A residential labeled dataset for smart meter data analytics.

Area of Science:

• Electrical Engineering
• Data Science
• Energy Systems

Background:

• Smart meter data is crucial for next-generation electrical grids, enabling energy-saving recommendations and predictive maintenance.
• Machine learning algorithms are key to these services but require well-labeled datasets for training and validation.
• Existing datasets often lack sufficient labels, hindering the development of advanced energy analytics.

Purpose of the Study:

• To introduce the SustDataED2 dataset, a novel, labeled dataset for energy research.
• To provide ground truth for evaluating machine learning models, especially for Non-Intrusive Load Monitoring (NILM).
• To facilitate the development of new energy data-based services.

Main Methods:

• Collected 96 days of aggregated and individual appliance energy consumption data from a single household.
• Sampled current and voltage waveforms at 12.8 kHz and individual appliance consumption at 0.5 Hz.
• Recorded precise ON-OFF transition timestamps for 18 appliances, ensuring data labeling.

Main Results:

• The SustDataED2 dataset offers 96 days of detailed household energy consumption.
• It includes high-resolution waveform data and labeled appliance transition timestamps.
• The dataset is available in accessible audio and CSV formats.

Conclusions:

• The SustDataED2 dataset addresses the scarcity of labeled data in energy research.
• It is a valuable resource for training and validating machine learning models for smart grid applications.
• This dataset will accelerate innovation in energy data-based services and NILM.