[Paper Review] PMU Placement for Line Outage Identification via Multiclass Logistic Regression

Taedong Kim, Stephen J. Wright|arXiv (Cornell University)|Sep 12, 2014

Power System Optimization and Stability11 references6 citations

TL;DR

This paper proposes a multiclass logistic regression (MLR) model trained on PMU-measured voltage phasor changes to identify single-line outages in power grids. By using a greedy heuristic for PMU placement on as few as 25% of buses, the method achieves near-perfect outage identification performance—over 99% accuracy with just 14 PMUs on the 118-bus system—demonstrating that sparse PMU deployment can match full-system observability for outage detection.

ABSTRACT

We consider the problem of identifying a single line outage in a power grid by using data from phasor measurement units (PMUs). When a line outage occurs, the voltage phasor of each bus node changes in response to the change in network topology. Each individual line outage has a consistent "signature," and a multiclass logistic regression (MLR) classifier can be trained to distinguish between these signatures reliably. We consider first the ideal case in which PMUs are attached to every bus, but phasor data alone is used to detect outage signatures. We then describe techniques for placing PMUs selectively on a subset of buses, with the subset being chosen to allow discrimination between as many outage events as possible. We also discuss extensions of the MLR technique that incorporate explicit information about identification of outages by PMUs measuring line current flow in or out of a bus. Experimental results with synthetic 24-hour demand profile data generated for 14, 30, 57 and 118-bus systems are presented.

Motivation & Objective

• To develop a reliable method for identifying single-line outages in power systems using synchronized PMU measurements.
• To address the challenge of optimal PMU placement in large transmission networks to maximize outage detection capability with limited resources.
• To improve upon existing PMU placement strategies by using outage signature discrimination as the primary optimization criterion.
• To demonstrate that sparse PMU deployment (as low as 25% of buses) can achieve near-full observability for line outage detection.
• To provide a practical, scalable solution for real-time outage identification using convex optimization and heuristic selection of PMU locations.

Proposed method

• Train a multiclass logistic regression (MLR) classifier on synthetic 24-hour load profiles to learn distinct voltage phasor signatures for each single-line outage scenario.
• Formulate the PMU placement problem as a regularized MLR optimization with Group LASSO penalties to promote sparsity and select key measurement buses.
• Apply a greedy heuristic to iteratively select PMU locations that maximize the minimum separation between outage class signatures in the projected measurement space.
• Use convex optimization techniques to train the MLR model efficiently, leveraging the full AC power flow model for accurate system representation.
• Integrate line current measurements into the MLR framework to enhance signature discrimination and improve detection robustness.
• Evaluate performance using IEEE 14, 30, 57, and 118-bus systems under realistic load conditions, comparing full- and partial-PMU deployment.

Experimental results

Research questions

• RQ1 Can multiclass logistic regression effectively distinguish between voltage phasor signatures of different single-line outages using PMU data?
• RQ2 What is the minimum number of PMUs required to achieve high-accuracy line outage identification in large power systems?
• RQ3 How does PMU placement strategy impact the classification performance of the MLR-based outage detection system?
• RQ4 To what extent can sparse PMU deployment (e.g., 25% of buses) maintain detection accuracy comparable to full-system instrumentation?
• RQ5 How do different heuristic strategies (greedy vs. Group LASSO) compare in selecting optimal PMU locations for outage identification?

Key findings

• With only 14 PMUs on the 118-bus system, the greedy heuristic achieved 99.3% probability of correctly identifying outages with confidence ≥0.9.
• On the 57-bus system, just 15 PMU locations enabled 98.3% identification accuracy for outages with confidence ≥0.9, outperforming Group LASSO by over 5%.
• Only 6 PMU locations were sufficient to identify 90% of outage events with high confidence (≥0.9), demonstrating strong performance with minimal instrumentation.
• The greedy heuristic consistently outperformed the Group LASSO heuristic in both PMU selection quality and classification accuracy across all test systems.
• Identification performance with PMUs on approximately 25% of buses was nearly indistinguishable from full-system instrumentation, with detection accuracy exceeding 99% in most cases.
• For the 14-bus system, 100% outage identification accuracy was achieved with only 3 PMUs using the greedy heuristic at τ=5×10⁻³.