Discussion on Distributed Genetic Algorithms for Designing Truss Structures

Distributed Genetic Algorithm (DGA) is one of optimization methods and very suitable to parallel computers. Therefore, DGA is a very effective tool in high performance computing. In this paper, we discussed DGAs for designing truss structures. Most of the real world problems in optimization problems have constraints. Since DGA is an algorithm for problems without constraints, the fitness function of DGA should be modified to handling constraints in problems with constraints. In this paper, we proposed a new method of handling constraints for DGAs. At the same time, the extended model of DGA called Environment Distributed GA (EDGA) is introduced. In EDGA, each island has the different environment and treats the different constraint in this paper. This setting keeps the diversity of the solutions and leads the good results. The proposed method is simulated for designing truss structure that is consisted of 640 design variables. Through the simulations, the followings were made clarified. The conventional handling method of the constraints of GAs cannot derive the solutions in GAs, since the feasible domain is very narrow. The proposed method operates effectively for designing truss structure. This tendency is applicable to also other problems.