| Real-world application problems often have complex constraints.These constraints make many infeasible regions and complex feasible regions exist in the objective space,which put forward higher requirements for the convergence,diversity and feasibility of constrained multiobjective evolutionary algorithms(CMOEA).How to efficiently obtain a set of efficient solutions for a constrained multi-objective optimization problem(CMOP)is a key challenge.However,the existing CMOEA are not satisfactory in solving CMOP with complex constraints.Especially when the feasible region is extremely small and scattered,the existing CMOEA cannot well balance the convergence,distribution and feasibility of the population.To address this issue,this thesis proposes two co-evolution-based constrained multiobjective evolutionary algorithms to deal with CMOP with complex feasible regions.The first algorithm enhances the search freedom of the auxiliary population in the solution space through a new aggregation method and the auxiliary population restart strategy,which can make the search of the auxiliary population in the solution space more sufficient and unconstrained.At the same time,the feasible solutions searched by the auxiliary population will effectively guide the main population to complete the evolution.This thesis names this algorithm as Enhanced Auxiliary Population Search for Constrained Multi-objective Evolutionary Optimization(EAPS).The second algorithm divides the solution process into three stages,namely the global search stage,the local search stage,and the feasible reusing stage.The first two stages do not consider constraints,and add the searched feasible solutions to the feasible pool,and the last stage reuses the feasible solutions in the feasible pool to guide the population to complete the final evolution.The algorithm achieves the effect of population adaptive adjustment to the problem through a novel adaptive stage switching method.This thesis names this algorithm as Global and Local Feasible Solution Search for solving Constrained Multi-Objective Optimization(GLS-CMOEA).EAPS with five state-of-the-art CMOEA(NSGAⅡ-CDP,CMOEA/D,ToR-NSGAⅡ,PPS,C-TAEA)on constrained test problems such as MW,C-DTLZ,DC-DTLZ,LIR-CMOP,DASCMOP,etc.The experimental results show that the proposed EAPS algorithm has strong competitiveness.The experimental performance of GLS-CMOEA is compared with five stateof-the-art CMOEA(NSGAⅡ-CDP,ToR-NSGAⅡ,PPS,CMOEA-MS,C-TAEA)in constrained test problems such as C-DTLZ,DC-DTLZ,MW,LIR-CMOP,DAS-CMOP,etc.By comparing the IGD and HV performance evaluation metrics,the proposed EAPS and GLS-CMOEA outperform the comparative algorithms on most test problems. |
