The Research On Constrained Multiobjective Evolutionary Optimization Method Based On State Transition Strategy

Constrained multiobjective optimization problems(CMOPs)are a class of problems with multiple conflicting objectives and constraints.Different from traditional multi-objective optimization problems(MOPs),due to the large number of constraints and complex mathematical form,the treatment of constraints has become the focus of research.The traditional unconstrained optimization methods have the defects of poor convergence and diversity,and can not balance the relationship between optimization objectives and constraints.Therefore,the research on constrained multi-objective evolutionary algorithms(CMOEAs)has important theoretical significance and practical value.In this paper,the constraint processing technology(CHT)in constrained multi-objective algorithm is deeply studied to solve the problem that the search is more difficult due to the constraints.A constrained multi-objective evolutionary algorithm(MOEA/D-STC)based on state transition strategy is proposed.The strategy introduces the framework of multi-objective optimization algorithm based on decomposition.The main research contents of this paper include: first,the algorithm collects the feasible solution proportion of the current population,constraint violation value and other effective information to determine whether the evolution process is in the constraint difficult situation,so as to grasp the status of the population in the evolution process,which plays a guiding role in the strategies of evolution,diversity maintenance and elite selection.Secondly,a new constraint handling strategy is proposed to solve the problem that the existing constraint processing technology is easy to make the population fall into local optimum or converge to unconstrained PF surface.According to the value of the judgment function,the strategy can deal with the constraints flexibly: when the decision population can't cross the obstacles caused by the constraints,the unconstrained optimization is carried out;When the population converges to the unconstrained PF surface,a strict constrained optimization strategy is adopted.Third,according to the improvement ?constraint method can flexibly update the threshold to select more feasible and better individuals.In order to avoid the population converging to local optimal and unconstrained Pareto optimal surface,in addition,a new fitness calculation method is used to improve the epsilon constrained dominance strategy and enhance the diversity of the population.In this experiment,MOEA/D-STC is compared with five most advanced single population CMOEAs(MOEA /D-CDP,MOEA/D-SR,MOEA/D-Epsilon,MOEA/D-IEpsilon,MOEA/DACDP)on three different test problem sets.The experimental results show that the proposed MOEA/D-STC algorithm has good convergence and diversity in dealing with constrained multi-objective problems,compared with the other five algorithms,it has a certain competitiveness.