Many-objective Optimization Algorithm Based On Conflict Probability

Many-objective optimization problems,i.e.,the problems to optimize more than 3 objectives,have become a research hotspot in the fields of computational intelligence.Classical multi-objective evolutionary algorithms,which can well solve the problems with fewer objectives,have met great challenges when tackling many-objective optimization problems.This thesis aims to study and design efficient objective space dimension reduction methods for solving many-objective optimization problems.The main research work and relevant achievement of the thesis are summarized as follows.(1)A new method to measure the conflict degree between objectives is proposed,called as conflict probability.By defining the objective conflict degree,a definition for conflict degree among the objectives is given.Based on this definition,a new method to measure objective conflict(i.e.,conflict probability)is proposed.Conflict probability can accurately measure the conflict degree among the objectives,which is suitable for different types of many-objective optimization problems.Conflict probability information constitutes the basis of objective reduction or objective space partitioning in many-objective optimization problems.(2)A many-objective evolutionary algorithm is proposed based on the decomposition and the redundancy on conflict probability information(CPI).A sorting method for all the objectives is proposed according to the conflict contribution rate of objectives.Thus,the objective reduction is obtained,calles as CPIOR.It is further combined to multiobjective evolutionary algorithm which is based on decomposition(MOEA/D)and then MOEA/D-CPIOR is proposed.The experimental results show that CPIOR is more accurate and more robust than other current algorithms,and MOEA/D-CPIOR can improve the optimization performance of MOEA/D on redundant many-objective optimization problems.(3)Objective space extraction on handling many-objective optimization problems is proposed.Based on the conflict contribution of objectives,a simplified method for extracting objective space is proposed to handle many-objective optimization problems.When compared to the objective space partitioning,objective space extraction can improve that the computing resources can be reasonably allocated into each objective sub-space,and thus the diversity performance and computational efficiency of the algorithm are enhanced,which have been verified in the experiments.(4)A self-adaptive objective space reduction method is proposed,which is designed based on the objective conflict information vector.Based on the conflict probability and the importance degree of objectives on the Pareto-optimal front,the improved conflict probability information is proposed to measure the objective conflict,and then the objective conflict information vector can be obtained.Moreover,a self-adaptive objective space reduction method is designed for many-objective optimization problems.This algorithm can adaptively distinguish the redundancy and the most essential objective sub-space of many-objective optimization problems.The objective reduction algorithm will handle the redundant problems,while the objective space extraction algorithm will deal with non-redundant problems.By this way,a self-adaptive objective space reduction method is run for many-objective optimization problems,and combined with evolutionary algorithm,then,the evolutionary algorithm for many-objective optimization problems is obtained,it can improve the optimization performance of multi-objective evolutionary algorithms.