Optimization Techniques for Data-Based Control and Machine Learning / von Katharina Bieker ; [Gutachter:innen Jun.-Prof. Dr. Sebastian Peitz, Prof. Dr. Sina Ober-Blöbaum, Prof. Dr. Stefan Klus]. Paderborn, 2023
Inhalt
- Introduction
- Theoretical Background
- Optimal Control and Model Predictive Control
- Data-Based Surrogate Modeling
- Multiobjective Optimization
- DeepMPC for Flow Control - A Motivating Example
- Utilizing Autonomous Models for Model Predictive Control
- The Basic Idea of the QuaSiModO Framework
- Error Bounds
- Numerical Experiments
- Numerical Experiments on Data Efficiency
- Treating l1-Regularized Problems via Multiobjective Continuation
- The Continuation Method
- Optimality Conditions for MOP-l1
- Predictor
- Corrector
- Changing the Activation Structure
- The Algorithm
- Numerical Results
- Towards High-Dimensional Problems
- Generalization to Piecewise Differentiable Regularization Terms
- Conclusion and Future Work
- List of Abbreviations
- Bibliography