Abstract
This paper presents a closed-form notion of controllability and observability for systems with communication delays, actuation delays, and locality constraints. The formulation reduces to classical notions of controllability and observability in the unconstrained setting. As a consequence of our formulation, we show that the addition of locality and communication constraints may not affect the controllability and observability of the system, and we provide an efficient sufficient condition under which this phenomenon occurs. This contrasts with actuation and sensing delays, which cause a gradual loss of controllability and observability as the delays increase. We illustrate our results using linearized swing equations for the power grid, showing how actuation delay and locality constraints affect controllability.
Introduction
In classical control theory, the concepts of controllability (the ability to drive a system to a desired state) and observability (the ability to reconstruct internal states from sensor data) are fundamental. These are typically measured using Gramians. However, traditional Gramians assume a centralized controller with instantaneous access to all system information.
Modern large-scale networked systems, such as smart power grids or autonomous vehicle fleets, operate under significant physical constraints. These include actuation and sensing delays, limited communication speeds, and locality constraints (where a controller only has access to information from nearby neighbors). This paper introduces a rigorous mathematical framework to quantify exactly how much performance is lost when these real-world constraints are imposed.
Methodology: System Level Synthesis and Constrained Gramians
To address these constraints, the authors utilize System Level Synthesis (SLS). Unlike traditional state-space methods, SLS focuses on "closed-loop maps"—the end-to-end responses of the system. This shift allows physical constraints to be mathematically represented as sparsity patterns or "support constraints" on the system matrices.
Algorithm 1 and the Rank Condition
The paper contributes two primary tools for system designers:
- Algorithm 1: A novel algorithm that decomposes complex, constrained optimization problems into independent sets, allowing for the derivation of a closed-form solution for the "controllability volume" under constraints.
- The Rank Condition (Theorem 9): A computationally efficient sufficient condition. It acts as a mathematical shortcut, allowing designers to check if a specific set of constraints will cause any performance loss without needing to perform the full, computationally expensive volume calculation. In testing, this check was up to six times faster than full calculation.
Key Results: The Impact of Locality and Delay
The framework was evaluated using linearized swing equations, a standard model for power grid dynamics, simulating a $3 \times 3$ grid of substations. The results revealed a stark difference between how locality and delays affect system performance.
The "No-Loss" Threshold for Locality
One of the most significant findings is that adding locality or communication constraints does not always degrade performance. There is often a "threshold" of connectivity; as long as controllers can communicate within a certain radius, the system performs as well as one with global information. As shown in Figure 1, for the power grid model, a locality radius of 6 or more resulted in zero loss of controllability volume when delays were absent.
The Unforgiving Nature of Delays
In contrast to locality, actuation delays cause a gradual and immediate loss of controllability. There is no threshold for delays; every incremental increase in delay strictly reduces the controllability volume. Figure 1 illustrates this clearly: as the actuation delay (y-axis) increases, the controllability volume (represented by color intensity) drops immediately, and the system quickly enters "uncontrollable" territory (hatched regions) for lower locality values.
Conclusion and Future Work
This work provides a bridge between physical network constraints and fundamental control metrics, proving that their new constrained metrics perfectly reduce to classical Gramian-based metrics when constraints are removed. While currently limited to linear systems, the framework offers a practical tool for engineers to optimize network design. Future research will focus on extending these results to non-linear systems and developing distributed methods to compute these metrics, eventually applying them to real-world problems like optimizing sensor placement for environmental monitoring.
Summary updated