Abstract:
We propose a data-driven control method for nonlinear dynamical systems based on the Koopman operator theory. Existing Koopman-based control methods apply linear optimal ...Show MoreMetadata
Abstract:
We propose a data-driven control method for nonlinear dynamical systems based on the Koopman operator theory. Existing Koopman-based control methods apply linear optimal control methods after system identification by approximating the original cost function in the Koopman space. Therefore, errors in system identification and cost approximation deteriorate the control performance. On the other hand, the proposed method directly maximizes the control performance with reinforcement learning, where a controller is modeled by a neural network that consists of a linear quadratic regulator and an encoder that embeds data into the Koopman space. We experimentally demonstrate the effectiveness of the pro-posed method over existing Koopman-based and reinforcement learning-based methods with two nonlinear dynamical systems.
Published in: 2021 60th IEEE Conference on Decision and Control (CDC)
Date of Conference: 14-17 December 2021
Date Added to IEEE Xplore: 01 February 2022
ISBN Information: