Regularity analysis of time series data: A new insight to mining value of power consumption profiles

Analyzing the regularity of power consumption data is critical for uncovering latent structure with practical value. The higher the regularity in power consumption data, the greater its analytical value. Existing entropy‑based measures rely on fixed pattern lengths and a single descriptive view, which can bias regularity estimates and limit generalization. We introduce variable‑length pattern vector entropy (VLPVE). VLPVE adaptively encodes patterns at multiple lengths, improving adaptability across time scales, and jointly quantifies regularity using both marginal pattern distributions and their temporal transition dynamics. We evaluate VLPVE on residential and commercial datasets at multiple temporal resolutions, focusing on three tasks: trend tracking of power consumption, anomaly detection, and cross-granularity regularity comparison. Across settings, VLPVE tracked the distribution of prediction errors over time, achieved 94% anomaly‑detection accuracy, and through multi‑granularity analysis, identified measurement schemes tailored to different consumer types. These results indicate that VLPVE provides a more faithful and transferable assessment of regularity in power consumption data.