Dynamic Constrained Multi-objective Optimization Algorithms Based On Decision Variable Classification

Dynamic constrained multi-objective optimization problems are widely used in practical applications,such as task planning,project scheduling and energy allocation.The objective functions and/or constraints of this kind of problem change with time,leading to the change of Pareto optimal solution set or front.Therefore,how to track the feasible Pareto optimal solutions quickly and accurately in the dynamic environment is the key issue to solve the dynamic constrained multi-objective optimization problem.In this thesis,based on the existing dynamic multi-objective optimization research,two dynamic constrained multi-objective evolutionary algorithms are proposed for dynamic constrained multi-objective optimization problems with time-varying objective functions and constraints.(1)A dynamic constrained multi-objective evolutionary algorithm based on decision variable classification is proposed for dynamic constrained multi-objective optimization problem.In order to make full use of the characteristics of decision variables to quickly track the feasible Pareto optimal solution set,a decision variable classification method is proposed considering the constraints.According to the influence of decision variables on convergence,distribution and constraint value,the decision variables are divided into four categories.When the environment changes,the hybrid response strategy implements the corresponding change response method for different types of decision variables to produce high-quality initial population.At the same time,an offspring generation method is proposed to effectively use convergence-related and distribution-related decision variables to generate offspring,so as to accelerate population to converge.Experimental results show that the proposed algorithm is superior to the other four dynamic constrained multi-objective evolutionary algorithms.(2)In order to improve the efficiency of solving dynamic constrained multi-objective optimization problems,a dynamic constrained multi-objective evolutionary algorithm based on multi-population coevolution is proposed.Based on the above work,an improved decision variable classification method is designed to divide decision variables into three types: constrained convergence,convergence and distribution variables,so as to improve the utilization efficiency of decision variables.When environmental changes are detected,Kalman filter prediction,center point prediction and precise controllable mutation operator are used to respond to the above three types of decision variables respectively,and initial population is generated in the new environment,so as to achieve the purpose of tracking Pareto optimal front in dynamic environment quickly and accurately.In addition,in order to accelerate population to converge to the feasible region in a static environment,a multi-population coevolution framework is proposed.According to the classification results of decision variables,the population is divided into two sub-populations,and the interaction of feasibility,convergence and distribution is achieved through the construction of completed solutions,so as to improve the evolutionary efficiency.Experimental results show that the proposed algorithm can efficiently solve dynamic constrained multi-objective optimization problems.The thesis includes 19 figures,8 tables and 103 references.