Robust simplex algorithm for online optimization

Abstract

A new optimization algorithm is introduced for online optimization applications. The algorithm was modified from the popular Nelder-Mead simplex method to make it noise aware and noise resistant. Simulation with an analytic function is used to demonstrate its performance. The algorithm has been successfully tested in experiments, which showed that the algorithm is robust for optimization problems with complex functional dependence, high cross-coupling between parameters, and high noise. Advantages of the new algorithm include high efficiency and that it does not require prior knowledge of the parameter space such as an initial conjugate direction set.