基于IGA与增强D-S证据理论的工业过程故障检测

打开文本图片集

中图分类号：TQ02 文献标志码：A DOI：10.3969/j.issn.1003-9015.2025.03.013

文章编号：1003-9015（2025）03-0503-19

Abstract：Toaddress thechalengesoffeature extraction incomplex industrial processdata，the need forfault detection methods to assume prior data distribution types，and the weak generalization capability of fault detection methods，we propose an industrial processfault detection algorithm based on Information Gain Adaptive Feature Selection（IGA）and enhanced D-S（Dempster-Shafer） evidence theory.First，information gain was calculated using a decision tree，and the numberoffeatures wasadaptivelyselected byseting a thresholdvalue through cross-validation，achieving data dimensionality reductionand feature extraction.Subsequently，a ternary statistic monitoring group was introduced to achieve linear fault detection，while autoencoders and one-class SVM methods were employed for nonlinear fault detection.Finally，enhanced D-S evidence theory was used to fuse theimproved weighted average evidence body through a new comprehensive metric，correcting Basic Probability Assignment（BPA）parameters to obtainthe fusion result and more reliable detection results.Verification using data from achiler unit and the Tennesse Eastman（TE）process indicated that the IGA and enhanced D-S evidence theory method achieved higher accuracy and robustnessthan standalone linear or nonlinear detection algorithms.Moreover，its application does not depend on specific data distribution assumptions，making it versatile and effectively applicable in industrial process fault detection.

Key words： information gain；feature extraction；evidence theory；fault detection；industrial process

1前言

近年来，随着工业化进程的发展，各类工业生产过程变得复杂化。（剩余19053字）