Data Collection for Modelling Behavioural during EmergenciesProf. dr. ir. Serge HoogendoornProject leader VICI programma “Liberate”

Transport and Traffic Modelling

• Developing theory and models that can predict network traffic operations in case of an evacuation

• Conditional on disaster dynamics, information and evacuation instructions, traffic information & management, etc.

• Taking into account all relevant behavioural aspects and uncertainties therein

Usage of models

• Evacuation potential and plan assessment

• Optimisation of instructions, evacuation management and control

A Traffic Engineer’s Perspective…Quantitative modelling and simulationEstablishing models that are predictively valid…

Example application: model based optimisation of evacuation instructions

• Describe optimisation problem as a bi-level problem

• Optimisation approach schedules and routes such that all critical links are fully used (assuming high compliance)

• Reduction of computational complexity by innovative fixed-point problem formulation for fast computation of optimal solutions

A Traffic Engineer’s Perspective…Quantitative modelling and simulation…and use them to optimise evacuation instructions

Major behavioural differences per phase

• Threat and warning phase characterised by denial (in general) to reduce ‘feeling of discomfort’

• Denial includes response to warnings, but may be improved by relevant information

• Impact phase is characterised by disbelief, denial, often due to overload of sensory information

• Although necessary, evacuation is difficult due to (mental, emotional) state of evacuee

But there are also big differences in behaviour between evacuees…

Modeling considerations: survival psychology insightsBehaviour Dynamical Framework in case of CalamitiesJohn Leach (1994) - Recommended reading!

T H R EAT

W AR N I N G

I MP ACT

R ECO VER Y

R ESCU E

Modeling considerations: survival psychology insightsBehaviour Dynamical Framework in case of CalamitiesLeaders, followers, and blockers…

G A T H E R I N F O R M A T I O N

J UD G E S I T UA T I O N

R A T I O N AL B E H A V I O UR

V E R I F Y P E R C E P TI O NS T E R E O -T Y P I C A L

B E H A V I O UR

B L O C K I N GH Y P O - A N D

H Y P E R -A C T I V IT Y

P A R A L Y SI S

1 0 - 1 5 %

1 0 - 1 5 %

7 0 - 8 0 %

Modelling efforts Bounded Rationality Framework of EVAQQuantifying behavioural aspects in mathematical and simulation models

Demand modelling Traffic assignment Traffic Operations

Sequential probability function:

• Travelers decide to stay or leave each time period

• Decision is based on characteristics of disaster, household, instructions, information, etc.

Hybrid route choice modelling:

• Pre-trip route choice based on expectations or compliance to instructions

• En-route information may lead to adaptation of route choice during trip

Advanced queuing modelling:

• Speeds and capacities are dependent on road conditions and weather conditions and control measures

• Inclusion of spil lback, capacity drop, etc.

Example controlled experimentEvacuation of Euroborg stadium Testing applicability of lane reversal concepts

E F F E C T E N O P C A P A C I T E I T I N F R A A F H A N K E L I J K V A N E M O T I E E N A F L E I D I N G

Example Empirical Data Driver Behaviour AdaptationEmpirical (and Experimental) data collection results…using remote sensing and driving simulator experiments

• Due to distraction, capacity per lane reduces with 30-50%

• Changes further include increased headway, reduced speed, and increased reaction time

• Driver simulator experiments show impacts on driving behaviour for evacuation situations

• More aggressive behaviour, unstable flows, faster = slower

Characteristic Incident Fog EmergencyFree speed - - +Max acceleration 0 - +Min headways + - -

Route choice behaviour data collection…Results of web-based survey show impact of different attributesTowards modelling behavioural aspects of evacuation…

• Dynamic evacuation scenario, choice to stay / leave, mode and route choice

Route choice behaviour data collection…Results of web-based survey show impact of different attributesTowards modelling behavioural aspects of evacuation…

Data collection for emergency response behaviourData is crucial, but how to ensure they are valid? Advanced stated preference as candidate solution direction…

• Virtual environments have been proposed as means to collect data for emergency conditions

• Examples:

• Driving simulator experiments (Yang)

• Virtual Reality tool BART (Kobes)

• Validation shows validity of virtual environments

• Provides means to study impact of signage and information provision

Innovative Deployment of VR environmentsEvacuation behaviour data collectionTowards modelling behavioural aspects of evacuation…

• Evacuation from a city by unfamiliar visitors using driver simulator environment

• Three wayfinding strategy types found (random exploration, follow pre-planned route, directional heading)

• Experience turned out to be quite immersive, given expressed feeling of anxiety

• Causes for stress varied, but time limit seemed to be most prominent

• Relation between task performance and stress level

Innovative Deployment of VR environmentsEvacuation behaviour data collectionTowards modelling behavioural aspects of evacuation…

Innovative Deployment of VR environmentsStudying Herding Behaviour in Evacuation Decision MakingTowards modelling behavioural aspects of evacuation…

Studying Evacuation and Travel Behaviour

• Environment appears immersive according to participants responses

• Preliminary data analysis shows impacts of information and interaction effects

• Sequential choice modelling predicting decision to stay or leave based on different attributes (earthquake, news bulletin, number of people seen leaving)

• Probability to leave is determined by stochastic utility of staying:

with

• Information has impact of decision to leave

• Importance of herding for large share of population, in line with theory

• Currently working on latent class models to see if we can identify three groups

ASC = 0.80,β1= −0.43, and β

3= −0.60

Ui(t) = ASC + β

1⋅n

leaving(t)+ β

2⋅earthquake + β

3⋅news(t)+ ε

Closing remarksData collection for modelling behaviour in case of emergencies The use of virtual environments…

• Need for modelling for assessment and robust optimisation

• Use of survival psychology knowledge as basis for (conceptual) modelling

• Perspectives of using VR environments for conceptual and mathematical modelling, and data collection for calibration and validation

• Depending on quality of the VR environment, experimental set-up and the background of participants, relatively valid data can be collected

• How to acquire data with absolute validity? Data fusion? Improving quality of simulator environment (see work Hans van Lint)

Including (Behavioural) Uncertainty in OptimisationRobust optimisation of evacuation instructionsDealing with limited knowledge and other sources of uncertainty

Anticipate on uncertain compliance levels

• Efficiency always increases upon anticipating compliance correctly when optimising

• Low compliance level show large improvements

• Low compliance with anticipation outperforms high compliance without anticipation

• Instructions optimised on lower compliance level appear less sensitive

Robust optimisation framework allows to include uncertainty in disaster dynamics