Sensitivity Analysis of Evacuation Simulations 07. November 2014, WOST
Contents Introduction Project Motivation and Background Basics of Evacuation Simulation Pathfinder simulation Input and Output Parameters Simulation Results FDS/Evac simulation Software Features and I/O Parameters Simulation Results Conclusions
Project Project: Sensitivity Analysis of Evacuation Simulations Objective Increase Quality of Evacuation Simulation - improve understanding of the effects of model parameters Project Participants Verein zur Förderung der Ingenieurmethoden im Brandschutz (ViB) e.v. (Association to Promote Engineering Methods in Preventive Fire Protection): Funding : Modelling, simulation and evaluation of results dynardo GmbH: Software optislang, Training, Support
Motivation and Background Motivation Today: increasing public attention for safe evacuation of persons from buildings and public events Simulation of evacuation - efficient method to find best escape concepts - increase safety Relatively new simulation discipline: Importance and effects of input parameters not well known Stochastic sensitivity analyses are appropriate to improve understanding of the models and parameters
Basics of Evacuation Simulation Model basics Floors where persons can move given by building geometry Persons modelled as autonomous agents Action and interaction of persons: equations of motion Action: controlled by available exits or doors, Interaction with other persons and obstacles - forces on persons, Properties described by physical and psychological parameters - decision-making heuristics Discretisation in space and time Explicit time integration, molecular dynamics
Basics of Evacuation Simulation Test Model Test model: geometry and setup Arbitrary room geometry on one floor Includes typical features: Long corridor Multiple doors and exits to select Bifurcation of escape paths
Input and Output Parameters Pathfinder: Input and Output Parameters Input Parameters for Occupant Properties Input Parameters for Geometry
Simulation Results Stochastic Sensitivity Analysis Advanced Latin Hypercube Sampling, 500 samples Occupant properties: normal distribution Geometric parameters: uniform distribution Output parameter: Evacuation Time (max.) 3 Sensitivity Studies Study 1: Only occupant parameters vary Study 2: Only geometric parameters vary Study 3: Both types of parameters vary
Simulation Results Video Reference Model Pathfinder
Simulation Results Pathfinder: Results of 3 Studies CoP Results of Study 1: Sensitivity to occupant parameters CoP Results of Study 2: Sensitivity to geometric parameters
Introduction Pathfinder simulation FDS/Evac simulation Conclusions Simulation Results Pathfinder: Results of 3 Studies MOP Results of Study 1: Sensitivity to occupant parameters MOP Results of Study 2: Sensitivity to geometric parameters
Simulation Results Pathfinder: Results of 3 Studies CoP Results of Study 3: Sensitivity to both occupant and geometric parameters Conclusion Sensitivity of max. evacuation time to occupant properties is much higher than to geometric door and exit dimensions
Software Features and I/O Parameters Software FDS/Evac Fire Dynamics Simulator by NIST/USA and VTT/Finland, widely-used fire simulation software Non-commercial FDS/Evac: evacuation module Combination of fire simulation and evacuation simulation possible Input Parameters Sensitivity study for occupant parameters only Constant geometric dimensions Options for specific agent behaviour: follow others, herding 500 samples in < 1 hour on 8 CPU cores
Software Features and I/O Parameters Video Reference Model FDS/EVAC
Simulation Results FDS/Evac sensitivity analysis CoP Results of FDS/Evac: Sensitivity to occupant parameters MOP, 3D-Response Surface for FDS/Evac: Sensitivity to occupant parameters
Conclusions Occupant parameters are more important than geometric dimensions in usual ranges Correct data for the expected parameter ranges will increase reliability of simulations Most important occupant parameters: walking speed, body dimensions and acceleration time Modelling of psychologic phenomena like herding is very important Evacuation simulation is an ideal field of application for stochastic analyses due to the short simulation times