Machine learning techniques for the prediction of polymerization kinetics and polymer properties

Abstract

In the current study, the ability of two data‐driven machine learning tools, the extreme learning machine (ELM) and the adaptive neuro‐fuzzy inference system (ANFIS), to predict the polymerization rate and melt flow index of linear low‐density polyethylene produced in a gas phase process was investigated. The level of interaction between the input variables (ethylene, 1‐butene, isopentane pressures, and reaction temperature) on the outputs (melt flow index and activity) was also examined. It was found that both outputs are impacted by the presence of isopentane as an induced condensing agent. Various statistical indicators, including the coefficient of correlation (R2) and root mean square error (RMSE), were used to quantitatively evaluate both developed models. The ANFIS model outperformed the developed ELM model in terms of predicting the MFI and the catalyst activity. A sensitivity analysis of the ANFIS and ELM models showed that all the input variables under investigation had a sizable impact on the responses and none of them could have been discarded. The present study showed that machine learning tools could be employed to adequately develop empirical models to predict polymerization kinetics as well as the final polymer properties.