ACS-F2

ACS-F2 dataset Database of appliance consumption signatures

Realization
Christophe Gisler
Antonio Ridi
Jean Hennebert
Contact
Jean Hennebert
jean.hennebert@hefr.ch
Content
dataset of electric signatures of equipments and categories to be used for machine learning tasks.
Keywords
  • Electric signatures
  • Database
  • Industry 4.0
  • Machine Learning
  • Time series

ACS-F2 is the second version of a database of electricity consumption signatures of devices in different categories. The database contains multi-dimensional time series of electricity related characteristics including current, active power and reactive power. The time series can be used to model the different equipment and perform machine learning based tasks such as classification or prediction. The dataset can be used to benchmark industrial types of models used to monitor electricity consumptions of equipments in an industry 4.0 setting.

By means of plug-based low-end sensors measuring the electrical consumption at low frequency, typically every 10 seconds, we made two acquisition sessions of one hour on about 225 home appliances divided into 15 categories and 6 consumption measures. We now give free access to this ACS-F2 database. The proposed test protocols will help the scientific community to objectively compare new machine learning algorithms.

Overview

In order to acquire the various electrical signals of the target appliances, we first had to set several acquisition modalities and parameters :

  • Acquisition sampling frequency: 10-1 Hz – Most approaches use high sampling frequency in order to capture the electrical noise generated by the appliance and use this noise to distinguish the different categories of appliances. In our approach, we use a larger time space in order to build what we call an electrical consumption signature of an appliance.
  • Acquisition duration: 2 sessions of 1 hour each – According to the sampling frequency, one hour of acquisition is a reasonable value to build an electrical consumption signature for an appliance, so that all possible running states are recorded. These two acquisition sessions are performed for each appliance of each category, at a different time and with a different use profile. It is essential to have a common and varied use of the equipment being acquired.
  • Number of categories: 15 – We selected the following categories :
    • Fridges & freezers,
    • TVs (LCD),
    • Hi-Fi systems (with CD players),
    • Laptops,
    • Computer stations (with monitors),
    • Compact fluorescent lamps (CFL),
    • Microwaves,
    • Coffee machines,
    • Mobile phones (via battery charger),
    • Printers,
    • Fans,
    • Shavers,
    • Monitors,
    • Incandescent Lamps,
    • Kettles.

    Those categories well represent most common appliances in home or office

  • Number of appliance instances per category: 15.
  • Number of appliances per acquisition device (i.e. PLOGG): 1 – We want to record disaggregated signals.

We measured the consumption in terms of :

  • real power (W),
  • reactive power (var),
  • RMS current (A),
  • frequency (Hz),
  • RMS voltage (V),
  • phase of voltage relative to current (φ).

Data Format

Two different data formats are available. An XML data structure has been designed for storing the raw observations, some meta-data and the ground truth values of the appliance categories.

A MAT data structure is also available. Six different headers introduce the lists of data for every feature, i.e. :

# name: rmsVolt
# type: matrix
# rows: 1
# columns: 360

Protocols

The proposed test protocols will help to objectively compare new algorithms.

Test Protocol 1.0 – Intersession

In the test protocol 1.0 (or intersession protocol) for appliance recognition, all instances of acquisition session 1 must be taken in the train (resp. test) set. In other words, cardinalities of both train and test sets are equal. For a given recording, it is allowed to use the whole duration of the signal, namely 1 hour. Classification results must be presented in the form of a confusion matrix and the overall recognition rate.

Test Protocol 2.0 – Unseen Instances

In the test protocol 2.0 (or unseen instances protocol) for appliance recognition, all instances of both sessions are taken to perform a k-fold cross-validation, where k = 15. In other words, we randomly partitioned all instances into k = 15 subsets. The cross-validation process consists then in taking successively each of the k-folds for testing and the remaining for the training. As for test protocol 1.0, for a given recording, it is allowed to use the whole duration of the signal, namely 1 hour. Classification results must be presented in the form of a confusion matrix and the overall recognition rate averaged over the k-folds.

Acknowledgement

This work has been supported by the grant Smart Living Green-Mod from the Hasler Foundation in Switzerland. Link : Hasler Foundation.

Terms of use

The owner of the ACS-F2 database is iCoSys Institute, University of Applied Sciences HES-SO//Fribourg, Engineering and Architecture Faculty, Pérolles 80, CH-1700 Fribourg, Switzerland.

The data is supplied with no guarantee of accuracy or usability. Therefore, we can not be liable of any loss or damage resulting of the use of the database. We can not guarantee to maintain the ACS-F2 database. The ACS-F2 database is for non-commercial research only.

Reference for citation

If you use our ACS-F2 database, we kindly ask you to cite our work:

  • A. Ridi, C. Gisler, and J. Hennebert, “ACS-F2 – A new database of appliance consumption signatures,” in Soft Computing and Pattern Recognition (SoCPaR), 2014 6th International Conference of, 2014, pp. 145-150.
    [Website] [PDF] [Bibtex]
    @conference{ridi2014socpar,
    Abstract = {We present ACS-F2, a new electric consumption signature database acquired from domestic appliances. The scenario of use is appliance identification with emerging applications such as domestic electricity consumption understanding, load shedding management and indirect human activity monitoring. The novelty of our work is to use low-end electricity consumption sensors typically located at the plug. Our approach consists in acquiring signatures at a low frequency, which contrast with high frequency transient analysis approaches that are costlier and have been well studied in former research works. Electrical consumption signatures comprise real power, reactive power, RMS current, RMS voltage, frequency and phase of voltage relative to current. A total of 225 appliances were recorded over two sessions of one hour. The database is balanced with 15 different brands/models spread into 15 categories. Two realistic appliance recognition protocols are proposed and the database is made freely available to the scientific community for the experiment reproducibility. We also report on recognition results following these protocols and using baseline recognition algorithms like k-NN and GMM.},
    Author = {A. Ridi and C. Gisler and J. Hennebert},
    Booktitle = {Soft Computing and Pattern Recognition (SoCPaR), 2014 6th International Conference of},
    Doi = {10.1109/SOCPAR.2014.7007996},
    Isbn = {9781479959358},
    Keywords = {machine learning, electric signal, appliance signatures},
    Month = {Aug},
    Pages = {145-150},
    Publisher = {IEEE},
    Title = {{ACS}-{F}2 - {A} new database of appliance consumption signatures},
    Pdf = {http://www.hennebert.org/download/publications/socpar-2014-ACS-F2-a-new-databas-of-appliance-consumption-signatures.pdf},
    Year = {2014},
    Pdf = {http://www.hennebert.org/download/publications/socpar-2014-ACS-F2-a-new-databas-of-appliance-consumption-signatures.pdf},
    Bdsk-Url-2 = {http://dx.doi.org/10.1109/SOCPAR.2014.7007996}}

Download

The ACS-F2 Database is available now ! It’s free to download for non-commercial research.

Download : ACS-F2 Database