Lookup NU author(s): Dr Moritz von Stosch
Full text for this publication is not currently held within this repository. Alternative links are provided below where available.
© 2017 Nova Science Publishers, Inc. Developing a process model is essentially an exercise of translation of existing sources of knowledge into a compact mathematical representation. In some cases, a deep physical understanding exists that is best expressed in the form of a mechanistic model. For other cases, the absence of a comprehensive knowledge base is compensated by measured process data that can be best modeled by empirical methods such as ANNs. For the vast majority of problems both types of knowledge coexist motivating their integration into hybrid mechanistic-ANN models. Many previous studies have shown that the hybrid ANN-mechanistic approach is advantageous in relation to the one or other modeling framework alone. In this chapter, some of the key concepts that support the hybrid ANN-mechanistic modeling are overviewed. Firstly, the main hybrid structures (serial, parallel-competitive, parallel-cooperative and recurrent structures with information feedback) are reviewed. Then the underlying structure-dependent identification methods are covered, with a focus on ANNs training constrained by an existing mechanistic model. Finally, two case studies are presented. The first case study reports a static serial ANN-mechanistic model that describes the burst phenomena in controlled drug release. The second case study presents a serial ANN-mechanistic model for dynamic modeling and optimization of a bioprocess.
Author(s): Azevedo C, Lee R, Portela RMC, Von Stosch M, Oliveira R
Publication type: Book Chapter
Publication status: Published
Book Title: Artificial Neural Networks in Chemical Engineering
Print publication date: 01/01/2017
Acceptance date: 02/04/2016
Publisher: Nova Science Publishers, Inc.
Place Published: New York
Library holdings: Search Newcastle University Library for this item