基于LSSVM的威布尔分布形状参数估计

固态介质击穿寿命特性通常用威布尔分布来描述,形状参数卢反应了固态介质的失效特征,因而需要精确估计β值.提出了在小样本情况下基于最小二乘支持向量机(LSSVM)的参数评估方法,并给出了LSSVM在MOS电容与时间有关的击穿寿命分布评估中的应用实例,并与常规的最小二乘评估方法相比,得到的结果表明LSSVM的评估精度更高(均方误差更小)、鲁棒性更好,在小样本情况下能更精确地确定威布尔分布的形状参数.

Shape Parameter Evaluation of Weibull Distribution Based on Least Squares Support Vector Machine

Weibull distribution is often used to describe the lifetime of parts and widely applied to characterize the breakdown statistics of solid dielectric samples.Weibull shape parameter β is a crucial parameter as it characterizes the failure rate trend of solid dielectric,hence,it should be estimated accurately.A new method was proposed based on least square support vector machine (LSSVM) to determine the shape parameters β according to the scarcity of failure data.This method was applied to the reliability estimation of the time dependent dielectric breakdown (TDDB) life data of MOS capacitors.Least square regression (LSR)was also compared with it.The obtained results show that LSSVM has higher accuracy(smaller mean square error) and better robustness than LSR in determining Weibull slope β with small sample.