High performance simulation-based optimization environment for large scale systems

Modelling large scale systems with natural and artificial components requires storage of voluminous amounts of knowledge/information as well as computing speed for simulations to provide reliable answers in reasonable time. Computing technology is becoming powerful enough to support such high performance modelling and simulation. This dissertation proposes a high performance simulation based optimization environment to support the design and modeling of large scale systems with high levels of resolution. The proposed environment consists of three layers--modeling, simulation and searcher layer. The modeling layer employs the Discrete Event System Specification (DEVS) formalism and shows how it provides efficient and effective representation of both continuous and discrete processes in mixed artificial/natural systems necessary to fully exploit available computational resources. Focusing on the portability of DEVS across serial/parallel platforms, the simulation layer adopts object-oriented technology to achieve it. DEVS is implemented in terms of a collection of classes, called containers, using C++. The searcher layer employs Genetic Algorithms to provide generic, robust search capability. In this layer, a class of parallel Genetic Algorithms, called Distributed Asynchronous Genetic Algorithm (DAGA), is developed to provide the speed required for simulation based optimization of large scale systems. This dissertation presents an example of DEVS modeling for a watershed, which is one of the most complex ecosystems. The example shows a well-justified process of abstraction from traditional differential equation models to DEVS representation. An approach is proposed for valid aggregation of spatially distributed systems to reduce the simulation time of watershed models. DEVS representation and spatial aggregation assure relative validity and realism with feasible computational constraints. Throughout the dissertation, several examples of GA optimization are presented to demonstrate the effectiveness of the proposed optimization environment in modeling large scale systems