基于置信规则库推理的证据滤波报警器设计

Alarm Design of Evidence Filtering Based on Belief Rule Base Inference

为了降低过程变量不确定性对报警结果的影响,给出一种基于置信规则库(BRB)推理的证据滤波报警器设计方法。将通常的绝对阈值转换为模糊隶属度形式的相对阈值,并用其完成从过程变量到报警证据的转化;根据证据距离得到它们之间的相互支持度。采用条件化证据线性更新规则实现滤波,对于更新权重的求取,采用置信规则库建立历史和当前时刻报警证据支持度与更新权重之间的非线性关系,并通过规则推理在线求取更新权重。最后通过过程变量仿真实例的误报漏报率统计分析,说明所提BRB方法能够充分利用专家知识实现更新权重的在线调节,性能更加优越。

In order to reduce the influence of process variable uncertainly on alarm results, a design method of evidence filtering alarm based on belief rule base（BRB） inference was proposed in this paper. The usual absolute threshold was converted into the relative threshold of the fuzzy membership form, with which the transformation from the process variable to the alarm evidence was completed. According to the evidence distance, the mutual support between them was obtained. Conditional evidence linear updating rules were adopted to achieve the filtering and gain the updating weight. BRB was used to establish the non-linear relationship between historical and current time alarm evidence support and updating weight. With rule-based inference, the updating weights were obtained online. The results of the statistical analysis of false and missing alarm rates of process variables show that, the proposed BRB method has a better performance by making full use of expert knowledge to realize the online regulation of updating weight