Smart non-intrusive appliance identification using a novel local power histogramming descriptor with an improved k-nearest neighbors classifier
Yassine Himeur, Abdullah Alsalemi, Faycal Bensaali, Abbes Amira
SS MART NON - INTRUSIVE APPLIANCE IDENTIFICATION USING ANOVEL LOCAL POWER HISTOGRAMMING DESCRIPTOR WITH ANIMPROVED K - NEAREST NEIGHBORS CLASSIFIER
A P
REPRINT
Yassine Himeur ∗ , Abdullah Alsalemi, Faycal Bensaali Department of Electrical EngineeringQatar UniversityDoha, Qatar ya[email protected];[email protected];[email protected]
Abbes Amira
Institute of Artificial IntelligenceDe Montfort UniversityLeicester, United Kingdom [email protected]
February 10, 2021 A BSTRACT
Non-intrusive load monitoring (NILM) is a key cost-effective technology for monitoring powerconsumption and contributing to several challenges encountered when transiting to an efficient,sustainable, and competitive energy efficiency environment. This paper proposes a smart NILMsystem based on a novel local power histogramming (LPH) descriptor, in which appliance powersignals are transformed into 2D space and short histograms are extracted to represent each device.Specifically, short local histograms are drawn to represent individual appliance consumption signa-tures and robustly extract appliance-level data from the aggregated power signal. Furthermore, animproved k-nearest neighbors (IKNN) algorithm is presented to reduce the learning computationtime and improve the classification performance. This results in highly improving the discriminationability between appliances belonging to distinct categories. A deep evaluation of the proposedLPH-IKNN based solution is investigated under different data sets, in which the proposed schemeleads to promising performance. An accuracy of up to 99.65% and 98.51% has been achieved onGREEND and UK-DALE data sets, respectively. While an accuracy of more than 96% has beenattained on both WHITED and PLAID data sets. This proves the validity of using 2D descriptors toaccurately identify appliances and create new perspectives for the NILM problem. K eywords Non-intrusive load monitoring · appliance identification ·
2D representation · local power histograms · feature extraction · improved k-nearest neighbors. Buildings are responsible on more than 32 percent of the overall energy consumed worldwide, and this percentageis expected to be doubled by 2050 as a result of the well-being improvement and wide use of electrical appliancesand central heating/cooling systems [1]. Specifically, this is due to population growth, house comfort enhancementand improvement of wealth and lifestyle. To that end, reducing wasted energy and promoting energy-saving inbuildings have been nowadays emerged as a hot research topic. One of the cost-effective solutions is via encouraging ∗ Sustainable Cities and Society 67, 102764, 2021 a r X i v : . [ c s . C Y ] F e b PREPRINT - F
EBRUARY
10, 2021energy-efficiency behaviors among building end-users based on analyzing energy consumption footprints of individualappliances. Therefore, tailored recommendations can be generated to help end-users improve their behavior [2].In this regard, load monitoring of appliances can not only provide the end-users with their fine-grained consumptionfootprints, but it promptly contributes in promoting sustainability and energy efficiency behaviors [3]. Moreover, it cansignificantly contribute in elaborating and developing reliable smart-grid demand management systems. On the otherhand, load consumption monitoring principally encompasses two wide groups, namely intrusive load monitoring (ILM)and non-intrusive load monitoring (NILM), respectively. ILM necessitates to install smart-meters at the front-end ofeach electrical appliance aiming at collecting real-time energy consumption patterns. Even though this strategy presentshigh performance in accurately gathering appliance-specific data, it requires a heavy lifting with high-cost installation,where a large number of sun-meters are installed. In addition an intrusive transformation of the available power grid isessential [4]. On the contrary, no additional sub-meter required when the NILM strategy (named energy disaggregationas well) is adopted to infer device-specific consumption footprints since the latter are immediately extracted from themain load using feature extraction and learning models [5].In this context, the NILM issue has been investigated for many years and extensive efforts are still paid to this problematicbecause of its principal contributions to improve energy consumption behavior of end-users [6, 7]. Specifically, it canhelp in achieving a better comprehending of consumers’ consumption behavior through supplying them with specificappliance data. Therefore, put differently, the NILM task indirectly aims at (i) promoting the energy efficiency behaviorof individuals, (ii) reducing energy bills and diminishing reliance on fossil fuel, and (iii) reducing carbon emissions andimproving environmental conditions [8].Two crucial stages in NILM are the feature extraction and inference and learning procedures. The feature extractionstep aims at deriving pertinent characteristics of energy consumption signals to help representing appliances from thesame category with similar signatures while differencing between power signals from different classes [9]. On the otherhand, the inference and learning step is essentially reserved to train classifiers in order to identify appliances and extractappliance-level power footprints [10]. It can be achieved either by using conventional classification models, such asartificial neural networks (ANN), support vector machine (SVM), k-nearest neighbors (KNN), etc. or novel classifiers,including deep neural networks (DNNs). Consequently, the identification of electrical devices simultaneously operatingthrough an interval of time in a household is the central part of the NILM architecture. Its performance is highlydependent on the deployed feature extraction and inference model. To that end, the development of robust schemesbelonging to these two modules attracts a considerable interest in recent years [11, 12].In this paper, recent NILM systems are first reviewed based on the principal components contributing into theimplementation of such architectures including feature extraction and learning models. In this respect, techniquespertaining to three main feature extraction categories are described among them graph signal processing (GSP), sparsecoding features and binary encoding schemes. Following, a discussion of their limitations and drawbacks is alsopresented after conducting a deep comparison of their performances and properties. Moving forward, a non-intrusiveappliance identification architecture is proposed, which is mainly based on a novel local power histogramming (LPH)descriptor. The latter relies on (i) representing power signals in a 2D space, (ii) performing a binary power encoding insmall regions using square patches of × samples and (iii) returning back to the initial 1D space through extractinghistograms of 2D representations. Following, an improved k-nearest neighbors (IKNN) is introduced to effectivelyidentify appliance-level fingerprints and reduce the computation cost. This has resulted in very short appliance signaturesof 256 samples, in which each power signal is represented with a unique histogram, and thus leads to a better applianceidentification performance at a low computational complexity. Moreover, it is worth-noting that to the best of theauthors’ knowledge, this paper is the first work that discusses the applicability of 2D local descriptors for identifyingelectrical appliances using their power consumption signals. Overall, The main contributions of this paper can besummarized as follows:• We present a comprehensive overview of recent trends in event-based NILM systems along with describingthe their drawbacks and limitations.• We propose a novel NILM framework based on an original 2D descriptor, namely LPH, which can beconsidered as an interesting research direction to develop robust and reliable NILM solutions. Explicitly, afterconverting appliance power signals into 2D space, the appliance identification becomes a content-based imageretrieval (CBIR) problem and a powerful short description is extracted to represent each electrical device.According, LPH operates also as a dimensionality reduction, where each resulted appliance signature has only256 samples.• We design a powerful IKNN model that efficiently aids in recognizing appliances from the extracted LPHfingerprints and reducing significantly the computational cost.2 PREPRINT - F
EBRUARY
10, 2021• We evaluate the performance of the proposed LPH-IKNN based NILM system on four different data setswith distinct sampling frequency rates and in comparison with various recent NILM systems and other 2Ddescriptors.The remainder of this paper is structured as follows. An overview of NILM systems is introduced in Section 2 alongwith a discussion of their drawbacks and limitations. In Section 3, the main steps of the proposed NILM system based onthe LPH descriptor and IKNN are described in details. The performance results of the exhaustive empirical evaluationconducted in this framework are presented and thoroughly discussed in Section 4, in which different comparisons areconducted with state-of-the-art works. Finally, Section 5 concludes the paper, discusses the important findings andhighlights the future works.
NILM frameworks can be categorized into two major groups. The first one called non-event-based approaches, whichfocus on using algorithms without depending on the training/learning procedures (using data from a particular building).They can segregate the main power signal collected from the overall circuit into various appliance-level fingerprints.An explicit example of this kind of techniques that have been typically studied is related the deployment of statisticalanalysis, including hidden Markov models (HMM) [13], higher-order statistics (HOS) [14] or probabilistic models[15]. The second group deals with methods allowing the identification of state changes occurred in power consumptionsignals using different types event detectors, classifiers, and further implementing appropriate techniques to calculatean individual load usage fingerprint for each electrical device. In this section, we focus on describing recent NILMsystems pertaining to the second category because the proposed framework is an even-based NILM framework.Explicitly, this category of NILM systems deploys two principal components. The first one is a feature descriptor toextract pertinent characteristics of electrical appliances, while the second is a learning algorithm that can help in detectingand classifying each device based on its features. Conventional NILM methods have been basically concentrated onextracting features related to steady-states and transient-states, in addition to the adoption of conventional machinelearning (ML) classifiers. On the other side, novel strategies are introduced in recent years to deal with the NILM issuebased on the use of new signal analysis procedures and innovative learning models. This class of NILM frameworks isdefined as non-conventional, they are classified into four principal sub-categories as follows:
Graph signal processing (GSP):
A trending research field aiming at describing stochastic characteristics of powersignals based on graph theory. In [16], a graph-based method for identifying individual appliances has been introducedafter detecting appliance events. This results in a better detection of appliance-level fingerprints and further a reductionof time computation compared to conventional graph-based techniques. In [17], various multi-label graphs have beendeveloped to detect individual devices based on a semi-supervised procedure. In [18], NILM performance have beenenhanced via the use of a generic GSP-based technique, which is build upon the application of graph-based filters. Thisresults in a better detection of on/off appliance states, via the mitigation of electric noise produced by appliances.
Sparse coding features:
In this category, the NILM framework is treated as a blind source separation problemand recent sparse coding schemes are then applied to split an aggregated power consumption signal into specificappliance based profiles [19]. In [20], a co-sparse analysis dictionary learning is proposed to segregate the total energyconsumption into a device-level data and significantly shorten the training process. In [21], a deep learning architectureis used for designing a multi-layer dictionary of each appliance rather than constructing one-level codebook. Obtainedmulti-layer codebooks are then deployed as features for the source-separation algorithm in order to break down theaggregated energy signal. In [22], an improved non-negative matrix factorization is used to pick up perceptibly valuableappliance-level signatures from the aggregated mixture.
Binary descriptions:
Most recently, binary descriptors have been investigated for the classification and fault detectionof 1D signals such as electroencephalogram (EEG), electrocardiogram (ECG), and myoelectric signals [23]. For powerconsumption signals, this concept is novel. The only few works that can be found in the literature are mainly focusingon representing the power signal in a novel space and directly being used to train the convolutional neural network(CNN). In [24], power fingerprints are derived by estimating the similarity of voltage-current (V-I) shapes, encoding itusing a binary dictionary and then extracting image graphical footprints that are directly fed to a self-organizing map(SOM) classifier, which is based on neural networks. In [25], V-I binary representation is employed through convertingthe normalized V-I magnitude into binary matrices using a thresholding process before being fed to a CNN. Morespecifically, this approach relies on binary coding of the V-I edges plotted in the new representation. These data are3
PREPRINT - F
EBRUARY
10, 2021then fed into an ML classifier in order to identify each appliance class. In [26], a color encoder is proposed to drawV-I signatures that can also be translated to visual plots. These footprints are then fed to a deep learning classifier toidentify each electrical appliance. In [27], a siamese-neural network is employed aims at mapping the V-I trajectoriesinto a novel characteristic representation plan.
Time-frequency analysis:
Time-frequency analysis is an imperative research topic, in which much attention has beendevoted to it in the past and even nowadays. It is applied in several applications among them energy efficiency [28],NILM or energy disaggregation [29] and power consumption anomaly detection [30]. In [31, 32], a novel NILMdescriptor is proposed based on the fusion of different time-domain descriptors. In [5], a novel time-scale analysis isadopted based on the use of multi-scale wavelet packet trees (MSWPT) and a cepstrum-based event detection schemeto glean appliance-level power consumption patterns from the aggregated load.
In the building energy sector, KNN has been widely deployed in the literature for different purposes, such as energydisaggregation [33] and anomaly detection [34, 35, 36] although it has some issues, e.g. the sensitivity of theneighborhood size k could significantly degrade its performance [37, 38]. To that end, an improved version is proposedin [39] to address this issue, named generalized mean distance-based k-nearest neighbor. Specifically, multi-generalizedmean distances are introduced along with the nested generalized mean distance that rely on the properties of thegeneralized mean. Accordingly, multi-local mean vectors of a specific pattern in every group are estimated throughdeploying its class-specific k nearest neighbors. Using the obtained k local mean vectors per group, the related k generalized mean distances are estimated and thereby deployed for designing the categorical nested generalized meandistance. Similarly, in [40], the authors introduce a local mean representation-based KNN aiming at further improvingthe classification performance and overcoming the principal drawbacks of conventional KNN. Explicitly, they select thecategorical KNN coefficients of a particular pattern to estimate the related categorical k-local mean vectors. Following,a linear combination of the categorical k-local mean vectors is used to represent the particular pattern. Moving forward,in order to capture the group of this latter, group-specific representation-based distances between the particular patternand the categorical k-local mean vectors are then considered.Moreover, in [41], two locality constrained representation-based KNN rules are presented to design an improved KNNclassifier. The first one is a weighted representation-based KNN rule, in which the test pattern is considered as a linearaggregation of its KNN samples from every group, while the localities of KNN samples per group are represented as theweights constraining their related representation elements. Following, a classification decision rule is used to calculatethe representation-based distance between the test pattern and the group-specific KNN coefficients. On the other side,the second rule is a weighted local mean representation-based KNN, where k-local mean vectors of KNN coefficientsper group are initially estimated and then utilized to represent the test pattern. On the other hand, aiming at improvingthe performance of existing KNN classifiers and making them scalable and automatic, granular ball computing has beenused in various frameworks. This is the case of [42], where a granular ball KNN (GBKNN) algorithm is developed,which could perform the classification task on large-scale data sets with low computation. In addition, it provides asolution to automatically select the number k of clusters. In addition to the use of KNN and its variants, K-means clustering (KMC) is another important data clustering method.It has been widely investigated to classify similar data into the same cluster in large-scale data sets for differentapplications, such as appliance identification [43], anomaly detection [44], cancer detection [45], and social mediaanalysis [46]. Despite the simplicity of KMC, its performance was not convincing in some applications. To that end,different variants have been proposed in the literature to design efficient, scalable and robust KMC classifiers. Forexample, in [47], to overcome the vulnerability of the conventional KMC classifier to outliers and noisy data, a tri-levelk-means approach is introduced. This was possible through updating the cluster centers because data in a specificdata set usually change after a period of time. Therefore, without updating the cluster centers it is not possible toaccurately represent data in every cluster. While in [48], the authors focus on improving both the accuracy and stabilityof the KMC classifier. This has been achieved by proposing a k-means scheme based on density Canopy, which aimsat solving the issue corresponding to the determination of the optimal number k of clusters along with the optimuminitial seeds. Specifically, the density Canopy has been utilized as a pre-processing step and then its feedback has beenconsidered as the cluster number and initial clustering center of the improved KMC technique. Similarly, in [49], anincremental KMC scheme is introduced using density estimation for improving the clustering accuracy. Explicitly, thedensity of input samples has been firstly estimated, where every primary cluster consists of the center points having a4 PREPRINT - F
EBRUARY
10, 2021density superior than a given threshold along with points within a specific density range. Following, the initial clusterhas been merged with reference to the distance between the two cluster centers before dividing the points without anycluster affection into clusters nearest to them.On the other hand, in some specific data sets, e.g. real-world medical data sets, data samples could pertain to morethan one cluster simultaneously while traditional KMC methods do not allow that since they are developed based on anexclusive clustering process. Therefore, an overlapping k-means clustering (OKMC) scheme is proposed in [50] toovercome that issue, which have intrinsically overlapping information. Similarly, in [51], the authors introduce a hybridclassifier that aggregates k-harmonic means and OKMC to address the sensitivity problem of the latter to initial clustercentroids.
Despite the fact that the outlined event-based NILM systems have recently been widely examined in the state-of-the-art,they can be affected by certain problems and limitations, which impede the development of powerful NILM architecturesand even increase the difficulty to implementing real-time NILM systems. Moreover, most of these issues have notyet been overcome. For example, most of existing solutions suffer from a low disaggregation accuracy. Therefore,these approaches need deeper investigation in order to improve their performance. Moreover, they are usually builtupon detecting transient states, which can limit their detection accuracy if multiple appliances are turning on/offsimultaneously. In addition, most of the reviewed NILM systems are only validated on one category of data with aunique sampling frequency. This restricts the applicability of these techniques on different data repositories. On theother hand, most of the existing classifier have some issue to accurately identify appliance-level data especially if thevalidation data set is imbalanced.To overcome the aforementioned limitations, we present, in this framework, a novel non-intrusive load identification,which relies on (i) shifting power fingerprints into 2D space, (ii) deriving binary characteristics at local regions, (iii)representing the extracted features in the decimal field, and (iv) going back to 1D space via capturing novel histogramsof the 2D representations. Following, these steps can help in designing a robust identification approach, which hasvarious benefits; (i) via transforming the appliance signatures into 2D space, novel appliance footprints are developedthat describe each appliance fingerprint in another way and texture descriptions are derived from local regions usingsquare kernels; (ii) the proposed strategy helps in identifying appliances at accurately without depending on thedevices’ states (i.e. steady or transient); (iii) the proposed scheme can support real-time applications because it canbe run at a low computation cost. Specifically, it acts as a dimensionality reduction component as well, where shortcharacteristic histograms having only 256 samples are collected at the final stage to represent every appliance, and(iv) an improved KNN algorithm has been developed to overcome the issues occurring with imbalanced data sets andimprove the appliance identification performance. Moreover, our 2D descriptor can be trained via simple machinelearning algorithms without the need to deploy deep leaning models, which usually have a high computation complexity.
This section focuses on presenting the principal steps of the proposed appliance identification system, which relies onthe application of an original 2D descriptor. Accordingly, the flowchart of the proposed NILM system is portrayed inFig. 1. It is clear that the 2D-based load identification system represents the fundamental part of the NILM system.
In recent years, 2D local feature extraction schemes have received significant attention in various research topics,including image and video processing [52], breast cancer diagnosis [53], face identification [54] and fingerprintrecognition [55]. They are generally deployed to derive fine-grained characteristics after partitioning the overall 2Drepresentation into various local regions using small kernels. Explicitly, a local feature extraction can be applied at eachlocal region of the 2D representation to draw pertinent features about the neighborhood of each key-point. The multiplefeatures derived from several regions are then fused into a unique, spatially augmented characteristic vector, in whichthe initial signals are effectively represented.
For the event detection step, various event detection schemes are proposed in the state-of-the-art. Event detectiontechniques are split into three main groups [56]: specialized heuristics, probabilistic models and matched filters [57, 58].In this framework, the pre-processed aggregated power is segregated into different sections using the edge detector5
PREPRINT - F
EBRUARY
10, 2021module [59] implemented in the NILMTK platform [60]. Accordingly, the on/off events of electrical devices aregenerally picked up via the analysis of power level variations in the aggregated signal. This event detector has beenelected because of its simplicity and availability of its source code in the NILMTK platform.
The proposed appliance identification scheme relies mainly on transforming the appliance consumption signals into 2Dspace and therefore treating the appliance recognition task as a CBIR problem. With this in mind, all image descriptorscould be utilized to extract the fine-grained properties of the obtained 2D power signal representations.In that respect, the proposed LPH-based feature extraction scheme transforms appliance signals to image representations.Following, an examination local regions around each power sample is performed using a block partition procedureto collect local features. Explicitly, LPH descriptor is introduced for abstracting histogram-based descriptions of the2D representations of power observations. Accordingly, LPH performs a binary encoding of power blocks throughcomparing the central power sample of each block with its neighbors.Fig. 2 explains the flowchart of the proposed LPH description scheme. A comparison of each central power observationis conducted with its power neighboring in a kernel of N × N power samples through subtracting the central powervalue from the neighboring power patterns. Following, a binary encoding procedure is applied where the positive valuesof the subtractions are moved to 1, on the flip side, the negative values are considered as 0. Next, a binary sequence isthen acquired by means of a clockwise-based comparison process. Consequently, the gathered binary samples representthe corresponding LPH codes. Moving forward, the overall binary sequences are gleaned from all the regions (kernels)to form a binary array, which in turn, is converted to the decimal field. Specifically, each binary sequence extractedfrom a specific block is converted to decimal (as it is illustrated in Fig. 2). Lastly, a histogramming procedure is appliedon the resulted decimal array, in which an LPH histogram is extracted to represent the initial power signal. The wholesteps of the proposed LPH descriptor are summarized in Algorithm 1.Figure 1: Flowchart of the proposed NILM framework.6 PREPRINT - F
EBRUARY
10, 2021Moving forward, a histogram of 256 samples is derived to represent each appliance signature, which has significantlylower number of samples compared to the initial signal. Accordingly, LPH helps also in reducing the dimensionality ofthe appliance power signals. Therefore, this leads to efficaciously reducing the computation cost of our NILM system.
This stage is responsible on predicting the labels of each power consumption observations P ( t ) that belongs to a specificmicro-moment group. Consequently, the class identification step of SAD-M2 is applied in two stages using a 10-foldvalidation, i.e. the training and test. In the first one, device load usage fingerprints are learned along with their labelsgenerated based on the rule-based algorithm described previously. Accordingly, folds of the database are utilizedrandomly in each training phase while the remaining fold is employed for the test purpose.Moving forward, selecting the value of K is of utmost importance for KNN model. However, power abnormalitydetection data sets suffer from the imbalanced classes issue, in which some classes include more consumptionobservations (i.e. majority classes) than other classes (i.e. minority classes). Accordingly, a salient drawback ofconventional KNN schemes is related to the fact that if K is a fixed, user-defined value, the classification output will bebiased towards the majority groups in most of the application scenarios. Therefore, this results in a miss-classificationproblem.To avoid the issue encountered with imbalanced data set, some works have been proposed with the aim of optimizingthe value of K , such as [61, 62]. However, they are very complex to implement and can significantly increase thecomputational cost, which hinders developing real-time abnormality detection solutions. In contrast, in this paper, weintroduce a simple yet effective improvement of KNN, which can maintain a low computational cost. It is applied asexplained in Algorithm 2.Overall, the proposed improved KNN helps in improving the appliance identification performance through enhancingthe classification accuracy and F1 score results in addition to reducing the execution time as it will be demonstrated inthe next step. Therefore, this could help in developing real-time abnormality detection solutions. We concentrate in this section on presenting the outcomes of an extensive empirical evaluation conducted on fourreal-world data sets, namely UK-DALE [63], GREEND [64], PLAID [65] and WHITED [66]. They are which arevastly deployed to validate NILM and load identification frameworks in the state-of-the-art.Figure 2: Block diagram of the LPH descriptor: Example using a patch of size × (N = 8).7 PREPRINT - F
EBRUARY
10, 2021
Algorithm 1:
The principal steps of the proposed LPH descriptor deployed to derive LPH features from the M appliance power signals. Result: B LP H : The histogram of local power histograms (LPH)a. Define the array Y ( i, j ) of the appliance power signatures, where i presents the index of appliance powersequences, and j stands for the index of the samples in every sequence; while i ≤ M ( with M the total number of appliance signatures in the overall database ) doStep 1. Normalize and transform the appliance signature Y ( i, :) into 2D space (image representation), asexplained in Fig. 3. Step 2.
Calculate the LPH values of each power pattern ( u c , v c ) in each specific kernel of size S × S , bycomparing the central power pattern with its neighbor as follows: LP H n,S ( u c , v c ) = N − (cid:88) n =1 b ( j n − j c )2 n (1)where j c refers to the central power sample, j n represents the n th surrounding power neighbor in a patch ofsize S × S and N = S − . Moving forward, a binary encoding function b ( u ) is generated as: b ( u ) = (cid:26) u ≥
00 if u < (2) Step 3.
Glean the binary samples
LP H n,S ( u c , v c ) generated from every kernel and therefore transform theobtained binary data into a decimal field in order to design a new decimal array I D (as it is explained in Fig. 2). Step 4.
Perform a histogramming procedure on the obtained decimal matrix for extracting an LPH histogram H LP H ( n, S ) , which is measured from each patch. Thus, the resulted histogram is then used as a texturefeature vector to represent the initial appliance signature. Finally, after conducting the histogramming process,a description histogram H LP H ( n, S ) is produced, which has N patterns (i.e. with relation to the N binarysamples generated by N power sample neighbors of each block of data). H LP H ( n, S ) = hist ( I D ) = [ H , H , · · · , H N ] (3) Step 5.
Normalize the resulted histogram to make the value of each bin in the range [0,1]. B iLP H = Normalize( H LP H ( n, S )) = b , b , · · · , b N = H (cid:80) N m =1 H , H (cid:80) N m =1 H , · · · , H N (cid:80) N m =1 H N (4) end Figure 3: Conversion of 1D signal into 2D representation.8
PREPRINT - F
EBRUARY
10, 2021
Algorithm 2:
IKNN algorithm used to classify appliances based on their LPH signatures.
Result:
Predicting class labels of test samplesRead the training appliance histograms extracted using LPH in Algorithm 1. while j ≤ J ( with J is the number of test appliance histograms to be identified ) doStep 1: Compute the information-entropy of every appliance histogram b that is deployed to estimate itsinformation gain. Thus, it operates as the weight of appliance histograms power consumption observations toallocate priorities to each of them; E ( B ) = − n (cid:88) i =1 a i log ( a i ) (5)where B is the training ensemble, | B | as the number of training data, a i = | c i , B | / | B | and a i refers to theprobability that an random histogram in B pertains to class c i Step 2:
Define the k values of the training ensemble; Step 3:
Partition the training ensemble into m sub-groups; Step 4:
Estimate the mean value of each sub-group to derive its center;
Step 5:
Identify the sub-group that is closest to a test histogram b j via estimating the Euclidean distancebetween each test observation and the center of each sub-group as follows: d j ( b c i , b j ) = (cid:113) ( b c i ) − ( b j ) (6)where c i represents the central instance of the i th sub-group and i = 1 , , · · · , m . Step 6:
Estimate the weighted-Euclidean distance wd j between the test histogram b j and every histogram inthe closest sub-group as follows: wd j ( b i , b j ) = (cid:113) w i ( b i − b j ) (7)Therefore, this results in determining the k nearest neighbors; Step 7:
Compute the weighted class probability of the test histogram b j as follows: c ( b j ) = arg max c ∈ C k (cid:88) i w i δ ( c, c ( y i )) (8)where y , y , · · · , y k refer to the k nearest neighbors of the test histogram b j , C denotes the finite set of theappliance class labels, δ ( c, c ( y i )) = 1 if c = c ( y i ) and δ ( c, c ( y i )) = 0 otherwise. end4.1 data set description The three power consumption repositories considered in this framework are gleaned at distinct sampling rates (i.e. 1/6Hz, 30 kHz and 44 kHz) to perform a thorough evaluation study and inform the effectiveness of the proposed solutionwhen the sampling rate of the recorded appliance consumption signals varies.Under UK-Dale, power usage footprints have been gathered for a long time period ranging 2 to 4 years at both samplingfrequencies of 1/6 Hz and 16 kHz (for aggregated data). In order to assess the performance of proposed scheme, weexploit the consumption fingerprints gleaned from a specific household at 1/6 Hz, which encompasses nine appliancecategories and each category includes a large number of daily consumption signatures. Moving forward, power tracesof six different appliances collected under GREEND [64] are also considered, in which a sampling frequency of 1 Hzhas been used to record energy consumption footprints for a period of more than six months. Under PLAID, the powersignatures of 11 device groups have been recorded on the basis of a frequency resolution of 30 kHz. Moreover, loadusage footprints of the WHITED have been gleaned with reference to 11 appliance classes at a frequency resolutionof 44 kHz. The properties of each data set, their appliance categories and the number of observed appliance/days arerecapitulated in Table 1.
Aiming at evaluating the quality of the proposed appliance identification objectively, various metric are considered,including the accuracy and F1 score, normalized cross-correlation (NCC) and histogram length. The accuracy is9
PREPRINT - F
EBRUARY
10, 2021Table 1: Properties of power consumption data sets considered in this framework, i.e. appliance classes and theirnumber for both PLAID and WHITED, and appliance classes and number of observed days for both UK-DALE andGREEND.
UK-DALE
GREEND
PLAID WHITED class days class app class app1 Dishwasher 183 machine w/freezer lamp introduced to measure the ratio of successfully recognized devices in the testbed, but it is nonetheless not enough toevaluate the performance of an appliance identification system giving that alone it is not regarded as a reliable measure.This is mainly the the case of imbalanced data sets, in which the power samples are not uniformly distributed (e.g. inthis framework, both PLAID and WHITED data sets are imbalanced). To reinforce the objectivity of the evaluationstudy, F1 score is also recorded as well, which is considered as a fairly trustworthy metric in such scenarios. Explicitly,F1 score describes the specified as the harmonic average of both the precision and recall measures.
Accuracy = T P + T NT P + F P + T N + F N (9)where
T P , T N , F P and
F N depict the number of true positives, true negatives, false positives and false negatives,respectively. F score = 2 × precision × recallprecision + recall (10)where [ precision = T PT P + T F ] and [ recall = T PT P + F P ] .Additionally, normalized cross-correlation (NCC) has been deployed to measure the similarity of the raw appliancesignatures and LPH histograms derived form original power signals. NCC is also described via calculating the cosine ofthe angle θ between two power signals (or extracted characteristic histograms) x and y : N CC = Cos ( θ ) = x · y | x | | y | = (cid:80) i x i · y i (cid:112)(cid:80) i x i (cid:112)(cid:80) i y i , − ≤ N CC ≤ (11) It is of utmost importance to comprehend at the outset how LPH histograms varies from the initial appliance powersignatures. Accordingly, this subsection focuses on investigating the nature of the relation between appliance powersignatures that pertain to the same appliance class. In addition, this can aid in understanding the way LPH histogramscould improve the discrimination ability between appliances belonging to different classes and on the other flipincreasing the similarity ratio between appliance from the same group.To that end, six appliance signatures s , s , · · · , s have been considered randomly from every device category of theUK-DALE data set. Moving forward, the NCC performance has been measured between these signatures to evidentlydemonstrate why the LPH can results in a better correlation between the signatures of the same device category. Fig. 4outlines obtained NCC matrices, which are calculated between the six raw power signals (left side) and the LPH featurevectors (right side), respectively. Both raw power signals and LPH vectors are gleaned from four device groups, definedas the washing machine, fridge w/ freezer, coffee machine and radio. It can be shown from the plots in the left side ofFig. 4 that NCC rates are quiet low and vary randomly. Specifically, it is hard to identify a certain interval specifying10 PREPRINT - F
EBRUARY
10, 2021(I) Coffee machine(II) Radio(III) Fridge w/ freezer(IV) Washing machineFigure 4: Correlation arrays computed for: (a) raw appliance signatures belonging to the same appliance category and(b) their LPH histograms.the limits of the NCC rates. On the other side, when measuring the correlation between LPH vectors as indicated in theright side of Fig. 4, NCC values outperform those obtained from the raw power signals. Overall, NCC rates gleanedfrom LPH vectors are generally more than 0.97 for all appliance groups investigated in this correlation study.Fig. 5 portrays an example of six appliance signals extracted from UK-DALE database, their encoded 2D representationsand final histograms collected using the LPH descriptor. It is worth noting that each appliance signal is represented11
PREPRINT - F
EBRUARY
10, 2021by a unique image in 2D space and further by a specific histogram in the final step. It has been clearly seen that viatransforming the appliance signals into 2D space, they have been considered as image, where we can use any 2D featureextraction scheme to collected pertinent features. Moreover, through adopting the 2D representation, every powersample has been encircled by eight neighboring samples instead of only two neighbors in the 1D space. Therefore,more opportunities have been available for extracting fine-grained characteristics from each device signature in reliableway. Consequently, it could help effectively correlate between devices that pertain to the same device group, and incontrast, it increases the distance between devices corresponding to distinct categories. In addition, the LPH-basedload identification system does not relies on capturing the appliances’ states (steady or transient). This represents andadditional benefit of the proposed solution, which could recognize each electrical device without the need to collectionstate information.Moreover, even the neighborhood is temporary distant in 2D space but this gives us various possibilities to encode thepower signal. Therefore, this results in better correlation and discrimination abilities and hence a better classificationperformance of the power signals. In the contrary, if the signal is processed in the original 1D space, the possibilitiesfor encoding the signal are very limited. Thus, the correlation and discrimination abilities lose their efficacy since theclassifiers frequently make confusion in identifying appliances using the features generated in this space.
We present in this subsection the performance of the proposed appliance identification system based LPH-IKNN incomparison with different classifiers, namely conventional KNN, DT, SVM, DNN and EBT. Specifically, Table 2reports the accuracy and F1 scores collected under UK-DALE, PLAID and WHITED data set, in which a 10-fold crossvalidation is adopted.It is clearly shown that the proposed IKNN classifier based on both Euclidean distance and weighted Euclideandistance outperforms the other classification models, it provides the best results on the three data sets considered in thisframework. For instance, it achieves 98.50% and 98.49 F1 score under UK-DALE while the performance has slightlypropped for both the PLAID and WHITED data sets. Accordingly, 96.85% accuracy and 96.18% F1 score have beenobtained under PLAID and 96.5% and 96.04% have been attained under WHITED. This might be justified by the riseof the sampling frequency in both PLAID and WHITED data sets, where data have been gleaned at 30 kHz and 44kHz, respectively, in contrast to UK-DALE, in which consumption footprints have been gathered at a resolution of1 Hz. Moreover, it is important to notice that the LPH descriptor serves not only as a feature extraction descriptorbut as a dimensionality reduction approach as well. Explicitly, for each appliance signal, the resulting LPH vectorencompasses only 256 samples while the initial appliance signals include much higher samples (e.g. 22491, 57600and 30000 samples WHITED, UK-DALE and PLAID, respectively). This drives us to determine that the proposedLPH-IKNN solution operates better under low frequencies. All in all, the performance obtained with LPH-IKNN arevery promising because they are all superior than 96% for all the data sets considered in this study.On the other side, it is worth noting that the proposed LPH descriptor can be trained using simple ML algorithmswithout the need to deploy deep leaning models, which usually have a high computation complexity. In this direction,it was obvious that conventional classifiers, e.g. LDA, DT, EBT, SVM and KNN outperforms significantly the DNNclassifier, especially under UK-DALE and GREEND data sets.
The promising results of the proposed LPH obtained under the three data sets considered in this study has pushed us toinvestigate the performance of other 2D descriptors in comparison with our solution. Accordingly, in this section, weinvestigate the performance of three other feature extraction schemes.•
Local directional patterns (LDP) : After transforming the power signal into 2D space, for each pattern of thepower array, an 8-bit binary sequence is derived using LDP [67]. The latter is measured via the convolutionof small kernels from the power array (e.g. × ) with the Kirsch blocks in 8 different orientations. Fig. 6portrays an example of the Kirsch blocks used in LDP.• Local ternary pattern (LTeP) : Unlike LPH, LTeP does not encode the difference of power patterns in everykernel into 0 or 1, but encode them into other quantization values using a thresholding process [68]. Letconsider thr is the threshold parameter, s c presents the central power pattern in a patch of × , and s n standsfor the neighbor patterns, every central pattern s (cid:48) c can be encoded as follows: s (cid:48) c = (cid:40) s n > s c + thr s n > s c − thr and s n < s c + thr − s n < s c − thr (12)12 PREPRINT - F
EBRUARY
10, 2021 P o w e r ( W ) P o w e r ( W ) P o w e r ( W ) P o w e r ( W ) P o w e r ( W ) P o w e r ( W ) (b) (c)(a)(d) (e) (f) I. Example of power signals from the UK-DALE data set.
50 100 15050100150 50 100 15050100150 50 100 1505010015050 100 15050100150 50 100 15050100150 50 100 15050100150 (a) (b) (c)(d) (e) (f)
II. Image representation of LPH encoding of the power signals.
Samples A m p li t ude Samples A m p li t ude Samples A m p li t ude Samples A m p li t ude Samples A m p li t ude Samples A m p li t ude (a)(d) (b)(e) (c)(f) III. Histograms of LPH representations of the power signals.Figure 5: Example of appliance signals, their 2D LPH representations and their LPH histograms from the UK-DALEdata set: (a) television, (b) Network Attached Storage (NAS), (c) washing machine, (d) dishwasher, (e) notebook and (f)coffee machine.•
Local Transitional Pattern (LTrP) : It compares the transitions of intensity changes in small local regions (e.g.kernels of × ) in different orientations in order to binary encode the 2D representations of appliance powersignals. Specifically, LTrP generates a bit (0/1) via the comparison the central power pattern of a × patchwith only the intensities of two neighbors related to two particular directions [69].• Local binary pattern (LBP):
Is a texture descriptor that presents a low computation complexity along with acapability to capturing a good part of textural patterns of 2D representations. LBP represents micro-patterns inpower matrices by an ensemble of simple computations around each power sample [70].13
PREPRINT - F
EBRUARY
10, 2021Table 2: Performance of the proposed appliance identification system using the LPH descriptor in terms of the accuracyand F1 score with reference to various classifiers.
Classifier Classifier UK-DALE
GREEND
PLAID WHITEDparameters Accuracy F1 score
Accuracy F1 score
Accuracy F1 score Accuracy F1 scoreLDA / 93.71 93.53
49 31.15 34.16 28.36DNNs 50 hidden layers 71.69 69.82
85 77.57 84.91 87.91SVM Quadratic kernel 93.93 93.81 distance + Euclideandistance
Figure 6: Kirsch kernels utilized in the LDP approach.•
Binarized statistical image features (BSIF):
It constructs local descriptions of 2D representations via effectivelyencoding texture information and extracting histograms of local regions. Accordingly, binary codes for powerpatterns are extracted via the projection of local power regions onto a subspace, where basis-vectors werelearnt using other natural images [71].Table 3 along with Fig. 7 portray the performance of LPH in comparison with the aforementioned 2D descriptors, amongthem LBP, LDP, LTeP, BSIF and LTrP with regard to the histogram length, accuracy and F1 score. The results havebeen obtained by considering the IKNN for all descriptors (K=5). I has been evidently shown that high performance hasbeen obtained with all the descriptors under UK-DALE. Explicitly, all the descriptors have achieved an accuracy and F1score of more than 96%. On the other hand, LDP and LTeP descriptors marginally surpass the LPH under this repository.On the contrary, the performance of the other descriptors have been highly dropped under PLAID and WHITED andonly LPH maintains good accuracy and F1 score results. For instance, LPH has attained 96.85% accuracy and 96.48F1 score under PLAID and 96.55% accuracy and 96.34% F1 score under WHITED. In this context, under PLAID,LPH outperforms BSIF, LBP, LTrP, LTeP and LDP in terms of the accuracy by more than 6%, 5%, 11%, 5.5% and 7%,respectively. While in terms of the F1 score, it outperforms them by 7%, 5.5%, 15%, 7% and 10%, respectively.Conversely, the performance variation reported under the different data sets is due to frequency resolutions variation, inaddition because UK-DALE records appliance power consumption for multiple days (i.e. it collects the consumption14
PREPRINT - F
EBRUARY
10, 2021Table 3: Performance of the LPH-based descriptor vs. other 2D descriptors with reference to the histogram length,accuracy and F1 score.
Descriptor Histogram UK-DALE
GREEND
PLAID WHITEDlength Accuracy F1 score
Accuracy F1 score
Accuracy F1 score Accuracy F1 scoreLDP 56 UK − DALE GREEND PLAID WHITED80859095100
Data sets A cc u r a cy ( % ) LDPLTePLTrPLBPBSIFLPH (a) UK − DALE GREEND PLAID WHITED80859095100
Data sets A cc u r a cy ( % ) LDPLTePLTrPLBPBSIFLPH (b)
Figure 7: The performance of LPH descriptor compared to other 2D feature extraction schemes in terms of (a) accuracy,and (b) F1 score.from the same devices but for distinct days for a long period) while PLAID and WHITED data sets observe differentdevices from distinct manufacturers (brands) and which are belonging to the same device category.Moving forward, we have evaluated the computation cost of the proposed appliance identification scheme based ondifferent 2D descriptors in order to demonstrates its applicability in real-time scenarios. Accordingly, the computationtime for the training and test phases of our approach have been computed using MATLAB 9.4. The computationalcosts are computed on a laptop having a Core i7-85500 with 32 GB RAM and 1.97 GHz. Table 4 depicts the obtainedcomputational times (in sec) with regard to various 2D descriptors under the three data sets adopted in this framework.Accordingly, it has been clearly seen that the appliance identification based LPH achieves the lowest computationaltime in comparison with the other descriptors for both the training and test stages under the three data sets. Moreover,the test time of LPH based solution under PLAID and WHITED is less than 1 sec, which can proves that it is possibleto implement it for real-time applications since most of the transmitter can transmit data with a sampling rate of morethan 1 sec. On the other flip, the test time of the LPH based solution has increased under UK-ALE to more than 2 secbecause in this case long daily consumption signatures are analyzed. In contrast to PLAID and WHITED, where shortappliance fingerprints from are considered.
Table 5 recapitulates the results of various existing load identification frameworks under REDD data set, in comparisonwith the proposed LPH-IKNN solution and with reference to different parameters, among them the description oflearning model, its type, number of the device categories and accuracy performance. It has been clearly seen that theLPH-IKNN framework outperforms all other architectures considered in this study. Moreover, LPH-IKNN has a lowcomputational cost, which can make it a candidate for real-time applications. On the other side, it is worth noting thatthe proposed method is evaluated under three distinct power repositories with different sampling rates, in which itpresents promising results. In contrast, each of the other methods is only validated under one data set, therefore, this15
PREPRINT - F
EBRUARY
10, 2021Table 4: Computational time (in sec) of the proposed appliance identification based on different 2D descriptors.
Time (in sec)2D descriptors UK-DALE
GREEND
PLAID WHITEDtraining test training test training test training testLDP 25.18 3.71
Table 5: Performance of the proposed LPH-IKNN based load identification system vs. existing solutions with referenceto different criteria.
Framework Approach Learning increases the credibility of the our study and proves that it could be deployed under different scenarios without caringabout the sampling rate.All in all, it is of significant importance to mention that the proposed LPH-IKNN has been validated using four differentdata sets (i.e. UK-DALE, GREEND, PLAID and UK-DALE) including different kinds of power signatures recorded(i) with distinct frequency resolutions, and (ii) under different scenarios. For instance, under both UK-DALE andGREEND, we collect the power consumption signatures that vary through the time for a set of appliances (i.e. eachdaily consumption trace represents a power signature); while under both PLAID and WHITED, for each appliancecategory, the power traces are gleaned from different manufacturers. In this regard, validating our solution under thesedata sets using different scenarios, has helped in (i) showing its high performance although the frequency resolutionhas been changed, and (ii) proving its capability to be implemented in real-application scenarios since it can identifyappliance-level data even if they are from different manufacturers and although the power signatures change from a dayto another.
In this paper, a novel method for performing accurate appliance identification and hence improving the performance ofthe NILM systems has been presented. The applicability of a local 2D descriptor, namely LPH-IKNN, to recognizeelectrical devices has been successfully validated. Consequently, other types of 2D descriptors can be investigated inorder to further improve the identification accuracy, such as local texture descriptors, color histograms, moment-baseddescriptors and scale-invariant descriptors. This line of research is full of challenges and plenty of opportunities areavailable. Moving forward, in addition to the high performance reached, the LPH-IKNN based appliance identificationsscheme has shown a low computational cost because of the use of a fast 2D descriptor along with the IKNN, which usesa smart strategy to reduce the training and test times. Furthermore, LPH-IKNN acts also as a dimensionality reduction,in which very short histograms have been derived to represent appliance fingerprints.On the other hand, although LPH-IKNN has shown very promising performance, it still has some drawbacks amongthem is its reliance on a supervised learning process. Explicitly, this could limit its application in some scenarios, where16
PREPRINT - F
EBRUARY
10, 2021it might be difficult to collect data to train the proposed model. To that end, it is part of our next future work to changethe learning process by building an improved version of this LPH-IKNN using an unsupervised learning approach.Moreover, another option is by adding a transfer learning module to eliminate the need to collect new data for trainingour system if the sampling frequency of collected data is changed. Moreover, IKNN classifier could be replaced by anyother improved algorithm that enables an automatic selection of the k value to simplify the use of LPH-IKNN in realapplication scenarios. In this context, the GBKNN classifier discussed in Section 2.2.1 seems to be a good alternativethat could be investigated in our future work.On the other hand, due to the size of appliance identification based data sets is not very large, it will be of significantimportance to investigate the use of other feature extraction methods in our future work, which are very convenientfor small data sets, e.g. rough set based techniques [78, 79]. The latter helps also in attribute reduction and featureselection and hence it could further reduce the computational cost of the appliance identification task to support real-timeapplications. Finally, it will also be part of our future work to focus on developing a powerful recommender system,which can use the output of the LPH-IKNN based NILM system to detect abnormal power consumption in buildingsbefore triggering tailored recommendations to help end-users in reducing wasted energy. Acknowledgements
This paper was made possible by National Priorities Research Program (NPRP) grant No. 10-0130-170288 from theQatar National Research Fund (a member of Qatar Foundation). The statements made herein are solely the responsibilityof the authors.
References [1] E. Elattar, N. Sabiha, M. Alsharef, M. Metwaly, A. Abd-Elhady, I. Taha, Short term electric load forecasting usinghybrid algorithm for smart cities, Applied Intelligence (2020) 1–21 doi:10.1007/s10489-020-01728-x .[2] C. Sardianos, I. Varlamis, C. Chronis, G. Dimitrakopoulos, A. Alsalemi, Y. Himeur, F. Bensaali, A. Amira, Theemergence of explainability of intelligent systems: Delivering explainable and personalized recommendations forenergy efficiency, International Journal of Intelligent Systems 36 (2021) 656–680.[3] C. Sardianos, I. Varlamis, G. Dimitrakopoulos, D. Anagnostopoulos, A. Alsalemi, F. Bensaali, Y. Himeur,A. Amira, Rehab-c: Recommendations for energy habits change, Future Generation Computer Systems 112 (2020)394 – 407.[4] A. Alsalemi, Y. Himeur, F. Bensaali, A. Amira, C. Sardianos, C. Chronis, I. Varlamis, G. Dimitrakopoulos, Amicro-moment system for domestic energy efficiency analysis, IEEE Systems Journal.[5] Y. Himeur, A. Alsalemi, F. Bensaali, A. Amira, Robust event-based non-intrusive appliance recognition usingmulti-scale wavelet packet tree and ensemble bagging tree, Applied Energy 267 (2020) 114877.[6] L. Pereira, N. Nunes, An empirical exploration of performance metrics for event detection algorithms in non-intrusive load monitoring, Sustainable Cities and Society 62 (2020) 102399.[7] H. Liu, C. Yu, H. Wu, C. Chen, Z. Wang, An improved non-intrusive load disaggregation algorithm and itsapplication, Sustainable Cities and Society 53 (2020) 101918.[8] A. Alsalemi, Y. Himeur, F. Bensaali, A. Amira, C. Sardianos, I. Varlamis, G. Dimitrakopoulos, Achieving domesticenergy efficiency using micro-moments and intelligent recommendations, IEEE Access 8 (2020) 15047–15055.[9] S. Welikala, C. Dinesh, M. P. B. Ekanayake, R. I. Godaliyadda, J. Ekanayake, Incorporating appliance usagepatterns for non-intrusive load monitoring and load forecasting, IEEE Transactions on Smart Grid 10 (1) (2019)448–461.[10] K. He, D. Jakovetic, B. Zhao, V. Stankovic, L. Stankovic, S. Cheng, A generic optimisation-based approach forimproving non-intrusive load monitoring, IEEE Transactions on Smart Grid 10 (6) (2019) 6472–6480.[11] S. W. Park, L. B. Baker, P. D. Franzon, Appliance identification algorithm for a non-intrusive home energy monitorusing cogent confabulation, IEEE Transactions on Smart Grid 10 (1) (2019) 714–721.[12] M. Ma, W. Lin, J. Zhang, P. Wang, Y. Zhou, X. Liang, Toward energy-awareness smart building: Discover thefingerprint of your electrical appliances, IEEE Transactions on Industrial Informatics 14 (4) (2018) 1458–1468.[13] S. Makonin, F. Popowich, I. V. Baji´c, B. Gill, L. Bartram, Exploiting hmm sparsity to perform online real-timenonintrusive load monitoring, IEEE Transactions on Smart Grid 7 (6) (2016) 2575–2585.17
PREPRINT - F
EBRUARY
10, 2021[14] J. D. S. Guedes, D. D. Ferreira, B. H. G. Barbosa, C. A. Duque, A. S. Cerqueira, Non-intrusive appliance loadidentification based on higher-order statistics, IEEE Latin America Transactions 13 (2015) 3343–3349.[15] T. Y. Ji, L. Liu, T. S. Wang, W. B. Lin, M. S. Li, Q. H. Wu, Non-intrusive load monitoring using additive factorialapproximate maximum a posteriori based on iterative fuzzy c-Means, IEEE Transactions on Smart Grid 10 (6)(2019) 6667–6677.[16] K. He, L. Stankovic, J. Liao, V. Stankovic, Non-intrusive load disaggregation using graph signal processing, IEEETransactions on Smart Grid 9 (3) (2018) 1739–1747.[17] D. Li, S. Dick, Residential household non-intrusive load monitoring via graph-based multi-label semi-supervisedlearning, IEEE Transactions on Smart Grid 10 (4) (2019) 4615–4627.[18] B. Zhao, K. He, L. Stankovic, V. Stankovic, Improving event-based non-intrusive load monitoring using graphsignal processing, IEEE Access PP (2018) 1–1.[19] J. Z. Kolter, S. Batra, A. Y. Ng, Energy disaggregation via discriminative sparse coding, in: Proceedings of the23rd International Conference on Neural Information Processing Systems - Volume 1, NIPS’10, Curran AssociatesInc., USA, 2010, pp. 1153–1161.[20] S. Singh, A. Majumdar, Analysis co-sparse coding for energy disaggregation, IEEE Transactions on Smart Grid10 (1) (2019) 462–470.[21] S. Singh, A. Majumdar, Deep sparse coding for non-intrusive load monitoring, IEEE Transactions on Smart Grid9 (5) (2018) 4669–4678.[22] A. Rahimpour, H. Qi, D. Fugate, T. Kuruganti, Non-intrusive energy disaggregation using non-negative matrixfactorization with sum-to-k constraint, IEEE Transactions on Power Systems 32 (6) (2017) 4430–4441.[23] M. Hammad, S. Zhang, K. Wang, A novel two-dimensional ecg feature extraction and classification algorithmbased on convolution neural network for human authentication, Future Generation Computer Systems 101 (2019)180 – 196.[24] L. Du, D. He, R. G. Harley, T. G. Habetler, Electric load classification by binary voltage-current trajectorymapping, IEEE Transactions on Smart Grid 7 (1) (2016) 358–365.[25] J. Gao, E. C. Kara, S. Giri, M. Bergés, A feasibility study of automated plug-load identification from high-frequency measurements, in: 2015 IEEE Global Conference on Signal and Information Processing (GlobalSIP),2015, pp. 220–224.[26] Y. Liu, X. Wang, W. You, Non-intrusive load monitoring by voltage-current trajectory enabled transfer learning,IEEE Transactions on Smart Grid 10 (5) (2019) 5609–5619.[27] L. D. Baets, C. Develder, T. Dhaene, D. Deschrijver, Detection of unidentified appliances in non-intrusive loadmonitoring using siamese neural networks, International Journal of Electrical Power & Energy Systems 104 (2019)645 – 653.[28] R. G. Junker, A. G. Azar, R. A. Lopes, K. B. Lindberg, G. Reynders, R. Relan, H. Madsen, Characterizing theenergy flexibility of buildings and districts, Applied Energy 225 (2018) 175–182.[29] Y. Himeur, A. Alsalemi, F. Bensaali, A. Amira, Improving in-home appliance identification using fuzzy-neighbors-preserving analysis based qr-decomposition, in: International Congress on Information and CommunicationTechnology, Springer, 2020, pp. 303–311.[30] X. Wang, S.-H. Ahn, Real-time prediction and anomaly detection of electrical load in a residential community,Applied Energy 259 (2020) 114145.[31] Y. Himeur, A. Alsalemi, F. Bensaali, A. Amira, Effective non-intrusive load monitoring of buildings based on anovel multi-descriptor fusion with dimensionality reduction, Applied Energy 279 (2020) 115872.[32] Y. Himeur, A. Alsalemi, F. Bensaali, A. Amira, Efficient multi-descriptor fusion for non-intrusive appliancerecognition, in: 2020 IEEE International Symposium on Circuits and Systems (ISCAS), IEEE, 2020, pp. 1–5.[33] X. Shi, H. Ming, S. Shakkottai, L. Xie, J. Yao, Nonintrusive load monitoring in residential households withlow-resolution data, Applied Energy 252 (2019) 113283.[34] Y. Himeur, A. Elsalemi, F. Bensaali, A. Amira, Smart power consumption abnormality detection in buildingsusing micro-moments and improved k-nearest neighbors, Intenational Journal of Intelligent Systems (2020) 1–25.[35] J. Mulongo, M. Atemkeng, T. Ansah-Narh, R. Rockefeller, G. M. Nguegnang, M. A. Garuti, Anomaly detectionin power generation plants using machine learning and neural networks, Applied Artificial Intelligence 34 (1)(2020) 64–79. 18
PREPRINT - F
EBRUARY
10, 2021[36] Y. Himeur, A. Alsalemi, F. Bensaali, A. Amira, A novel approach for detecting anomalous energy consumptionbased on micro-moments and deep neural networks, Cognitive Computation 12 (6) (2020) 1381–1401.[37] S. Mehta, X. Shen, J. Gou, D. Niu, A new nearest centroid neighbor classifier based on k local means usingharmonic mean distance, Information 9 (9) (2018) 234.[38] H. A. Abu Alfeilat, A. B. Hassanat, O. Lasassmeh, A. S. Tarawneh, M. B. Alhasanat, H. S. Eyal Salman, V. S.Prasath, Effects of distance measure choice on k-nearest neighbor classifier performance: A review, Big data 7 (4)(2019) 221–248.[39] J. Gou, H. Ma, W. Ou, S. Zeng, Y. Rao, H. Yang, A generalized mean distance-based k-nearest neighbor classifier,Expert Systems with Applications 115 (2019) 356–372.[40] J. Gou, W. Qiu, Z. Yi, Y. Xu, Q. Mao, Y. Zhan, A local mean representation-based k-nearest neighbor classifier,ACM Transactions on Intelligent Systems and Technology (TIST) 10 (3) (2019) 1–25.[41] J. Gou, W. Qiu, Z. Yi, X. Shen, Y. Zhan, W. Ou, Locality constrained representation-based k-nearest neighborclassification, Knowledge-Based Systems 167 (2019) 38–52.[42] S. Xia, Y. Liu, X. Ding, G. Wang, H. Yu, Y. Luo, Granular ball computing classifiers for efficient, scalable androbust learning, Information Sciences 483 (2019) 136–152.[43] K. T. Chui, K. Tsang, S. Chung, L. Yeung, Appliance signature identification solution using k-means clustering,in: IECON 2013-39th Annual Conference of the IEEE Industrial Electronics Society, IEEE, 2013, pp. 8420–8425.[44] J. Henriques, F. Caldeira, T. Cruz, P. Simões, Combining k-means and xgboost models for anomaly detectionusing log datasets, Electronics 9 (7) (2020) 1164.[45] T. Saba, Recent advancement in cancer detection using machine learning: Systematic survey of decades, compar-isons and challenges, Journal of Infection and Public Health 13 (9) (2020) 1274–1289.[46] A. Alsayat, H. El-Sayed, Social media analysis using optimized k-means clustering, in: 2016 IEEE 14th Interna-tional Conference on Software Engineering Research, Management and Applications (SERA), IEEE, 2016, pp.61–66.[47] S.-S. Yu, S.-W. Chu, C.-M. Wang, Y.-K. Chan, T.-C. Chang, Two improved k-means algorithms, Applied SoftComputing 68 (2018) 747–755.[48] G. Zhang, C. Zhang, H. Zhang, Improved k-means algorithm based on density canopy, Knowledge-based systems145 (2018) 289–297.[49] W. Lu, Improved k-means clustering algorithm for big data mining under hadoop parallel framework, Journal ofGrid Computing (2019) 1–12.[50] J. J. Whang, I. S. Dhillon, D. F. Gleich, Non-exhaustive, overlapping k-means, in: Proceedings of the 2015 SIAMInternational Conference on Data Mining, SIAM, 2015, pp. 936–944.[51] S. Khanmohammadi, N. Adibeig, S. Shanehbandy, An improved overlapping k-means clustering method formedical applications, Expert Systems with Applications 67 (2017) 12–18.[52] H. Tao, Detecting smoky vehicles from traffic surveillance videos based on dynamic features, Applied Intelligence50 (2019) 1–16. doi:10.1007/s10489-019-01589-z .[53] A. Kumar, S. K. Singh, S. Saxena, A. K. Singh, S. Shrivastava, K. Lakshmanan, N. Kumar, R. K. Singh, Comhisp:A novel feature extractor for histopathological image classification based on fuzzy svm with within-class relativedensity, IEEE Transactions on Fuzzy Systems (2020) 1–1.[54] D. Gong, Z. Li, W. Huang, X. Li, D. Tao, Heterogeneous face recognition: A common encoding featurediscriminant approach, IEEE Transactions on Image Processing 26 (5) (2017) 2079–2089.[55] D. Valdes-Ramirez, M. A. Medina-Pérez, R. Monroy, O. Loyola-González, J. Rodríguez, A. Morales, F. Herrera,A review of fingerprint feature representations and their applications for latent fingerprint identification: Trendsand evaluation, IEEE Access 7 (2019) 48484–48499.[56] N. Batra, R. Kukunuri, A. Pandey, R. Malakar, R. Kumar, O. Krystalakos, M. Zhong, P. Meira, O. Parson, Towardsreproducible state-of-the-art energy disaggregation, in: Proceedings of the 6th ACM International Conference onSystems for Energy-Efficient Buildings, Cities, and Transportation, BuildSys ’19, ACM, New York, NY, USA,2019, pp. 193–202.[57] N. Batra, R. Kukunuri, A. Pandey, R. Malakar, R. Kumar, O. Krystalakos, M. Zhong, P. Meira, O. Parson, Ademonstration of reproducible state-of-the-art energy disaggregation using nilmtk, in: Proceedings of the 6thACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation, BuildSys’19, ACM, New York, NY, USA, 2019, pp. 358–359.19
PREPRINT - F
EBRUARY
10, 2021[58] T. Lu, Z. Xu, B. Huang, An event-based nonintrusive load monitoring approach: Using the simplified viterbialgorithm, IEEE Pervasive Computing 16 (4) (2017) 54–61.[59] N. Batra, R. Kukunuri, A. Pandey, R. Malakar, R. Kumar, O. Krystalakos, M. Zhong, P. Meira, O. Parson, Ademonstration of reproducible state-of-the-art energy disaggregation using nilmtk, in: Proceedings of the 6thACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation, BuildSys’19, Association for Computing Machinery, New York, NY, USA, 2019, pp. 358–359.[60] N. Batra, J. Kelly, O. Parson, H. Dutta, W. Knottenbelt, A. Rogers, A. Singh, M. Srivastava, Nilmtk: An opensource toolkit for non-intrusive load monitoring, 2014.[61] W. Liu, S. Chawla, Class confidence weighted knn algorithms for imbalanced data sets, Vol. 6635, 2011, pp.345–356. doi:10.1007/978-3-642-20847-8_29doi:10.1007/978-3-642-20847-8_29
