Source-linked AI summary
GREEND: An Energy Consumption Dataset of Households in Italy and Austria
Andrea Monacchi, Dominik Egarter, Wilfried Elmenreich, Salvatore D'Alessandro, Andrea M. Tonello
TL;DR
GREEND addresses the need for household energy datasets that support assessment of home-energy-management solutions before deployment. The paper constructs and documents a device-level dataset from Austrian and Italian households, then applies it to load disaggregation, occupancy detection, and appliance usage mining. The dataset supports these benchmark applications, while its limited length led the authors to omit seasonal information from the appliance-usage model.
Problem
Researchers and designers need suitable household energy-consumption datasets to develop and assess real-world home energy-management solutions.
Method
The paper presents GREEND, documents its household measurement campaign and scenarios, and applies state-of-the-art methods to three energy-data tasks.
Results
GREEND supports applications in load disaggregation, occupancy detection, and appliance usage mining.
Takeaways & Limitations
GREEND provides openly usable household device-consumption data and campaign software for researchers and engineers working on domestic energy-management systems.
Takeaways & Limitations
Because the dataset was limited in length at the time of writing, the appliance-usage modeling omitted seasonal information.
Abstract
from arXiv · showhide
Home energy management systems can be used to monitor and optimize consumption and local production from renewable energy. To assess solutions before their deployment, researchers and designers of those systems demand for energy consumption datasets. In this paper, we present the GREEND dataset, containing detailed power usage information obtained through a measurement campaign in households in Austria and Italy. We provide a description of consumption scenarios and discuss design choices for the sensing infrastructure. Finally, we benchmark the dataset with state-of-the-art techniques in load disaggregation, occupancy detection and appliance usage mining.
1 Introduction
GREEND addresses the need for publicly available household energy data to develop and assess real-world energy-management solutions. It presents a device-level dataset from Austria and Italy, documents its measurement campaign and scenarios, and benchmarks applications in load disaggregation, occupancy detection, and appliance usage mining.
- Renewable generation and electric-vehicle diffusion create stability challenges that motivate demand response and coordinated device operation.
- User activity information can improve control strategies by accounting for inhabitants’ routines and reducing discomfort.
- GREEND is a publicly available device-level power-usage dataset collected in households in Austria and Italy.The paper also releases measurement-campaign code and metadata describing consumption scenarios.
- The paper benchmarks GREEND with state-of-the-art techniques for load disaggregation, occupancy detection, and appliance usage mining.
2 Other consumption datasets
Existing household energy datasets differ in electrical signals, sampling characteristics, geography, duration, and household coverage. These differences affect which applications they support and motivate GREEND’s Austria–Italy measurement campaign.
- Energy-consumption datasets are classified by sampling frequency and measured signals such as active, reactive, apparent power, energy, frequency, phase angle, voltage, and current.
- Many existing datasets were collected in the USA at 120V, whereas European countries commonly use 230V.
- Dataset duration and household coverage constrain analysis of consumption behavior and seasonal patterns.A high number of households supports statistical analysis, while short campaigns cannot capture seasonal consumption behavior.
- GREEND is described as the first 1Hz consumption dataset for Austria and Italy.
- Real-environment collection matters because the context in which appliances are used contributes to demand complexity.
3 Dataset for Italy and Austria
GREEND’s measurement campaign was designed around diverse households and device-level active-power data for energy-management research in Austria and Italy. Its infrastructure combines distributed sensing, configurable storage, and publicly released campaign software.
- The campaign targets energy-consumption reduction in Austria’s Carinthia region and Italy’s Friuli-Venezia Giulia region.
- Active power is sampled at 1Hz and timestamped with UTC Unix time to support load disaggregation and contextual matching.
- Household selection promotes diverse dwelling and consumer scenarios, including apartments, detached houses, terraced houses, couples, students, retirees, and families.
- The campaign was designed to last one year to observe and model seasonal consumption behavior.
- The deployment uses an ARM-based platform, battery, and Plugwise kit with nine Zigbee sensing outlets measuring connected loads.
- Data storage supports local CSV files, remote MySQL storage, combined backup, and SFTP upload, with SFTP selected for the campaign.
4 Case studies
GREEND is evaluated through three case studies: load disaggregation, occupancy detection, and appliance usage modeling. These studies show how the dataset supports appliance-state estimation, consumption-based occupancy inference, and probabilistic prediction of user-driven device usage.
- 4.1 Non-Intrusive appliance load monitoring: Particle-filter load disaggregation models appliance states with hidden Markov models and combines them into a fractional hidden Markov model for total household demand.The evaluation uses appliance models and a decision maker to estimate appliance operation from aggregated power consumption.
- 4.1 Non-Intrusive appliance load monitoring: GREEND supports load-disaggregation experiments whose complexity varies with appliance types and the number of aggregated power loads.The dataset covers both on/off devices, such as water kettles, and multi-state devices, such as dishwashers.
- 4.2 Occupancy detection: Occupancy detection applies NIOM to two months of house #5 measurements using 15-minute slots, separate weekday and weekend cases, and a nighttime baseline.The analysis uses total power consumption to estimate inactivity-period baselines and derives occupancy probabilities.
- 4.2 Occupancy detection: The derived occupancy probabilities agree with residents’ reported routines, including weekday departures and returns and greater weekend daytime presence.Residents reported waking around 6:45 AM, leaving around 8:00 AM, returning around 6:00 PM on weekdays, and returning around midnight after weekend outings.
- 4.3 Appliance usage modeling: A Bayesian network models appliance usage by representing device-related random variables in a directed acyclic graph with conditional probability distributions.The network structure follows prior work; expectation maximization learns parameters and junction-tree inference performs exact inference.
- 4.3 Appliance usage modeling: Coffee-making patterns are regular over the day, and the modeled couple tends to wake earlier on weekdays, consistent with household interviews.The usage forecast is based on posterior probabilities for starting the coffee machine in house #0.
5 Conclusions and future work
GREEND is an openly available household consumption dataset from Austria and Italy, accompanied by software resources and demonstrated across several energy-management analysis tasks.
- GREEND provides open consumption data from selected household devices measured in Austria and Italy.
- The release includes campaign collection scripts, a Raspberry Pi SD image, and processing scripts for extracting edges and consumption events.
- The dataset was used to develop and assess techniques for load disaggregation, occupancy detection, and appliance usage mining.
- Future extensions target aggregated consumption for whole-demand modeling and renewable-generation data for forecasting energy generation.
- Ready-to-use regional models are planned for import into simulation tools for smart-grid applications.