Genetic Algorithm Parameter Effect on 3D Truss Optimization with Discrete Variable
Advanced Journal of Graduate Research , Volume 5, pp 61-70; doi:10.21467/ajgr.5.1.61-70
Abstract: The Genetic Algorithm is one of the advanced optimization techniques frequently used for solving complex problems in the research field, and there are plenty of parameters which affect the outcome of the GA. In this study, a 25-bar truss with the nonlinear constraint is chosen with the objective to minimize the mass and variables being the discrete area. For the same, GA parameter like Selection Function, Population Size, Crossover Function, and Creation Function are varied to find the best combination with minimum function evaluation. It is found that the Uniform selection gives the best result irrespective of the creation function, population size or crossover functions. But this is at the cost of a large number of function evaluations, and the other selection function fails to reach the global optimum and has a smaller number of function evaluation count. If the analysis of selection function is done one at a time, it is seen that all Cases performs better in Roulette but, Case A which is non-integer type with 200 population size being computationally cheaper than Case B and C of population size 300. In the Tournament selection, Case A, B with smaller population size and Case C with higher population size performs better. Case C performs better at Remainder selection with smaller population size, and Case A and B for Stochastic Uniform with higher population size. And, it is clear that the function evaluation count increases with the population size in every Case from this study.
Keywords: optimization / GA / discrete / truss / population size / crossover / Uniform / Function Evaluation / Smaller
Click here to see the statistics on "Advanced Journal of Graduate Research" .