Adaptive submodel selection in hybrid models

Abstract

Hybrid modeling seeks to address problems associated with the representation of complex systems using "single-paradigm" models: where traditional models may represent an entire system as a cellular automaton, for example, the set of submodels within a hybrid model may mix representations as diverse as individual-based models of organisms, Markov chain models, fluid dynamics models of regional ocean currents, and coupled population dynamics models. In this context, hybrid modelers try to choose the best representations for each component of a model in order to maximize the utility of the model as a whole. Even with the flexibility afforded by the hybrid approach, the set of models constituting the whole system and the dynamics associated with interacting models may be most efficient only in parts of the global state space of the system. The immediate consequence of this possibility is that we should consider adaptive hybrid models whose submodels may change their representation based on their own state and the states of the other submodels within the system. This paper uses a simple example model of an artificial ecosystem to explore a hybrid model which may change the form of its component submodels in response to their local conditions and internal state relative to some putative optimization choices. The example demonstrates the assessment and actions of a "monitor" agent which adjusts the mix of submodels as the model run progresses. A simple mathematical structure is also described and used as the basis for a submodel selection strategy, and alternative approaches are briefly discussed.