基于改进SPRT的多重响应序贯分支筛选法

IMSB screening method based on improved SPRT

针对仿真实验中多重响应序贯分支法(MSB)不能验证单个因子对所有响应效应显著性这一不足,在改进Wa1d序贯概率比检验(SPRT)的基础上提出了一种新的多重响应序贯分支法(简称IMSB).该方法在仅增加少量仿真次数并在保证统计效力的基础上,即可获取单个因子对所有响应的效应显著性信息.蒙特卡洛对比仿真表明IMSB是切实可行的,在完全保证仿真实验效力的同时准确获取了因子对所有响应的效应显著性判断,且较传统的针对单个响应的SB法减少了大量的仿真次数,该方法在大型复杂仿真系统的后仿真分析中具有良好的实际应用价值.

To address the weakness of multi-response sequential bifurcation（MSB） of being incapable of identifying significance of a given factor for all types of response,this paper proposed an improved MSB method,called IMSB for factor screening in discrete-event simulation.This IMSB,by ameliorating Wald ＇s sequential probability ratio test（SPRT）,can discern any individual factor effect as being important or unimportant for all types of responses with only a small amount of replicates increases and the statistical efficacy is guaranteed.The comparative analysis in Monte Carlo simulation between SB,MSB and IMSB indicates that all these methods give appropriate type-I error rates and powers,and that both MSB and IMSB requires much fewer replicates than SB does.Moreover,unlike MSB,an individual factor effect whether important for a given response is available in IMSB.IMSB can be applied to post-simulation analysis of large-scale simulation,which makes it attractive for a wide variety of simulation applications.