Dynamic Modeling of Multioperating Conditions for a Three-Converter Gas Holder System in Steel Industry by Using Time-Varying Dynamic Bayesian Networks

The multiple joint Linz-Donawitz converter gas (LDG) holder systems are usually employed to alleviate the LDG fluctuation in steel enterprises. A dynamic modeling method based on time-varying dynamic Bayesian network (TVDBN) is proposed for a three-converter gas holder system (TCGHS). Considering the connectivity, the difference in pressure, and the interlocking rules between different gas holders, the operating conditions are defined in this article by physical mechanism analysis of the operation of this gas holder system. Based on them, the TVDBN's structural model with a mixture of continuous and discrete variables is constructed to describe their transition process through the change of the network structures and the uncertainty relationship of the process variables. The Gaussian regression network is designed for depicting the quantitative uncertain relationship between the gas holder level and gas generation as well as consumption flow rate under each operating condition. Furthermore, as several operating conditions occur rarely due to the operation mechanism, an online parameter learning method is proposed to learn these rare operating conditions according to new coming data. The simulation experiments are performed for a TCGHS of a steel plant in China based on actual data. The results show that the proposed method can not only accurately identify the joint operating conditions of the gas holders and their conversion process but also obtain higher prediction accuracy of gas holder levels compared with the state-of-the-art comparative methods, providing sound support for LDG scheduling.