Intelligent prediction and early warning of abnormal conditions for fluid catalytic cracking process

Fluid catalytic cracking (FCC) is a key unit in the petrochemical production process with frequently encountered abnormal conditions and great safety challenge due to its complex and harsh production environment. The prediction and early warning of abnormal conditions in FCC process is able to improve the safety and stability of production process and avoid the occurrence of severe accidents. In this paper, a data-driven and knowledge-based fusion approach (DL-SDG) is proposed for prediction and early warning of abnormal conditions in FCC process. Firstly, the key variable is identified as prediction target of the process through the calculation of centrality in complex network. Secondly, Spearman ranking correlation coefficient is used for the selection of feature variables to reduce the input data dimension and improve the prediction accuracy of the deep learning (DL) model. Then, the long short-term memory network with attention mechanism and convolution layer is applied to predict the future trend of the key variable. Finally, the signed directed graph (SDG) model deduces the propagation path of abnormal conditions based on the predicted results of key variable to facilitate handling the anomaly in time. The proposed method was successfully applied to a typical FCC unit in a petrochemical enterprise with an excellent performance.