Use cases for multiagent simulation in passenger and freight transport

Sebastian Hörl

21 November 2022

ABMT

IRT SystemX

• Public/private research institute situated in Paris
   
• Focus on fostering digital transition in a range of fields from transport, cybersecurity to circular economy
   
• Transferring research results and tools into active application by development and provision of industry platforms
   
• Various collaborative projects with multiple French companies (Renault, SNCF, ...) and academic partners (Université Paris Saclay, CentraleSupélec, Université Gustave Eiffel)
   
• Participation in European projects

MATSim @ IRT SystemX

• On-demand mobility simulation
  
   

• First/last mile mobility simulation

Passenger transport

Freight transport

• First/last mile delivery
  
  
  
  
  
  
  
  
  
   
• VRP Solvers
  • Development of electric VRP for JSprit

http://www.loc.gov/pictures/item/2016800172/

The street in 1900

The street today

https://commons.wikimedia.org/wiki/File:Atlanta_75.85.jpg

• Autonomous Mobility
  
  
• Mobility as a Service
  
  
• Mobility on Demand
  
  
• Electrification
  
  
• Aerial Mobility

Julius Bär / Farner

The street of tomorrow?

I. Transport simulation

Classic transport planning

• Zones
• Flows
• Peak hours
• User groups

Aggregated

Agent-based models

0:00 - 8:00

08:30 - 17:00

17:30 - 0:00

0:00 - 9:00

10:00 - 17:30

17:45 - 21:00

22:00 - 0:00

• Discrete locations
• Individual travelers
• Individual behaviour
• Whole day analysis

Disaggregated

Icons on this and following slides: https://fontawesome.com

MATSim

• Flexible, extensible and well-tested open-source transport simulation framework
   
• Used by many research groups and companies all over the world
   
• Extensions for parking behaviour, signal control, location choice, freight, ...

matsim-org/matsim-libs

MATSim

Synthetic demand

MATSim

Mobility simulation

Synthetic demand

MATSim

Decision-making

10:00 - 17:30

17:45 - 21:00

22:00 - 0:00

Mobility simulation

Synthetic demand

MATSim

Decision-making

Mobility simulation

Synthetic demand

MATSim

Decision-making

Mobility simulation

Analysis

Synthetic demand

https://pixabay.com/en/zurich-historic-center-churches-933732/

II. AMoD in Zurich

Cost structures?

User preferences?

System impact?

Cost Calculator for automated mobility

Stated preference survey

MATSim simulation

1

2

3

What do we know about automated taxis?

What do we know about automated taxis?

Bösch, P.M., F. Becker, H. Becker and K.W. Axhausen (2018) Cost-based analysis of autonomous mobility services, Transport Policy, 64, 76-91

What do we know about automated taxis?

Felix Becker, Institute for Transport Planning and Systems, ETH Zurich.

VTTS

AMoD

Car by Adrien Coquet from the Noun Project

Bus by Simon Farkas from the Noun Project

Wait by ibrandify from the Noun Project

VTTS

AMoD

Car by Adrien Coquet from the Noun Project

Bus by Simon Farkas from the Noun Project

Wait by ibrandify from the Noun Project

Model structure

Cost calculator

Plan modification

Discrete Mode Choice Extension

Mobility simulation

Prediction

Price

Trips

- Utilization

- Empty distance, ...

- Travel times

- Wait times, ...

Fleet sizing with dynamic demand

Fleet sizing with dynamic demand

Fleet sizing with dynamic demand

Visualisation

Automated taxi

Pickup

Dropoff

Hörl, S., F. Becker and K.W. Axhausen (2020) Automated Mobility on Demand: A comprehensive simulation study of cost, behaviour and system impact for Zurich

Hörl, S., F. Becker and K.W. Axhausen (2020) Automated Mobility on Demand: A comprehensive simulation study of cost, behaviour and system impact for Zurich

13% reduction in vehicles

100% increase in VKT

Hörl, S., F. Becker and K.W. Axhausen (2020) Automated Mobility on Demand: A comprehensive simulation study of cost, behaviour and system impact for Zurich

100% increase in VKT

Hörl, S., F. Becker and K.W. Axhausen (2020) Automated Mobility on Demand: A comprehensive simulation study of cost, behaviour and system impact for Zurich

Other aspects

Hörl, S., C. Ruch, F. Becker, E. Frazzoli and K.W. Axhausen (2019) Fleet operational policies for automated mobility: a simulation assessment for Zurich, Transportation Research: Part C, 102, 20-32.

Fleet control

Operational constraints

Spatial constraints

Intermodality

Pooling

III. Demand data

https://pixabay.com/en/paris-eiffel-tower-night-city-view-3296269/

Synthetic travel demand

Synthetic travel demand

Synthetic travel demand

0:00 - 8:00

08:30 - 17:00

17:30 - 0:00

0:00 - 9:00

10:00 - 17:30

17:45 - 21:00

22:00 - 0:00

Synthetic travel demand

Synthetic travel demand

Person ID    

Age

Gender

Home (X,Y)

1

43

male

(65345, ...)

2

24

female

(65345, ...)

3

9

female

(65345, ...)

Synthetic travel demand

Person ID    

Activity

Start

End

Loc.

523

home

08:00

(x,y)

523

work

08:55

18:12

(x,y)

523

shop

19:10

19:25

(x,y)

523

home

19:40

(x,y)

Synthetic travel demand

Synthetic travel demand

Synthetic travel demand

Synthetic travel demand

Synthetic travel demand

+

=

Synthetic travel demand

+

=

Current use cases

Current use cases

Current use cases

Current use cases

Current use cases

Current use cases

Current use cases

Current use cases

Current use cases

eqasim-org/eqasim-java

eqasim-org/ile-de-france

Current use cases

eqasim-org/sao_paulo

eqasim-org/california

Balac, M., Hörl, S. (2021) Synthetic population for the state of California based on open-data: examples of San Francisco Bay area and San Diego County, presented at 100th Annual Meeting of the Transportation Research Board, Washington, D.C.

Sallard, A., Balac, M., Hörl, S. (2021) Synthetic travel demand for the Greater São Paulo Metropolitan Region, based on open data, Under Review

Emissions in Paris

Grand Paris Express

Automated taxis in Paris

Automated taxis in Paris

IV. Logistics

Photo by Tom Fisk from Pexels

LEAD Project

• H2020 Project from 2020 to 2023
   
• Six living labs with different innovative logistics concepts
  • Lyon, The Hague, Madrid, Budapest, Porto, Oslo
  • One partner for implementation and one for research each
     
• Development of a generic modeling library for last-mile logistics scenario simulation and analysis

Lyon Living Lab

• Peninsula Confluence between Saône and Rhône
   
• Interesting use case as there are limited access points
   
• Implementation of an urban consolidation center (UCC) to collect the flow of goods and organize last-mile distribution
  • using cargo-bikes
  • using electric robots
  • and others
     
• Analysis and modeling through
  • Flow estimation through cameras
  • Simulation of future scenarios
     
• Focus on parcel deliveries due to data availability

Modeling methodology

Agent-based simulation of Lyon

Demand in parcel deliveries

Distribution by operators

KPI Calculation

• Main question: What is the impact on traffic and population of implementing an Urban Consolidation Center in Confluence?
   
• Focus on B2C parcel deliveries due to data availability

Methodology: Parcel demand

+

Based on sociodemographic attributes of the households, parcels are generated for the city on an average day.

Gardrat, M., 2019. Méthodologie d’enquête: le découplage de l’achat et de la récupération des marchandises par les ménages. LAET, Lyon, France.

Methodology: Parcel demand

+

Based on sociodemographic attributes of the households, parcels are generated for the city on an average day.

Gardrat, M., 2019. Méthodologie d’enquête: le découplage de l’achat et de la récupération des marchandises par les ménages. LAET, Lyon, France.

Methodology: Parcel demand

+

Based on sociodemographic attributes of the households, parcels are generated for the city on an average day.

Gardrat, M., 2019. Méthodologie d’enquête: le découplage de l’achat et de la récupération des marchandises par les ménages. LAET, Lyon, France.

Methodology: Parcel demand

+

Based on sociodemographic attributes of the households, parcels are generated for the city on an average day.

Gardrat, M., 2019. Méthodologie d’enquête: le découplage de l’achat et de la récupération des marchandises par les ménages. LAET, Lyon, France.

Methodology: Parcel demand

Gardrat, M., 2019. Méthodologie d’enquête: le découplage de l’achat et de la récupération des marchandises par les ménages. LAET, Lyon, France.

+

Based on sociodemographic attributes of the households, parcels are generated for the city on an average day.

• Based on synthetic population, find average number of purchases delivered to a household defined by socioprofesional class, age of the reference person and household size per day.

Methodology: Parcel demand

Gardrat, M., 2019. Méthodologie d’enquête: le découplage de l’achat et de la récupération des marchandises par les ménages. LAET, Lyon, France.

+

Based on sociodemographic attributes of the households, parcels are generated for the city on an average day.

• We now the average number of parcels, but we do not now the distribution of the number of parcels for a household on an average day.
   
• We know it must be non-negative, and we know the mean.
   
• Without additional data, we assume maximum entropy distribution, which is Exponential in this case.

Methodology: Parcel demand

Gardrat, M., 2019. Méthodologie d’enquête: le découplage de l’achat et de la récupération des marchandises par les ménages. LAET, Lyon, France.

+

Based on sociodemographic attributes of the households, parcels are generated for the city on an average day.

Methodology: Parcel demand

Gardrat, M., 2019. Méthodologie d’enquête: le découplage de l’achat et de la récupération des marchandises par les ménages. LAET, Lyon, France.

+

Based on sociodemographic attributes of the households, parcels are generated for the city on an average day.

Methodology: Parcel demand

Gardrat, M., 2019. Méthodologie d’enquête: le découplage de l’achat et de la récupération des marchandises par les ménages. LAET, Lyon, France.

+

Based on sociodemographic attributes of the households, parcels are generated for the city on an average day.

Methodology: Parcel demand

Gardrat, M., 2019. Méthodologie d’enquête: le découplage de l’achat et de la récupération des marchandises par les ménages. LAET, Lyon, France.

+

Based on sociodemographic attributes of the households, parcels are generated for the city on an average day.

Methodology: Route optimization

Using a heursitic routing solver, an optimal distribution scheme per distribution center is obtained to arrive at a lower bound estimate for the distance covered for the deliveries.

JSprit

First case study

Hörl, S. and J. Puchinger (2021) From synthetic population to parcel demand: Modeling pipeline and case study for last-mile deliveries in Lyon, Working paper.

Solve VRP-TW based on generated parcels and household presence

JSprit

First case study

Hörl, S. and J. Puchinger (2021) From synthetic population to parcel demand: Modeling pipeline and case study for last-mile deliveries in Lyon, Working paper.

Solve VRP-TW based on generated parcels and household presence

JSprit

Scaling up: City-wide baseline

Using shares of customer preference, parcels are assigned to operators and the closest distribution centers.

Scaling up: City-wide baseline

Using shares of customer preference, parcels are assigned to operators and the closest distribution centers.

Scaling up: City-wide baseline

Using shares of customer preference, parcels are assigned to operators and the closest distribution centers.

Methodology: Supply side

Vehicle routing and fleet composition problem

• Minimize total cost of delivering all parcels for each center
  • Distance cost influenced by fuel / electricity cost
  • Fixed cost (per day) influenced by salaries and vehicle cost
     
• Derive
  • Distances driven and related emissions
  • Fleet composition for each center

Novel approximation method

Visualisation platform

UCC platform

V. Calibration

Photo by Tom Fisk from Pexels

Challenge

• Agent-based models have become a common tool in transport planning
   

• Reproducibility: Are simulation results replicable by other researchers?
   

• Relevancy: Are the results corresponding well to reality?
   

• Reuse: Can we easily adapt and advance the modeling system?

Modeling pipeline

• Goal: Provide a modeling pipeline from raw data to the final results.

• Should allow to automatically repeat the modeling process.
• Should allow to replace individual modeling components.

Modeling pipeline

Modeling pipeline

Simulation

Synthetic demand

Simulation

Mobility simulation

Synthetic demand

Simulation

Decision-making

10:00 - 17:30

17:45 - 21:00

22:00 - 0:00

Mobility simulation

Synthetic demand

Simulation

Decision-making

Mobility simulation

Synthetic demand

Simulation

Decision-making

Mobility simulation

Synthetic demand

Simulation

Decision-making

Mobility simulation

Synthetic demand

Simulation

Decision-making

Mobility simulation

Synthetic demand

Simulation

Decision-making

Mobility simulation

Analysis

Synthetic demand

Calibration problem

• Find parameters (mode choice model, network capacities, ...) that minimize mismatch with reference data

Calibration loop

Calibration loop

Stopping criterion

• Currently, simple process:
  • Smooth trajectory of mode shares over horizon S
     
  • Approximate derivative over horizon H
     
  • Define threshold

Example

• Parameters: ASCs, VOT Car, scaling factor on network capacity
• Target: Mode share by distance
• Reference: EGT 2010 Île-de-France
• Objective: L2 norm
• Algorithm: CMA-(1,1)-ES

Example

• Parameters: ASCs, VOT Car, scaling factor on network capacity
• Target: Mode share by distance
• Reference: EGT 2010 Île-de-France
• Objective: L2 norm
• Algorithm: CMA-(1,1)-ES

Framework

• Designed as a modular framework
  • Wraps around different simulators
  • Modular structure of objectives
  • Various optimization algorithms
  • (Unit tested)
  • (Restartable)

sebhoerl/boptx

Framework

• Gradient approximation

  • SPSA
  • FDSA

• Evolutionary search

  • CMA-ES / Elitist CMA-ES
  • xNES / Elitist xNES

• Accelleration

  • odpyts

• Others

  • Uniform sampling
  • Nelder-Mead
  • Differential Evolution
  • scipy optimize interface

• Coming up

  • Batch Bayesian Optimization (Kriging) with various acquisition functions
  • Multi-fidelity BBO

Framework

• Gradient approximation

  • SPSA
  • FDSA

• Evolutionary search

  • CMA-ES / Elitist CMA-ES
  • xNES / Elitist xNES

• Accelleration

  • odpyts

• Others

  • Uniform sampling
  • Nelder-Mead
  • Differential Evolution
  • scipy optimize interface

• Coming up

  • Batch Bayesian Optimization (Kriging) with various acquisition functions
  • Multi-fidelity BBO

Questions?

https://slides.com/sebastianhorl/abmt-ethz-21nov2022