Application of adaptive neuro-fuzzy inference system for the prediction of the yield distribution of the main products in the steam cracking of atmospheric gasoil

Abstract

The capabilities of atmospheric gasoil as a feedstock of steam crackers were determined through several experiments in a pilot plant. The operating variables that were considered in these experiments were the coil outlet temperature (COT), the feed flow rate and the steam ratio. To investigate the dependence of the yield of the produced olefins on the operating variables, a model based on the adaptive neuro-fuzzy inference system (ANFIS) network, was developed. The developed model utilises the Sugeno inference system and is based on a partitioning algorithm. The operating variables are considered the inputs to the model and the yields of the products and the rate of coke formation are the outputs. The results of the model were compared with the experimental data and the results from a rigorous kinetic model. The values of the statistical parameters R2, RMSE and MRE reveal that the ANFIS model is accurate. A sensitivity analysis was performed to determine the effects of changes in the operating variables on the yields of the products and the rate of coke formation. The analysis indicated that the residence time in the reactor has the greatest impact on the yields of the products. Also the temperature is the second important operating condition that has a strong effect on the yield of the product. Furthermore, the results revealed the important role of propylene in formation of the secondary products. Moreover, an optimisation algorithm was utilised to determine the maximum attainable values of the yields of ethylene, propylene and butadiene, which were found to be 35.14%, 19.75% and 7.11%, respectively.