16 January 2023

To ABM or not to ABM

​

Luis (Pilo) Willumsen

Nommon Solutions and Technologies

Willumsen Advisory Services

Transport Modellers Forum

To ABM or not to ABM?

When is it desirable to change modelling approach?

​

• When the existing model cannot generate the indicators of performance you need to support decisions
• When the model generates the performance indicators but you believe they are biased and cannot be trusted to make decisions

​

• “Greater realism” is not in itself a reason except in research.
• ”Greater forecasting accuracy”, must be demonstrated in forecasting practice.

2

Therefore

A diagnostic is necessary for your Strategic Model

• Start with the problems you need to address,
• The decisions that need to be made (interventions, policy options) and
• What performance indicators are needed to support these decisions
• Does the current (aggregate) model fails to produce the right KPIs? or
• You do not trust them for good reason?

​

• Is it possible to improve the current Strategic Model to deliver the required KPIs with reasonable precision and complement it with judgement?
• If not, what kind of model is required bearing in mind:
• Expected improvement in decision making
• Resources (time, people, IT, money) needed

To ABM or not to ABM

​

DALL-E

3

To ABM or not to ABM

Current Challenges

Overarching Objective:

Equity, Intra and Inter generations

4

Equity

Fairer access to jobs and opportunities is a must and our models should always generate appropriate equity impacts

The evolving environmental crisis requires models that record the impact of interventions on the route to NET ZERO

​

Our current models can consider and deliver outputs that are relevant to making decisions affecting these issues.

To ABM or not to ABM

5

New Technologies

• Virtual alternatives to travelling will continue to grow eliminating some trips and generating new ones
• “Demand Responsive Public Transport” poses a major modelling challenge
• Only a small proportion of these new technologies will help social objectives

​

• Ride Sharing Services
• Multi-point delivery vehicles
• They all require modelling the supply of mobility services:
• Some adjustments and simplifications are possible, but full treatment requires Agent Based Modelling

​

​

To ABM or not to ABM

​

6

Uncertainty

​

• Model, Data, Technologies, Behaviour Change, AI, etc…
• We can reduce Base Year model errors but not true Future Uncertainties
• Treating Uncertainty as variations on a Central Case is often misleading, even using Stochastic Risk Analysis

​

• Scenario Planning is a first step in handling Uncertainty: THEN
• Need major overhaul of decision making including CBA
• Incorporate the value of Resilience and Flexibility, and
• Monitor evolving trends to adapt quickly

​

​

• Good current classic models are good enough for this treatment of Uncertainty

​

To ABM or not to ABM

​

7

An example of extending a classic model

The task

To extend and improve a large Strategic Model to represent in a simplified way all new mobility technologies

Demand Responsive Transport poses the most difficult challenges:

• Importance of Fleet Size
• Relocation of units to serve customers, this implies empty movements
• Walking, Waiting and Detour times for DRT
• Level of service provided at any location and time

​

• Scheduled services are easy: Cable Car, Hyperloop, but AirTaxis are not

​

​

9

General considerations�

• As it is an aggregate, zonal based model of time periods (minimum one hour) we must accept limitations:
• Trips will be non-integer and between zones
• No accurate representation of ride-sharing routing
• Fleet size may not be sufficient to deliver a good Level of Service LOS
• Generalised costs must include IVT, Walk, Wait and Diversion Time plus an ASC that reflects willingness to share a ride
• Diversion distance and time, plus empty unit re-location will only be approximations and cannot be accurately assigned as network flows
• Unit or Fleet availability will only be approximate

To ABM or not to ABM

​

10

A possible simplification

To ABM or not to ABM

11

Supply side

• We need to account for:
1. Additional empty VKT for re-location of units, all new modes
2. Additional VKT for Diversions in the case of ride-sharing
3. Additional personal travel time due to Diversions for ride-sharing
4. Additional Walk Time for car club/e-scooter depending on fleet availability
5. Additional Wait and/or Diversion Time depending on fleet availability
• c, d and e affect the Level of Service and can be adjusted when fleet is insufficient
• ”Back of a v. small envelope” estimations of necessary fleet size could be based on Vehicle Kilometres Travelled, Passenger Kilometres Travelled and assumed occupancy rates plus empty re-location.

​

To ABM or not to ABM

​

12

DRT Fleet size

To ABM or not to ABM

​

13

Decisions, decisions…

Requirements for New Model Forms

Candidate model forms are Agent Based Models and/or Activity Based models

The new formulations must model behaviour more realistically and provides the right KPIs but also:

• really deliver answers that an improved classic model cannot
• they can be calibrated and validated to a sufficient standard,
• the data collection required is manageable,
• have good convergence, stability and repeatability properties
• produce interpretable outputs
• produce reasonable results in forecasting mode based on projectable inputs
• run fast enough to be usable for forecasting under uncertainty (several scenarios)

DALLE-E

DALL-E

15

How to decide to move to AgBM

The extensions suggested for classic models would not work well if DRT is expected to become a significant market share, say > 5%

However, it could be useful to estimate what would be needed to reach that level and whether the result would be worthwhile

Alternatively, one can develop Agent Based Models that use more conventional treatment of Tours, Destination, Mode and TOD choices; Assignment may be aggregate or microsimulation

This can be handled using existing software including open source

​

It is still a major undertaking

16

Activity Based Modelling

Defining characteristics of an ABM

​

1. Travel is derived from Activity participation;
2. Detailed modelling of sequencing activities and tours to serve them;
3. Individual’s activities are both planned, scheduled and executed in the household (family) context; (Note this is rarely done in ABMs as it is too difficult)
4. Activities are spread throughout a 24-hour period in a continuous manner, rather than simple categorization of ‘peak’ and ‘off peak’ events; (Note that for activity re-scheduling a week would be better)
5. Travel and location choices are restricted in time and space, and by personal constraints.

​

17

General structure of an ABM

18

Preparing transition to ABM or AgBM

1. Identify questions that the current model fails to answer
2. Explore whether enhancements will provide the right KPIs; an enhancement may involved a hybrid approach: some agent modelling where needed (say MaaS) and conventional the rest
3. Explore the potential of new data sources to reduce data collection costs
4. Select software and possibly a “donor model” and
5. Design one or more pilot tests of the preferred approach in forecasting mode to ascertain advantages and limitations compared to conventional models and
6. Report together with new validation tests.

​

​

19

Requirements for Demand Responsive Transport

DALLE-E

This table would be different for other applications

20

CONTACT

Luis.Willumsen@nommon.es

www.nommon.es

​