Numerical method for global optimisation: dynamic encoding algorithm for searches

Abstract

A fast and reliable non-gradient optimisation method is proposed. The basic structure of the proposed numerical method is a binary matrix in which each row represents an encoded string for the corresponding parameter. The operations of adding a binary digit and performing an increment or a decrement to a selected row make successive approximations of the local minima, and the restart is undertaken from an unexplored initial matrix as an escape scheme. To verify the feasibility of the proposed method for global optimisation, several test functions are optimised as a benchmark, and the parameter identification of an induction motor is carried out on experimental data. Both the results of the benchmarking and the parameter identification show that the proposed non-gradient optimisation method is capable of fast and reliable global search for various problems.