Evaluating Emergency Evacuation Events Using Building WiFi Data

摘要

Building operators are required to conduct periodic drills to ensure smooth evacuations in the event of emergencies. However, quantitative evaluation of the drill for adherence to building codes is largely manual and error-prone. Further, unplanned evacuations are seldom documented, let alone evaluated. This paper explores the use of building WiFi data for quantitative evaluations of both planned and unplanned evacuation events. We collect and analyze WiFi connectivity logs spanning a period of 180 days from 14 buildings in a large University campus. For our first contribution, we isolate WiFi data for known planned evacuation drills, conduct floor-level analysis to eliminate noise associated with transient WiFi connections or persistently connected devices, and highlight the anatomy of evacuations across multiple representative buildings each with differing number of levels, exit layouts, and occupant types. Armed with a detailed understanding of the anatomy of a planned evacuation, our second contribution develops a novel method to automatically identify evacuation events from WiFi data; we use it to detect 29 unplanned evacuations, and corroborate them against documented records where available. Our third contribution deduces quantitative measures to compare planned and unplanned evacuations, in terms of evacuation speed and occupancy levels, and further quantifies the man-days of productivity loss arising from unplanned evacuation events across campus. We believe our work is the first to show that building evacuations can be evaluated systematically and accurately at scale using WiFi data, both to corroborate current manual records and to gain new insights.