An efficient k-NN-based rao optimization method for optimal discrete sizing of truss structures

This study proposes a new method called k-nearest neighbor comparison (k-NNC) to address the computational cost issue of truss optimization with discrete variables using metaheuristic algorithms. The k-NNC judges a new design candidate is worth evaluating by comparing its k available closest designs (k-nearest neighbors) with another design in the population. The new design will be eliminated without evaluating it if the majority of the k nearest neighboring designs are inferior to the one being compared. The k-NNC is combined with Rao algorithms along with Deb's constraint handling rules and the rounding technique to be suitable for constrained optimization problems with discrete variables. The new optimization Rao algorithms based on k-NNC are used in five truss optimization examples, including both planar trusses and spatial trusses, to evaluate their effectiveness. The numerical results demonstrate that the proposed k-NNC-based Rao algorithms outperform the original Rao algorithms in terms of computational costs. Moreover, the overall performance of k-NNC-based Rao algorithms is similar to or better than that of some state-of-the-art metaheuristic algorithms conducted on the same examples.