Data-Driven Model Reduction for Optimal Control of Large-scale Dynamical Systems
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
In the thesis, we investigate data-driven model reduction for optimal control of large-scale dynamical systems. Optimal control problems play an important role in many engineering applications. However, computational cost is the bottleneck for obtaining optimal control of large-scale dynamic systems. Model reduction which approximates the large-scale complex model accurately by a smaller reduced order model can largely reduce the computation cost. In the thesis, a new model reduction approach, the Loewner Framework, is investigated for generating reduced order models for linear-quadratic optimal control problems. The Loewner Framework is a data-driven model reduction method which can construct the reduced order models from measurements directly. The property gives Loewner framework more flexibility compared with other model-driven methods. Besides, the Loewner framework is an interpolation-based method that requires much less computation cost than SVD-based model reduction methods for large-scale dynamical systems. In this thesis, the iterative error system approximation approach which has an aposteriori error bound is developed for the model reduction of optimal control problems. On several optimal control problems involving CD player, damped Euler-Bernoulli beam and water pollution problem, Loewner framework shows unique performance compared with other methods like balanced truncation and rational Krylov method.
Description
Advisor
Degree
Type
Keywords
Citation
Zhang, Qiang. "Data-Driven Model Reduction for Optimal Control of Large-scale Dynamical Systems." (2019) Master’s Thesis, Rice University. https://hdl.handle.net/1911/105343.