Multi Objective Optimization using Non-Dominated Sort Genetic Algorithm with Artificial Neural Network for Reactive Dividing Wall Column

In this study, multi-objective optimization of reactive dividing wall column is presented. Production of methyl acetate from acetic acid and methanol is taken as case study. Machine learning approach is introduced in this work by means of artificial neural network and genetic algorithm. Required data generation, input and output variable fixation to model neural network is done from the sensitivity analysis. Based on the dataset, neural network model is trained by Lavenberg–Marquardt algorithm and predict purity and TAC of column with high accuracy. Further parametric constrained optimization of systems has been done using multi-objective genetic algorithm and set of pareto optimal solution is generated. Based on gray relational analysis, best optimal point found out. After optimization the system gives significant reduction on TAC and enhancement in purity. Results shows reactive dividing wall column reduces total annual cost around 17.77%. All the results in present work is validated with exiting literature and also cross validated with ASPEN plus.