A Novel Variable Relational Network Construction Method for Plant-wide Monitoring in Chemical Production

With the development of the modern industries, the plant-wide monitoring of the chemical production process is growing significantly. The accurate and effective production variable relational network construction is the basis of the plant-wide monitoring. Compared with the conventional methods depending on the prior production knowledge, the paper proposes a data-driven method integrated the maximal information coefficient and Granger causality test for analyzing the time-vary data to construct the variable relational network. Based on the constructed network and parameters obtained by Bayesian network, not only can the normal or abnormal production states be monitored by statistics, but the fault roots are diagnosed by Bayesian reasoning. The effectiveness and practicality of the proposed monitoring scheme is validated by TE process. The obtained monitoring results can help operators understand the abnormal production state better and guide decision-makers to take the suitable actions for improving the production in chemical process.