An Evaluation of Complex Adaptive Evolvable System Simulation

Abstract

Building models has been a human activity for centuries. Through the ages man has discovered patterns in how objects function together in the exchange of information, materials, and energy. We know these groups of objects and how they behave as 'systems'. As we study systems, we have learned that systems can be extraordinarily complex. Additionally, complex systems can be adaptive. Computer software exists to help us design, simulate to ultimately understand these Complex Adaptive Systems. However, there exists a type of system that has not been as thoroughly explored. These systems are known as Complex Adaptive Evolvable Systems (CAES). In this thesis I describe what makes these systems unique, and provide a list of properties a simulation tool should possess to successfully design and simulate such a system. I then evaluate several popular modeling platforms (StarLogo TNG, NetLogo, VensimPLE, Simile, and AnyLogic) against these properties. The results show that none of these provide significant support for CAES simulation. However, Simile and AnyLogic both provide an extensible framework that could support such simulation if additional development is performed. I then conclude with how such development could be approached and provide an alternative solution if these platforms (Simile, AnyLogic) cannot be extended to support evolvable components. This alternative approach is a new language, built as a hybridization of current System Dynamic and Agent Basted Modeling frameworks, that includes components with attributes that support evolvability.