Investigation and optimization of olefin purification in methanol-to-olefin process based on machine learning approach coupled with genetic algorithm

In this study, the goal was to develop a robust model for prediction of the performance of a purification section of methanol to olefin (MTO) process based on machine learning approach. The optimum operating conditions were determined by performing a genetic algorithm as an optimization technique. Finding the optimum conditions caused a considerable decrease both in fixed capital investment (FCI) and working capital investment (WCI). To do this, the separation section of MTO process was investigated and modelled through artificial intelligence (AI). This separation section of MTO process is comprised of three columns: C2-stripper, deethanizer, and C3-stripprer. For each column, three operative parameters (number of stages, reflux ratio, and pressure) were selected and investigated. The performance of columns was assessed through monitoring the purity of products in top stream and energy consumption. The experimental layout was designed using response surface methodology (RSM). And the obtained data were used to develop artificial neural network (ANN) models for each column. Several structures of ANNs were investigated to select the optimum model parameters. The observations show good agreement between the real and predicted data.