[Paper Review] Outage Detection in Partially Observable Distribution Systems using Smart Meters and Generative Adversarial Networks

Yuxuan Yuan, Kaveh Dehghanpour+2 more|arXiv (Cornell University)|Dec 10, 2019

Electricity Theft Detection Techniques20 references4 citations

TL;DR

This paper proposes a GAN-based, unsupervised outage detection method for partially observable distribution systems using smart meter (SM) data. By decomposing the network into zones via BFS and training GANs to model normal SM data distributions, the method detects outages through anomaly scoring; it achieves over 90% F1-score and AUC for medium-to-large outages, with robustness to external outages and data scarcity.

ABSTRACT

In this paper, we present a novel data-driven approach to detect outage events in partially observable distribution systems by capturing the changes in smart meters' (SMs) data distribution. To achieve this, first, a breadth-first search (BFS)-based mechanism is proposed to decompose the network into a set of zones that maximize outage location information in partially observable systems. Then, using SM data in each zone, a generative adversarial network (GAN) is designed to implicitly extract the temporal-spatial behavior in normal conditions in an unsupervised fashion. After training, an anomaly scoring technique is leveraged to determine if real-time measurements indicate an outage event in the zone. Finally, to infer the location of the outage events in a multi-zone network, a zone coordination process is proposed to take into account the interdependencies of intersecting zones. We have provided analytical guarantees of performance for our algorithm using the concept of entropy, which is leveraged to quantify outage location information in multi-zone grids. The proposed method has been tested and verified on distribution feeder models with real SM data.

Motivation & Objective

• Address the challenge of outage detection in distribution systems with limited smart meter coverage, where full observability is unattainable.
• Overcome data imbalance issues caused by the scarcity of outage events compared to normal operation data.
• Capture complex temporal-spatial dependencies in high-dimensional SM data without relying on labeled outage data.
• Enable accurate outage localization in multi-zone networks by leveraging zone coordination and information-theoretic zone decomposition.
• Develop a robust, unsupervised framework that does not require prior knowledge of outage patterns or labeled training data.

Proposed method

• Apply a breadth-first search (BFS)-based algorithm to partition the distribution network into overlapping zones that maximize outage location information.
• Train a generative adversarial network (GAN) in an unsupervised manner on historical SM data within each zone to learn the joint distribution of normal operation.
• Compute an anomaly score for real-time SM measurements by quantifying deviation from the learned normal distribution using the GAN’s discriminator output.
• Use a zone coordination mechanism to infer the true outage location across intersecting zones, leveraging interdependence and spatial consistency.
• Leverage entropy-based analytical guarantees to ensure the zone decomposition captures maximal outage location information.
• Define the anomaly score as a function of the discriminator’s output, with higher values indicating potential outage events.

Experimental results

Research questions

• RQ1 How can outage detection be effectively performed in partially observable distribution systems where only a subset of nodes are equipped with smart meters?
• RQ2 Can a GAN-based unsupervised model reliably detect outage events by learning the normal distribution of SM data without requiring labeled outage data?
• RQ3 How can the spatial and temporal dependencies in high-dimensional SM data be effectively modeled for real-time outage detection?
• RQ4 What is the impact of external outages on the anomaly score of a zone, and can the method distinguish between internal and external disturbances?
• RQ5 How does the proposed zone decomposition and coordination mechanism improve outage localization accuracy in multi-zone systems?

Key findings

• The proposed method achieves an F1-score of 0.9326 and AUC of 0.9363 for large outage cases (150kW, 80 customers), indicating high detection accuracy.
• For medium outages (100kW, 50 customers), the F1-score reaches 0.9175 and AUC is 0.9179 across all zones, demonstrating strong performance.
• In small outage cases (40kW, 20 customers), the F1-score is 0.7223 and AUC is 0.7641, showing detectability even for minor disturbances.
• The anomaly score remains stable (mean ≈1.263) under normal conditions and increases significantly (mean ≈1.33) during outages, enabling clear distinction.
• Outages occurring outside a zone do not significantly alter the anomaly score of that zone, with Δζ (change in anomaly score) remaining near zero, confirming robustness.
• The zone coordination mechanism successfully localizes outages across intersecting zones, with precision and recall exceeding 0.9 for medium and large events in most zones.