Research On Many-Objective Evolutionary Algorithm

Multi-objective optimization problem is very common in practical engineering applications,and it's one of the main research areas.Multi-objective evolutionary algorithm is a better way to deal with multi-objective optimization problems by maintaining the population and continuously searching in the decision space to obtain a set of approximate Pareto optimal solutions.By analyzing related theories of multi-objective optimization and existing multi-objective evolutionary algorithms,this paper proposes a decomposition-based multi-objective evolutionary algorithm with pareto front grid and a multi-population multi-objective evolutionary algorithm based on contributing objectives of decision variables.The main research work is as follows:(1)The constrained decomposition approach with grids can reflect the neighbor structure of the solution by establishing a grid system.It performs well compared with other decomposition methods,especially on multi-objective problems with complex Pareto front.However,its performance depends on the grid segmentation parameters,resulting in a waste of computing resources.This paper proposes a decomposition-based multi-objective evolutionary algorithm with pareto front grid(PFGMOEA).On the basis of ensuring the advantages of grid decomposition,the grid segmentation parameters can be set to a small value.This paper defines a new concept of Pareto Front Grid,using individuals within the Pareto frontier grid to guide the search of the current population.With the statistical estimation method,a new nadir point selection strategy is given to improve the problem of slow population convergence caused by the large grid system area generated in the early stage of evolution.In order to test the performance of the algorithm,PFGMOEA and other state-of-the-art multi-objective evolution algorithms were tested on 16 widely used benchmark functions.The results prove the effectiveness of the PFGMOEA proposed in this paper.(2)Decision variables can be divided into convergence-related or diversity-related according to their control attributes.By analyzing the contribution objective of the convergence-related decision variables,a multi-population multi-objective evolutionary algorithm based on cont ributing objectives of decision variables DVCOEA is proposed.By analyzing the main convergence direction of each convergent decision variable,an analysis method of the contribution objective of the decision variable is given.Then the decision variables are grouped according to the contribution objective.In order to ensure the optimization effect,the algorithm uses a multi-population multi-objective framework for optimization.Two different optimization strategies are designed for individuals in sub-populations and external archives to optimize population convergence and diversity,respectively.Finally,the DVCOEA algorithm and two algorithms are experimentally compared on the benchmark function.From the experimental results,it can be seen that the DVCOEA algorithm is effective on solving large-scale multi-objective and many-objective optimization problems.