TOWARDS A FRAMEWORK FOR SIMULATION-BASED EVALUATION OF PERSONAL DECISION SUPPORT SYSTEMS FOR FLOOD EVACUATION.

Flood model. We use Dynamic Rapid Flood Spreading Model (DRFSM) for flood dynamics simulation. The model provides data on inundation dynamics in form of impact zones’ flooding.Demand model. Transport demand model gives the information on volumes of transportations between different areas in the city. Agent-based traffic model. Microscopic agent-based traffic model determines a vehicular traffic dynamics in urban areas and belongs to the class of time-stepped models.Dynamic road graph model. Flooding dynamics is translated to traffic model as a dynamical roads’ closing.

Agent based modelling simulation, GIS tools

For the experimental study an area of Vasilievsky Island in St.Petersburg (Russia) was chosen. The population of the island is approximately 200 thousands of people. The island has four bridges on the eastern part) and one large road, the Western High-Speed Diameter (WHSD), will be constructed in several years to the western part of the island.

Scenario 1: No decision support (baseline). Representing a process of the vehicular evacuation without any support. Each agent move to the closest exit defined by the geographical distance. Agent without support uses the route with minimal length and do not take into account information on congestions. Scenario 2: Ordinary navigators. Simple navigation system has a statistical data on a current roads load. Agents with such system installed find optimal routes by minimization of estimated travel time in congested road network. Scenario 3: Learning navigators. the more advanced navigation system is used. This kind of system allows agents to commit their knowledge to the shared storage. When an agent discovers the flooded road all agents using the system discover this too automatically, and they take this new knowledge into account in their further routing. The knowledge sharing capability could be represented in contemporary navigation systems as a capability to declare information about accidents, road closures. Scenario 4: Navigators with full information about flooding. The full information on flood dynamics is accessible for agents who use the navigation system. It means that all the agents discover information on flooded areas at the moment they became flooded. This case can be projected to reality as a system which collects actual data from the sensors.

Design and development of the framework for an agent-based simulation of social systems in dynamic geospatial environment

Future work: In future we plan to improve the traffic model by implementation of multilane roads, by more precise capturing of a human behavior (trust, panics). Second future plans’ direction refers to implementation of additional scenarios and DSS types. Scenario 5. Navigation system with flood dynamics forecast. Scenario 6. The scenario extends the last one with forecast of road load dynamics. The forecast can be obtained by a simulation with traffic models or by use of central collaborative information storage. Scenario 7. based on the system-optimal approach to planning the evacuation Scenario 8. with capability for agents to communicate with their friends by mobile phones in order to share the information on flood dynamics.

It was shown that more knowledge about environment dynamics leads to better service in both homogenous and heterogeneous cases. With decreasing the gap between knowledge acquisition about the environmental dynamics (flood and traffic) from first to fourth scenario we observe increase of a number of successfully escaped agents. It was shown that the knowledge sharing between agents can be a good alternative to the navigation system acquiring knowledge about the environment from a lot of sensors.

The framework should be feasible for the task of evaluation of decision support systems for vehicular flood evacuation. Investigation of the influence of different navigation decision support services on evacuation process on the base of agent-based simulation.