Multi-objective optimization of semi-batch copolymerization reactors using adaptations of genetic algorithm

Abstract

The polymerization of styrene-acrylonitrile (SAN) random copolymers in semi-batch reactors is optimized using multiple objective functions that are often conflicting and non-commensurate in nature. These include the average composition of the copolymer product, its number-average molecular weight, its polydispersity index, and the conversion of monomers attained in the reactor. Two decision/control variables are used, namely, the rate of continuous addition of a monomer-solvent-initiator mixture (having a specified and fixed composition) and the history of the temperature in the reactor. The elitist non-dominated sorting genetic algorithm, NSGA-II, is adapted and used for decision variables that are functions of time (trajectory optimization). This robust, AI (artificial intelligence)-based technique, enables the solution of far more complex optimization problems than those reported in the literature. A set of several non-dominating (equally good) Pareto optimal solutions was obtained. These provide insights into the conflicting nature of the objective functions. An engineer (decision maker) can then use his judgment (often intuitive) to choose the preferred solution from among these possibilities.