AI-Ready Smart Meter Electricity Consumption Dataset for Time-Series Analysis and Anomaly Detection

This dataset provides preprocessed smart meter electricity consumption data for research in time-series analysis, anomaly detection, and energy consumption modeling. It is derived from the London Smart Meter Dataset and contains 167,811,461 valid observations from 5,561 households, recorded at 30-minute intervals.

The original data were cleaned and standardized to improve consistency and support further analysis. Invalid energy readings were removed, while valid zero-consumption values were preserved. Additional features were generated to capture household consumption patterns over time, including temporal information, lag values, consumption changes, and rolling statistics. These features were calculated using only current and historical observations to avoid introducing future information.

The dataset is available in Apache Parquet format, which supports efficient storage and processing of large time-series data. It can be used for studying household electricity consumption patterns, developing and evaluating anomaly detection methods, exploring time-series modeling techniques, and conducting research on smart energy systems.

The dataset does not include verified anomaly labels. Researchers can therefore use it for unsupervised anomaly detection, exploratory analysis, or experiments involving separately defined anomaly labels.

By providing cleaned consumption records and precomputed historical features, this dataset aims to reduce data preparation effort and offer a practical starting point for further research in energy analytics and machine learning.