Publications Details

Derivative-free optimization via evoluationary algorithms guiding local search (EAGLS) for minlp

Griffin, J.D.; Fowler, K.R.; Gray, G.A.; Hemker, T.; Parno, M.D.

Derivative-free optimization approaches are commonly used for simulation-based design problems when objective function and possibly constraint evaluations have a black-box formulation. A variety of algorithms have been developed over the last several decades to address the inherent challenges such as computationally expensive function evaluations, low amplitude noise, nonsmoothness, nonconvexity, and disconnected feasible regions. Hybrid methods are emerging within the direct search community as new tools to overcome weaknesses while exploiting strengths of several methods working together. In this work, we extend the capabilities of a parallel implementation of the generating set search (GSS) method, which is a fast local derivative-free approach, to handle integer variables. This is achieved with a hybrid approach that uses a genetic algorithm (GA) to handle the integer variables. Promising points are selected as starting points for the GSS local search with the integer variables held fixed before being passed back to the GA for the standard selection, mutation and crossover operations for the next iteration. We provide promising numerical results on three mixed integer problems; one based on the design of a compression spring, a simulation-based problem from hydrology, and a standard problem taken from the literature. © 2011 Yokohama Publishers.