A structured modeling approach for dynamic hybrid fuzzy-first principles models

Hybrid fuzzy-first principles models can be attractive if a complete physical model is difficult to derive, These hybrid models consist of a framework of dynamic mass and energy balances, supplemented with fuzzy submodels describing additional equations, such as mass transformation and transfer rates. In this paper, a structured approach for designing this type of model is presented. The modeling problem is reduced to several simpler problems, which are solved independently: hybrid model structure and sub-process determination, subprocess behavior estimation, identification and integration of the submodels to form the hybrid model. The hybrid model is interpretable and transparent. The approach is illustrated using data from a (simulated) fed-batch bioreactor. (C) 2002 Elsevier Science Ltd. All rights reserved.