[Paper Review] A Pattern Recognition Method for Partial Discharge Detection on Insulated Overhead Conductors

Ming Dong, Jessie Sun+1 more|arXiv (Cornell University)|May 5, 2019

High voltage insulation and dielectric phenomenaMaterials Science6 references3 citations

TL;DR

This paper proposes a novel pattern recognition method combining Seasonal and Trend decomposition using Loess (STL) for signal preprocessing and Support Vector Machine (SVM) with a Gaussian radial basis kernel for classifying partial discharge (PD) activities on insulated overhead conductors. The approach achieves high classification accuracy on the VSB Kaggle dataset, effectively detecting PD events that standard protection systems miss due to lack of overcurrent signatures.

ABSTRACT

Today,insulated overhead conductors are increasingly used in many places of the world due to the higher operational reliability, elimination of phase-to-phase contact, closer distances between phases and stronger protection for animals. However, the standard protection devices are often not able to detect the conductor phase-to-ground fault and the more frequent tree/tree branch hitting conductor events as these events only lead to partial discharge (PD) activities instead of causing overcurrent seen on bare conductors. To solve this problem, in recent years, Technical University of Ostrava (VSB) devised a special meter to measure the voltage signal of the stray electrical field along the insulated overhead conductors, hoping to detect the above hazardous PD activities. In 2018, VSB published a large amount of waveform data recorded by their meter on Kaggle, the world's largest data science collaboration platform, looking for promising pattern recognition methods for this application. To tackle this challenge, we developed a unique method based on Seasonal and Trend decomposition using Loess (STL) and Support Vector Machine (SVM) to recognize PD activities on insulated overhead conductors. Different SVM kernels were tested and compared. Satisfactory classification rates on VSB dataset were achieved with the use of Gaussian radial basis kernel.

Motivation & Objective

• To address the challenge of detecting partial discharge (PD) activities on insulated overhead conductors that do not trigger overcurrent and thus evade standard protection devices.
• To develop a robust pattern recognition method capable of identifying subtle PD signals from stray electrical field voltage measurements.
• To leverage machine learning techniques on real-world waveform data from the VSB Kaggle dataset to improve PD detection reliability.
• To evaluate and compare different SVM kernels for optimal classification performance on PD signal classification.

Proposed method

• The method applies Seasonal and Trend decomposition using Loess (STL) to decompose raw voltage signals into seasonal, trend, and remainder components, enhancing signal clarity for PD detection.
• The remainder component from STL decomposition is used as the primary input for subsequent classification, isolating transient PD-related anomalies.
• A Support Vector Machine (SVM) classifier is trained on the decomposed signal features, with multiple kernel types (e.g., linear, polynomial, Gaussian radial basis) evaluated.
• The Gaussian radial basis function (RBF) kernel is selected as the optimal kernel based on performance metrics, demonstrating superior classification accuracy.
• The method is trained and validated on the VSB Kaggle dataset, which contains real-world voltage waveform recordings from insulated overhead conductors.
• Feature extraction and classification are performed in a supervised learning framework, with model performance assessed using standard classification metrics.

Experimental results

Research questions

• RQ1 Can STL decomposition effectively extract meaningful features from noisy stray electrical field voltage signals for partial discharge detection?
• RQ2 Which SVM kernel yields the highest classification accuracy for distinguishing PD events from normal operating conditions in the VSB dataset?
• RQ3 Can the proposed method reliably detect partial discharge activities that do not produce overcurrent, thus evading conventional protection systems?
• RQ4 How does the combination of STL and SVM compare to other signal processing and classification techniques for this specific application?

Key findings

• The proposed method achieved high classification accuracy on the VSB Kaggle dataset, demonstrating strong performance in detecting partial discharge events.
• The Gaussian radial basis function (RBF) kernel in the SVM classifier outperformed other kernel types, yielding the best classification results.
• STL decomposition effectively isolated transient PD-related components from the raw voltage signals, improving signal-to-noise ratio for classification.
• The method successfully identified partial discharge activities that are undetectable by standard overcurrent protection due to the absence of fault current.
• The approach is robust to the variability in real-world signal data, as evidenced by consistent performance across the VSB dataset.