ANN-assisted optimization-based design of energy-integrated distillation columns

The optimal design of chemical processes is of essential importance for an increased sustainability. However, the resulting non-convex mixed-integer nonlinear programming (MINLP) problems cannot directly be solved to global optimality. Therefore, different alternatives have been proposed, which either build on the application of a simulation-based optimization by means of a metaheuristic or the global optimization of a surrogate model, both requiring extensive simulations. The current work proposes a novel alternative approach for a surrogate-assisted hybrid optimization, which exploits a local deterministic optimization of a full MINLP problem to generate a compact artificial neural network (ANN) model that allows for the direct optimization on a reduced search space. In order to provide a sufficient accuracy of the ANN while targeting the global optimum of the design problem, a tailored mixed adaptive sampling is introduced. Application of the algorithm is illustrated for the optimal design of a distillation-based separation of benzene, toluene, and xylene with different means for energy integration.

Subjects

artificial neural networks

distillation

energy integration

optimization

sampling

MLE@TUHH

DDC Class

500: Science

Options

ANN-assisted optimization-based design of energy-integrated distillation columns