Deep Neural Networks for ODE-based Dynamic Optimization in Chemical Processes

Designed a deep neural network framework to solve ODE-based dynamic optimization problems in chemical processes, using a Runge-Kutta approximation to learn continuous control trajectories for reactors and ethanol production.

The approach outperformed ant-colony, biogeography-based, and gradient-based optimization, setting a new best solution for the ethanol-production problem (objective value 20430.672 vs. 20429.9).