Abstract: In this paper, an Adaptive Direct search Simulated Annealing （ADSA） algorithm is developed to overcome poor stability, low convergence rate of Direct search Simulated Annealing （DSA） algorithm for high dimensional global optimization problems. ADSA improves capability of escaping from local extreme and rapidly exploring in adjacent space, by formulating adaptive optimization module and new point generation mechanism with dynamic adjustment search based on iterative temperature. Cauchy distribution state generator enhances the computational efficiency and avoids trapping into local optimum for complex multimodal functions, which is good at searching in wide space. Feasibility-based rule is introduced into ADSA to handle constraints. The algorithm is validated using five complex functions and four standard engineering design problems commonly in the literatures. The results indicate that ADSA is more prominent than DSA, and can effectively solve the high dimensional optimization problems.

Key words: simulated annealing, global optimization, constrained optimization, feasibility-based rule

摘要： 针对直接搜索模拟退火算法求解高维优化问题存在稳定性差、收敛成功率低现象，提出一种自适应的直接搜索模拟退火算法。该算法通过构造基于迭代温度动态调整搜索范围的新点产生方式和自适应寻优模块，增强了算法跳出局部极值和加快邻域搜索的能力，利用柯西分布状态发生函数的大范围遍历特点，弥补了直接搜索模拟退火算法求解高维多峰值问题易陷入局部解和计算效率低的不足。结合可行规则法处理约束问题，典型高维函数和工程优化设计实例的测试结果表明，该算法能够有效求解高维优化问题，整体性能较直接搜索模拟退火算法有显著提高。

关键词: 模拟退火, 全局优化, 约束优化, 可行规则