Power Distribution Grid Enhancement via Online Feedback Optimization

TL;DR

Power distribution grids face voltage constraints as distributed energy resources grow. The paper compares local droop-based Volt/VAr control with coordinated approaches, notably Online Feedback Optimization (OFO), and demonstrates that OFO can nearly reach the centralized ORPF benchmark while requiring only voltage magnitude measurements and limited grid information. Using a CIGRÉ LV benchmark grid and real Dataport data, the authors show OFO increases feasible active-power transfer by about 9% in steady-state and achieves a 10.5% improvement in a real feeder experiment compared with droop control. These results indicate that OFO offers a practical path to near-optimal grid enhancement, enabling higher DER hosting capacities without expensive infrastructure upgrades.

Abstract

The rise in residential photovoltaics and other distributed energy sources poses unprecedented challenges for the operation of power distribution grids. When high amounts of active power are injected into the grid by such power sources, the overall power flow is often limited because of voltages reaching their upper acceptable limits. Volt/VAr control aims to raise this power flow limit by controlling the voltage using reactive power. This way, more active power can be transmitted safely without physically reinforcing the grid. In this paper, we use real consumption and generation data on a low-voltage CIGRÉ grid model and an experiment on a real distribution grid feeder to analyze how different Volt/VAr methods can enhance grid capacity, i.e., by how much they can improve the grid's capability to transmit active power without building new lines. We show that droop control enhances the grid but vastly underutilizes the reactive power resources. We discuss how the effectiveness of droop control can be partially improved by employing machine-learning techniques to tune the droop coefficients, but we demonstrate that local control laws are inherently unable to achieve optimal grid enhancement. In contrast, methods that coordinate the use of reactive power resources across the grid, such as Online Feedback Optimization (OFO), can enhance the grid to its full potential. A numerical study performed on data from an entire year using a realistic grid model suggests that OFO can enable another 9\% of maximum active power injections compared to droop control. To achieve that, OFO only requires voltage magnitude measurements, minimal model knowledge, and communication with the reactive power sources. A real-life experiment provides a demonstration of the practical feasibility of the proposed approach and enhanced the grid by another 10.5\% compared to droop control.

Figures (9)

• Figure 1: The CIGRÉ low-voltage distribution grid with European layout. The residential subnetwork is highlighted and was used for this work. The dark blue icons indicate the locations of PV power infeed. Figure adapted from cigre_grid.
• Figure 2: One exemplary day (July 2, 2018) from the resulting dataset, after assigning the available Dataport data to the load buses in the CIGRÉ LV grid. All other buses have a constant load and generation of 0 W.
• Figure 3: Training points for the ML-tuned droop as obtained by solving the ORPF problem, and the optimized droop curves.
• Figure 4: Block diagram of the OFO controller. Two dual variables integrate constraint violations and are used to update the control inputs toward the optimum of the ORPF problem based on closed-loop measurements. The sensitivity matrix $H$ is the only required model information. Figure taken from ortmann2020experimental.
• Figure 5: Voltages and reactive power injections throughout a sunny summer day for the 2035 scenario for each of the considered controllers.