Non-parametric stochastic subset optimization for optimal-reliability design problems

The stochastic subset optimization (SSO) algorithm has been recently proposed for design problems that use the system reliability as objective function. It is based on simulation of samples of the design variables from an auxiliary probability density function, and uses this information to identify subsets for the optimal solution. This paper presents an extension, termed Non-Parametric SSO, that adopts kernel density estimation (KDE) to approximate the objective function through these samples. It then uses this approximation to identify candidate points for the global minimum. To reduce the computational effort an iterative approach is established whereas efficient reflection methodologies are implemented for the KDE.

► An extension to the SSO stochastic optimization algorithm is presented, termed NP-SSO.
► Focus is on design problems using reliability as objective function.
► The objective function is approximated through kernel density estimation (KDE).
► This is iteratively established, exploiting samples of the design variables.
► Appropriate reflection methodologies are discussed for the KDE.