Hybride Modellierung für die Vorhersage und Überwachung des Wachstums von Mikroalgen

Produktbeschreibung

Microalgae hold significant potential for biofuel, biomaterial, and bio-based chemical production, necessitating
advanced modeling approaches for optimizing cultivation. This dissertation evaluates machine learning (ML) models,
specifically Long Short-Term Memory (LSTM) and Support Vector Regression (SVR), against traditional Monod
and Haldane models for predicting microalgae growth under varying light conditions in outdoor flat-panel
airlift photobioreactors.

Contents
Abbreviations IX
1. Introduction 1
1.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2. Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3. Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4. Overview of the dissertation . . . . . . . . . . . . . . . . . . . . 6
2. Literature Review 9
2.1. Factor affecting the microalgae growth . . . . . . . . . . . . . . 9
2.1.1. Light . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.2. Nutrients . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.3. Temperature . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.4. pH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.5. Salinity . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2. Types of microalgae cultivation system . . . . . . . . . . . . . . 12
2.3. Open systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.1. Types of Open Systems . . . . . . . . . . . . . . . . . . 13
2.3.2. Raceway ponds . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.3. Circular ponds . . . . . . . . . . . . . . . . . . . . . . . 13
2.4. Closed Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4.1. Types of Closed Systems . . . . . . . . . . . . . . . . . . 15
2.4.2. Flat panel PBRs . . . . . . . . . . . . . . . . . . . . . . 15
2.4.3. Tubular PBRs . . . . . . . . . . . . . . . . . . . . . . . 15
2.4.4. Verticle column PBRs . . . . . . . . . . . . . . . . . . . 16
2.5. Kinectic models for the prediction and optimization of microalgae
growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5.1. Models accounting for light intensity effect . . . . . . . . 17
2.5.2. Models accounting for light intensity and temperature effect 21
2.5.3. Models accounting for light intensity and substrate effect 24
V2.6. Machine learning approaches to enhance microalgae growth and
productivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.6.1. Prediction of microalgae growth or productivity using neuralnetwork-based and non-neural-network-based approaches . 27
2.7. Significance of hybrid modeling approaches . . . . . . . . . . . . 33
2.7.1. Hybrid modeling application in biotechnological processes 34
2.7.2. Hybrid modeling application in chemical engineering . . . 36
3. Methodology 41
3.1. Identification of Research Gaps . . . . . . . . . . . . . . . . . . 41
3.1.1. Challenges in Data for Microalgae Cultivation . . . . . . . 41
3.1.2. Challenges for modeling of microalgae cultivation . . . . . 42
3.2. Strategies for Addressing Gaps and Focus of the Dissertation . . 42
3.2.1. Potential solution for the dataset enhancement . . . . . . 42
3.2.2. Potential solutions for modeling rigorousness enhancement 44
3.3. Research Questions, Hypothesis and Methodology . . . . . . . . 45
4. Improving Microalgae Growth Modeling of Outdoor Cultivation
with Light History Data using Machine Learning Models: A
Comparative Study 49
4.1. Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.1.1. Description of cultivation . . . . . . . . . . . . . . . . . 50
4.1.2. Data preprcoessing . . . . . . . . . . . . . . . . . . . . . 51
4.1.3. Training and Test Dataset . . . . . . . . . . . . . . . . . 51
4.2. Traditional models . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.2.1. Average light intensity . . . . . . . . . . . . . . . . . . . 52
4.2.2. Monod and Haldane model . . . . . . . . . . . . . . . . 54
4.3. Machine learning models . . . . . . . . . . . . . . . . . . . . . . 54
4.3.1. Support vector regression . . . . . . . . . . . . . . . . . 55
4.3.2. LSTM . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.3.3. Analysis of the trained machine learning models . . . . . 56
4.4. Demonstration of applications . . . . . . . . . . . . . . . . . . . 57
4.4.1. Biomass softsensor . . . . . . . . . . . . . . . . . . . . . 57
4.4.2. Harvest strategy . . . . . . . . . . . . . . . . . . . . . . 57
VI5. Hybrid Modeling for the Prediction and Monitoring of Microalgae Processes 59
5.1. Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.1.1. Description of Dataset . . . . . . . . . . . . . . . . . . . 60
5.1.2. Data Preprocessing . . . . . . . . . . . . . . . . . . . . 61
5.1.3. Training and Test Dataset . . . . . . . . . . . . . . . . . 62
5.2. Traditional and Machine Learning Models . . . . . . . . . . . . . 63
5.2.1. Monod Model . . . . . . . . . . . . . . . . . . . . . . . 63
5.2.2. LSTM Model . . . . . . . . . . . . . . . . . . . . . . . . 64
5.2.3. Biomass prediction using Runge-Kutta . . . . . . . . . . 66
5.3. Hybrid Modeling for the prediction of biomass concentration . . . 67
5.3.1. Hybrid Model Approach 1 . . . . . . . . . . . . . . . . . 67
5.3.2. The integrator cell . . . . . . . . . . . . . . . . . . . . . 68
5.3.3. Hybrid Model Approach 2 . . . . . . . . . . . . . . . . . 68
6. Results 71
6.1. Evaluation of Microalgae Growth Modeling in Outdoor Cultivation:
Impact of Light History Data and Machine Learning Approaches . 71
6.1.1. Comparison of machine learning models . . . . . . . . . . 72
6.1.2. Light acclimation impact on specific growth rate . . . . . 73
6.1.3. Applications of the models . . . . . . . . . . . . . . . . . 74
6.2. Hybrid Modeling for the Prediction of Microalgae Growth . . . . 77
6.2.1. Evaluation of Sequence Lengths in LSTM . . . . . . . . . 77
6.2.2. LSTM varying light sequence length and train-test batch
ratios . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.2.3. LSTM-based softsensor for the prediction of microalgal
biomass . . . . . . . . . . . . . . . . . . . . . . . . . . 80
6.3. Evaluation of hybrid model . . . . . . . . . . . . . . . . . . . . 81
6.3.1. Performance evaluation of LSTM as a residual predictor of
specific growth . . . . . . . . . . . . . . . . . . . . . . . 81
6.3.2. Evaluation of Hybrid model performance across varying
training and test batches for LSTM residual predictor and
biomass prediction . . . . . . . . . . . . . . . . . . . . . 83
6.3.3. Hyperparameter optimization . . . . . . . . . . . . . . . 86
VII7. Discussion 89
7.1. Machine Learning and Traditional Growth Models in Microalgae
Growth Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . 89
7.1.1. Comparison between machine learning and traditional models 89
7.1.2. Interpretation of Machine Learning Models’ Performance . 90
7.1.3. Interpretation of Light Acclimation and Respiration in Microalgae Growth . . . . . . . . . . . . . . . . . . . . . . 90
7.1.4. Implications of Model Applications . . . . . . . . . . . . 91
7.1.5. Interpretation of Sequence Lengths in LSTM . . . . . . . 91
7.1.6. Interpretation of LSTM Model Performance with Varying
Light Sequence Lengths and Train-Test Batch Ratios . . . 92
7.2. Hybrid Model Sensitivity to Training and Testing Data Variability 92
7.3. Limitations of Hybrid model . . . . . . . . . . . . . . . . . . . . 94
7.3.1. Integration with ML models . . . . . . . . . . . . . . . . 94
7.3.2. Vulnerability of Hybrid Models and LSTM Models to Data,
Normalization, and Scaling . . . . . . . . . . . . . . . . 95
7.4. Comparative analysis of LSTM and Hybrid model . . . . . . . . . 95
8. Summary and Outlook 97
8.1. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
8.2. Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
A. Appendix A 101
A.1. Light Attenuation Model . . . . . . . . . . . . . . . . . . . . . . 101
A.2. Evaluation of Hybrid model performance across varying training
and test batches for LSTM residual predictor . . . . . . . . . . . 101
Bibliography 109
Declaration 131

Keywords: Bedeutung von Mikroalgen, Kultivierungssysteme, Lichtanpassung, Kinetische Modellierung, Maschinelles Lernen, Hybride Modellierung, Runge-Kutta-Löser, Importance of Microalgae, Cultivation Systems, Light Acclimation, Kinetic Modeling, Machine Learning, Hybrid Modeling, Runge-Kutta Solver