Sequential simulated annealing for multimodal design optimization

This paper proposes a sequential simulated annealing that aims to seek the global optimum and other important relative optima by sequential executions of regular simulated annealing. Each simulated annealing attains a relative optimum, and uses exclusively downhill designs at the final stage of the annealing process to define a hyperspherical optimum-containing region. In subsequent simulated annealing, designs will not be allowed to enter hyperspherical regions already containing an attained optimum, thereby forcing the search for another optimum. The sequential simulated annealing offers an efficient and effective tool for simultaneously attaining the global and other best optimal designs. The efficiency and effectiveness of the sequential simulated annealing are demonstrated in a number of multimodal problems.