Clustering Algorithm Based on Genetic Simulated Annealing Optimization
While traditional genetic algorithms exhibit significant individual diversity during early iterations, the classic roulette wheel selection mechanism causes offspring production to correlate directly with parental fitness values. This often leads to premature convergence as superior individuals dominate the population prematurely. During later stages, fitness values tend to converge, diminishing the reproductive advantages of elite individuals and stalling evolutionary progress. The algorithm incorporates fitness scaling where temperature-controlled annealing maintains balanced selection pressure during high-temperature phases (early iterations), while intensified scaling at lower temperatures amplifies fitness differences to accentuate elite advantages. This hybrid approach leverages complementary strengths of simulated annealing and genetic algorithms to overcome premature convergence, with customized genetic encoding and fitness functions specifically designed for clustering problems to ensure efficient global convergence.