Simulation-based optimization for design parameter exploration in hybrid system

Abstract

A considerable interest in the design parameter optimization of hybrid systems has arisen in many engineering, science, and military disciplines. These hybrid systems are employed by interactions between discrete event subsystems and continuous subsystems. Such interactions often lead to some challenges in optimization. They are difficult to obtain the closed form of the objective function and/or have to use simulation to evaluate the objective function. In this kind of situation, simulation may be an attractive approach to optimize the function without findings its closed form, although a naive application of classical optimization methods does not work. The hybrid simulation models, however, are computationally expensive due to the substantial complexity in their behavior characteristics. This difficulty in computation presents an obstacle to finding for optimal solutions efficiently using simulation evaluations. This dissertation proposes an efficient simulation-based optimization method of hybrid systems to assist in making high-quality solutions and speeding up convergence. The proposed method combines a metamodel-based method and a metaheuristic search-based method. The metamodel replaces an original simulation model cost-effectively by approximation techniques. The role of the metamodel is to provide good initial candidates and reduced search space for the metaheuristic optimizer. The metamodel construction for hybrid systems, however, is not straightforward due to their nonlinear, discontinuous behavior characteristics caused by interactions between their continuous subsystems and the discrete event subsystem. This dissertation contributes to construct a piecewise concept for identifying a complex nonlinearity in hybrid system behavior. The metamodel necessarily includes approximation errors, even if the proposed method emphasizes computational efficiency of the metamodel. Such approximation errors often cause that some of candidates recommended by me...