Publications

Detailed Information

Model-based Experimental Design for Computationally Efficient Parameter Estimation of Fed-batch Bioreactors : 반회분식 생물공정의 파라미터 추정 계산 효율 향상을 위한 모델 기반 실험계획법

DC Field Value Language
dc.contributor.advisor이종민-
dc.contributor.author김정훈-
dc.date.accessioned2019-10-21T02:35:47Z-
dc.date.available2019-10-21T02:35:47Z-
dc.date.issued2019-08-
dc.identifier.other000000156444-
dc.identifier.urihttps://hdl.handle.net/10371/162049-
dc.identifier.urihttp://dcollection.snu.ac.kr/common/orgView/000000156444ko_KR
dc.description학위논문(박사)--서울대학교 대학원 :공과대학 화학생물공학부,2019. 8. 이종민.-
dc.description.abstractIdentification of batch dynamical systems is a tricky task because of its complexity and nonlinearity.
If the macroscopic structure of a model is available, one can utilize Model-based Design of Experiments (MBDOE) method to facilitate the identification process, more precisely, the parameter estimation.
However, a few crucial problems arise in utilizing MBDOE for estimating parameters of batch dynamical systems.
First, the whole design depends on the initial estimate of the parameters.
Second, the gigantic size of the problem prevents one from obtaining reliable solution in practical amount of time.
Third, correlation between the parameters inhibits calculation process of MBDOE.
In this thesis, we propose two new schemes of MBDOEs that solve issues of the existing MBDOE schemes.
The first MBDOE modifies the existing on-line MBDOE into a form that can be efficiently used in large models, solving initial parameter dependency issue, computation time and sensitivity matrix singularity issue.
The second MBDOE improves the existing anti-correlation MBDOE into a form suitable for iterative experiments and causes no numerical instability.
Finally, we apply the combined scheme of proposed methodologies to the microalgal bioreactor model to demonstrate its use, as well as study various issues that can occur when the algorithm is applied in actual cases.
-
dc.description.abstract회분식 및 반회분식 반응기 모델은 매우 복잡하고 비선형성이 크기 때문에, 파라미터 추정이 매우 어렵다.
모델에 대한 구조가 알려져 있는 상태라면, 파라미터 추정을 위해서 모델 기반 실험계획법(MBDOE)를 사용할 수 있다.
하지만 이 MBDOE를 회분식 반응기의 파라미터 추정에 적용할 경우 여러 가지의 치명적인 문제점이 발생하게 된다.
첫 번째, MBDOE의 결과가 초기 파라미터 추정치에 따라 달라진다.
두 번째, 문제 자체의 크기가 너무 커서 한정된 시간 안에 믿을 만한 해를 구하기가 불가능하다.
세 번째, 파라미터들간의 상관성 때문에 수치적으로 안정된 MBDOE 계산을 수행 하는 것이 어렵다.
본 논문에서는 이러한 기존의 MBDOE 기법의 문제점들을 해결하는 두 가지의 새로운 MBDOE 기법을 제안한다.
첫 번째 MBDOE는 기존의 온라인 MBDOE를 그 크기가 큰 모델에도 효율적으로 적용 가능한 형태로 개선하여 초기 파라미터에 대한 의존성 문제, 계산 시간 문제와, sensitivity matrix의 불안정성 문제를 해결한다.
두 번째로 제안한 MBDOE는 기존의 anti-correlation MBDOE을 더 개선시켜서 반복 실험에 적당하고 수치적으로 안정한 형태로 발전시킨다.
마지막으로, 이렇게 제안된 두 가지의 방법론을 반회분식 미세조류 모델의 파라미터 추정 문제에 실제로 적용하여, 알고리즘의 사용 방법을 실제적으로 증명하고, 적용 과정에서 발생할 수 있는 다양한 문제들에 대해 탐구하였다.
-
dc.description.tableofcontents1. Introduction . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Identification of batch processes and experimental designs . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Issues of existing MBDOEs . . . . . . . . . . . . . . 4
1.2.1 Dependence on the initial parameter estimate . 4
1.2.2 Numerical size of the problem . . . . . . . . . 4
1.2.3 Correlation between the parameters . . . . . . 5
1.3 Current approaches to the issues . . . . . . . . . . . . 6
1.3.1 Dependence on the initial parameter estimate . 6
1.3.2 Numerical size of the problem . . . . . . . . . 7
1.3.3 Correlation between the parameters . . . . . . 8
1.4 Scope of the study . . . . . . . . . . . . . . . . . . . 10
1.5 Outline of the thesis . . . . . . . . . . . . . . . . . . 11
2. Preliminaries . . . . . . . . . . . . . . . . . . . . . . . 12
2.1 Model-based design of experiments (MBDOE) . . . . 12
2.1.1 Basic formulation . . . . . . . . . . . . . . . 12
2.1.2 Issues seen in detail . . . . . . . . . . . . . . 14
2.2 On-line MBDOE . . . . . . . . . . . . . . . . . . . . 21
2.3 Anti-correlation MBDOE . . . . . . . . . . . . . . . 25
3. Parameter subset selective on-line MBDOE . . . . . . 27
3.1 Objective of the methodology . . . . . . . . . . . . . 27
3.2 Theoretical formulation . . . . . . . . . . . . . . . . 28
3.2.1 Parameter subset selection . . . . . . . . . . . 28
3.2.2 Optimal input calculation . . . . . . . . . . . 33
3.2.3 Implementation and parameter re-estimation . 34
3.3 Demonstration . . . . . . . . . . . . . . . . . . . . . 34
3.3.1 Model description and problem settings . . . . 36
3.3.2 Result . . . . . . . . . . . . . . . . . . . . . . 37
3.3.3 Comparison for different number of subset parameters . . . . . . . . . . . . . . . . . . . . 44
3.3.4 Effect of model conditions and hyper-parameters
on the performance of the scheme . . . . . . . 48
4. Successive complementary anti-correlation MBDOE . 50
4.1 Objective of the method . . . . . . . . . . . . . . . . 50
4.2 Theoretical formulation . . . . . . . . . . . . . . . . 53
4.2.1 Initial experimental design . . . . . . . . . . 53
4.2.2 Complementary design formulation . . . . . . 53
4.2.3 Iteration and termination . . . . . . . . . . . 58
4.3 Case study . . . . . . . . . . . . . . . . . . . . . . . 59
4.3.1 Model description . . . . . . . . . . . . . . . 59
4.3.2 Solution method . . . . . . . . . . . . . . . . 60
4.3.3 Result . . . . . . . . . . . . . . . . . . . . . 61
4.4 Remarks on the choice of hyper parameters . . . . . 73
4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . 75
5. Application to a microalgal fed-batch bioreactor . . . 79
5.1 Necessity of the combined scheme . . . . . . . . . . 79
5.2 Overall scheme of the study . . . . . . . . . . . . . . 82
5.3 Model description . . . . . . . . . . . . . . . . . . . 85
5.4 Parameter subset selective on-line MBDOE . . . . . . 89
5.4.1 Simulation settings . . . . . . . . . . . . . . . 89
5.4.2 Result . . . . . . . . . . . . . . . . . . . . . . 91
5.5 Successive complementary anti-correlation MBDOE . 102
5.5.1 Simulation settings . . . . . . . . . . . . . . . 102
5.5.2 Result . . . . . . . . . . . . . . . . . . . . . . 103
5.6 Comparison to the D-optimal-only case . . . . . . . . 113
5.7 Remarks . . . . . . . . . . . . . . . . . . . . . . . . 117
5.7.1 Choice of the solution method . . . . . . . . . 117
-
dc.language.isoeng-
dc.publisher서울대학교 대학원-
dc.subjectBatch process-
dc.subjectSystem identification-
dc.subjectParameter estimation-
dc.subjectModel-based design of experiment-
dc.subject.ddc660.6-
dc.titleModel-based Experimental Design for Computationally Efficient Parameter Estimation of Fed-batch Bioreactors-
dc.title.alternative반회분식 생물공정의 파라미터 추정 계산 효율 향상을 위한 모델 기반 실험계획법-
dc.typeThesis-
dc.typeDissertation-
dc.contributor.AlternativeAuthorKim, Jung Hun-
dc.contributor.department공과대학 화학생물공학부-
dc.description.degreeDoctor-
dc.date.awarded2019-08-
dc.identifier.uciI804:11032-000000156444-
dc.identifier.holdings000000000040▲000000000041▲000000156444▲-
Appears in Collections:
Files in This Item:

Altmetrics

Item View & Download Count

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share