Modeling Ethylene and Propylene Yield For Cracking Furnace Based On A Kind of New Recurrent Neural Network

This paper employs a kind of novel neural network, recurrent network with dynamic biases, to model the yields of ethylene and propylene for an industrial cracking furnace. The process information of the furnace is introduced to adapt the furnace´s feedstock changes and running phase by the dynamic biases. Comparision with the models based on other algorithms is conducted. The model based on this approach is presented to demonstrate satisfactory result.