Comparative Analysis of Dimensionality Reduction Techniques for Partial Discharge Dataset

It is well know that partial discharges (PDs) in medium/high voltage equipment are indicators of insulation degradation. In order to prevent fault conditions, it is considered necessary to monitor their activity. Dimensionality reduction techniques are used in the analysis of PD datasets due to the high complexity and volume of the data, which often include hundreds of thousands of waveform samples and multidimensional PRPD patterns. This paper investigates and reports a short review about the main dimensionality reduction techniques such as Principal Component Analysis (PCA), Singular Value Decomposition (SVD), t-distributed Stochastic Neighbor Embedding (t-SNE), and autoencoder applied to PD dataset obtained from on-site measurement. These methods are evaluated based on their ability to retain the underlying structure of the data, computational efficiency, and usefulness for visualization and clustering. The comparative analysis highlights the trade-offs between interpretability and performance, providing insights into the selection of appropriate techniques for PD data analysis.