Skip to content
Conference Open access

Reduction of Electrical Equipment Monitoring Time Series While Preserving Extreme and Rapidly Changing Segments

Sep 2026 · 2026 International Ural Conference on Electrical Power Engineering (UralCon) · pp. 721-726 · 0 citations · 9 references

Abstract

Continuous monitoring of electrical equipment generates long time series that inflate storage needs and slow down visualisation, diagnostics, and data analysis. Standard uniform downsampling or interval averaging reduces data volume but often suppresses short-duration extremes and abrupt changes - events that are critical for condition assessment. We propose ERCPR, a lightweight method that combines a uniformly distributed temporal backbone with priority sampling of observations having high local rate of change, high curvature, or large deviation from a rolling median. Robust percentile normalisation makes the method scale-independent. Using 69680 oil-temperature measurements from the public ETTm1 dataset, we compared ERCPR with uniform sampling, Piecewise Aggregate Approximation (PAA), and uniform Min-Max selection. At 90% reduction, ERCPR achieved RMSE of 0.576 °C (about 9.6% higher than PAA's 0.526 °C) but reduced the maximum reconstruction error by 44.8% (5.48 °C vs. 9.92 °C) and preserved 92.0% of extremes and 14.0% of rapid changes - versus 0.0% for PAA. Our results show that ERCPR offers a practical trade-off between high compression and preservation of diagnostically important events, especially when worst-case errors matter. The method runs in under 0.3 seconds for 70 k points, making it suitable for preprocessing in real-world monitoring systems.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.