Many recent data-driven control approaches for linear time-invariant systems are based on finite-horizon prediction of output trajectories using input-output data matrices. When applied recursively, this predictor forms a dynamic system representation...
Many recent data-driven control approaches for linear time-invariant systems are based on finite-horizon prediction of output trajectories using input-output data matrices. When applied recursively, this predictor forms a dynamic system representation subject to stability analysis, which we refer to as the data-driven representation. It is observed that this representation is generally non-minimal, containing latent poles in addition to the original poles of the underlying system. In this dissertation, we show that these latent poles are guaranteed to be stable by employing the Moore-Penrose inverses of the data matrices, regardless of their number or the underlying system's stability. We also analyze the effect of noisy data on the stability of the data-driven representation, ensuring that the latent poles remain stable under sufficiently small noise. This result obviates the need to eliminate the latent poles through procedures that resort to low-rank approximation in data-driven control and analysis.
We apply the main results to two distinct problems; data-driven output feedback control and inversion. First, for data-driven output feedback control, a non-minimal state-space realization of the system is constructed from its data-driven representation, where the state consists of the past inputs and outputs. Exploiting the stability guarantee of the latent poles, we examine the stabilizability and detectability of this realization. Leveraging these properties, a data-driven output feedback linear quadratic regulator (LQR) controller is designed.
Second, we extend the principles of the main result to data-driven inversion, or unknown input estimation. Likewise, we identify the existence of latent poles in this inverse setting and demonstrate that their stability is ensured by the use of the Moore-Penrose inverse of the data matrix. This enables asymptotic input estimation for minimum-phase systems, without requiring knowledge of the initial input trajectory. Building on this, we implement data-driven disturbance observer (DD-DOB), which estimates the input disturbance and rejects it simultaneously. Furthermore, we present a method to arbitrarily assign the latent poles of data-driven representations using data-driven inversion.
In addition, we investigate the impact of the depth of data matrices—which also determines the number of the latent poles—on data-driven control, particularly on the performance against online and offline measurement noises.