Many/Multi-Objective Evolutionary Algorithms Using Heuristic Information

The evolutionary algorithm is a classical swarm intelligence optimization method,which is widely used to solve multi-objective optimization problems because of its advantages of simple operation,good universality,insensitivity to function properties and population-based search.With the advent of the era of big data,the number of objectives and the scale of decision variables of multi-objective optimization problems are getting larger and larger,and the structure of functions involved is also becoming more complex than ever.The cost of traditional evolutionary algorithms increases exponentially and the search efficiency decreases significantly.In addition,when some additional restrictions are put on decision variables,some problems maybe have too tight constraints,which make it difficult for the existing algorithms to obtain a sufficient number of feasible solutions,and then decrease the efficiency of the algorithm.In view of the above deficiencies,based on the framework of evolutionary algorithms and the structural characteristics of the problem,some efficient evolutionary operators are developed,and a novel constraint handling scheme is also provided.As a result,these strategies are adopted to improve the performance of the proposed algorithms.The main work is as follows:In multi-objective optimization algorithms,non-dominated sorting always produces a large amount of computation,and in many-objective optimization,the case become more worse due to too many non-dominated solutions.In order to overcome the shortcomings,an evolutionary algorithm using objective probability selection is proposed.Firstly,the selection scheme can effectively reduce the computational cost caused by non-dominated sorting,since only a small number of objectives are selected probabilistically and used to sort.Secondly,according to different objective selection,an external archive set is set up to retain elite individuals.In external archive set,the non-dominated sorting is executed on all objectives and dominated solutions are deleted.Finally,for the crossover operation,the population is divided into two sub-populations according to individual quality,and the crossover operator is designed by using high-quality individuals.The crossover operator using heuristic information can effectively produce potential better offspring.Compared with the current popular multi-objective evolutionary algorithm on benchmark test sets,the simulation results show that the Pareto optimal solution set obtained by the proposed algorithm has advantages in individual distribution and quality.Constraint handling is one of the key issues in constraint multi-objective optimization.Most existing algorithms usually retain feasible solutions through selection operators.In order to obtain a sufficient number of feasible solutions to optimization problems with tight constraints,a new crossover operator is designed by embedding constraint handing method.Firstly,the population is divided into three subpopulations by the number of constraints that each individual satisfies,namely,the set of fully satisfied,partially satisfied and unsatisfied constraints.In these subpopulations,different heuristic techniques are used to design crossover operators which are beneficial to generate high-quality individuals.Then the objective value and constraint violation are used to evaluate individuals,and high-quality individuals are put into the next generation of population.Finally,the proposed algorithm is compared with the state-of-art algorithms on benchmark test sets,and the results show that the proposed heuristic crossover technique is effective in obtaining high-quality feasible solutions.