A new model-based multi-objective Memetic algorithm and its convergence analysis
Wei Jing-xuan · Control theory & applications · 2008
The multi-objective optimization problem is converted into a constrained optimization problem.Based on the constraint dominance principle,a new selection strategy is proposed for the converted problem to remove the drawback in most algorithms taking Pareto dominance as selection strategy but ignoring preference information.Memetic algorithm is one of the most efficient algorithms for optimizing multi-objective problems,incorporating local search into evolution- ary computation.The new multi-objective Memetic algorithm combines the genetic algorithm with simulated annealing algorithm by introducing the C-metric to improve the global search ability.The convergence of this algorithm is proved with related theories of probability.Simulation results demonstrate the ability of the new algorithm in finding the uniformly distributed and widely-spread non-trivial solutions on the entire Pareto front.