基于信息散度的补充样本加权融合评估

Weighted Bayesian Fusion Evaluation Based on Information Divergence of Prior Sample

小子样试验评估时,为避免大量先验信息湮没实际飞行试验的信息,融合先验补充样本时通常会根据工程经验对补充样本量作一定限制,对补充样本量的选取并没有量化的理论准则。分析了先验分布与实际试验样本服从分布的差异,通过分布差异进行信息散度计算,用信息散度确定先验样本权重。提出了考虑先验信息可信度的加权方法,以进行Bayes估计。最后提供了正态逆Gama分布参数的加权Bayes估计方法。理论分析和仿真说明,本文的加权方法是合理的。

In the test evaluation for small sample size, some prior samples are often fused for providing more accurate estimation value. The prior sample size is generally restricted according to engineering background for fear that the fusion result is decided by the prior sample mainly if too many prior sam- ples are fused. However, the restriction for actual test sample size is decided without theoretical analy- sis. The difference between the subjected distributions of the prior sample and actual sample was inves- tigated. Based on the distribution difference, the information divergence was calculated. The weight of prior sample was decided by the information divergence. A weighted method considering the credibility of the prior information was proposed for Bayesian estimation algorithm and the weighted estimation of normal inverse Gama distribution parameters was provided. Taking the estimation of the dispersion of the guided weapon falling points as an example, the conclusion is drawn that our method is credible.