Machine learning aided multi-objective optimization and multi-criteria decision making: Framework and two applications in chemical engineering

Abstract

To accelerate data-driven studies for various optimization applications in chemical engineering, a comprehensive machine learning aided multi-objective optimization and multi-criteria decision making (abbreviated as ML aided MOO-MCDM) framework is proposed in the present paper. The framework comprises a total of seven steps; firstly, study the application and its input-output datasets to identify objectives, constraints and required ML models; secondly, select ML model(s) for some or all objectives and constraints; thirdly, train the chosen ML model(s), including finding optimal hyperparameter values in each of them using an advanced/global optimization algorithm; fourthly, formulate the MOO problem for the application; fifthly, select a MOO method and develop/test the program; sixthly, solve the formulated MOO problem with the developed/tested MOO program many times and review the Pareto-optimal solutions obtained; lastly, perform MCDM using several methods and choose one Pareto-optimal solution for implementation. The proposed ML aided MOO-MCDM framework is useful for process design and operation of chemical and related processes. It is shown to be beneficial for the optimization of two complex chemical processes, which are supercritical water gasification process aiming for H2-rich syngas with lower greenhouse gas emissions, and combustion process in a power plant targeting for higher energy output and lower pollution of the environment.