Determining effective evacuation strategies based on WiFi data in Buildings

Abstract

Evacuation strategies are critical in preventing casualties during emergency evacuations in buildings. As large-scale gatherings and high crowd densities in buildings occur more often, the need of relevant and effective evacuation strategies emerges. However, the domain of research that tries to identify possible ways to improve evacuation, i.e. prescriptive domain, is underlooked. Several studies successfully improve evacuation by optimizing existing evacuation scenarios in buildings. A shortcoming of these studies is that they often focus on one strategy and scenario in particular. Therefore, one should opt for a more generic approach to evaluate the effectiveness of evacuation strategies under different circumstances. A way to mitigate uncertainties in evacuation is by using data. Recent studies use a data-driven approach, in which data is used as an input to calibrate and enhance the evacuation strategy. A promising source of data is WiFi data. WiFi data captures movement patterns of building occupants and can be translated to population and building characteristics. Therefore, WiFi data offers the creation of evacuation scenarios in which evacuation strategies can be practically tested. This study aims to (1) evaluate the efficiency of evacuation strategies in buildings under different circumstances, and (2) determine effective evacuation strategies given WiFi data as an input. Therefore, this study presents a new exploratory agent-based approach to evaluate evacuation strategies, and moreover, presents an approach to incorporate input data to practically test evacuation strategies in a given building. To do so, this study used three methodological approaches, namely ExploratoryModeling and Analysis (EMA), Agent- Based Modeling (ABM) and Data Mining. EMA is used to experiment with the created agent-based evacuation model. EMA addresses the effect of uncertainties on the evacuation time, and if evacuation strategies are effective and robust in different circumstances. This study showed that in the created model of the TU Delft TPM faculty building, guiding evacuation strategies, such as dynamic signs and using evacuee staff members turned out to be an effective option if the familiarity in the building is low. However, as the familiarity increases the relative effectiveness of these strategies becomes negligible. In case of increasing familiarity, bottleneck improvement strategies, such as wider exits or stairs and obstacle placement, decrease the total evacuation time consequently. Moreover, this study concluded that an exploratory approach for evacuation models is promising, as the effectiveness of evacuation strategies is very dependent on the evacuation scenario. As a result, this study is able to evaluate these scenarios beforehand and to determine the effect on the total evacuation time. In this study the uncertainties crowd density, familiarity with the building, compliance with given instructions, and the exit capacity are leading in influencing the total evacuation time. The latter was found as a newly modelled uncertainty for evacuation scenarios.