A data-driven industrial alarm decision method via evidence reasoning rule

Abstract In order to deal with the generalized uncertainty of the process variable, under the framework of Dempster–Shafer theory of evidence (DST), a data-driven approach without any probabilistic assumption is presented via the dynamic form of the evidence reasoning (ER) rule. Firstly, the process variable is transformed into the corresponding alarm evidence according to referential evidential matrix constructed by casting historical samples. Secondly, the ER rule is proposed to recursively combine the current and historical alarm evidence to generate the global alarm evidence for alarm decision. In the process of recursive fusion, the forgetting strategy is introduced to calculate the reliability factors of the current and historical alarm evidence; the genetic algorithm is designed to optimize the importance weights of evidence. Finally, numerical experiment and industrial case are given to show that the proposed method has a better performance than the classical methods and the initial conditional evidence updating method.