An agent-based simulator for indoor crowd evacuation considering fire impacts

Highlights

• FREEgress is developed to simulate dynamic fire impacts on indoor fire evacuation.

• Impacts of heat, temperature, toxic gas and smoke particles are incorporated.

• Fire impacts on evacuees' mobility, navigation and health conditions are simulated.

• FREEgress is verified by comparing its performance to existing simulation tools.

• Impacts of three key factors on evacuation outcomes are assessed using FREEgress.

Abstract

Fire emergencies impose significant threats to building occupants. During evacuation, fire has significant impacts on evacuees' behaviors, by e.g., changing their route availability, disturbing their perception of the environment due to reduced visibility, impairing their mobility that is usually associated with severe injuries, and causing significant mental stress that may lead to complicated and unpredictable navigation decisions. Despite the detrimental effects of fire on crowd evacuation, most existing building evacuation simulation models and tools do not account for the impacts of fire on the evacuation process; at most they rely on oversimplified assumptions and simulation settings. In this study, a new fire evacuation simulation model, named FREEgress (Fire Risk Emulated Environment for Egress), is developed to simulate the dynamic influences of heat, temperature, toxic gas and smoke particles on evacuees' mobility, navigation decision making and health conditions. FREEgress (1) introduces evacuee agents who are aware of and able to assess the fire hazards, and can make fire risk-informed navigation decisions; and (2) models the interactions between evacuee agents and the dynamic fire emergency environments and the consequent evacuation process. The verification of FREEgress is conducted by comparing its simulation results with two existing simulation tools, SAFEgress and FDS + Evac. In addition, a case study using FREEgress is carried out to simulate the evacuation in a museum for 30 different fire emergency scenarios. The simulation results are analyzed to assess the impacts of three important factors, namely initial fire location, evacuation delay time and evacuee behavior, on the evacuation process and evacuation outcomes. The case study demonstrated the potential value of FREEgress to support both the safety design of new buildings and maintenance and emergency management of constructed facilities.

Introduction

Fire emergencies impose critical threats to buildings and their occupants. Public fire departments across the U.S. attended 499,000 fires in buildings in 2018, which caused 2910 deaths and 12,700 injuries [1]. During fire emergencies, hazardous fire conditions and unsuccessful evacuation attempts can expose occupants to significant risks [2,3]. Evacuation simulation is an effective approach to reproduce occupants' evacuation behavior during building fire emergencies, which is fundamentally important for advancing the understanding about occupants' navigation decision-making during evacuation, and for developing appropriate measures to facilitate the evacuation process and hence reduce the risks occupants may be faced with [4].

There is an increasing volume of literature in recent decades that has focused on developing models for simulating crowd evacuation during building fire emergencies. These models can be broadly categorized into three groups based on simulation techniques, namely particle system models, cellular automata models and agent-based models [5]. A typical example of particle system models is the social force model proposed by Helbing [6]. Although particle system-based simulations can successfully simulate typical phenomena (such as panic) and observe self-organization behaviors (e.g., faster is slower and mass behavior) in pedestrian dynamics, they cannot reproduce subtleties of individual behaviors (e.g., walking in pairs) [7]. Moreover, they neglect to consider occupants' decision making and oversimplify their navigation process [8]. Cellular automata models are widely adopted by many commercial simulation tools, such as Building EXODUS [9], Simulex [10], and CAFÉ [11]. These models reproduce many collective behaviors (such as clogging and arching) and are suitable for large-scale computer simulations, but they have limited realism in representing occupants' decision making and dynamic environment change [7]. Nor can these models represent the impact of pedestrians' injuries or that of high-density crowds [8]. Agent-based models consider each evacuee as an autonomous agent, who can perceive surrounding environments, exchange information with other agents, make informed evacuation decisions, and implement evacuation strategies accordingly. Examples of agent-based models for crowd evacuation include Vicrowd [12], HiDAC [13], MASSEgress [14], SAFEgress [15] and Pathfinder [16]. These models can not only simulate the intelligent and heterogeneous agents and environments but also capture emergent phenomena (such as crowd congestion) and complex human behaviors (such as competitive behavior, queuing behavior and herding behavior) [4]. Therefore, these models have been popularized in the latest literature. While various existing agent-based models have incorporated many principles of human behavior and significantly advanced the efficacy of building fire evacuation simulation, most existing models have thus far ignored the impacts of fire hazards on human behavior and consequently on the outcomes of evacuation. Fire has significant impact on evacuees' egress behaviors in several aspects [3,17]. First, evacuees, by instinct, would choose a route that can avoid high temperature and heat; second, heavy smoke can reduce the visibility and therefore cause occupants to slow down, while the toxic gases can impair occupants' mobility and even lead to severe injuries and failure of evacuation. In extreme cases, fire hazards can cause significant mental stress that may lead evacuees to make complicated and unpredictable navigation decisions.

Despite the significant effects of fire in crowd evacuation, most existing building simulation models and tools do not account for these impacts or rely on oversimplified assumptions and simulation settings. The lack of realistic simulation of fire impacts is especially critical. Modeling fire impacts is a challenging issue considering the fact that fire and smoke develops and spreads, and their influence on occupants is highly dynamic and spatiotemporal-specific. Although several commercial or academic simulation tools have attempted to incorporate the impacts of fire in evacuation simulation, including Building Exodus [9], FDS + Evac [18], FireGo [19] and AIEval [20], fire impacts are highly oversimplified and usually underestimated in these tools, owing to the particle system or cellular automata-based structure of these tools [21] or their simplified qualitative rule-based reasoning mechanism [7]. Failure to appropriately account for the fire impacts has largely prevented fine-grained modeling of evacuees' navigation decision-making and behaviors, leading to inaccurate prediction of evacuation process and outcomes.

Motivated by this gap, this study aims to develop a new simulation model, FREEgress (Fire Risk Emulated Environment for Egress), to incorporate the various impacts of fire on evacuees into the evacuation simulation, by (1) introducing evacuee agents, who are aware of and able to assess the fire hazards, and can make fire risk-informed navigation decisions; (2) modeling interactions between evacuee agents and the dynamic fire emergency environments and the consequent evacuation process. FREEgress inherits major features of SAFEgress [15], its earlier version which is proven effective in simulating both human and social behaviors in the evacuation process [21]. By appropriately accounting for fire impacts in the agent-based modeling of fire evacuation, FREEgress aims to achieve more realistic and fine-grained simulation of evacuees' navigation decision-making and navigation behaviors by incorporating dynamic fire impacts, and ultimately achieve more accurate simulation and prediction of crowd evacuation processes and outcomes for various building fire emergency scenarios.