[Paper Review] Design of Resource Agents with Guaranteed Tracking Properties for Real-Time Control of Electrical Grids

Andrey Bernstein, Niek J. Bouman+1 more|arXiv (Cornell University)|Nov 27, 2015

Smart Grid Energy Management8 references17 citations

TL;DR

This paper proposes a bounded accumulated-error property for resource agents in real-time electrical grid control, ensuring that implemented power setpoints track requested setpoints on average. By using error-diffusion techniques, the method guarantees c-bounded accumulated error, which improves convergence to optimal operation, reduces renewable curtailment, and enhances system robustness in microgrids with heterogeneous, uncertain, or discrete resources.

ABSTRACT

We target the problem of controlling electrical microgrids with little inertia in real time. We consider a central controller and a number of resources, where each resource is either a load, a generator, or a combination thereof, like a battery. The controller periodically computes power setpoints for the resources based on the estimated state of the grid and an overall objective, and subject to safety constraints. Each resource is augmented with a resource agent that a) implements the setpoint requests sent by the controller on the resource, and b) translates device-specific information about the resource into a device-independent representation and transmits this to the controller. We focus on the resource agents and their impact on the overall system's behavior. Intuitively, for the system to converge to the objective, the resource agents should be obedient to the requests from the controller, in the sense that the actually implemented setpoint should be close to the requested setpoint, at least on average. This can be important especially when a controller that performs continuous optimization is used (for the sake of performance) to control discrete resources (which have a discrete set of implementable setpoints). We formalize obedience by defining the notion of $c$-bounded accumulated-error. We then demonstrate its usefulness, by presenting theoretical results (for a simple scenario) and simulation results (for a more realistic setting) that indicate that, if all resource agents in the system have bounded accumulated-error, the closed-loop system converges on average to the objective. Finally, we show how to design resource agents that provably have bounded accumulated-error for various types of resources, such as resources with uncertainty (e.g., PV panels) and resources with a discrete set of implementable setpoints (e.g., on-off heating systems).

Motivation & Objective

• To address the challenge of controlling heterogeneous, uncertain, or discrete electrical resources in real-time microgrid control.
• To formalize obedience of resource agents to controller requests through a bounded accumulated-error metric.
• To design resource agents that provably achieve c-bounded accumulated-error for improved system-wide performance.
• To demonstrate that bounded error ensures convergence to optimal operation and energy balance in virtual power plant applications.

Proposed method

• Introduce the c-bounded accumulated-error property as a formal metric for resource agent obedience to controller requests.
• Design resource agents that use error-diffusion techniques to map continuous setpoints to discrete or uncertain implementable setpoints.
• Apply the method to resources with discrete control (e.g., on/off heaters) and uncertain sources (e.g., PV panels) via convex and belief set representations.
• Use a hierarchical agent framework (Grid Agents and Resource Agents) where RAs translate device-specific constraints into abstract, device-independent representations.
• Implement a robust continuous optimization at the Grid Agent level, relying on RA feedback to maintain system feasibility and performance.
• Validate the method via theoretical analysis under simplified assumptions and simulation in a realistic low-voltage microgrid scenario with variable solar irradiance.

Experimental results

Research questions

• RQ1 How can resource agents be designed to ensure that their implemented setpoints track the controller’s requested setpoints on average?
• RQ2 What formal property guarantees that the closed-loop system converges to the optimal objective despite discrete or uncertain resources?
• RQ3 How does bounded accumulated error improve system performance in real-time grid control with high renewable penetration?
• RQ4 Can a systematic design method be developed for resource agents to provably achieve bounded accumulated error across diverse resource types?

Key findings

• The use of error-diffusion in resource agents bounds the accumulated error, preventing unbounded growth seen in naive projection-based methods.
• With bounded accumulated error, the time-averaged implemented setpoint converges to the optimal setpoint, ensuring energy balance and improved virtual power plant performance.
• The system exhibits reduced sensitivity to the grid agent’s gradient-descent step size, enhancing robustness.
• Renewable utilization improves significantly: PV curtailment is reduced, and power output tracks the maximum possible under variable irradiance.
• In simulations with a 300 ms square-wave solar irradiance profile, bounded error agents achieved near-optimal power delivery and stable switching behavior.
• The bounded accumulated-error property ensures that the total energy produced or consumed converges to the requested amount, which is critical for long-term grid stability and market applications.