No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
RELEVANT FEATURE SELECTION FOR HOME APPLIANCES RECOGNITION
Abstract
All over the world, the residential sector represents an important part in electrical energy consumption, and thus, is an opportunity to address substantial savings in terms of energy and money. In order to attempt this objective , a relevant knowledge of the appliances used in residential buildings is needed in order to better control or monitor energy consumption. This will be made possible through an effective automatic recognition of the home appliances. In this context, the main objective of this work is to be able to describe appliances as best as possible in order to recognize them individually, using features deduced from current and voltage measurements recorded at the grid connection point. In this paper, methods for selecting the most relevant features allowing the recognition of home appliances are proposed. The set up of a database of sampled measurements recorded on various home appliances types is also introduced.
Supplementary resource (1)
Data
November 2017
... This is the case for the features provided by deep convolutional neural networks (CNN), which can suffer from the robustness issue with adversarial examples [13,14], which require the use of attention mechanisms [15]. However, only a few works investigate in detail the role and the meaning of Feature Selection (FS) methods in NILM problems when addressed through a pattern recognition approach [16][17][18][19][20]. ...
... We can describe the different HEAs using 90 features extracted at each voltage period summarized in Table 1. The detail of their computation was introduced in [19], based on the latest IEEE 1459-2010 standard for the definition of single phase physical components under non-sinusoidal conditions [28,29]. From the voltage v(t) and current i(t) signals, we compute the Fourier coefficients v ak , v bk , i ak and i bk , using the following formulas: ...
... • Finally, the global and harmonic power factors F p and F pk : Table 1. Summary of the proposed electrical features [19]. The features colored in red meet the additivity criterion given by Equation (13). ...
Non-Intrusive Load Monitoring (NILM) refers to the analysis of the aggregated current and voltage measurements of Home Electrical Appliances (HEAs) recorded by the house electrical panel. Such methods aim to identify each HEA for a better control of the energy consumption and for future smart grid applications. Here, we are interested in an event-based NILM pipeline, and particularly in the HEAs’ recognition step. This paper focuses on the selection of relevant and understandable features for efficiently discriminating distinct HEAs. Our contributions are manifold. First, we introduce a new publicly available annotated dataset of individual HEAs described by a large set of electrical features computed from current and voltage measurements in steady-state conditions. Second, we investigate through a comparative evaluation a large number of new methods resulting from the combination of different feature selection techniques with several classification algorithms. To this end, we also investigate an original feature selection method based on a deep neural network architecture. Then, through a machine learning framework, we study the benefits of these methods for improving Home Electrical Appliance (HEA) identification in a supervised classification scenario. Finally, we introduce new transfer learning results, which confirm the relevance and the robustness of the selected features learned from our proposed dataset when they are transferred to a larger dataset. As a result, the best investigated methods outperform the previous state-of-the-art results and reach a maximum recognition accuracy above 99% on the PLAID evaluation dataset.
... In the literature, most of the state-of-the-art methods disaggregate the observed energy curve measured at the building electrical panel by computing signal features that allow the recognition of the HEA signature [10][11][12]. Previous works aim at computing an effective set of electrical features that are used in a machine learning framework [13][14][15]. A recent study [16] proposes a deep learning approach based on CNN applied on a so called "VI trajectory" for which the authors report an average F-measure of 77.60% on the publicly available PLAID dataset [17]. ...
... Thus, investigating HEA identification allows us to comparatively assess in a practical scenario the use of several STFT-based TF representations with different TF concentration and to establish links with our previous works [15,18,19] based on "handcrafted" physically meaningful electrical features computed from the active and reactive powers. Our Time-Frequency Representation (TFR) are investigated with several deep CNN architectures. ...
... For a finite length signal, with a window h defined on [−K/2, K/2], the bounds of the summation can be replaced by K min = max(0, n − K/2) et K max = min(M − 1, n + K/2 − 1). Hence, Equation (15) can be computed by: ...
Since decades past, time-frequency (TF) analysis has demonstrated its capability to efficiently handle non-stationary multi-component signals which are ubiquitous in a large number of applications. TF analysis us allows to estimate physics-related meaningful parameters (e.g., F 0 , group delay, etc.) and can provide sparse signal representations when a suitable tuning of the method parameters is used. On another hand, deep learning with Convolutional Neural Networks (CNN) is the current state-of-the-art approach for pattern recognition and allows us to automatically extract relevant signal features despite the fact that the trained models can suffer from a lack of interpretability. Hence, this paper proposes to combine together these two approaches to take benefit of their respective advantages and addresses non-intrusive load monitoring (NILM) which consists of identifying a home electrical appliance (HEA) from its measured energy consumption signal as a "toy" problem. This study investigates the role of the TF representation when synchrosqueezed or not, used as the input of a 2D CNN applied to a pattern recognition task. We also propose a solution for interpreting the information conveyed by the trained CNN through different neural architecture by establishing a link with our previously proposed "handcrafted" interpretable features thanks to the layer-wise relevant propagation (LRP) method. Our experiments on the publicly available PLAID dataset show excellent appliance recognition results (accuracy above 97%) using the suitable TF representation and allow an interpretation of the trained model.
... The common thread in all these research works is that the detection is done in a univariate context by only considering the active power signal, whereas prior literature on NILM focuses on identifying an effective f eatures' set that defines a unique HEA signature [22][23][24] . Fig. 2 illustrates three power time series related to the same scenario of several HEAs switched on and off using our own acquisition system [25] , and three power time series derived from the Building-Level fully-labeled Dataset for Electricity Disaggregation (BLUED) [26] . ...
... FS is a process commonly used for classification tasks to reduce overfitting and to avoid the curse of dimensionality problem [39,40] . The selected subset contains the minimal number of features that contribute most to the classification accuracy [24,40] . In this section, we aim at only using FS for event detection purposes and not for classification. ...
... TP ( 24: 25: ...
Non Intrusive Load Monitoring (NILM) approaches refer to the analysis of the aggregated electrical signals of Home Electrical Appliances (HEAs) in order to identify their operating schedules. It has emerged as a promising solution to help residential consumers to reduce their electricity bills through a breakdown of energy consumption. NILM methods are either event-based or non event-based. This categorization depends on whether or not they rely on the detection of HEAs’ significant state transitions (e.g., On/Off or state change) in power consumption signals. This paper focuses on event-based approaches and especially in multivariate change detection algorithms. It aims at highlighting the benefits brought by a multivariate approach for change detection using the appropriate electrical features. We first suggest to extend four existing change
detection algorithms in the multidimensional case. The studied detection algorithms are first detailed and compared to each other and to their existing scalar versions through numerical simulations. Then, a new feature selection algorithm for change detection is presented and assessed when combined with the most efficient detector among the four investigated ones. Finally, the feature
selection method for detection purposes is applied to two different NILM case studies. The first one uses power features derived from the BLUED current and voltage measurements and the second one is based on current and voltage measurements acquired using our own acquisition system. Compared to the classical scalar approach, the results show that the multivariate approach brings a significant performance improvement when the features selected by the
proposed algorithm are used.
... The inclusion and reinforcement of power theories as a useful tool for identifying loads is another important issue. Houdi et al. [64] performed the reduction of features deduced from voltage and current measurements, targeting applications in a digital system using a low-cost processor with a low storage capacity. From ninety features derived from the IEEE 1459-2010 [16] standard, RMS voltage, RMS current, and the apparent, reactive, distortion, and active powers, the authors compared the success rate of the whole feature set with: In feature selection, the goal is to choose a subset of features that provides better distinctive performance than when using all of them. ...
... This latter creates new features from transformations of the original features into lowdimensional data. For instance, in the NILM literature, some research efforts use PCA [31,64], Linear Discriminant Analysis (LDA) [65], Autoencoder [66], and t-distributed stochastic neighbor embedding (t-SNE) [67]. ...
The load disaggregation concept is gaining attention due to the increasing need for optimized energy utilization and detailed characterization of electricity consumption profiles, especially through Nonintrusive Load Monitoring (NILM) approaches. This occurs since knowledge about individualized consumption per appliance allows to create strategies striving for energy savings, improvement of energy efficiency, and creating energy awareness to consumers. Moreover, by using feature extraction to devise energy disaggregation, one can achieve accurate identification of electric appliances. However, even though several literature works propose distinct features to be utilized, no consensus exists in the literature about the most appropriate set of features that ensure high accuracy on load disaggregation. Thus, beyond presenting a critical analysis of some significant features often selected in the literature, this paper proposes identifying the most relevant ones considering collinearity and machine learning algorithms. The results show that high-performance metrics can be achieved with fewer features than usually adopted in the literature. Moreover, it is demonstrated that the Conservative Power Theory can offer the most representative features for appliance identification, leading to efficient power consumption disaggregation.
... In this paper, we present a low cost and easy to install device for current and voltage acquisitions in the NILM framework, based on an Arduino MKR Zero microcontroller. The goal of such a device is to acquire voltage and current measurements that will be used to detect the switching on/off of HEAs [21] and to identify them once their electrical signatures have been set up [22]. Energy monitoring devices available in the market as well as prototypes set up by the NILM research community were presented with special emphasis on their main technical characteristics. ...
Non Intrusive Load Monitoring (NILM) of Home Electrical Appliances (HEAs) has received a major interest in the last twenty years due to the increase of electricity consumption in the residential sector. One NILM key component is the acquistion of voltage and current signals at a high sampling rate, to ease HEAs recognition. In this paper, we discuss the design of a cheap and easy to install current and voltage measurement system based on an Arduino MKR Zero microcontroller. This system allows to monitor the usage of HEAs and thus give to consumers information on their consumption and help them to manage it. The hardware architecture of the proposed system is first described. It includes the current and voltage conditioning circuits. Then, a detail of the software implemented in the microcontroller is given. It consists mainly of an interruption routine that ensures the acquisition regularity and a background task for storing data into a micro SD card. Finally, we show in the last section several individual HEAs power consumption profiles and a whole house power consumption profile obtained using the designed acquisition device.
... The former use an event detection algorithm to locate and identify significant HEA events (e.g. On/Off or state transitions) in the power consumption curve, thanks to the HEAs' signatures available in the HEA features database [7]. The latter disaggregate every sample of the load curve to provide the contribution of each identified HEA in the total consumption [3]. ...
Energy disaggregation is a key enabling technology for residential power usage monitoring, which benefits various applications such as carbon emission monitoring and human activity recognition. However, existing methods are difficult to balance the accuracy and usage burden (device costs, data labeling and prior knowledge). As the high penetration of smart speakers offers a low-cost way for sound-assisted residential power usage monitoring, this work aims to combine a smart speaker and a smart meter in a house to liberate the system from a high usage burden. However, it is still challenging to extract and leverage the consistent/complementary information (two types of relationships between acoustic and power features) from acoustic and power data without data labeling or prior knowledge. To this end, we design COMFORT, a cross-modality system for self-supervised power usage monitoring, including (i) a cross-modality learning component to automatically learn the consistent and complementary information, and (ii) a cross-modality inference component to utilize the consistent and complementary information. We implement and evaluate COMFORT with a self-collected dataset from six houses in 14 days, demonstrating that COMFORT finds the most appliances (98%), improves the appliance recognition performance in F-measure by at least 41.1%, and reduces the Mean Absolute Error (MAE) of energy disaggregation by at least 30.4% over other alternative solutions.
This research aims at studying the performance of k-nearest neighbor classification when applying different distance measurements. In this work, we comparatively study 11 distance metrics including Euclidean, Standardized Euclidean, Mahalanobis, City block, Minkowski, Chebychev, Cosine, Correlation, Hamming, Jaccard, and Spearman. A series of experimentations has been performed on eight synthetic datasets with various kinds of distribution. The distance computations that provide highly accurate prediction consist of City block, Chebychev, Euclidean, Mahalanobis, Minkowski, and Standardize Euclidean techniques.
Automatic timbre characterization of audio signals can help to measure similarities
between sounds and is of interest for automatic or semi-automatic databases indexing.
The most effective methods use machine learning approaches which require qualitative
and diversified training databases to obtain accurate results. In this paper, we introduce a
diversified database composed of worldwide non-western instruments audio recordings
on which is evaluated an effective timbre classification method.
A comparative evaluation based on the well studied Iowa musical
instruments database shows results comparable with those of state-of-the-art methods.
Thus, the proposed method offers a practical solution for automatic ethnomusicological
indexing of a database composed of diversified sounds with various quality. The relevance of audio
features for the timbre characterization is also discussed in the context of non-western
instruments analysis.
The article tackles the issues related to the identification of electrical appliances inside residential buildings. Each appliance can be identified from the aggregate power readings at the meter panel. The possibility of applying a temporal multilabel classification approach in the domain of non-intrusive load monitoring is explored (non-event based method). A novel set of meta-features is proposed. The method is tested on sampling rates based on the capabilities of current smart meters. The proposed approach is validated over a dataset of energy readings at residences for a period of a year for 100 houses containing different sets of appliances (water heater, washing machines, etc.). This method is applicable for the demand side management of households in the current limitation of smart meters, from the inhabitants or from the grid operator’s point of view.
This paper presents a current sensor-based home appliance and its state recognition method for intelligent outlets. Our system has three main functions: remote control, monitoring, and power supply schedule management. This research focuses particular on the monitoring function. To recognize the appliance and the state of the appliance, we extract ten features based on a measured current signal. In the experiment, we gather a number of signals with various appliances, and find that three features I rms , I avg , and I peak yield valid recognition results of 84.3%, 86.4%, and 90.3% for classifying the state of the appliance into three categories. Moreover, sufficient recognition rates of 97.4%, 97.7%, and 99.0% are obtained by consideration of three candidates.
We are developing a novel home network system based upon the integration of information and energy. The system aims to analyze
user behavior with a power-sensing network and provide various life-support services to manage power and electric appliances
according to user behavior and preferences. This paper describes an electric appliance recognition method using power-sensing
data measured by CECU (Communication and Energy Care Unit) which is an intelligent outlet with voltage and current sensors to integrate legacy appliances (which are incompatible
with a communications network) with the home network. Furthermore, we demonstrate a prototype home energy management system
and examples of services based upon appliance recognition.
Smart grid privacy encompasses the privacy of information extracted by analysing smart metering data. In this paper, we suggest that home electrical power routing can be used to moderate the home's load signature in order to hide appliance usage information. In particular, (1) we introduce a power management model using a rechargeable battery, (2) we propose a power mixing algorithm, and (3) we evaluate its protection level by proposing three different privacy metrics: an information theoretic (relative entropy), a clustering classification, and a correlation/regression one; these are tested on different metering datasets. This paper sets the ground for further research on the subject of optimising home energy management with regards to hiding load signatures.
Variable and feature selection have become the focus of much research in areas of application for which datasets with tells or hundreds of thousands of variables are available. These areas include text processing of internet documents, gene expression array analysis, and combinatorial chemistry. The objective of variable selection is three-fold: improving the prediction performance of the predictors, providing faster and more cost-effective predictors, and providing a better understanding of the underlying process that generated the data. The contributions of this special issue cover a wide range of aspects of such problems: providing a better definition of the objective function, feature construction, feature ranking, multivariate feature selection, efficient search methods, and feature validity assessment methods.
ABSTRACT Principal Components Analysis (PCA) is the predominant linear dimensionality reduction technique, and has been widely applied on datasets in all scientiflc domains. We consider, both theoretically and empirically, the topic of unsuper- vised feature selection for PCA, by leveraging algorithms for the so-called Column Subset Selection Problem (CSSP). In words, the CSSP seeks the\best"subset of exactly k columns from an m£n data matrix A, and has been extensively stud- ied in the Numerical Linear Algebra community. We present a novel two-stage algorithm for the CSSP. From a theoretical perspective, for small to moderate values of k, this algorithm signiflcantly improves upon the best previously-existing re- sults [24, 12] for the CSSP. From an empirical perspective, we evaluate this algorithm as an unsupervised feature selec- tion strategy in three application domains of modern,sta- tistical data analysis: flnance, document-term data, and ge- netics. We pay particular attention to how this algorithm
In this work, musical instrument recognition is consid- ered on solo music from real world performance. A large sound database is used that consists of musical phrases ex- cerpted from commercial recordings with different instru- ment instances, different players, and varying recording conditions. The proposed recognition scheme exploits class pairwise feature selection based on inertia ratio maximization. More- over, new signal processing features based on octave band energy measures are introduced that prove to be useful. Classification is performed using Gaussian Mixture Mod- els in a one vs one fashion in association with a data rescal- ing procedure as pre-processing. Experimental results show that substantial improvement in recognition success is thus achieved.
Feature selection is applied to reduce the number of features in many applications where data has hundreds or thousands of features. Existing feature selection methods mainly focus on finding relevant features. In this paper, we show that feature relevance alone is insufficient for efficient feature selection of high-dimensional data. We define feature redundancy and propose to perform explicit redundancy analysis in feature selection. A new framework is introduced that decouples relevance analysis and redundancy analysis. We develop a correlation-based method for relevance and redundancy analysis, and conduct an empirical study of its efficiency and effectiveness comparing with representative methods.
This paper presents the top 10 data mining algorithms identified by the IEEE International Conference on Data Mining (ICDM) in December 2006: C4.5, k-Means, SVM, Apriori, EM, PageRank, AdaBoost, kNN, Naive Bayes, and CART. These top 10 algorithms are among the most influential data mining algorithms in the research community. With each algorithm, we provide a description of the algorithm, discuss the impact of the algorithm, and review current and further research on the algorithm. These 10 algorithms cover classification, clustering, statistical learning, association analysis, and link mining, which are all among the most important topics in data mining research and development. Yes Yes
This paper addresses the problem of classifying large databases of musical instrument sounds. An efficient algorithm is proposed for selecting the most appropriate signal features for a given classification task. This algorithm, called IRMFSP, is based on the maximization of the ratio of the between-class inertia to the total inertia combined with a step-wise feature space orthogonalization. Several classifiers -flat gaussian, flat KNN, hierarchical gaussian, hierarchical KNN and decision tree classifiers -are compared for the task of large database classification. Especially considered is the application when our classification system is trained on a given database and used for the classification of another database possibly recorded in completely different conditions. The highest recognition rates are obtained when the hierarchical gaussian and KNN classifiers are used. Organization of the instrument classes is studied through an MDS analysis derived from the acoustic features of the sounds.
Consumer systems for home energy management can provide significant energy saving. Such systems may be based on nonintrusive appliance load monitoring (NIALM), in which individual appliance power consumption information is disaggregated from single-point measurements. The disaggregation methods constitute the most important part of NIALM systems. This paper reviews the methodology of consumer systems for NIALM in residential buildings.
This paper analyses the interactions between the incremental
changes of current harmonic phasors injected by a power system NL
(nonlinear load) and the resulting variations of harmonic voltages and
powers. It proves that it is not always possible, following simple
measurements of individual harmonic powers, to decide if a certain
harmonic current is harmful or useful. It is suggested that the NL
“distortion” be evaluated with the help of a quantity called
non60 Hz (nonfundamental) apparent power
Non intrusive load monitoring (NILM) performance evaluation : A unified approach for accuracy reporting
- S Makonin
- F Popowich
S. Makonin, F. Popowich, "Non intrusive load
monitoring (NILM) performance evaluation :
A unified approach for accuracy reporting",
Energy Efficiency, Vol. 8, No. 4, pp. 809-814,
2015.