Problem-specific Multi-objective Evolutionary Algorithms

Multi-objective optimization problems are a kind of problems optimizing simultaneously several conflicting objectives and keeping a balance between the diversity and the convergence of solutions.They are generally divided into constrained and unconstrained problems.Each objective is decided by decision variables,and there exists no any solution to simultaneously optimize all objectives.Therefore,it's necessary to find a compromise solution set for all objectives.In addressing the constrained optimization problem in the real-world scenario,we can say that obtaining a feasible solution takes precedence over optimizing the objective function.The main challenge in constrained optimization is simultaneously handling the constraints as well as optimization of the objective functions.The thesis gives novel multi-objective evolutionary algorithms based on problem information.1.For multi-objective optimization problems,based on problem information and MOEA/D frame,a multi-objective evolutionary algorithm is developed.Firstly,a specific sub-function is separated from a series of objectives,which is applied to provide an approximate search direction and speed the convergence of the algorithm.Then,the crowding degree scheme,as in NSGA-II,is used to select potential promising solutions in the process of iterations such that Pareto solution set has more uniform and extensive distribution.Finally,a novel multi-objective evolutionary algorithm is presented by embedding these schemes into MOEA/D.The simulation results show the proposed algorithm is feasible and efficient.2.For constrained multi-objective optimization problems,an evolutionary algorithm is proposed using a new feasibility measure.Firstly,fitness keeping with feasibility is designed to choose the individual with good feasibility as much as possible.Secondly,a specific sub-function is separated from a series of objectives or constraints,which is applied to provide an approximate search direction and speed the convergence of the algorithm.Thirdly,differential evolution is used to make the population search the optimal solutions.Then,the crowding degree scheme,as in NSGA-II,is used to select potential promising solutions in the process of iterations such that Pareto front is uniform as much as possible.Finally,a novel multi-objective evolutionary algorithm is presented by embedding these schemes.The simulation illustrates the effectives of the proposed algorithm.