University of South Florida
Fractional Calculus Approach For Learning Unknown Dynamic System
Abstract
dc:descriptionThe rapidly growing interest in data-driven modeling and fractional-order systems highlights a keychallenge in modern science and engineering: capturing long-memory effects and nonlinear behaviors in real-world dynamical processes. Conventional integer-order methods, while powerful, often overlook the history-dependent nature of many phenomena—ranging from viscoelastic materials to anomalous diffusion and hereditary feedback systems. This dissertation addresses that gap by blending fractional calculus with cutting-edge machine learning to create robust, memory-aware modeling and control frameworks. Beginning with an extension of Dynamic Mode Decomposition (DMD) to fractional-order systems, we introduce Fractional Dynamic Mode Decomposition (F-DMD) as a data-driven tool that leverages Mittag- Leffler functions and fractional discretizations. By accommodating power-law relaxation and nonlocal behaviors, F-DMD significantly broadens the applicability of DMD-based Koopman operator methods. We then augment fractional-order modeling with Occupation Kernel Regression (OKR) to handle nonlinearities in both memory and state-space dynamics, offering a versatile system identification approach for incomplete or noisy datasets. Next, we shift attention to control and develop a new paradigm where Kolmogorov-Arnold Networks (KANs)—which feature trainable activation functions—are explicitly coupled with fractional dynamics. Dubbed fractional KANs, these architectures enable precise control of memory-dependent processes while retaining interpretability. We provide theoretical guarantees on the existence, uniqueness, and stability of solutions, confirmed through comprehensive numerical experiments and a practical mobile robot path-tracking application. Finally, we analyzed Fractional Kolmogorov-Arnold Network (fKAN) equipped with trainable, fractionalorthogonal Jacobi basis functions to tackle time-delay systems. We present a rigorous Lyapunov-based stability analysis that underscores the resilience of this memory-rich framework under parametric uncertainty and delayed feedback. The strategies and theoretical insights presented here establish a new foundation. This foundation supports reliable, efficient data-driven techniques. These techniques can model, control, and analyze complex systems. Such systems often show deep memory effects. They also involve nonlinear interactions. This work opens promising paths for future studies. Potential areas include machine learning and control theory. Applications may extend to other fields as well.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy (Ph.D.)
- Level thesis:degree_level
- Dissertation
- Discipline thesis:degree_discipline
- Mathematics and Statistics
- Grantor
- University of South Florida
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Chen, Haowei(Alice)
- Contributors dc:contributor
-
- Joel Rosenfeld, Ph.D.
- Tansel Yucelen, Ph.D.
- Dmitry Khavinson, Ph.D.
- Seung-Yeop Lee, Ph.D.
Subjects
dc:subject × 7Identifiers
dc:identifier.*- Repository record dc:identifier
- https://digitalcommons.usf.edu/etd/10929