| Summary: | 碩士 === 國立中正大學 === 化學工程研究所 === 85 === In the research,we adopt Saccharomyces diastaticus LORRE-316 as
the strain of the ethanol fermentation that some
characteristics,like high final ethanol
concentration,high ethanol tolerance,high cell density,fast
fermentation and cell self-flocculation.The parameter
estimation of multiobjective optimization is
applied to set up the growth model of the strain.The method
helps to make the dynamic action more accurate.
Besides,the neural network learning model is an
effective method to deal with more complicated dynamic model
that can be established by our experimental
data,training,as well as learning without
knowing other specific dynamic model.
No matter parameter estimation of multiobjective optimization or
neural network is used to set up the action
model,we will get more accurate result by increasing
our learning category.The increasing in our learning category
can also modify our dynamic procedure making the
established model meet the practical situation
more.Because of the complication of biochemical
reaction ,we have to follow the manipulative range as our
learning example.Thus,good learning
results can be presented no matter the parameter
estimation or neural network is utilized.Final,we apply the
optimal feed strategies of the fed-batch
process to prove the model we establish can
accurately estimate the growth tendency.The yield of the
optimization strategy is higher and the time of the
fermentation is shorter than that of the constant flow
rate strategy.As a result,optimal feed strategies of the fed-
batch fermentation is proved to be necessary.
|
|---|
