Machine Learning for Scalable and Optimal Load Shedding Under Power System Contingency

TL;DR

This work tackles the challenge of real-time, network-aware load shedding under contingencies by introducing a decentralized learning-for-OLS framework. Offline, per-bus neural networks map local measurements to local OLS dual variables $\hat{\alpha}_i$, enabling autonomous online decisions via KKT-type relations without central coordination. The approach is analyzed for identifiability using DC power-flow factors and principal angles, and validated on IEEE 118-bus and ACTIVSg2000 Texas-scale systems, achieving low prediction errors (e.g., $\mathbb{E}[|\hat{\alpha}_i-\alpha_i|]/\mathbb{E}[\alpha_i]$ around 0.2–0.7%) and accurate load-shedding amounts. The results demonstrate scalable, communication-efficient resilience enhancement for large grids, with potential extensions to topology-aware and risk-aware designs.

Abstract

Prompt and effective corrective actions in response to unexpected contingencies are crucial for improving power system resilience and preventing cascading blackouts. The optimal load shedding (OLS) accounting for network limits has the potential to address the diverse system-wide impacts of contingency scenarios as compared to traditional local schemes. However, due to the fast cascading propagation of initial contingencies, real-time OLS solutions are challenging to attain in large systems with high computation and communication needs. In this paper, we propose a decentralized design that leverages offline training of a neural network (NN) model for individual load centers to autonomously construct the OLS solutions from locally available measurements. Our learning-for-OLS approach can greatly reduce the computation and communication needs during online emergency responses, thus preventing the cascading propagation of contingencies for enhanced power grid resilience. Numerical studies on both the IEEE 118-bus system and a synthetic Texas 2000-bus system have demonstrated the efficiency and effectiveness of our scalable OLS learning design for timely power system emergency operations.

Figures (12)

• Figure 1: A piece-wise quadratic cost function for the active power flexibility (both reserve and load shedding) per bus $i$.
• Figure 2: Illustration of the proposed scalable OLS design.
• Figure 3: The principal angles between the two subspaces spanned by ${\mathbf{d}}_{{\cal{L}}_i}^{(k)}$ and ${\mathbf{d}}_{{\cal{L}}_i}^{(k')}$, for the case of (a) 1-D subspaces with $\beta_1>0$; and (b) 2-D subspaces with $\beta_1=0$ and $\beta_2>0$.
• Figure 4: The scatter plots of the local measurements at Bus 34 for the 5 contingencies considered for the IEEE 118-bus system based on a combination of only two measurements indexed by: (a) 5 and 13; (b) 13 and 14; and (c) 9 and 16.
• Figure 5: The median value of training time (in seconds) for each bus in the 118-bus system.