Game Theoretic Personality Changing MOPSO method for Multi-Objective Optimization

Abstract

Abstract Multi-objective particle swarm optimization is an extension/generalization of the particle swarm optimization for optimizing more than one objective simultaneously. Since particle swarm optimization is inspired by animal behavior, mainly bird flocking and fish schooling, and also the competitive nature among objectives which can be observed in multi-objective optimization, it is intuitive to look into this problem from a game theoretic point of view. In this thesis we have introduced a novel method called Personality changing multi-objective particle swarm optimization (PC-MOPSO) which is based on game theory and is an enhancement to standard-MOPSO [1]. We apply PC-MOPSO to the specific problems of two-objective optimization. In PC-MOPSO each particle has two personalities and each personality is trying to maximize its payoff by making a rational decision/choice. As a result of this iterative process the final optimal solutions are achieved. Four standard test functions used for evaluation and simulation of results show that the proposed method is capable of handling ZDT1 problem with 2.79% higher accuracy than the standard-MOPSO and also performed competitively in other problems. One interesting observation which is unique to the new method is, unlike in other MOPSO methods which have utilized an external repository to find final solutions, in this method solutions are achieved in a distributed manner. This can be seen as similar to most evolutionary algorithms.