Improving the Efficiency of Stochastic Nelder-Mead Simplex Method For Simulation Optimization

• Main Authors: Lian, Fu Shi, 連福詩
• Other Authors: Chang, Kuo Hao
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/22286602355238503339

碩士 === 國立清華大學 === 工業工程與工程管理學系 === 103 === Stochastic Nelder-Mead simplex method (SNM) is a direct search algorithm in simulation optimization, which can deal with some problems which are unsmooth or whose gradient does not exist. Comparing with the Nelder-Mead simplex algorithm (NM), SNM used an effective method for determining the sample size, and effective local search and global search architecture. It solved two problems of the NM algorithm: (1) lacking of sample size scheduling, (2) the optimal quality cannot be determined. However, the sample size assigned for each variable in each iteration are the same in SNM method, leading to lower operational efficiency. This article determines the number of experiments assigned to each variable in each iteration by using OCBA (Optimal Computing Budget Allocation) method to improve the efficiency of the algorithm of SNM. Meanwhile, this paper proposes a modification for the reflection step to reduce the probability of contraction. Experimental results show that the proposed I-SNM algorithm can effectively improve efficiency.