A novel approach for modeling cracking furnace severity

Due to the strong nonlinear and dynamic behavior of the naphtha cracking furnace, the ordinary feed forward neural network was not suitable to model its severity index. Principal component analysis (PCA) method and dynamic neural network structure were therefore employed at this situation. The model was built on the industrial process data instead of simulation data, and the process operation condition selected for modeling cross over the normal operation range. Comparison with the models based on other algorithms was conducted. The relative parameters that affect the furnace severity were analyzed. The results show that the model based on this approach is suitable in a wide range of operation conditions, and can be used to supervise the process dynamic.