Research On Constrained Multi-objective Evolutionary Algorithm Based On Coevolution

Multi-objective optimization problems are heavily involved in scientific and engineering applications.Since the 21 st century,evolutionary algorithms have been more commonly used in solving multi-objective optimization problems because of their simplicity and efficiency.However,for the complex constrained multi-objective optimization problems that commonly exist in practical applications,when the scale of decision variables and constraints become more and more complex,the current algorithms are often difficult to meet the convergence and diversity more effectively..To this end,this paper discusses and studies the evolutionary algorithms for solving complex constrained multi-objective optimization problems as follows:In order to deal with complex constrained multi-objective problems more effectively,this paper proposes a two-stage strategy-based constrained multi-objective co-evolutionary algorithm(TSC-CMOEA).The algorithm is mainly developed and improved for the two main problems of population convergence and distribution balance and computational efficiency.The algorithm divides the search process into two stages: in the first stage,the co-evolutionary strategy is used to generate better solutions by cooperating with the population corresponding to the auxiliary problem and the population corresponding to the original problem to generate a better solution,so as to quickly cross the infeasible region and approach Pareto.Frontier;in the second stage,the auxiliary problem is discarded when the auxiliary problem fails,and only the evolution of the main population is maintained to save computing resources and strengthen convergence.The combination of the co-evolutionary strategy and the two-stage evolution strategy makes the population converge to the feasible and non-dominated domains more efficiently and quickly.Experiments are conducted on4 sets of benchmarks(C-DTLZ,MW,LIRCMOP,and DOC),and TSC-CMOEA compares algorithms in solving complex constrained multi-objective problems.For the practical engineering application of micro-grid optimization scheduling,the optimization problem often not only has complex constraints,but also has a very high decision dimension.However,the existing constrained multi-objective algorithms are not efficient in solving constrained multi-objective problems with large scale decision variables.As the scale of constrained multi-objective optimization problems increases gradually,it is necessary to study efficient algorithms for solving large-scale constrained multi-objective optimization problems.Therefore,this paper proposes a reference vector-guided domination coevolutionary multi-objective algorithm(CLS-RVEA)for solving constrained large-scale multi-objective problems.Cls-rvea uses a set of uniform reference vectors to mark a fixed number of domain solutions during evolution,thus constructing corresponding subpopulations.In the evolution of each subpopulation,individuals were selected using the newly constructed environmental selection based on Angle penalty distance and dominance relationship.This can increase the selection pressure.At the same time,the above methods are embedded in the framework of coevolution.Through the effect of coevolution strategy,the population can effectively cross over the infeasible region and approach the Pareto frontier efficiently.Experiments on three benchmark test sets(MW,LIRCMOP and CLSMOP)with 100,500 and 1000 decision variables show that the algorithm is effective in constrained large-scale multi-objective optimization.