Maxim Dulebenets

By Maxim A. Dulebenets

Evolutionary Algorithms have been extensively used for solving stochastic, robust, and dynamic optimization problems of a high complexity. Selection mechanisms play a very important role in design of Evolutionary Algorithms, as they allow identifying the parent chromosomes, that will be used for producing the offspring, and the offspring chromosomes, that will survive in the given generation and move on to the next generation. Selection mechanisms, reported in the literature, can be classified in two groups: (1) parametric selection mechanisms, and (2) non-parametric selection mechanisms. Unlike parametric selection mechanisms, non-parametric selection mechanisms do not have any parameters that have to be set, which significantly facilitates the Evolutionary Algorithm parameter tuning analysis. This study presents a comprehensive analysis of the commonly used non-parametric selection mechanisms. Comparison of the selection mechanisms is performed for the machine scheduling problem. The objective of the presented mathematical model is to determine the assignment of the arriving jobs among the available machines, and the processing order of jobs on each machine, aiming to minimize the total job processing cost. Different categories of Evolutionary Algorithms, which deploy various non-parametric selection mechanisms, are evaluated in terms of the objective function value at termination, computational time, and changes in the population diversity. Findings indicate that the Roulette Wheel Selection and Uniform Sampling selection mechanisms generally yield higher population diversity, while the Stochastic Universal Sampling selection mechanism outperforms the other non-parametric selection mechanisms in terms of the solution quality.

Part of the book: Nature-inspired Methods for Stochastic, Robust and Dynamic Optimization