Gaussian behaviors: representations and data-driven control

Abstract

We propose a modeling framework for stochastic systems based on Gaussian processes. Finite-length trajectories of the system are modeled as random vectors from a Gaussian distribution, which we call a Gaussian behavior. The proposed model naturally quantifies the uncertainty in the trajectories, yet it is simple enough to allow for tractable formulations. We relate the proposed model to existing descriptions of dynamical systems including deterministic and stochastic behaviors, and linear time-invariant (LTI) state-space models with Gaussian process and measurement noise. Gaussian behaviors can be estimated directly from observed data as the empirical sample covariance under the assumption that the measured trajectories are from independent experiments. The distribution of future outputs conditioned on inputs and past outputs provides a predictive model that can be incorporated in predictive control frameworks. We show that subspace predictive control (SPC) is a certainty-equivalence control formulation with the estimated Gaussian behavior. Furthermore, the regularized data-enabled predictive control (DeePC) method is shown to be a distributionally optimistic formulation that optimistically accounts for uncertainty in the Gaussian behavior. To mitigate the excessive optimism of DeePC, we propose a novel distributionally robust control formulation, and provide a convex reformulation allowing for efficient implementation.

Tasks

Reproductions