Abstract:

High-dimensional, non-linear, and non-Gaussian distributions of measured data in batch processes directly influence accuracy of detecting abnormal data. In order to integrate information of multi-source abnormal detection and increase detection accuracy, a method was proposed on the basis of multi-evidence fusion decision. With introduction of the Dempster-Shafer evidence theory, the main focal element was used to identify fake evidence and to recompute weight of evidences. The re-calculation on weight of evidences improved handling conflict evidences, reduced influence of conflict evidences on multi-evidence fusion decision, and enhanced detection accuracy of abnormal measured data. Furthermore, an abnormal detection model was constructed from multi-evidence fusion decision and was applied to decision-making of abnormal data detection in batch processes. The experimental results show that the proposed method can combine multi-evidence information and analyze conflict evidence effectively. Thus abnormal data detection for batch processes is achieved with low false and missing detection rates.

Key words: batch processes, Dempster-Shafer theory, conflicting evidence, multi-evidence decision, abnormal measured data detection

摘要：

间歇过程测量数据的高维、非线性、非高斯分布特征直接影响过程测量数据异常检测的准确性，为了融合多源数据异常检测信息，提升间歇过程测量数据异常检测精度，提出了一种基于多证据融合决策的间歇过程测量数据异常检测方法，该方法通过引入证据理论（Dempster-Shafer，D-S），采用主焦元判别伪证据和重新计算证据权重改进冲突证据处理方法，减小了冲突证据对多证据融合决策结果的影响，提高了间歇过程测量数据异常检测的准确率。构建了基于多证据融合的测量数据异常检测模型并将其应用到间歇过程测量数据异常检测决策判决中。实验结果表明，该方法能够融合多证据信息，有效地处理冲突证据，实现了间歇过程测量数据异常检测，降低了误检和漏检率。

关键词: 间歇过程, D-S证据理论, 冲突证据, 多证据决策, 测量数据异常检测