This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: 1 City University of Hong Kong City University of Hong Kong City University of Hong Kong City University of Hong Kong Part 6: Problems and Difficulties Problem 1: Premature Convergence and Genetic Drift s Stochastic errors in sampling caused by small population sizes s Genetic Drift : Population converges on a single peak without differential advantage s Premature Convergence : Loss of population diversity before optimal value has been found s Reason: Loss of niche and species 2 Solution: Niche Maintenance s Preselection : offspring replaces the inferior parent if its fitness is better than that of the inferior parent s Diversity is maintained because strings tended to replace strings similar to themselves s Crowding : individual is compared to a random subpopulation of crowding factor members. The individual with the highest similarity (bit-by-bit similarity) is replaced by the new string Niche Maintenance s Fitness Sharing : a function defined to determine the neighborhood and degree of sharing for each string in the population. s Triangular sharing function s ||•|| can be hamming difference or Euclidean distance. s An individual’s derated fitness is computed as: ≤- = else if 1 ) ( share ij share ij ij d d d s σ σ where j i ij x x d- =...
View Full Document
- Spring '11
- ε, City University of Hong Kong, Partition coefficient, Niche Maintenance, Genetic Algorithms and Simulated Annealing