Data assimilation in agent based simulation of smart environments using particle filters

Agent-based simulations are useful for studying people's movement and to help making decisions in situations like emergency evacuation in smart environments. These agent-based simulations are typically used as offline tools and do not assimilate real time data from the environment. With more and more smart buildings equipped with sensor devices, it is possible to utilize real time sensor data to dynamically inform the simulations to improve simulation results. In this paper, we propose a method to assimilate real time sensor data in agent-based simulation of smart environments based on Particle Filters (PFs). The data assimilation aims to estimate the system state, i.e., people's location information in real time, and use the estimated states to provide initial conditions for more accurate simulation/prediction of the system dynamics in the future. We develop a PF-based data assimilation framework and propose a new resampling method named as component set resampling to improve data assimilation for multiple agents. The proposed framework and method are demonstrated and evaluated through experiments using a sparsely populated smart environment.