Optimal control of a fed-batch fermenter using parameterized data-driven models

Abstract

An optimal control approach is proposed for semi-batch processes based on parameterized data-driven (PDD) model structures. Orthonormally parameterized input trajectories, initial states and process parameters are inputs to the model, which predicts output trajectories in terms of Fourier coefficients. Two model structures (linear and quadratic) are incorporated into PDD modeling approach, and a previously proposed model structure of artificial neural networks (ANNs) is considered for comparison. Proposed PDD modeling approach using newly proposed model structures is capable of capturing nonlinear and time-varying behavior inherent in fed-batch systems fairly accurately, and results of operating trajectory optimization using all models are found to be comparable to the results obtained using exact first principles model.