Efficiently Implementing Genetic Optimization with Nonlinear Response History Analysis of Taller Buildings

Abstract

Nonlinear response history analysis is an important tool for accurately determining the performance of tall buildings under severe earthquake loading. When a standard genetic algorithm is used in conjunction with nonlinear response history analysis, it is desirable to use smaller generation sizes because of the computational effort to analyze individual designs. A study was conducted to evaluate how different genetic algorithm techniques influence the reliability and efficiency of the algorithm when used with nonlinear response history analysis and small generation sizes. The system used in the study was a nine-story buckling restrained braced frame that was optimized to minimize brace areas under individual earthquake records. A baseline study showed that a typical genetic algorithm did not converge to the same best design for different random number sequences (seed numbers). Forced diversity improved the reliability of the algorithm such that it converged to the same optimum, regardless of initial seed number. Adaptive mutation decreased the required number of generations when coupled with a noncrossover constraint. Consecutive identical generations were found to predict convergence and provide a basis for an exit criterion.