基于K-L信息距离的多源信息融合法

Multi-source information fusion based on K-L information distance

为解决小子样条件下岩土参数概率分布推断的难题,并克服基于专家信息的融合法不可避免地带有主观随意性的弊端,引入信息论中K-L信息距离的概念,基于先验信息可信度,提出一种新的多源信息融合方法。利用K-L信息距离作为参数分布之间距离的度量,定义先验分布差异率,确定融合权重,进而根据Bayes原理得到后验分布,优化岩土参数分布概型。工程实例分析结果表明,该法计算简单,且克服了推断过程中的主观随意性。计算结果显示该法所得融合分布的方差比已有成果所得方差偏小,说明该法可实现统计意义上的参数概型优化,为岩土参数设计值的合理选取提供了参考。

In order to solve the problem of geotechnical parameters distribution inferring on the condition of small sample and get over the inevitable shortcoming of subjective randomness in those fusion methods based on expert＇s experience,the notion of K-L information distance in the information field was introduced;and multi-source information fusion method was proposed on the basis of the credibility of prior distributions.Using K-L information distance as a measurement of distances between parameter distributions,＂the rate of prior distribution difference＂was defined and the fusion weight was determined.Further,the posterior distribution was obtained by using the Bayes principle;and the fitting probability models of geotechnical parameters were optimized. It is shown in the process of a project case that the method suggested is simple while no subjective factors are included in the statistical inference.The results show that the variance of the suggested fusion distribution is smaller than ones of the existing results,which illuminates that an optimum fitting probabilistic model for parameters in statistical sense can be resulted from the proposed method;and it provides a reference for choosing design values of geotechnical parameters reasonably.