Description |
-
Dynamic response surface model (DRSM) methodology is a generalization on response surface models (RSM) that greatly facilitates the ability to determine reaction stoichiometries and rate information for a black box system. Activities reported in this thesis implement new strategies to tackle two of the weaknesses within the existing methodology: oscillatory end behavior and subjective determination ... read moreof significant singular values. Regularization techniques were explored as remedies for the first weakness, while both empirical and statistical methods were explored to resolve the latter weakness. L2 regularization, also known as ridge regression, was shown to reduce the oscillation-based error by over 50% for a targeted problem species in a pharmaceutical reaction system provided by Pfizer Inc. Reduction in oscillatory behavior stabilizes model extrapolation and downstream rate calculations. An f-test strategy that compared variances of singular values was determined best for objective determination of significant singular values, which is necessary to establish the number of expected independent reactions occurring in the black box system. A new iteration of the methodology is presented incorporating these improvements to overcome the two targeted weaknesses.read less
|
This object is in collection