Agent-based global search algorithms employ a set of search agents to traverse a given design space in pursuit of an optimum solution, and are normally accepted as the most reliable methods for finding a global optimum in a complex design space. However, when non-linear constraints are present in the optimization problem, modifications have to be made to such algorithms to allow the handling of these constraints. The gravitational search algorithm (GSA) is a recent addition to the family of global search methods but, to date, little research has been presented focussed on dealing with the handling of constrained optimization using GSA. To that end, this paper presents a constraint handling method specifically for use with GSA called separation-sub-swarm (3S), that splits the primary swarm into a feasible and infeasible swarm where the sub-swarms optimize either the constraints or the true objective function, which has the advantage of being algorith-mically independent so is applicable to any agent-based search algorithm. This algorithm is applied here first to constrained analytical optimizations, and shown to be very effective and efficient. It is further applied to a transonic aerodynamic shape optimization problem and a problem that is subject to research by the AIAA Design Optimization Discussion group, again showing impressive results.
展开▼
