Search for: All records

This paper proposes a data-driven framework to solve time-varying optimization problems associated with unknown linear dynamical systems. Making online control decisions to steer a system to the solution trajectory of a time-varying optimization problem is a central goal in many modern engineering applications. Yet, the available methods critically rely on a precise knowledge of the system dynamics, thus requiring ad-hoc system identification and model refinement phases. In this work, we leverage tools from behavioral theory to show that the steady-state transfer function of a system can be computed from control experiments without knowledge or estimation of the system model. Such direct computation allows us to avoid the explicit model identification phase, and is significantly more tractable than the direct model-based computation. We leverage the data-driven representation to design a controller inspired from a gradient-descent method that drives the system to the solution of an unconstrained optimization problem, without any knowledge of time-varying disturbances affecting the model equation. Results are tailored to cost functions that are smooth and satisfy the Polyak-Lojasiewicz inequality. Simulation results illustrate the technical findings.