Real-Time Evacuation Prediction Based on Simulations and Machine Learning Model

Agent-based evacuation models provide useful data of the evacuation process, but they are not primarily designed for use during an emergency. The article will explore the possibilities of utilizing simulation data from the Pathfinder software in real-time through a machine learning model. The performance of the artificial neural network model will be tested across three fundamental scenarios: bidirectional movement of occupants in a corridor, unidirectional movement through a narrow bottleneck, and on unidirectional movement stairs. The machine learning model will be trained on data compiled from a set of simulations and validated against real experimental data.