Investigate on self-adaptive non-revisiting genetic algorithm

Abstract

Non-revisiting genetic algorithm has significant improvement compared with seven states of art algorithms. In this project, we want to investigate the behaviour of a self-adaptive approach to non-revisiting genetic algorithm. The Self-adaptive approach used in this project is to integrate strategy variables (crossover rate, crossover operators and crossover point) into chromosomes such that strategy variables undergo the same evolutionary process as chromosomes. Theoretically, better individuals are generated by the better values of strategy variables. Therefore it is more likely to inherit these "good" strategy variables to the chromosome. In this project, a number of self-adaptive strategies using different strategy variables combinations are tested with 34 famous benchmark functions. By t-test, the strategy of using crossover rate or crossover point only does not lead to significant improvement in performance, whereas using crossover rate and crossover operators give better result than the original non-revisiting genetic algorithm. 12 out of 34 functions show significant improvement by using crossover rate and crossover operators as strategy variables.