Research Of Multiobjective Evolutionary Algorithms

Multi-objective Optimization Problem is a new discipline, originated in many practical design of complex systems, modeling and planning.Since their evolutionary nature of Genetic Algorithms, makes it suitable for handling multi-objective optimization problem, Pareto optimal front is not sensitive to the shape and continuity than the traditional method of solving complex problems can be much larger.The basic features of Genetic Algorithm is a multi-direction and global search, the operation targeted a group of individuals, which means from population to population, making Genetic Multi-objective Optimization Algorithm to run a Multi-objective Optimization Problem can find several Pareto optimal solutions, while the traditional optimization methods may require many times the operator to achieve this effect, Genetic Multi-objective Optimization Algorithm to find good convergence and distribution of Pareto optimal solution is very effective when, is an effective Multi-objectiveOptimization Problem.1984, Schaffer proposed VEGA, the difference between VEGA and simple genetic algorithm is a circular mechanism, for each objective function run genetic operations, and produce the final result.Subsequently raised by Pareto ranking ideological Goldberg had MOGA (Fonseca and Fleming, 1993), NPGA (Nafpliotis and Horn,1994) and NSGA (Srinivas and Deb,2000) and other algorithms. MOGA based on the individual to sort the order of values, and assign different fitness value, the same sequence have the same fitness value, well preserved, a better individual.NPGA uses a binary tournament selection mechanism, so that individuals can be randomly selected to compare with more set to be only the fittest survive. NSGA algorithm is the first species to classify, and then click the fitness value assigned.Followed by the introduction of the elitist mechanism (Elitism), produced SPEA (Zitzler and Thiele,1999), SPEA2 (Zitzler et al.2001), PESA (Come et al.2000) and NSGA2 (Deb et al.2002) and other algorithms.NSGA2 comes from NSGA is introduced based on an external population, in practice need to determine the order of value, needs to calculate the crowding distance of individuals. SPEA try to adapt the value of their individual populations by the individual and the external relations of domination decisions, and then use binary tournament selection mechanism to select the individual, the individual which has a smaller fitness is better.SPEA2-comes from SPEA improves the fitness distribution mechanism and the introduction of a minimum neighbor density estimation mechanism of the search process for a more accurate guide, while the introduction of new file truncation method to ensure the preservation of the edge solution.PESA, which has the biggest feature is the use of hyper-box to keep the population diversity, with compression factor to complete the selection and population update mechanism, likely to preserve the Pareto optimal front in the sparse areas in the individual, to maintain a good population diversity sex.This article from the experimental point of view, for the mentioned eight different algorithms to give a comprehensive comparison. Select a few specific test functions used widely to compare all the convergence rate, the population diversity and other properties, and then draw from the. strengths of each method, propose a final improvement in the performance of genetic multi-objective optimization algorithm. IV...