A reliable and computationally cheap approach for finding robust optimal solutions

Current robust optimisation techniques can be divided into two main groups: algorithms that rely on previously sampled points versus those that need additional function evaluations to confirm robustness of solutions during optimisation. This paper first identifies and investigates the drawbacks of these two methods: unreliability for the first and excessive computational cost for the second. A novel approach is then proposed to alleviate the drawbacks of both methods. The proposed method considers the number of suitable, previously sampled points in the parameter space as a key metric to decide whether a solution can be assumed to be a robust solution when relying on previously sampled points. This factor is treated as a constraint that prevents solutions with low numbers of suitable, previously sampled points from participating in the improvement of the next population. To prove the effectiveness of the proposed algorithm, the proposed method is implemented for Particle Swarm Optimisation (PSO) and applied to several test functions from the literature. The results show that the proposed approach is able to effectively improve the reliability of algorithms that rely of previously sampled points without the need for extra function evaluations.