摘要：AbstractCausal network modeling is an important part of alarm root cause analysis in industrial process. The transfer entropy is an effective method to model the causal network. However, there are some problems in determining the prediction horizon of transfer entropy. To solve the problems, a modified transfer entropy, which consider about the prediction horizon from one variable to another and to itself simultaneously, is proposed to improve the capacity of causality detection. Moreover, based on the data-driven and process knowledge modeling methods, an approach combining the modified transfer entropy with superficial process knowledge is designed to correct false calculations and optimize causal network models. Two case studies including a stochastic process and Tennessee Eastman process are carried out to illustrate the feasibility and effectiveness of the proposed approach.
