将Matlab中全局优化工具箱中的模式搜索求解器与割峰函数结合,提出一个两阶段模式搜索算法。首先通过模式搜索求解器求解包含多个极小值的优化问题,返回结果后,在当前迭代点处定义割峰函数,继而采用模式搜索求解器进一步极小化辅助函数寻找比当前结果更好的下降点。该算法简单易行,数值结果表明新算法提高了模式搜索求解器获得全局解的效率。
We presented a two stage pattern search method, which combined the cut-peak function and the pattern search solver in Matlab. A simple cut-peak function and choice function were defined at solution returned by pattern search solver.By minimizing the choice function, a global descent of the original objective function was assured. Since the pattern search method did not require the gradient of the choice function, smoothing technique was not employed. The new algorithm was simple to implement and numerical results indicated that the new method improved the efficiency of finding the global minimization.