
Conclusions
Context
I
Station-based bike sharing
I
Work presented here: from EUNOIA EU-funded project
I
Idea of simulating joint use of joint resources
I
Implemented for use in Zurich, Barcelona and London
I
Bike Sharing important in London and Barcelona, planned in
Zurich
1 / 18
Introduction
MATSim Implementation
Results
Conclusions
Important properties of bike sharing systems
I
Small fleet in comparison to population
I
Strong interaction between users (capacity)
Demand imbalanced
I
I
I
I
asymmetric flows
aversion for uphill riding
Important aspect: bike redistribution
2 / 18
Introduction
MATSim Implementation
Results
Conclusions
Requirements for Bike Sharing Simulation
I
I
Should mainly be useful to operators
Microsimulation allows to experiment with redistribution
strategies
I
I
I
Disaggregate representation: allows to actually simulate
relocating trucks and reactive strategies
Represents reactions of demand to the quality of the
redistribution
Allows to represent incentives for users to mitigate imbalance
3 / 18
Introduction
MATSim Implementation
Results
Conclusions
Introduction
MATSim Implementation
Results
Conclusions
Elements
I
I
Representation of infrastructure: stations with capacity and
initial number of bikes
Routing
I
I
I
I
should allow for different station combinations for a same OD
should allow access and egress to and from public transport
Scoring: fares, slopes
General requirement: modular
I
Allow arbitrary redistribution agents to be plugged in
4 / 18
Introduction
MATSim Implementation
Results
Conclusions
Routing
I
Two solutions for allowing different stations to be used for a
same OD
I
I
I
randomize routing
within-day adaptation
Solution retained: randomize routing
I
I
random station around origin and destination selected
with evolution, demand should spread over stations according
to capacity, the same way car traffic spreads over routes
I
Within-day would be more desirable, but not implemented for
time constraints
I
Need for slope information for scoring: bike routed on the
network (multimodal extension)
5 / 18
Introduction
MATSim Implementation
Results
Conclusions
Public Transport Routing
I
add bike sharing to possible access and egress modes for
public transport
I
extend search radius to 5 km for bike sharing
Performance!
I
I
I
I
I
All public transport stops in the area considered valid
Routing to those stops is expensive (graph shortest path)
Solution here: extensive caching
Long term dream: real multimodal router!
6 / 18
Introduction
MATSim Implementation
Results
Conclusions
Scoring
I
Disutility of slope
I
I
Need for a link-level disutility (routing)
Most route choice models in the litterature: route-level
I
I
I
I
Maximum or average gradient
Used parameter from a route choice model from the
litterature, using cumulated altitude gain
1m uphill gradient is equivalent to 13m on flat land
Fares
I
I
I
Typical Bike Sharing fares work by steps
First 30 min free
> 90% of the trips in the free period
7 / 18
Introduction
MATSim Implementation
Results
Conclusions
Mobsim
...
Person Agents
Information/Queues
Relocating Agents
give/take bikes
Bike Sharing Manager
reads / writes
Infrastucture Data
notifies
Listenners
8 / 18
Introduction
MATSim Implementation
Results
Conclusions
Introduction
MATSim Implementation
Results
Conclusions
Scenario
I
Full planned system for Zurich
I
Optimal redistribution
I
Compare with and without aversion for uphill riding
9 / 18
Introduction
MATSim Implementation
Results
Conclusions
Infrastucture
10 / 18
Introduction
MATSim Implementation
Results
Conclusions
Infrastucture
11 / 18
Introduction
MATSim Implementation
Results
Conclusions
2500
2000
1500
1000
500
0
Number bikes in use
3000
3500
En Route Throughout Day
0
5
10
15
20
25
30
Time (h)
12 / 18
Introduction
MATSim Implementation
Results
Conclusions
0.6
0.4
0.2
personal bike
bike sharing
0.0
Proportion
0.8
1.0
Trip Durations
0
10
20
30
40
50
60
Duration (min)
13 / 18
Introduction
MATSim Implementation
Results
Conclusions
Outflow Excess
I
left: no aversion for uphill riding
I
right: aversion for uphill riding
14 / 18
Introduction
MATSim Implementation
Results
Conclusions
Outflow
I
left: no aversion for uphill riding
I
right: aversion for uphill riding
15 / 18
Introduction
MATSim Implementation
Results
Conclusions
Inflow
I
left: no aversion for uphill riding
I
right: aversion for uphill riding
16 / 18
Introduction
MATSim Implementation
Results
Conclusions
Introduction
MATSim Implementation
Results
Conclusions
Notes
I
Sampling complicated
I
No redistribution strategy experimented with
I
Toy prototypes show that the architecture for redistribution
works
17 / 18
Introduction
MATSim Implementation
Results
Conclusions
Outlook
I
Multi-agent simulation allows to actually test redistribution
strategies
I
But getting data about how it is done in reality hard
I
Microsimulation allows to isolate effects (eg. effect of slope)
18 / 18
Questions?