Digitally Evolving Models for Dynamically Adaptive Systems

Abstract

Developing a Dynamically Adaptive System (DAS) requires a developer to identify viable target systems that can be adopted by the DAS at runtime in response to specific environmental conditions, while satisfying critical properties. This paper describes a preliminary investigation into using digital evolution to automatically generate models of viable target systems. In digital evolution, a population of self-replicating computer programs exists in a user-defined computational environment and is subject to instruction-level mutations and natural selection. These "digital organisms" have no built-in ability to generate a model - each population begins with a single organism that only has the ability to self-replicate. In a case study, we demonstrate that digital evolution can be used to evolve known state diagrams and to further evolve these diagrams to satisfy system critical properties. This result shows that digital evolution can be used to aid in the discovery of the viable target systems of a DAS.