Stochastic multistage algorithms for multimodal structural optimization

This paper proposes five multistage multimodal optimization algorithms for simultaneously seeking multiple optima in a nonconvex design problem. In a single stage algorithm, multistate simulated annealing or sequential local searches are used for simultaneously seeking the global optimum and other local optima. A two-stage algorithm employs a random search followed by topographical near-optimum identification used for singling out near-optimum designs in each extreme-containing valley from a number of sampled designs. In the three stage algorithms, a multistate simulated annealing or a random search is followed by the near-optimum identification tool and subsequent local searches. Three illustrative problems are used to test five proposed multistage multimodal optimization algorithms.