Anomaly Detection in Power Grids via Context-Agnostic Learning

TL;DR

The paper tackles real-time anomaly detection in power-grid SCADA data under dynamic topology and load conditions. It introduces GridCAL, a context-agnostic learning framework that maps real-time line-branch flows to a baseline context using a three-step process: baseline context selection, injection-correction, and inverse projection based on Line Outage Distribution Factors (LODF). A two-part algorithm computes graph-distance weights for historical data and applies the context-agnostic mapping to learn a unified statistical model, enabling robust detection with a threshold learned via cross-validation. Empirical evaluation on the 2383-bus Polish grid demonstrates improved accuracy and efficiency over state-of-the-art methods, indicating strong practical potential for real-time anomaly detection and localization in large-scale power systems.

Abstract

An important tool grid operators use to safeguard against failures, whether naturally occurring or malicious, involves detecting anomalies in the power system SCADA data. In this paper, we aim to solve a real-time anomaly detection problem. Given time-series measurement values coming from a fixed set of sensors on the grid, can we identify anomalies in the network topology or measurement data? Existing methods, primarily optimization-based, mostly use only a single snapshot of the measurement values and do not scale well with the network size. Recent data-driven ML techniques have shown promise by using a combination of current and historical data for anomaly detection but generally do not consider physical attributes like the impact of topology or load/generation changes on sensor measurements and thus cannot accommodate regular context-variability in the historical data. To address this gap, we propose a novel context-aware anomaly detection algorithm, GridCAL, that considers the effect of regular topology and load/generation changes. This algorithm converts the real-time power flow measurements to context-agnostic values, which allows us to analyze measurement coming from different grid contexts in an aggregate fashion, enabling us to derive a unified statistical model that becomes the basis of anomaly detection. Through numerical simulations on networks up to 2383 nodes, we show that our approach is accurate, outperforming state-of-the-art approaches, and is computationally efficient.

Figures (6)

• Figure 1: Flow signal before any processing (top), after only inverse projection (middle), and after injection-correction and inverse projection (bottom) on line 89 (#10 among current edgeset).
• Figure 2: Flow signal before (top) and after (bottom) inverse projection on line 24 (#1 among current edgeset)
• Figure 3: Flow signal before any processing (top), after only inverse projection (middle), and after injection-correction and inverse projection (bottom) on line 93 (#14 among current edgeset)
• Figure 4: Comparing performance of the proposed method with existing methods under the measure of AUC (left) and F-measure (right). The simulations were performed for a dataset with relatively small load variability and no load/generation shift scenarios.
• Figure 5: Comparing performance of the proposed method with existing methods under the measure of AUC (left) and F-measure (right). The simulations were performed for a dataset with relatively large load variability but without load/generation shift scenarios.