Elsevier

Energy and Buildings

Volume 151, 15 September 2017, Pages 98-106
Energy and Buildings

Comprehensive feature selection for appliance classification in NILM

https://doi.org/10.1016/j.enbuild.2017.06.042Get rights and content

Highlights

  • Concise and updated review of features for appliance classification in NILM.

  • Proposal of a systematic feature selection algorithm to identify a unique appliance signature.

  • The systematic feature selection improves the classification accuracy vs. using all the features.

Abstract

Since the inception of non-intrusive appliance load monitoring (NILM), extensive research has focused on identifying an effective set of features that allows to form a unique appliance signature to discriminate various loads. Although an abundance of features are reported in literature, most works use only a limited subset of them. A systematic comparison and combination of the available features in terms of their effectiveness is still missing. This paper, as its first contribution, offers a concise and updated review of the features reported in literature for the purpose of load identification. As a second contribution, a systematic feature elimination process is proposed to identify the most effective feature set. The analysis is validated on a large benchmark dataset and shows that the proposed feature elimination process improves the appliance classification accuracy for all the appliances in the dataset compared to using all the features or randomly chosen subsets of features.

Introduction

A breakdown of energy consumption at the appliance level is not only an essential requirement for energy providers, in designing practical demand response algorithms (e.g., taking into account human behavioral uncertainties, or targeting specific user and appliance groups), but it also benefits residential customers by providing them the necessary information for improving their energy consumption efficiency [1]. Non-intrusive (appliance) load monitoring (NILM) techniques are cost-effective solutions to obtain such information. The general framework of NILM starts from input measurements of total electricity consumption to eventually disaggregate it into the individual contributions of each load.

A crucial step in NILM is feature extraction, which applies signal processing techniques to extract features from voltage (V) and current (I) measurements. The ultimate goal of the feature extraction step is to derive a signature (using a feature or combination of features) that can uniquely identify the individual appliances. The performance of any NILM system depends on the uniqueness of the appliance signature compared to that of other devices. Hence, identification of such signature is crucial in improving the load discrimination capability of a NILM system. Although NILM has been the subject of research for over two decades, so far a systematic selection of the various electrical features proposed for effective discrimination of loads has not yet been presented. Identifying the most meaningful set of electrical parameters to distinguish all appliances still remains one of the major challenges in NILM [42]. In this paper, we tackle this issue and contribute with:

  • 1.

    a concise and up-to-date review of the features reported in recent NILM literature (Section 2) and

  • 2.

    a systematic signature identification algorithm based on a comprehensive dataset with diverse appliances and various households (Section 3.1).

Section snippets

State of the art on feature extraction

The seminal work by Hart [2], has inspired extensive research on extracting features and developing discriminating algorithms for NILM purposes. Zeifman and Roth [3] and Zoha et al. [4] provide an extensive overview of features and algorithms that were proposed before 2012. In this section, a concise and updated review is provided that incorporates the latest developments in the state-of-the-art on feature extraction.

The type of features that can be extracted from voltage (V) and current (I)

Methodology

In this section, a systematic approach is presented to identify and combine features in such a way that the ability to distinguish various loads is maximized.

A comprehensive list of steady and transient state features, including references to their extraction steps is outlined in Table 2 and Appendix A.

The feature selection process starts from all features and iteratively eliminates the least important ones until a reduced subset is obtained. Note that the goal is not to reduce number of

Results and discussion

In this section, we demonstrate our systematic feature selection process and present the per appliance based analysis using a comprehensive dataset explained next.

Conclusion

The effectiveness of a NILM algorithm to distinguish between appliances largely depends on determining a set of discriminative features. Various research has focused on suggesting and extracting such features to classify appliances. As a first contribution, we provided a systematic listing and comparison between features that have been proposed.

As a second contribution, we constructed and suggested an optimal subset of features to be used in a computational model which can achieve top

Future work

The majority of the proposed NILM solutions are tested on private datasets with a limited number of appliances. Additionally, the existing literature categorizes appliances into three categories based on their operational characteristics: ON/OFF appliances, multiple state appliances, and variable load appliances. However, when it comes to identifying a unique signature for each appliance, such categorization is not very effective because it ignores the front-end circuit topology of the

References (55)

  • J.T. Powers et al.

    Using a rule-based algorithm to disaggregate end-use load profiles from premise-level data

    IEEE Comput. Appl. Power

    (1991)
  • M. Baranski et al.

    Nonintrusive appliance load monitoring based on an optical sensor

    2003 IEEE Bologna Power Tech Conference Proceedings

    (2003)
  • M. Baranski et al.

    Detecting patterns of appliances from total load data using a dynamic programming approach

    Fourth IEEE International Conference on Data Mining, 2004, ICDM’04

    (2004)
  • M. Baranski et al.

    Genetic algorithm for pattern detection in nialm systems

    IEEE International Conference on Systems, Man and Cybernetics, 2004, vol. 4

    (2004)
  • A.G. Ruzzelli et al.

    Real-time recognition and profiling of appliances through a single electricity sensor

    2010 7th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks (SECON)

    (2010)
  • M.B. Figueiredo et al.

    An Experimental Study on Electrical Signature Identification of Non-Intrusive Load Monitoring (NILM) Systems

    (2011)
  • F. Sultanem

    Using appliance signatures for monitoring residential loads at meter panel level

    IEEE Trans. Power Deliv.

    (1991)
  • D. Srinivasan et al.

    Neural-network-based signature recognition for harmonic source identification

    IEEE Trans. Power Deliv.

    (2006)
  • C. Laughman et al.

    Power signature analysis

    IEEE Power Energy Mag.

    (2003)
  • M. Berges et al.

    Learning systems for electric consumption of buildings

    ASCE International Workshop on Computing in Civil Engineering

    (2009)
  • M.E. Berges et al.

    Enhancing electricity audits in residential buildings with nonintrusive load monitoring

    J. Ind. Ecol.

    (2010)
  • M. Dong et al.

    Non-intrusive signature extraction for major residential loads

    IEEE Trans. Smart Grid

    (2013)
  • S.B. Leeb et al.

    Transient event detection in spectral envelope estimates for nonintrusive load monitoring

    IEEE Trans. Power Deliv.

    (1995)
  • S.R. Shaw et al.

    Nonintrusive load monitoring and diagnostics in power systems

    IEEE Trans. Instrum. Meas.

    (2008)
  • K.D. Lee et al.

    Estimation of variable-speed-drive power consumption from harmonic content

    IEEE Trans. Energy Convers.

    (2005)
  • W.L. Chan et al.

    Harmonics load signature recognition by wavelets transforms

    DRPT2000. International Conference on Electric Utility Deregulation and Restructuring and Power Technologies. Proceedings (Cat. No.00EX382)

    (2000)
  • C. Duarte et al.

    Non-intrusive load monitoring based on switching voltage transients and wavelet transforms

    2012 Future of Instrumentation International Workshop (FIIW) Proceedings

    (2012)
  • Cited by (138)

    View all citing articles on Scopus
    View full text