106 シミュレーテッドアニーリングプログラミングにおける温度スケジュールの自動化

抄録

Simulated Annealing Programming(SAP) is a method of automatic programming, which extended Simulated Annealing(SA) so that a tree structure could be treated as a solution of optimization problem. Because of the mechanism of accepting bad solutions probabilistically in the optimization process, SAP can generate the optimal solution without lapsing into local solution. In order to obtain the global optimization solution, SAP needs appropriate temperature schedule, and it requires much computational cost to determine the appropriate temperature schedule. In this research, we propose the method to automatically determine a appropriate temperature schedule. In the proposed method, a temperature schedule is determined based on a history of the acceptance rate. Through the numerical experiments, We found that the proposed method provided an effective temperature schedule.