Efficient Mathematical Modeling for Bioprocesses Based on Macroscopic Balancesand Neural Networks

摘要

The aims of this thesis are: to apply neural networks and gray box models to real-life systems which are relevant for (bio)chemical processes; to develop an (heuristic) identification procedure for neural networks, addressing issues like selecting proper identification experiments, calculation of the neural network parameters and determination of the neural network configuration; to develop a gray box modeling strategy which can be applied to a wide range of (bio)chemical processes; to understand the extrapolation properties of gray box models so that given the application domain of the model the smallest possible set of identification experiments can be chosen in order to save time and money; to compare neural networks with alternative black box modeling strategies such as polynomials; and to compare the gray box modeling strategy with black box strategies and white box strategies in order to clearly indicate possible advantages and to indicate their potential in the field of improvement of process operation.