The Research On Multi-objective Evolutionary Optimization Algorithms Based On Game Strategy

As a very important class of fundamental optimization problems,multi-objective optimization problem (MOP) is prevalent in scientific research andengineering practice, it has wide application prospects. Multi-objective evolutionaryalgorithm (MOEA) which is originated from biology has a good performance onMOPs. As a mainstream method, it is mainly based on nondominated sortingapproach. With the evolution of population, the number of levels of solutions isdeclined, the population diversity is lost gradually, the motivation of evolutionbecomes weaker, and all of these result in that the algorithm is prone to convergeslowly, sink in to local optimum, and get weak ability of global optimization and soon.Game Theory is a subject to study how to make decisions for the player to getthe maximum interests. It and MOP belong to the same problem of multi-agentoptimization, handling multi-objective optimization problem using game theory canimprove the performance of multi-objective evolutionary algorithm and expand thefield of engineering.This research focuses on multi-objective evolution optimization based on gametheory and its application of grid task scheduling, the main contents include:First, we introduce the multi-objective optimization problem and summarize thecurrent available methods for solving this problem, and then we explained someintelligence swarm algorithms. After, we explained in details the basics of gametheory, and we have found that game theory enhances the classical traditionalmulti-objective evolutionary algorithm.What’s more, through analysis of multi-objective evolutionary algorithm andvarious game models, we proposed a multi-objective genetic algorithm based onmixed strategy game theory. In this algorithm, every iteration is refereed as a game,a player selects a strategy and takes an action in order to achieve the maximalincome for the object he works on, that would generate a tensile force over thepopulation to move to the pareto solutions. Theoretical analysis and experimentalverification proves that the algorithm has better performance on convergence anddistribution.Finally, a multi-objective grid task scheduling algorithm based on MSG-MOEAwas proposed. According to the failure problem occurring in the computing resources in real grid environment, we proposed a task scheduling model based ontasks survivability and makespan, and designed a scheduling algorithm for thismodel. Simulation results show that the algorithm can effectively solvemulti-objective grid task scheduling problem.