Wang Jie, Zhou Zhi-Jie, Hu Chang-Hua, Tang Shuai-Wen, Cao You
IEEE Trans Cybern. 2022 Aug;52(8):8088-8100. doi: 10.1109/TCYB.2021.3051676. Epub 2022 Jul 19.
Evidential reasoning (ER) rule has been widely used in dealing with uncertainty. As an important parameter to measure the inherent property of evidence, the evidence reliability makes the ER rule constitute a generalized reasoning framework. In current research of the ER rule, the evidence reliability tends to be expressed in the form of quantitative value by certain methods or expert knowledge. The single quantitative value lacks the ability to describe the statistical property of reliability, which leads to unreasonable results. In this article, a new ER rule with continuous probability distribution of reliability denoted by ERr-CR is proposed. The combination of two pieces of evidence is discussed in detail, where the reliability is profiled as random variables with specific probability distribution. To characterize the output of ERr-CR, a novel concept of expectation of the expected utility is proposed. In addition, the ERr-CR is expanded to multiple pieces of evidence to show its universality. Further, the basic performances of the ERr-CR are explored to illustrate the rationality. Moreover, a case study of safety assessment of natural gas storage tanks (NGSTs) is conducted to show the potential applications of ERr-CR, which makes the proposed method more practical.
证据推理（ER）规则已被广泛应用于处理不确定性问题。作为衡量证据固有属性的一个重要参数，证据可靠性使ER规则构成了一个广义推理框架。在当前ER规则的研究中，证据可靠性往往通过某些方法或专家知识以定量值的形式表示。单一的定量值缺乏描述可靠性统计特性的能力，这导致结果不合理。本文提出了一种新的ER规则，称为ERr-CR，其可靠性具有连续概率分布。详细讨论了两条证据的组合，其中可靠性被描述为具有特定概率分布的随机变量。为了表征ERr-CR的输出，提出了期望效用期望的新概念。此外，将ERr-CR扩展到多条证据以展示其通用性。进一步探索了ERr-CR的基本性能以说明其合理性。此外，进行了天然气储罐安全评估的案例研究，以展示ERr-CR的潜在应用，这使得所提出的方法更具实用性。
