A Multi-objective Evolutionary Algorithm Based on Mixed Game Strategy

Non-dominated sorting multi-objective optimization algorithms can constantly lead to the population of Pareto front optimal. However, the non-dominated sorting strategy lacks high capability to explore the Pareto front in the evolutionary subsequent process. We introduce a mixed strategy game model into evolutionary algorithms in this paper. Based on this strategy, we propose a novel multi-objective evolutionary algorithm (MSG-MOEA). A player adopts a strategy against the rest of the players with a certain probability in their respective strategy space instead of some specific strategy. According to the results of the game earning, the player constantly updates this probability to maximize the interest of his own objective. Through the players' constant pursuit of the maximal interest, a kind of tension could be brought to the population, which would push forward the population to the Pareto front. The proposed approach has been used some test functions and metrics for validation which are taken from the standard multi-objective optimization evolutionary literature. The experiment results have been compared against the NSGAII algorithm, which is one of the most highly competitive EMO algorithms. Algorithm analysis and simulation results show that the proposed algorithm performs well in solving complex multi-objective optimization problems.