Benchmark datasets for controlled drift analysis under THD-based power factor degradation

This record provides a benchmark dataset of processed and THD-derived energy–power factor time series for controlled drift analysis. The benchmark was constructed from three publicly available datasets spanning distinct operating domains: the Steel Industry Energy Consumption dataset (industrial demand), the HH1000 dataset (aggregated residential demand from 1000 households), and the Single-Family House Germany dataset (residential household demand in Germany). The purpose of this collection is to support reproducible studies of controlled data degradation and drift sensitivity under harmonic-distortion-oriented transformations. Starting from the original source data, each dataset was processed into a unified analysis-ready format and then expanded into derived variants through systematic THD-based power factor degradation. This procedure enables structured train–test evaluations across multiple distortion levels while preserving the underlying energy-related signal and modifying the power-factor-related representation in a controlled manner. The benchmark is particularly suited for studies involving dataset shift, controlled concept drift, robustness of learning systems, reconstruction/error-surface analysis, and power-quality-aware machine learning. By combining industrial and residential contexts, the collection offers a heterogeneous experimental basis for comparing model behavior under progressively stressed operating conditions. Files included in this record correspond to processed and derived data products used for analysis and benchmarking. Users should cite both this Zenodo record and the original source datasets, whose licenses and attribution requirements remain applicable to the underlying data.