Estimating Varying Parameters in Dynamical Systems: A Modular Framework Using Switch Detection, Optimization, and Sparse Regression

TL;DR

This paper introduces a modular framework for estimating both discrete and continuous parameter variations in dynamical systems, combining switch detection, optimization, and sparse regression techniques, validated on diverse real-world examples.

Contribution

It presents a novel, flexible framework that integrates switch detection, optimization, and sparse regression for estimating varying parameters in dynamical systems.

Findings

Effective detection of parameter switches using binary segmentation.

Successful estimation of continuous parameter functions via sparse regression.

Robust performance demonstrated across multiple complex systems.

Abstract

The estimation of static parameters in dynamical systems and control theory has been extensively studied, with significant progress made in estimating varying parameters in specific system types. Suppose, in the general case, we have data from a system with parameters that depend on an independent variable such as time or space. Further, suppose the system's model structure is known, but our aim is to identify functions describing parameter-varying elements as they change with respect to time or another variable. Focusing initially on the subclass of problems where parameters are discretely switching piecewise constant functions, we develop an algorithmic framework for detecting discrete parameter switches and fitting a piecewise constant model to data using optimization-based parameter estimation. Our modular framework allows for customization of switch detection, numerical integration, and optimization sub-steps to suit user requirements. Binary segmentation is used for switch detection, with Nelder-Mead and Powell methods employed for optimization. To address broader problems, we extend our framework using dictionary-based sparse regression with trigonometric and polynomial functions to obtain continuously varying parameter functions. Finally, we assess the framework's robustness to measurement noise. We demonstrate its capabilities across several examples, including time-varying promoter-gene expression, a genetic toggle switch, a parameter-switching manifold, the heat equation with a time-varying diffusion coefficient, and the advection-diffusion equation with a continuously varying parameter.