Data-Driven Model Order Reduction for Continuous- and Discrete-Time Nonlinear Systems
By: Behrad Samari , Henrik Sandberg , Karl H. Johansson and more
Potential Business Impact:
Makes robots learn without knowing how they work.
Model order reduction simplifies high-dimensional dynamical systems by deriving lower-dimensional models that preserve essential system characteristics. These techniques are crucial to controller design for complex systems while significantly reducing computational costs. Nevertheless, constructing effective reduced-order models (ROMs) poses considerable challenges, particularly for dynamical systems characterized by highly nonlinear terms. These challenges are further exacerbated when the actual system model is unavailable, a scenario frequently encountered in real-world applications. In this work, we propose a data-driven framework for the construction of ROMs for both continuous- and discrete-time nonlinear dynamical systems with unknown mathematical models. By leveraging two sets of data collected from the system, referred to as two input-state trajectories, we first construct a data-based closed-loop representation of the system. We then establish a similarity relation between the output trajectories of the original system and those of its data-driven ROM employing the notion of simulation functions (SFs), thereby enabling a formal characterization of their closeness. To achieve this, we propose data-dependent semidefinite programs as sufficient conditions to simultaneously construct both ROMs and SFs, while offering correctness guarantees. We demonstrate that the obtained data-driven ROMs can be employed for synthesizing controllers that ensure the unknown system satisfies high-level logic properties. This is accomplished by first designing controllers for the data-driven ROMs and then translating the results back to the original system through an interface function. We evaluate the efficacy of our data-driven findings through four benchmark case studies involving unknown dynamics with highly nonlinear terms.
Similar Papers
Model Order Reduction from Data with Certification
Systems and Control
Builds simple models from data for complex systems.
Data-driven stochastic reduced-order modeling of parametrized dynamical systems
Machine Learning (CS)
Makes computer simulations run much faster.
Efficient data-driven regression for reduced-order modeling of spatial pattern formation
Pattern Formation and Solitons
Makes complex patterns appear faster on computers.