PLAN C (Planning with Large Agent-Networks against Catastrophes):
Planning responses to Catastrophes using a novel agent-based model simulation computational tool.
NYU’s PLAN C is an innovative tool for emergency managers, urban planners and public health officials to prepare and evaluate Pareto-optimal plans to respond to urban catastrophic situations.
PLAN C was designed and developed by the NYU Bioinformatics Group under the supervision of Professor Bud Mishra and with the interdisciplinary collaboration of a team of experts from the NYU Centre for Catastrophe Preparedness and Response (CCPR). The current version of the code and its architecture was primarily developed (in Java and C) by Giuseppe Narzisi under the guidance of Prof. Mishra and with the contributions from other researchers of the Bioinformatics Lab (listed later).
PLAN C uses a powerful large-scale computational multi-agent based disaster simulation framework involving thousands of agents. It has been able to simulate the complex dynamics of emergency responses in different urban catastrophic scenarios (e.g., chemical agent, bomb explosion, food poisoning, small pox). It can devise plans that optimize multiple objective functions (e.g., number of casualties, economic impact, time to recovery, etc.) in terms of their Pareto frontier in a high-dimensional space; for this purpose, it uses an evolutionary genetic search algorithm. It is designed to be easy to use and parametrize by relatively unsophisticated users. The technology can be easily transferred to any urban setting, to multiple computer platforms, and to different modes (offline or online) of planning.
An effective response simulation to catastrophes in an urban environment requires modeling the actions of a large number of actors, each endowed with their own skills, objectives, behaviors and resources, to be able to coordinate their efforts in order to mitigate the outcome of a disaster. This modeling and simulation effort aims to build a flexible, adaptable and general model, combining expertise from both public health and computational sciences, to improve preparedness and response capabilities of a city. Typically, training drills in the field and tabletop exercises are used to inform policy-makers to prepare responders for catastrophes of large magnitude. Usage of computer modeling simulation enables close to optimal design of such exercises to maximize their effectiveness across multiple dimensions. With the powerful computational reasoning and analysis technologies that have been implemented and developed in Plan C’s models, it is now possible to go beyond table-top exercises to help policy makers to consider, simulate and analyze the effect of a wide range of parameters, in several concomitant catastrophes.
Our effort in this direction has resulted in a system with the following combination of features:
The earliest scenario investigated by PLAN C was the 1998 food poisoning of a gathering of over 8000 people at a priest's coronation in Minas Gerais, Brazil leading to 16 fatalities. Multi-agent modeling was explored for this problem by allowing simplistic hospital and person agents to interact on a two-dimensional integer grid. As various complex, counter-intuitive and unanticipated behaviors emerged in the simulation of an extremely parameter sensitive system, it could immediately validate its potential use in agent-simulation-based analysis of catastrophes [MGD+05].
For the second scenario, PLAN C examined an attack with the nerve gas agent Sarin in Manhattan to further evaluate these tools and techniques. Its choice was based on the literature available about a similar attack that occurred in Matsumoto in 1994 and later, in Tokyo in 1995. Several important emergency response issues were addressed by repeated simulation: namely, when to proceed to a hospital, when to discharge a person from the hospital, number of on-site treatment units necessary, the importance of public awareness and law enforcement, the role of responder size and activation time, and the diffusion of information about hospitals and capacities, [MNN+06].
From a more theoretical and computer science perspective, we also studied the source of complexity of the emergent dynamics generated by the model in different catastrophic scenarios [NMN+06].
As part of our research effort, we also continue to work on a novel optimization tool that, when used in combination with the model, can help in deciding how to select better plans. As soon as our rigorous modeling, simulation and analysis approach matures, our next step is to go beyond the simple simulation and analysis of a response plan in order to directly search for “optimal” plans that can efficiently manage a crisis. A large scale emergency plan naturally involves multiple objectives: minimize the number of casualties (affected people), fatalities (mortalities), the average ill-health of the total population, average waiting time at the hospitals, etc., and maximize the average time taken by a person to die (so as to increase the chance for external help to arrive), utilization of resources at different locations (so that no one location runs out of critical resources), etc. There are also tangible economic, legal and ethical issues in disaster management, which stipulate many different classes of objective functions. NYU’s PLAN C models provide as their output the individual traces of all its agents and statistical information about the time-course of the global behavior. In this context, planning can be seen as the problem of adjusting the controllable parameters in the interaction between different classes of agents (hospitals, persons, on-site responders, ambulances, etc.) and available resources, in order to moderate the negative consequences of a catastrophic event. The emergency response planning problem is formulated by PLAN C as a multi-objective optimization problem where the input parameters of the model are the decision variables and the criteria for plan evaluation are the output objectives of the system. Multi-objective evolutionary algorithms have been explored to effectively tackle this problem, in order that suitable Pareto optimal plans can be generated automatically [NMM06].
This research effort has benefited by the generous support from the US Department of Homeland Security through a research Grant # 2204-GT-TX-0001.