Fermentation Process Modeling Data Preprocessing StrategiesUsing Neural Networks

摘要

This research effort explores the development of simple neural models for biological process development at thernpilot or laboratory level, representing, in effect, an effort to develop rapid prototyping of biological control schemes.rnThrough the use of a black box simulator of a biological process, backpropagation and the radial basis functionrnnetworks were developed to predict product yields. In view of the model requirements, typical evaluations of neuralrnnet performance, such as root mean square error, may prove to be inadequate or misleading. The preliminary resultsrnshow that using simple neural architectures to develop complex models is possible but requires sensitivity to therninteraction of the requirements for control and modeling, and a context specific awareness of the use of differentrndata preprocessing strategies on the control model.