The Causal Network Modeling And Alarm Root Cause Analysis Based On Modified Transfer Entropy

Alarm system is an important facility to ensure the safety of process industry production.The alarm flooding is one of the major problem to'degrade the performance of alarm system.The causal network modeling and root cause analysis can provide the key alarm information to operator to prevent them from being disturbed by alann flooding.Because of the good ability of causality detection and root cause inference,it has been attracted broad attention by many researches.To address the deficiency of transfer entropy when determining the prediction horizon of reference variable,a modified adjustable reference horizon transfer entropy(ARHTE)is proposed to optimize the prediction horizon of reference variable and improve the significance of transfer entropy.The experiment of stochastic process shows the veracity of proposed algoritlhm.Then the simple process knowledge is employed in removed the redundancy and false result of ARHTE and build the synthetical causal network modeling method combine with ARHTE.Finally,a new ARHTE index is proposed as the criterion to decide the alarm root and build the root cause analysis project based on causal network model.This paper takes the Tennessee Eastman(TE)process as example to build the causal network model of TE process.The experimental result show that the root cause analysis method can detect the alarm root accurately.Contrast with the traditional method,the proposed ARHTE method is reasonable and valid.