Time-Varying Optimization of LTI Systems Via Projected Primal-Dual Gradient Flows

Gianluca Bianchin, Jorge Cortés, Jorge I. Poveda, Emiliano Dall’Anese

Abstract

This paper investigates the problem of regulating in real time a linear dynamical system to the solution trajectory of a time-varying constrained convex optimization problem. The proposed feedback controller is based on an adaptation of the saddle-flow dynamics, modified to take into account projections on constraint sets and output-feedback from the plant. We derive sufficient conditions on the tunable parameters of the controller (inherently related to the time-scale separation between plant and controller dynamics) to guarantee exponential and input-to-state stability of the closed-loop system. The analysis is tailored to the case of time-varying strongly convex cost functions and polytopic output constraints. The theoretical results are further validated in a ramp metering control problem in a network of traffic highways.

Introduction

In many modern engineering applications—such as smart power grids, communication networks, or autonomous traffic systems—the target for a controller is not a fixed equilibrium point but a moving target defined by an optimization problem. For instance, a traffic management system aims to maximize throughput while maintaining congestion below safety thresholds, even as traffic volume and road capacity fluctuate.

This presents two primary challenges. First, the physical system (the "plant") possesses its own inertia and dynamics that must be respected. Second, real-world systems are subject to hard constraints on inputs (e.g., physical limits of actuators) and outputs (e.g., safety bounds). Existing methods often simplify the plant as a static entity or struggle to manage time-varying constraints and unknown disturbances simultaneously. This research bridges this gap by designing a controller that functions as an "online optimizer."

Methodology: Projected Primal-Dual Gradient Flows

The proposed framework utilizes a Projected Primal-Dual Gradient Flow controller. Unlike traditional approaches that solve an optimization problem offline and send the result to a controller, this architecture embeds the optimization algorithm directly into the feedback loop. The gradient steps of the solver constitute the actual control actions.

The feedback architecture connects an LTI plant with a dynamic controller that tracks the solution of a constrained time-varying optimization problem.

Key technical components include:

  • Regularized Lagrangian: The authors employ a regularized Lagrangian to ensure the dual variables are strongly concave. This allows the controller to achieve exponential convergence to an approximate solution, which is essential for tracking a continuously moving target.
  • Lipschitz-Continuous Projections: To ensure mathematical stability, the controller uses smooth, Lipschitz-continuous projections. This prevents the "jumpy" behavior typical of standard optimization projections that can destabilize physical systems.
  • Singular Perturbation Analysis: Stability is guaranteed through "time-scale separation." By deriving specific conditions on controller gains (tuning parameters $\epsilon$ and $\eta$), the authors ensure the controller is fast enough to track the optimum but slow enough to maintain plant stability.
Trajectories of the projected primal-dual gradient flow under different tuning parameters, illustrating convergence behavior within the feasible region defined by the boundary of Ω.

Application and Key Results

The framework was validated using a high-fidelity simulation of the Los Angeles highway network. Using the Cell-Transmission Model (CTM) for traffic flow, the task was ramp metering—controlling highway entry to maximize throughput while preventing congestion spillback at bottlenecks.

Key Findings: * Exponential Stability: The closed-loop system is proven to be Input-to-State Stable (ISS), meaning tracking error remains bounded relative to the rate of change of the optimization problem. * Performance: In traffic simulations, the controller significantly outperformed the classic reactive ALINEA controller and achieved performance comparable to Model Predictive Control (MPC). * Efficiency: Unlike MPC, which requires solving complex optimization problems at every time step, this controller uses simple gradient-like updates, making it highly efficient for large-scale systems.

Conclusion

By treating the controller as an online optimizer, this framework provides a robust method for steering LTI systems toward time-varying optimal states. While currently focused on linear plants and polytopic constraints, the success in nonlinear traffic simulations suggests potential for broader applications in complex, dynamic engineering environments.

Summary updated

Abstract

This paper investigates the problem of regulating in real time a linear dynamical system to the solution trajectory of a time-varying constrained convex optimization problem. The proposed feedback controller is based on an adaptation of the saddle-flow dynamics, modified to take into account projections on constraint sets and output-feedback from the plant. We derive sufficient conditions on the tunable parameters of the controller (inherently related to the time-scale separation between plant and controller dynamics) to guarantee exponential and input-to-state stability of the closed-loop system. The analysis is tailored to the case of time-varying strongly convex cost functions and polytopic output constraints. The theoretical results are further validated in a ramp metering control problem in a network of traffic highways.
BibTeX
@article{bianchin2022time,
  author = {Gianluca Bianchin and Jorge Cortés and Jorge I. Poveda and Emiliano Dall’Anese},
  title = {{Time-Varying Optimization of LTI Systems Via Projected Primal-Dual Gradient Flows}},
  journal = {IEEE Trans. Control. Netw. Syst.},
  year = {2022},
  url = {https://openreview.net/forum?id=QAB4jnOKEc},
}
Figure