估计中模糊先验信息的一类定量描述方法

A Method for Quantitatively Describing Fuzzy Prior Information in Bayesian Estimation

摘要:

在Bayes可靠性评估中，为了提高小样本条件下可靠性的精度，需要利用专家经验等信息.而可靠性工程专家习惯于将自己的意见用模糊信息来表述.基于模糊隶属函数，对专家模糊经验信息做出了定量描述，并在此基础上利用Bayes方法实现了语音选择器的专家信息与实验数据的有效融合.实例表明，在专家经验信息的置信区间较宽时，采用三角型模糊分布能有效提高可靠性评估的精度.而置信区间较窄时，正态型分布具有更好的融合效果.

Abstract:

In order to precisely evaluate reliability under small sample size, prior information such as expert judgment is needed in Bayesian reliability estimation. General Bayesian method cannot deal with expert judgment when it is fuzzy. With a quantitative description method of fuzzy prior distribution for microphone selectors introduced, based on fuzzy membership functions, experts’ prior information can be effectively merged with test data with Bayes method. Reliability evaluation shows that, the precision can be enhanced notably for data with small samples by using Bayes estimation with fuzzy prior distributions. Furthermore, triangle fuzzy prior distributions can be used to enhance the precision when the bandwidth of the fuzzy prior distributions are wide. And normal distributions are applicable to the circumstance when the fuzzy prior distributions are narrow.