A Method For Solving Large Scale Multi Objective Optimization Problems Based On Particle Swarm Optimization

Optimization problems are generally divided into single objective optimization problems and multi-objective optimization problems.In the multi-objective optimization problem,each objective function is usually in conflict with each other,that is,the optimization of one sub-objective function may lead to the deterioration of the performance of one or more other objective functions.Particle swarm optimization is an easy to implement and fast convergence of swarm intelligence optimization technology,so many scholars use it to solve multi-objective optimization problems.However,the traditional multi-objective particle swarm optimization algorithm is effective in dealing with small-scale multi-objective optimization problems.However,when the objective dimension is increased,the effect will decline,especially in the large-scale multi-objective optimization problem performance is poor,it is difficult to ensure that the solution set has good convergence and distribution.In order to choose the global optimal solution more reasonably and improve the convergence and diversity of the population,the first kind of large-scale multi-objective particle swarm optimization algorithm based on K-means clustering is proposed in this thesis.Firstly,K-means clustering is used to classify particles,and the individual with the highest non-dominant rank is selected as the guiding individual of this class.Each class of particles has a global-like optimal solution to guide the updating of particles,so that the global-like optimal solution can guide the particles to move evenly to the Pareto frontier.Then,the inertial weight and acceleration coefficient of particle renewal are dynamically selected according to the value of the objective function.Since it is a multi-objective problem,the value of the objective function is not unique.Assuming an ideal point,the Euclidean distance between the particle and the ideal point is calculated,and the appropriate acceleration coefficient is selected to measure the particle performance through the Euclidean distance.The position and velocity of particles are updated by the improved particle swarm updating formula.Finally,the three test functions are compared with the eight most advanced multi-objective optimization evolutionary algorithms,and the experimental results show that the proposed algorithms have good convergence and diversity.The traditional fitness value calculation method pays more attention to the convergence of particles.In order to consider the diversity of particles,the second method is the multi-objective particle swarm optimization algorithm based on individual evaluation method.In this thesis,the definitions of convergence distance and diversity distance are first given.Then,the fitness value calculation method is defined,the particles are sorted according to the new fitness value,and the particles with better performance are selected and put into the optimal solution set.Finally,the proposed algorithm is compared with eight advanced multi-objective optimization evolutionary algorithms in three test functions,which proves the superiority of the proposed algorithm in solving large-scale optimization problems.In this thesis,two kinds of multi-objective particle swarm evolution algorithms are developed to solve the shortcomings of the existing work,which greatly improves the convergence and diversity of particle swarm optimization algorithms,has strong applicability,and has excellent performance in solving large-scale multi-objective optimization problems.Therefore,the algorithm proposed in this thesis has important theoretical and application significance.