Generalizing Koopman Theory to allow for inputs and control

Joshua L. Proctor,Steven L. Brunton,J. Nathan Kutz

DOI: https://doi.org/10.48550/arXiv.1602.07647

2016-02-25

Abstract:We develop a new generalization of Koopman operator theory that incorporates the effects of inputs and control. Koopman spectral analysis is a theoretical tool for the analysis of nonlinear dynamical systems. Moreover, Koopman is intimately connected to Dynamic Mode Decomposition (DMD), a method that discovers spatial-temporal coherent modes from data, connects local-linear analysis to nonlinear operator theory, and importantly creates an equation-free architecture allowing investigation of complex systems. In actuated systems, standard Koopman analysis and DMD are incapable of producing input-output models; moreover, the dynamics and the modes will be corrupted by external forcing. Our new theoretical developments extend Koopman operator theory to allow for systems with nonlinear input-output characteristics. We show how this generalization is rigorously connected and generalizes a recent development called Dynamic Mode Decomposition with control (DMDc). We demonstrate this new theory on nonlinear dynamical systems, including a standard Susceptible-Infectious-Recovered model with relevance to the analysis of infectious disease data with mass vaccination (actuation).

Optimization and Control

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to extend the Koopman operator theory to allow the influence of inputs and controls, so as to be able to analyze systems with nonlinear input - output characteristics. Standard Koopman analysis and dynamic mode decomposition (DMD) methods cannot produce input - output models when dealing with controlled systems, and external forces will disrupt the dynamics and modes of the system. Therefore, the authors propose a new method - Koopman with inputs and control (KIC), aiming to overcome these limitations, so that the Koopman operator theory can be applied to a wider range of complex systems, especially those systems that contain external inputs and control signals.

Specifically, the goals of the paper include:

1. **Extend the Koopman operator theory**: Extend the Koopman operator theory from being only applicable to nonlinear dynamical systems without inputs to being able to handle systems with input and control signals. This involves defining a new Koopman operator that can handle the observation functions of state variables and input variables.

2. **Establish a connection with DMDc**: Demonstrate how KIC strictly connects and generalizes a recently developed method called dynamic mode decomposition with control (DMDc). DMDc is a DMD extension method specifically designed for handling input - output systems.

3. **Verify the effectiveness of the new theory**: Demonstrate the effectiveness and success of the KIC method through applications on nonlinear dynamical systems, including a standard susceptible - infected - recovered (SIR) model. In particular, in analyzing infectious disease data that includes large - scale vaccination (as a control signal), KIC shows its advantages.

4. **Provide a data - driven solution**: The KIC method is a data - driven, model - free method that can construct the best (in the least - squares sense) underlying dynamics and control of a given system from data. This makes KIC one of the attractive data - driven architectures in modern dynamical systems theory.

In summary, through the introduction of the KIC method, this paper fills the gap in the Koopman operator theory in dealing with nonlinear systems with input and control signals, providing new tools and perspectives for the analysis and control of complex systems.