[Paper Review] Anomaly Classification in Distribution Networks Using a Quotient Gradient System

Hamid Khodabandehlou, Iman Niazazari+2 more|arXiv (Cornell University)|May 14, 2018

Power Quality and HarmonicsEngineering28 references4 citations

TL;DR

This paper proposes a quotient gradient system (QGS)-trained two-stage partially recurrent neural network to classify anomalies in power distribution networks using high-fidelity PMU data. By transforming phasor measurements into a constraint satisfaction problem, QGS enables robust optimization for improved anomaly detection accuracy, outperforming conventional neural network classifiers in simulations with strong sensitivity to PMU count, reporting rate, noise, and data fusion timing.

ABSTRACT

The classification of anomalies or sudden changes in power networks versus normal abrupt changes or switching actions is essential to take appropriate maintenance actions that guarantee the quality of power delivery. This issue has increased in importance and has become more complicated with the proliferation of volatile resources that introduce variability, uncertainty, and intermittency in circuit behavior that can be observed as variations in voltage and current phasors. This makes diagnostics applications more challenging. This paper proposes using quotient gradient system (QGS) to train two-stage partially recurrent neural network to improve anomaly classification rate in power distribution networks using high-fidelity data from micro-phasor measurement units (PMUs). QGS is a systematic approach to finding solutions of constraint satisfaction problems. We transform the PMUs data from the power network into a constraint satisfaction problem and use QGS to train a neural network by solving the resulting optimization problem. Simulation results show that the proposed supervised classification method can reliably distinguish between different anomalies in power distribution networks. Comparison with other neural network classifiers shows that QGS trained networks provide significantly better classification. Sensitivity analysis is performed concerning the number of PMUs, reporting rates, noise level and early versus late data stream fusion frameworks.

Motivation & Objective

• To address the growing challenge of distinguishing between true anomalies and normal switching events in power distribution networks due to increasing integration of volatile renewable resources.
• To improve the reliability and accuracy of anomaly classification in the presence of voltage and current phasor variations caused by intermittent generation.
• To develop a systematic training framework for neural networks that can handle constraint satisfaction problems arising from real-time PMU data.
• To evaluate the robustness of the proposed method under varying operational conditions, including PMU density, reporting frequency, noise levels, and data fusion strategies.

Proposed method

• Transform high-fidelity PMU data from distribution networks into a constraint satisfaction problem (CSP) representing anomaly detection as an optimization task.
• Apply the quotient gradient system (QGS), a systematic method for solving CSPs, to train a two-stage partially recurrent neural network (PRNN).
• Use QGS to solve the resulting optimization problem by navigating the manifold of feasible solutions, ensuring convergence to valid classifications.
• Design a two-stage PRNN architecture to model temporal dependencies and improve classification of transient events in power systems.
• Integrate early and late data stream fusion frameworks to assess temporal processing impact on classification performance.
• Perform sensitivity analysis on key system parameters: number of PMUs, reporting rates, noise levels, and fusion timing.

Experimental results

Research questions

• RQ1 Can a QGS-trained neural network reliably classify anomalies in distribution networks with high accuracy under real-world variability?
• RQ2 How does the number of PMUs affect the classification performance of the proposed QGS-based method?
• RQ3 What is the impact of PMU reporting rate and noise levels on the robustness of the anomaly classification system?
• RQ4 How do early versus late data stream fusion strategies influence classification outcomes in the proposed framework?
• RQ5 Does the QGS-based training approach significantly improve classification accuracy compared to standard neural network training methods?

Key findings

• The QGS-trained neural network achieves significantly higher anomaly classification accuracy than conventional neural network classifiers in simulated distribution network scenarios.
• The method demonstrates robust performance across varying PMU counts, with classification accuracy maintained even at reduced sensor density.
• Higher PMU reporting rates improve classification reliability, with optimal performance observed at sampling intervals consistent with real-time monitoring standards.
• The system remains resilient to moderate noise levels, maintaining high classification accuracy under realistic measurement uncertainty.
• Late data stream fusion consistently outperforms early fusion in terms of classification accuracy, suggesting temporal context is critical for anomaly detection.
• Sensitivity analysis confirms the method’s stability and adaptability across diverse operational configurations, supporting practical deployment.