Specifying, detecting and analysing emergent behaviours in multi-level agent-based simulations

We introduce a method for analysing emergent behaviours in multi-agent simulations using complex events. Complex events are composed of interrelated events, and they can be defined at any level of spatio-temporal abstraction (equal to or above the lowest level of abstraction given by the model). Minimal types of complex events define sets, which are equated with particular emergent behaviours and can be detected in simulation. Since complex events are derived from the agent-based model itself, they provide significant benefits when compared with traditional state-aggregation methods. First, they provide a method of specifying emergent behaviour, so that such behaviour can be monitored. Second, they provide a mechanism that retains the underlying structure of that behaviour. This latter property supports analysis of the mechanisms at lower levels that give rise to emergent behaviours, and identification of patterns between levels. In other words, multi-agent simulations become less 'opaque' [1].