Proceedings of the 2011 Workshop on Agent-Directed Simulation
Agent-based simulation for large-scale emergency response: A survey of usage and implementation
ACM Computing Surveys (CSUR)
| Hi-index | 0.00 |
This work addresses research questions important to the Dynamic, Data-Driven Application Systems (DDDAS) community: specifically how to create, update and validate simulations instantiated from streaming sensor data. The use of Agent-Based Modeling simulations in an online context presents several challenges: simulations must demonstrate good runtime characteristics, yet in order to fit in our validation framework, the simulations must be modular and allow for alternative models of human behavior. We present a simulation of pedestrian movement we have developed according to our revised simulation design criteria, built using techniques from Design Patterns and Pattern Oriented Modeling. We present a thorough evaluation of the simulation in terms of model validation, simulation design and runtime characteristics. We present and evaluate methods for online validation of Agent-Based Models (ABM). We introduce an aggregate method for online creation and updating of ABM simulations and evaluate the approach against alternatives. We have developed answers to these questions through the design and implementation of a DDDAS application, the WIPER system. The Wireless Integrated Phone-based Emergency Response system is designed to provide emergency responders with timely information on the status of a city or region, as well as the capability to detect, follow and possibly predict crisis events. WIPER uses the DDDAS approach to process streaming information from a cellular phone service provider to detect and predict crisis events. We demonstrate that real time simulation of pedestrian crowds is possible with our Agent-Based Modeling simulation and present upper bounds on the population size that can be simulated in real time. We show that for certain validation measures, our validation approach yields 100% accuracy at selecting model type based on simulation output. The results of our updating approach demonstrates that aggregate updating outperforms one-to-one agent-to-referent reparameterization under certain conditions and provides empirical evidence suggesting the effects of naive realism.