Highly Parallel Simulated Annealing Applied to the Optimization of Structural Systems

模拟退火算法是解决复杂优化问题的概率优化方法之一，近年来开始应用于真实的世界问题。然而，该方法需要多次重复计算，因此其计算成本变得非常高。因此，对优化方法进行并行化研究是非常重要和迫切的。我们的研究主要集中在模拟退火的并行化和有效的并行模拟退火算法的发展。本文的主要结论如下：1）将温度并行模拟退火算法推广到连续优化问题，并提供了有效的性能; 2）该方法可以在恒定接受概率的基础上自动调整邻域范围;该算法可以自动确定连续优化问题模拟退火算法中的最高温度4）提出了一种基于不同邻域范围并行退火的邻域范围确定方法，该方法具有良好的搜索性能。

Simulated annealing is one of the probabilistic optimization methods for solving complicated optimization problems, and it is started to be used in real world problems recently. However, the method requires many repeated calculations and hence its computational cost becomes very high. Therefore, it is very important and urgent to parallelize the optimization method. Our research is focused on the parallelization of simulated annealing and the development of effective parallel algorithms for parallel simulated annealing. The followings are the conclusions.1) The extension of Temperature Parallel Simulated Annealing to continuous optimization problems is proposed and it provides effective performance.2) The neighborhood range can be automatically adjusted by the proposed method based on the constant probability of acceptance.3) The maximum temperature of the temperature schedule in simulated annealing for continuous optimization problems can be automatically determined by the proposed method where the temperature increases from the lowest temperature instead of cooling.4) A new method for determination of the appropriate neighborhood range based on the parallel annealing with different neighborhood ranges is proposed, and it yields good search performance.