Evolutionary Computation in Combinatorial Optimization

Fitness based selection procedures leave majority of population individuals idle, that is, they don’t take place in any recombination operation although some of them have above average fitness values. Based on this observation, a two-phase two-strategy genetic algorithm using a conventional strategy with multiple mutation operators in the first phase is proposed. In the second phase, those individuals that are not sufficiently recombined in the first phase are reconsidered within a second strategy and recombined using multiple mutation operators only. In the second strategy, mutation operator probabilities are adaptively determined based on the cumulative fitness-gain achieved by each mutation operator over a number of generations. The proposed genetic algorithm paradigm is used for the solution of hard numerical and combinatorial optimization problems. The results demonstrate that the proposed approach performs much better than the conventional implementations in terms of solution quality and the convergence speed.