Microsimulation & Risk Modelling Group

Mass Gathering Vulnerability Analysis

Broadly speaking, the current range of crowd modelling approaches approximate an average behaviour for their simulated persons walking through a space. In contrast, a microsimulation approach is based on identifying individual behaviours within the population. These are identified through multiple re-runs of the basic simulation, to build up an "average behaviour" on a large scale for multiple persons. This permits the determination of transient and extreme states of the system that are not available to other modelling techniques.

In terms of spatial modelling, our microsimulation approach does not require or enforce any particular physical layout of locations. For the purposes of navigation, a two-dimensional graph of connections can be associated with the abstract locations, giving them position and length, and allowing for navigation by individuals in the simulation. The approach is general enough that a simulation's spatial construction could be (in part) derived from GIS data.

Behaviours may be controlled through several means:

• Events that are scheduled either exactly once at a given point in time, or regularly on a multi-day schedule.
• The motivation module, which allows changes in response to rules defined in terms of characteristics, both internal and external to the individuals. These characteristics may, themselves, be changed over time as a result of interactions between individuals and their environment, or between individuals.
• As a response to other, arbitrary stimuli defined by the simulation modules, such as a change in infection status, arrival at a location, etc.

The simulation begins with an "instantiation": a complete description of what all the individuals are doing during the simulation is constructed from a parameterised input template. The instantiation is run multiple times in order to build up a picture of the overall behaviour of the simulation. The primary advantage is the determination of transient and extreme states of the system. This is not available to other modelling techniques.