Removing diversing and trapping in local solution problems in genetic algorithms

ABSTRACT PhD Thesis REMOVING DIVERSING AND TRAPPING IN LOCAL SOLUTION PROBLEMS IN GENETIC ALGORITHMS Ah KARCI Fırat University Graduate School of Natural and Applied Sciences Department of Electric-Electronics Engineering 2002, Page : 78 In this thesis, a method was proposed to generate a partially random initial population and generated population was called uniform population. The expectation of initial population is equal to median of solution space and deformity coefficient of population is zero and population is symmetric with respect to median of solution space. So, it can be called as uniform population. The aim of this thesis is to overcome the diversing from global solution and trapping in local solution problems. After application of this population to some multi-model function to find maximum point of them, uniform population have potential to be trapped in premature solution. Hence mutation and crossover operators were redefined for this population. In general, results of uniform population were better than or equal to results of random population. Island model of genetic algorithms was constructed by making use of structure of uniform population and it was called island uniform population. The process of island uniform population made number of iteration decay rapidly. The properties of uniform population: 1) have diversity, 2) there is no linear dependency between two consecutive generations, 3) there is no potential to be trapped in local solution, 4) there is no potential to go far away global solution, 5) increasing the diversity of population, 6) island uniform population gives better results, 7) the probability of all initial chromosomes tcrbe same is zero, 8) the probability of all pairs of chromosomes that each of them is from different island, to be same is zero, 9) population is symmetric with respect to median of solution space, 10) uniform population, 11) symmetric population with respect to median of solution space. Keywords: Evolutionary Computation, Genetic Algorithms, Random Initial Population, Schema Theory. LX