“Genetic algorithm with differntial temperature (GADT)” is a hybrid of genetic algorithm and simulated annealing. In GADT, it is guaranteed that the probability distribution of each individual's state converges to Gibbs distribution. Comparing with conventional genetic algorithms, GADT is easy to be analyzed as Markov chains. GADT can satisfy contradictional requirements “To get good value” and “To escape from local optimum easily” by giving differnent temperature to each individual. For numerical experiments, GADT was used to solve a travelling-salesman problem and results are shown.
キーワード : 遺伝的アルゴリズム，シミュレーテッドアニーリング，マルコフ連鎖，収束
Keywords: genetic algorithm, simulated annealing, Markov chains, convergence