Conference Paper

Automatic identification of electrical appliances using smart plugs

Authors:
To read the full-text of this research, you can request a copy directly from the authors.

Abstract

We report on the evaluation of signal processing and classification algorithms to automatically recognize electric appliances. The system is based on low-cost smart-plugs measuring periodically the electricity values and producing time series of measurements that are specific to the appliance consumptions. In a similar way as for biometric applications, such electric signatures can be used to identify the type of appliance in use. In this paper, we propose to use dynamic features based on time derivative and time second derivative features and we compare different classification algorithms including K-Nearest Neighbor and Gaussian Mixture Models. We use the recently recorded electric signature database ACS-Fl and its intersession protocol to evaluate our algorithm propositions. The best combination of features and classifiers shows 93.6% accuracy.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... (9) The development of communication and interoperability standards for appliances and devices. (10) Overcoming barriers and impediments to the adoption of smart grid technologies, methods, and services [1]. ...
... Researchers have paid little attention to the development of an ecosystem to support the DR. Most published work uses cloud computing to manually control load [9][10][11][12]. However, cloud computing has issues with latency, privacy, efficiency, power consumption, and security while processing sensitive data. ...
... Intrusive load monitoring is the term used to describe this technique. The goal of [10]'s research is to identify home appliances by using the Gaussian Mixture Model (GMM) and K-Nearest Neighbors classifiers (k-NN). The researchers made use of a cheap commercial smart plug (the Plugwise system). ...
Article
Full-text available
The demand response program is an important feature of the smart grid. It attempts to reduce peak demand, improve the smart grid efficiency, and ensure system reliability. Implementing demand-response programs in residential and commercial buildings requires the use of smart meters and smart plugs. In this paper, we propose an architecture for a home-energy-management system based on the fog-computing paradigm, an Internet-of-Things-enabled smart plug, and a smart meter. The smart plug measures in real-time the root mean square (RMS) value of the current, frequency, power factor, active power, and reactive power. These readings are subsequently transmitted to the smart meter through the Zigbee network. Tiny machine learning algorithms are used at the smart meter to identify appliances automatically. The smart meter and smart plug were prototyped by using Raspberry Pi and Arduino, respectively. The smart plug’s accuracy was quantified by comparing it to laboratory measurements. To assess the speed and precision of the small machine learning algorithm, a publicly accessible dataset was utilized. The obtained results indicate that the accuracy of both the smart meter and the smart plug exceeds 97% and 99%, respectively. The execution of the trained decision tree and support vector machine algorithms was verified on the Raspberry Pi 3 Model B Rev 1.2, operating at a clock speed of 600 MHz. The measured latency for the decision tree classifier’s inference was 1.59 microseconds. In a practical situation, the time-of-use-based demand-response program can reduce the power cost by about 30%.
... ALM can be executed using both intrusive and non-intrusive techniques. However, due to costly sub-metering installations, difficult upgrades settings, and customer privacy issues from one side, which hinder the former, as well as the integration of enhanced electrical and computer engineering tools from other side, which facilitates the latter; the non-intrusive approach is favored from both academic and industrial perspectives [9][10][11]. Non-intrusive Load Monitoring (NILM) technology is the practice of disaggregating household total electrical load measured at a single point into individual appliances signals, using the combination of an electrical acquisition system and signal processing algorithms [11,12]. ...
... 1. Customer perception: the most preferred classification considering the consumers' preference is to categorize appliances based on their role in the house, which refers to lighting appliance(s), kitchen/ domestic appliance(s), and heating/cooling appliance(s) [10,33,35]. 2. Operational state: this class reflects the number of appliance's steady state operations, which is preferable from the NILM perspective of load identification. ...
... This category consists of two-state appliances, multi-state appliances, continuously variable appliances, and permanent appliances. It is noted that active power properties of an appliance are utilized in order to define this class [10,14]. 3. Waveform features: the creation of this category is based on the characteristics of voltage/current waveforms which comprises resistive appliances, electronically fed appliances, electronic power control appliances, pump-operated appliances, and motor-driven appliances [36]. ...
Article
The enhanced utilization of Appliance Load Monitoring (ALM) in customer sites enabled by Home Energy Management Systems (HEMS) technologies, offers customized services and enables demand side flexibility in power systems. The significant integration of advanced electrical and computer engineering tools makes the nonintrusive approach of ALM a technically feasible solution to improve demand side energy utilization in the context of HEMS. This paper presents a comprehensive study conducted to reveal significant inevitabilities of a well organized Non-intrusive Load Monitoring (NILM) that aids Smart Home (SH) idea to be implemented. In fact, the viewpoint of this study is to discuss critical issues related to NILM prerequisite necessities, hindered the practical implication of this approach despite improvements during over 30 years. Accordingly, this work presents actual analyses in order to elucidate some arguments using state of the art procedures and results of a semi-synthetic data generator tool. In addition, with the aim of an achievable NILM, we analyze NILM applications from the stakeholders’ perspective to assist the choice of employed techniques. Consequently, by investigating crucial intentions of an effective NILM considering current standstill and future progression, the authors propose the Advanced NILM (ANILM) concept and describe its properties to provide an enhanced energy usage system in demand side. In order to meet its ambition, the paper uses a realistic point of view to pinpoint major obstacles toward NILM and elaborate various factors that will make it effectively feasible.
... Interestingly, the raw observations were simply normalized and used directly as features. A continuation of this work was presented by Ridi et al. in [11] where the signature database ACS-F1 was used, increasing the number of categories to 10 and showing a tuned up system performance of 93.8%. In the next Section this database will be presented. ...
... With this protocol, all the testing signatures come from appliances already seen in the training phase. At the time of writing this article, the best performances on the intersession protocol are reported in [11]. In this work, two classifiers are compared, namely k-NN and GMM systems showing respectively 88% and 93.8% correct category identification. ...
... In our procedure and as proposed in [10,11], we use as baseline coefficients the raw observation O as part of the features. We analyze here the impact of including information about the dynamics of the signal through the computation of the so-called delta and delta-delta or acceleration coefficients. ...
Conference Paper
We assess the feasibility of unseen appliance recognition through the analysis of their electrical signatures recorded using low-cost smart plugs. By unseen, we stress that our approach focuses on the identification of appliances that are of different brands or models than the one in training phase. We follow a strictly defined protocol in order to provide comparable results to the scientific community. We first evaluate the drop of performance when going from seen to unseen appliances. We then analyze the results of different machine learning algorithms, as the k-Nearest Neighbor (k-NN) and Gaussian Mixture Models (GMMs). Several tunings allow us to achieve 74% correct accuracy using GMMs which is our current best system.
... 3. We then use the Expectation-Maximisation (EM) iterative algorithm for computing the distribution of the Gaussian Mixture Models. A GMM is a parametric probability density function computed as a weighted sum of Gaussian component densities [107]: ...
... The spectrum of 0-3500 lux was evenly distributed into 35 zones of 100 lux each. [107]. In our case, after testing several distance metrics, the Euclidean distance has been chosen. ...
... Model framework (Hidden Markov Models) and definitionsIn specific, we use HMMs which are state-based stochastic approaches. Historically they have been applied in many domains such as speech recognition, handwriting recognition, biometric authentication and financial time series analysis[104; 105], but in the later years they have been increasingly used in building control, principally in the field of occupant pattern detection[90; 106] and appliance identification[107]. HMMs offer a robust probabilistic framework to model time series, where the system being modelled is assumed to be a Markov process with hidden states (unlike the standard Markov models, where states are directly visible). ...
Article
People in developed countries spend today most of their time inside buildings as part of the modern way of life. As a result, the building sector accounts for almost 40% of the total energy consumption and a big part of the energy bill goes to maintain the visual and thermal comfort of their occupants. At the same time, awareness is being raised during the last decades about the greenhouse gas emissions and the possibly irreversible effects of global warming; both linked to excessive use of non-renewable primary energy sources which still power most of the world, including our buildings. Thus, moderating the energy consumed in them is a top priority. However, this does not imply a horizontal cut in energy consumption that would result in a drop of user comfort. Instead, we suggest that improving energy efficiency in buildings while maintaining or even improving the user comfort is the optimal solution. It is indeed the core of this thesis that there is a great energy saving potential in refining the control of building systems such as electric lighting, heating, cooling and ventilation, which more than often consume a lot of energy without delivering the analogous amount of visual and thermal comfort. In this direction, this thesis proposes the development of a novel predictive control algorithm for the control of electrochromic glazing using a low cost sky scanner using a simple web camera. The developed algorithm demonstrated an average prediction accuracy of 92% and integrates and controls the blinds and electric lighting to maximise visual comfort taking into account outdoor and indoor conditions, presence and user actions. Measurements and extensive simulations showed that the elaborated algorithm improves thermal and visual comfort when compared to standard glazing coupled with blinds and exhibits acceptable levels of energy consumption for space heating and electric lighting. In the same subject of improving building control, a novel approach for controlling building systems by using state-based stochastic data-driven models to identify "season" is defined and developed. We reason that the season variable is unique to every building and it depends on weather conditions, user behaviour and building construction. The developed models identified "season" with an accuracy that ranged from 69 to 91% and it was shown through simulations that a controller based on Hidden Markov Models can reduce energy demand for heating and improve the thermal comfort of occupants in different building construction types. Finally, the use of Hidden Markov Models was further explored in this thesis by suggesting a novel model for the estimation of occupants' visual comfort in buildings. The proposed model is based on horizontal workplane illuminance measurements using ceiling-mounted sensors as well on vertical illuminance monitoring at the observer's eyes plane (pupillary illuminance) by means of wearable portable sensors. We argue that the proposed model improves greatly over the various existing discomfort glare indices and metrics and it is also convincingly demonstrated that it can be seamlessly integrated and used in building automation systems based on fuzzy logic.
... These data mostly correspond to the active power demanded by the artifacts [4], which are processed in a Matlab® environment applying, for example, the Multi-Layer Back-Propagation Neuronal Network (MLBPNN) algorithm that allows detecting the real power profile. Consumed by each electrical device [5], so then the concept of digital footprint of device consumption [6] is proposed. On the other hand, NILM systems have the advantage of not interfering with the home circuit with expensive monitoring devices [7]. ...
... The result of this test is shown in Table 2, where the precision of this algorithm with the previous cluster technique improves the results for the kettle going from 86.6% to 95.1%, in the heater it increases from 91.5% to 95.7% and in the electric oven it changes from 94.4% to 98.1%. These results are better than those reported by [6]. ...
Chapter
Full-text available
For future smart cities, smart homes will be required. The key elements are the smart use of energy and smart communication systems that are connected to homes. Along with this, the devices inside the house will need to be monitored and managed efficiently. One of the current proposals is the use of Home Energy Management Systems (HEMS) allowing to solve problems associated with efficient management, the economy of electrical energy, and failures/alarms regarding the operation and safety of appliances. This work proposes a model for the recognition of patterns of energy consumption in household appliances, based on the capture of electrical parameters through Smart Socket, using an intrusive method in the electric charge. The data acquisition system corresponds to an IoT platform that uses automatic meter reading elements, which, connected via Wi-Fi, send data to a cloud service. The results obtained allow a characterization of household appliance consumption profiles, with high levels of reliability and under multiple operating states. Because of the foregoing, the detection, monitoring, and control of household appliances connected to the electrical network allow the reduction of both household billing and CO2 emissions.
... As regards the NILM approaches based on machine learning, they are mainly based on the extraction of features, which will be used either to train a machine learning algorithm (e.g. Support Vector Machines (SVMs) [33], Artificial Neural Networks (ANNs) [34], Decision Trees (DTs) [35], Hidden Markov Models (HMMs) and their variants [36]- [38] and K-Nearest Neighbours (KNNs) [39]) or to define a set of rules or thresholds [40]- [43]. However, latest research in deep learning and big data has led to a significant increase in data-driven approaches using large scale datasets (e.g. ...
... electronic devices like laptops, personal computers, LCD screens, etc.). An example of each of the four device categories is illustrated in Fig. 3. To efficiently disaggregate the aggregated signal to appliance level the appliance signatures of one-dimensional power signal need to be transferred to a sequence of multidimensional feature vectors uniquely characterizing the appliance behaviour (except for proposed approaches performing disaggregation directly on the raw data [36]- [39]). The features extracted from the aggregated energy measurements strongly depend on the sampling frequency. ...
Article
The rapid development of technology in the elec- trical energy sector within the last 20 years has led to grow- ing electric power needs through the increased number of electrical appliances and automation of tasks. In parallel the global climate protection goals, energy conservation and efficient energy management arise interest for reduction of the overall energy consumption. These requirements have led to the recent adoption of smart-meters and smart-grids, as well as to the rise of Load Monitoring (LM) using energy disaggregation, also referred to as Non-Intrusive Load Monitoring (NILM), which enables appliance-specific energy monitoring by only observing the aggregated energy consumption of a household. The real-time information on appliance level can be used to get deeper insights in the origin of energy consumption and to make optimization, strategic load scheduling and demand management feasible. The three main contributions are as follows: First, a generalized up-to-date review of NILM approaches including a high-level taxonomy of NILM methodologies is provided. Second, previously published results are grouped based on the experimental setup which allows direct comparison. Third, the article is accompanied by a software implementation of the described NILM approaches.
... Generalmente los datos capturados del consumo de los artefactos se envían por internet y se almacenan en una plataforma web disponible en la nube. Estos datos en su mayoría corresponden a la potencia activa demandada por los artefactos [4], los cuales se procesan en ambiente Matlab® aplicando por ejemplo el algoritmo de Multi Layer Back-Propagation Neuronal Network (MLBPNN) que permite detectar el perfil de potencia real consumida por cada dispositivo eléctrico [5], así entonces se propone el concepto de huella digital del consumo del artefacto [6]. Por otra parte, los sistemas NILM tienen la ventaja de no intervenir el circuito del hogar con dispositivos costosos de monitoreo [7]. ...
... El resultado de esta prueba se muestra en Tabla II, en donde se logra apreciar que la precisión de este algoritmo con la técnica previa de clúster mejora los resultados para el hervidor pasando de un 86.6% a un 95.1%, en el calefactor se aumenta del 91.5% a un 95.7% y en el horno eléctrico cambia de 94.4% a 98.1%. Estos resultados son mejores que los reportados por Ridi et al. en [6]. ...
Article
Full-text available
For an efficient energy management by residential users, monitoring and control of connected household appliances is required. If the consumption pattern of each of these devices is identified, then the management will be more efficient, reducing both the billing and the CO2 emissions. This paper proposes a model for the recognition of energy consumption patterns in household appliances, based on the capture of electrical parameters in connected appliances, through Smart Socket with an Intrusive Load Monitoring approach. The data acquisition system corresponds to an-Internet of Things (IoT) platform that uses Automatic Meter Reading devices, connected to a IoT-gateway via Wi-Fi to send data to an application on the web. For the recognition of the patterns, machine learning techniques are used. Accuracy results on pattern identification are obtained about 91% after applying a backpropagation method in an Artificial Neural Network in time basis. Through this work, the prediction of consumer categories in household appliances, with high levels of reliability and under multiple operating states, is reached. These results enhance the efficient management of energy in a Smart Home and Smart Cities environment.
... It allows to detect electrical noise to classify electrical device in home by exploiting the electrical noise as additional parameters. Whereas, time-series measurements, which represent electrical signatures of different electrical devices, are used in [12] for their identification. ...
... Two main approaches, to face with automatic identification of electrical devices, emerged from the above-related works. One is based on the employment of additional monitoring devices either distributed [8,12] or centralized [9][10][11] which results expensive in terms of money for their installation and hardly scalable; the second one, that does not exploit any additional devices, is centered on energy measurements [6], but it lacks in the categorization and formalization of the adopted features. Our work stands out from the previous ones because (i) a set of features that characterize electrical devices are proposed and formalized (ii) a model, based on their combination, is used to identify and recognize devices when they are plugged into the circuit without additional monitoring devices, (iii) better performances in term of higher accuracy are reached. ...
... It allows for detecting electrical noise to classify electrical devices in homes by exploiting the electrical noise as an additional parameter. However, time series measurements, which represent electrical signatures of different electrical devices, are used in [22] for their identification. ...
... Two main approaches, to face with automatic identification of electrical devices, emerged from the above related works. One is based on the employment of additional monitoring devices either distributed [15,22] or centralized [19][20][21] which results expensive in terms of money for their installation and hardly scalable; the second one that does not exploit any additional devices, is centered on energy measurements [13], but it lacks in the categorization and formalization of the adopted features. Some of those works used aggregated traces (AT) of multiple devices and attempt to disaggregate energy usage, whereas other works, as in our case, used directly disaggregated traces (DT). ...
Article
Full-text available
Smart Homes (SHs) represent the human side of a Smart Grid (SG). Data mining and analysis of energy data of electrical devices in SHs, e.g., for the dynamic load management, is of fundamental importance for the decision-making process of energy management both from the consumer perspective by saving money and also in terms of energy redistribution and reduction of the carbon dioxide emission, by knowing how the energy demand of a building is composed in the SG. Advanced monitoring and control mechanisms are necessary to deal with the identification of appliances. In this paper, a model for their automatic identification is proposed. It is based on a set of 19 features that are extracted by analyzing energy consumption, time usage and location from a set of device profiles. Then, machine learning approaches are employed by experimenting different classifiers based on such model for the identification of appliances and, finally, an analysis on the feature importance is provided.
... While the energy management feature helps in monitoring the energy consumption, a device scheduling and control method surely gives authority to the users (consumers/prosumers) to manage their energy consumption. There are various possible scenarios for the need of device scheduling and control including the presence/absence of user, device standby power losses, device overloading, controlling energy bills and limiting energy usage [28,29]. The features discussed in this section also stem from the need for an efficient device scheduling and control. ...
... A major research effort is directed towards the analysis of smart plug data using Machine Learning and related algorithms. Some of the focus areas include Appliance identification, Presence detection, Demand Response management and Peak load management [21][22][23][24]27,29]. Another important issue that needs to be considered while implementing a smart plug is that despite the great advancement in Machine Learning and Artificial Intelligence, proper hardware to implement the Machine Learning algorithms is still in the development phase. ...
Article
Full-text available
In the current energy ecosystem, the need for a Hybrid Appliance Load Monitoring System (HALMS) to establish a smarter grid and energy infrastructure is undeniable. The increasing popularity of the Internet of Things (IoT) has suddenly pushed the demand for smart and connected devices. This review introduces the term smart plug as a device that uses IoT for establishing HALMS. These smart plugs are a handy solution to make the so-called 'dumb' devices smart. The strategy of smart plugs to enhance the energy management experience in connected spaces is presented. This study extensively highlights the current smart plug technologies and the relevant activities and limitations that need to overcome the requirements of HALMS.
... Nowadays, electrical appliances dominate the energy consumption in residential sector. The most measurements used today are blind [1], which means that the consumption of individual units is ignored as it is described in Fig. 1. This disadvantage cannot give any indication about detailed consumption that down to the used appliances. ...
... The classification is achieved using; random committee, Bayesian Network, J48, JRip, LogitBoost, Naive Bayes, Random Forest, and Random Tree the accuracy up to 95.5% obtained by Random Committee. In [16] the proposed appliances identification was built to recognize five classes of appliances, classification was performed using two classifier from two distinguished families, accuracy in this work up to 85%, This work was developed in [1] and the number of appliances categories was increased to 10 raw features that are analyzed using principal component analysis and the recognition of appliances achieved using SVM, 16 types of devices are investigated in this work given an accuracy up to 99.9%. Classification of ON/OFF appliances category was examined in [17], three classifiers namely Bayes Net, Random Forest and Hoeffding Tree are used to identify appliances through load signatures the proposed methods are implemented by using the WEKA software. ...
Conference Paper
Full-text available
Recognition of appliances’ signatures is an important task in energy disaggregation applications. To save and manage energy, load signatures provided by appliances can be used to detect which appliance is used. In this study, we use a low frequency database to identify appliances based on discrete wavelet transform for features extraction and data dimensionality reduction. Further that, the accuracy of several classifiers is investigated. This paper aims to prove the effectiveness of DWT in load signatures recognition. Then, the best classifier for this studied task is selected.
... The dataset used for the project is the second release of the Appliance Consumption Signature Fribourg database, ASC-F2, compiled by Gisler and Ridi et al and first demonstrated at the 6th International Conference on Soft Computing and Pattern Recognition in August 2014 [9]. Created through research in collaboration with the University of Applied Sciences Western Switzerland and Fribourg University's department of Informatics, ASC-F2 is an improvement on the first database, ASC-F1 which was released in 2013. ...
... Also in 2012, Zufferey et al used K Nearest Neighbour model (KNN) and the Gaussian mixture models (GMM) to achieve an accuracy of 85% in both models, using data collected from similar smart plugs [10]. Ridi, Gisler, and Hennebert, were able to achieve similar accuracies of 83.1% and 89.8% using the ASC-F2 database for KNN and GMM models respectively, when combined with a dynamic intersession protocol, where the first set of one hour sessions for each appliance is used to train the data, while the second set is used to test the data [9]. ...
Research
Full-text available
We examine classification models to identify household electric appliances from their electric consumption signatures taken in one hour sessions with a smart plug device.
... Features based on the signal evolution, as the first and second derivative, have been proposed in [39] [31]. Such features have shown to bring useful information in the classification task. ...
... Their best reported performance is 85% accuracy using k-NN. 1 www.tracebase.org 2 www.wattict.com A continuation of this work is presented by Ridi et al. [39] [31]. They report about MS-ILM and AS-ILM systems using the ACS-F1 database previously described [30]. ...
Article
Electricity load monitoring of appliances has become an important task considering the recent economic and ecological trends. In this game, machine learning has an important part to play, allowing for energy consumption understanding, critical equipment monitoring and even human activity recognition. This paper provides a survey of current researches on Intrusive Load Monitoring (ILM) techniques. ILM relies on low-end electricity meter devices spread inside the habitations, as opposed to Non-Intrusive Load Monitoring (NILM) that relies on an unique point of measurement, the smart meter. Potential applications and principles of ILMs are presented and compared to NILM. A focus is also given on feature extraction and machine learning algorithms typically used for ILM applications.
... As features input of the model, we compute a sequence of vectors using a sliding window procedure on top of the time series. More details about the modelling scheme are provided in [4] and [5]. We applied the two evaluation protocols proposed in [3] to benchmark our models. ...
... In order to evaluate the state-based model performances, two classifiers have been compared, namely k-Nearest Neighbor (k-NN) and GMM systems. Using the intersession protocol, we achieved respectively 88% and 93.8% correct category identification [5], while using the unseen instance protocol we achieved respectively 57% and 74% correct category identification [6]. Clearly the first protocol achieved better results than the second, given that in the latter case instances coming by appliances never seen before have to be classified. ...
... They used random committee algorithm for identifying the devices with an accuracy of 95%. (Ridi et al., 2013) have created ACS-F1 (Appliance Consumption Signature Fribourg-1) dataset containing 1 hour data for each of the 100 devices in 10 device types. The data was collected at a frequency of 1 sample every 10 seconds. ...
... As regards NILM approaches based on deep learning, the methods are mainly based on the extraction of features, which will be used either for training a Machine Learning (ML) algorithm (e.g. Support Vector Machines (SVMs) [32], Artificial Neural Networks 13 (ANNs) [33], Decision Trees (DTs) [34], Hidden Markov Models (HMMs) and their variants [35]- [37] and K-Nearest Neighbours (KNNs) [38]) or defining a set of rules or thresholds [39]- [42]. However, latest research in deep learning and big data lead to a significant increase in data-driven approaches using large scale datasets (e.g. ...
Thesis
Full-text available
The rapid development of technology in the electrical sector within the last 20 years has led to growing electric power needs through the increased number of electrical appliances and automation of tasks. In contrary, reduction of the overall energy consumption as well as efficient energy management are needed, in order to reduce global warming and meet the global climate protection goals. These requirements have led to the recent adoption of smart-meters and smart-grids, as well as to the rise of Non-Intrusive Load Monitoring. Non-Intrusive Load Monitoring aims to extract the energy consumption of individual electrical appliances through disaggregation of the total power consumption as measured by a single smart meter at the inlet of a household. Therefore, Non-Intrusive Load Monitoring is a highly under-determined problem which aims to estimate multiple variables from a single observation, thus is impossible to be solved analytical. In order to find accurate estimates of the unknown variables three fundamentally different approaches, namely deep-learning, pattern matching and single-channel source separation, have been investigated in the literature in order to solve the Non-Intrusive Load Monitoring problem. While Non-Intrusive Load Monitoring has multiple areas of application, including energy reduction through consumer awareness, load scheduling for energy cost optimization or reduction of peak demands, the focus of this thesis is especially on the performance of the disaggregation algorithm, the key part of the Non-Intrusive Load Monitoring architecture. In detail, optimizations are proposed for all three architectures, while the focus lies on deep-learning based approaches. Furthermore, the transferability capability of the deep-learning based approach is investigated and a NILM specific transfer architecture is proposed. The main contribution of the thesis is threefold. First, with Non-Intrusive Load Monitoring being a time-series problem incorporation of temporal information is crucial for accurate modelling of the appliance signatures and the change of signatures over time. Therefore, previously published architectures based on deep-learning have focused on utilizing regression models which intrinsically incorporating temporal information. In this work, the idea of incorporating temporal information is extended especially through modelling temporal patterns of appliances not only in the regression stage, but also in the input feature vector, i.e. by using fractional calculus, feature concatenation or high-frequency double Fourier integral signatures. Additionally, multi variance matching is utilized for Non-Intrusive Load Monitoring in order to have additional degrees of freedom for a pattern matching based solution. Second, with Non-Intrusive Load Monitoring systems expected to operate in realtime as well as being low-cost applications, computational complexity as well as storage limitations must be considered. Therefore, in this thesis an approximation for frequency domain features is presented in order to account for a reduction in computational complexity. Furthermore, investigations of reduced sampling frequencies and their impact on disaggregation performance has been evaluated. Additionally, different elastic matching techniques have been compared in order to account for reduction of training times and utilization of models without trainable parameters. Third, in order to fully utilize Non-Intrusive Load Monitoring techniques accurate transfer models, i.e. models which are trained on one data domain and tested on a different data domain, are needed. In this context it is crucial to transfer time-variant and manufacturer dependent appliance signatures to manufacturer invariant signatures, in order to assure accurate transfer modelling. Therefore, a transfer learning architecture specifically adapted to the needs of Non-Intrusive Load Monitoring is presented. Overall, this thesis contributes to the topic of Non-Intrusive Load Monitoring improving the performance of the disaggregation stage while comparing three fundamentally different approaches for the disaggregation problem.
... Literature already reports many approaches utilizing a wide variety of different sampling frequencies in the range of 0.1Hz up to 10kHz, showing that the choice of sampling frequency, features and method of edge-detection are having a significant influence on overall detection accuracy [14], [25]. However, to best of our knowledge, device specific sampling frequencies have not been reported in literature. ...
Conference Paper
This paper proposes a new appliance-driven selection of sampling frequencies for improving the energy disaggregation performance in non-intrusive load monitoring. Specifically, the methodology uses a machine learning model with parallel device detectors and optimized device dependent sampling rates in order to improve device identification. The performance of the proposed methodology was evaluated on a state-of-the-art baseline system and a set of publicly available databases increasing performance up to 6.7% in terms of estimation accuracy when compared to the baseline energy disaggregation setup without device dependent sampling rates.
... and short, repetitive characteristic substructures, which were also important in the sensor readings or image outline domains (Section 5.4). Note that the authors of the ACS-F1 dataset scored 93% [34] using a hidden Markov model and a manual feature set. Unfortunately their code is not available and the runtime was not reported. ...
Conference Paper
Full-text available
Time series (TS) occur in many scientific and commercial applications, ranging from earth surveillance to industry automation to the smart grids. An important type of TS analysis is classification, which can, for instance, improve energy load forecasting in smart grids by detecting the types of electronic devices based on their energy consumption profiles recorded by automatic sensors. Such sensor-driven applications are very often characterized by (a) very long TS and (b) very large TS datasets needing classification. However, current methods to time series classification (TSC) cannot cope with such data volumes at acceptable accuracy; they are either scalable but offer only inferior classification quality, or they achieve state-of-the-art classification quality but cannot scale to large data volumes. In this paper, we present WEASEL (Word ExtrAction for time SEries cLassification), a novel TSC method which is both fast and accurate. Like other state-of-the-art TSC methods, WEASEL transforms time series into feature vectors, using a sliding-window approach, which are then analyzed through a machine learning classifier. The novelty of WEASEL lies in its specific method for deriving features, resulting in a much smaller yet much more discriminative feature set. On the popular UCR benchmark of 85 TS datasets, WEASEL is more accurate than the best current non-ensemble algorithms at orders-of-magnitude lower classification and training times, and it is almost as accurate as ensemble classifiers, whose computational complexity makes them inapplicable even for mid-size datasets. The outstanding robustness of WEASEL is also confirmed by experiments on two real smart grid datasets, where it out-of-the-box achieves almost the same accuracy as highly tuned, domain-specific methods.
... The electric signals then observed are corresponding to one or few appliances, making the detection and identification task easier. Different machine learning modelling strategies are used to capture the specificities of electric signature with ILM, such as the use of Gaussian Mixture Models or Hidden Markov Models that reported good identification accuracy up to 95% [8,9]. ...
Article
Electricity load monitoring in residential buildings has become an important task allowing for energy consumption understanding, indirect human activity recognition and occupancy modelling. In this context, Non Intrusive Load Monitoring (NILM) is an approach based on the analysis of the global electricity consumption signal of the habitation. Current NILM solutions are reaching good precision for the identification of electrical devices but at the cost of difficult setups with expensive equipments typically working at high frequency. In this work we propose to use a low-cost and easy to install low frequency sensor for which we improve the performances with an active machine learning strategy. At setup, the system is able to identify some appliances with typical signatures such as a fridge. During usage, the system detects unknown signatures and provides a user-friendly procedure to include new appliances and to improve the identification precision over time.
... The use of this kind of devices allows to create the consumption fingerprint of the appliances, so it can later be used to automatically recognize them with no user interaction. Similar topics are dealt with in previous works such as [12] or [13]. ...
Conference Paper
Full-text available
Measuring and recording systems for the consumption of electrical energy which are connected to households, are essential in the optimization of energy use. Non-Intrusive Load Monitoring (NILM) is one of the most used techniques in the study of electrical consumption; these systems are based on the analysis of the load curve (the aggregated electrical consumption of the whole household). Thanks to a significant reduction in the price of sensors and sensor systems in recent years, it is possible to individually monitor each one of the devices connected to the grid. In this paper we compare different classifiers in order to find out which is the most appropriate for the identification of individual appliances attending to their consumption. In this way, we will know which electrical appliance is connected to a smart plug, helping to obtain more accurate and efficient load monitoring systems.
... In a study, the authors have used machine learning approaches for electric appliance classification based on low end sensor called PLOGGs to measure energy consumption. The authors have applied Gaussian Mixture Model (GMM) and K-Nearest Neighbor (KNN) techniques on derived features and observed 85% accuracy in results for identification [22,31]. Similarly, a research [23] performed by the same authors report an accuracy of 74% on unseen appliance identification. ...
Conference Paper
Energy utilization can be improved by precise plug load monitoring and control. Plug load energy consumption is nearly 30% of the total building energy consumption. Therefore, plug load identification is a key requirement for energy conservation in buildings. Intrusive load monitoring techniques identify loads precisely but have not been tested widely so far for their performance in changing operating conditions. Hence, the present research proposes a robust low frequency intrusive load monitoring technique to identify load accurately. A smart power strip using proposed load identification technique is designed and developed. Linear regression is applied on the acquired data to capture the behavioral trends of a particular device more explicitly and concisely. Further, weighted K-NN classifier is applied on the transformed data set for device. Experimental results show that the proposed algorithm performs better than the standard classifiers, and can offer tangible savings.
... Recent research studies have also been performed on the identification of the different home appliances. Ridi et al. [32] focused on analyzing classification algorithms, including K-Nearest Neighbor (KNN) and Gaussian Mixture Models (GMM), to recognize electric appliances automatically. The authors propose a system based on low-cost smart plugs that measure current periodically and that produce time series that characterize the consumption of an appliance. ...
Article
Full-text available
The Internet of Energy (IoE) represents a novel paradigm where electrical power systems work cooperatively with smart devices to increase the visibility of energy consumption and create safer, cleaner and sustainable energy systems. The implementation of IoE services involves the use of multiple components, like embedded systems, power electronics or sensors, which are an essential part of the infrastructure dedicated to the generation and distribution energy and the one required by the final consumer. This article focuses on the latter and presents a smart socket system that collects the information about energy price and makes use of sensors and actuators to optimize home energy consumption according to the user preferences. Specifically, this article provides three main novel contributions. First, what to our knowledge is the first hardware prototype that manages in a practical real-world scenario the price values obtained from a public electricity operator is presented. The second contribution is related to the definition of a novel wireless sensor network communications protocol based on Wi-Fi that allows for creating an easy-to-deploy smart plug system that self-organizes and auto-configures to collect the sensed data, minimizing user intervention. Third, it is provided a thorough description of the design of one of the few open-source smart plug systems, including its communications architecture, the protocols implemented, the main sensing and actuation components and the most relevant pieces of the software. Moreover, with the aim of illustrating the capabilities of the smart plug system, the results of different experiments performed are shown. Such experiments evaluate in real-world scenarios the system’s ease of use, its communications range and its performance when using HTTPS. Finally, the economic savings are estimated for different appliances, concluding that, in the practical situation proposed, the smart plug system allows certain energy-demanding appliances to save almost €70 per year.
... Note that the authors of the ACS-F1 dataset scored 93% [35] using a hidden Markov model and a manual feature set. Unfortunately their code is not available and the runtime was not reported. ...
Article
Full-text available
Time series (TS) occur in many scientific and commercial applications, ranging from earth surveillance to industry automation to the smart grids. An important type of TS analysis is classification, which can, for instance, improve energy load forecasting in smart grids by detecting the types of electronic devices based on their energy consumption profiles recorded by automatic sensors. Such sensor-driven applications are very often characterized by (a) very long TS and (b) very large TS datasets needing classification. However, current methods to time series classification (TSC) cannot cope with such data volumes at acceptable accuracy; they are either scalable but offer only inferior classification quality, or they achieve state-of-the-art classification quality but cannot scale to large data volumes. In this paper, we present WEASEL (Word ExtrAction for time SEries cLassification), a novel TSC method which is both scalable and accurate. Like other state-of-the-art TSC methods, WEASEL transforms time series into feature vectors, using a sliding-window approach, which are then analyzed through a machine learning classifier. The novelty of WEASEL lies in its specific method for deriving features, resulting in a much smaller yet much more discriminative feature set. On the popular UCR benchmark of 85 TS datasets, WEASEL is more accurate than the best current non-ensemble algorithms at orders-of-magnitude lower classification and training times, and it is almost as accurate as ensemble classifiers, whose computational complexity makes them inapplicable even for mid-size datasets. The outstanding robustness of WEASEL is also confirmed by experiments on two real smart grid datasets, where it out-of-the-box achieves almost the same accuracy as highly tuned, domain-specific methods.
... For sake of simplicity hereunder we will use the term "phase angle" for indicating the sinus of in radiant. We repeated the computation of dynamic coefficients, being their usefulness demonstrated in similar tasks [8]. The velocity and acceleration parameters, also called respectively delta and delta-delta coefficients, were computed. ...
Article
We asset about the analysis of electrical appliance consumption signatures for the identification task. We apply Hidden Markov Models to appliance signatures for the identification of their category and of the most probable sequence of states. The electrical signatures are measured at low frequency (10-1 Hz) and are sourced from a specific database. We follow two predefined protocols for providing comparable results. Recovering information on the actual appliance state permits to potentially adopt energy saving measures, as switching off stand-by appliances or, generally speaking, changing their state. Moreover, in most of the cases appliance states are related to user activities: the user interaction usually involves a transition of the appliance state. Information about the state transition could be useful in Smart Home / Building Systems to reduce energy consumption and increase human comfort.We report the results of the classification tasks in terms of confusion matrices and accuracy rates. Finally, we present our application for a real-time data visualization and the recognition of the appliance category with its actual state.
... They find applications in domains such as speech and handwriting recognition, biometric authentication and financial time series analysis [4; 5]. More recently they have been used in building control in the fields of occupant pattern detection [6; 7] and appliance identification [8]. In HMM, the system being modelled is assumed to be a Markov process with hidden states but with visible output tokens which depend on the (hidden) state. ...
Article
In this article we present a novel approach to model visual comfort based on supervised state-based machine learning with Hidden Markov Models and one easy-to-obtain variable (illuminance measurements at the horizontal work-plane; E desk). Data mining was performed on sensor data recorded for two years in a single-occupant office room and the developed model classifies workplane illuminances into 3 states: comfort; discomfort because of low light; discomfort because of excessive light. Results show that a training period of 4 to 8 months of recorded data leads to a visual comfort identification (classification) accuracy of 100%. When training the model using 4-month data, an overall 92% accuracy can be achieved (75% for the 'discomfort because of low light' state). Following further analysis of this occupant-adapted model, we discuss the confidence ('normalised relative likelihood') with which the model classifies illuminances in one of three different states as a function of the E desk. We argue that the resulting metrics are an ideal input which can be readily used into automatic lighting controllers based on fuzzy logic. Last, the model's performance is compared and validated against state-of-the art classifiers such as Bayesian and k-Nearest Neighbors.
... We applied three machine learning algorithms: k-Nearest Neighbours (k-NN), Gaussian Mixture Models (GMM) and Hidden Markov Models (HMM). In [16] we applied the Intersession protocol to the ACS-F1 database. As machine learning algorithm we used k-NN and GMM, tuning respectively the number of Neighbours and the number of Gaussians. ...
Article
The automatic identification of appliances through the analysis of their electricity consumption has several purposes in Smart Buildings including better understanding of the energy consumption, appliance maintenance and indirect observation of human activities. Electric signatures are typically acquired with IoT smart plugs integrated or added to wall sockets. We observe an increasing number of research teams working on this topic under the umbrella Intrusive Load Monitoring. This term is used as opposition to Non-Intrusive Load Monitoring that refers to the use of global smart meters. We first present the latest evolutions of the ACS-F database, a collections of signatures that we made available for the scientific community. The database contains different brands and/or models of appliances with up to 450 signatures. Two evaluation protocols are provided with the database to benchmark systems able to recognise appliances from their electric signature. We present in this paper two additional evaluation protocols intended to measure the impact of the analysis window length. Finally, we present our current best results using machine learning approaches on the 4 evaluation protocols.
... The system monitors device loads and is used for behavior and pattern prediction to identify a range of applications. In [34][35], the authors use smart plugs for load identifications. This information is useful for building management systems to conserve energy. ...
Article
Full-text available
Building automation is a trusted technology to ensure user comfort in buildings. Today the challenge is not only to provide comfort, but to do it while having a close eye on the energy performance of the building. Distributed, renewable energy sources, advances in IT and tighter legislative constraints offer and demand changes in the way we operate buildings. We look at the core fields that enable building operation: monitoring of operation data, controls, automation and the management organization to show that innovation can be achieved by integrating cross-domain knowledge and getting inspirations from other domains, where existing approaches and technologies can be used and transferred to the domain of building automation.
... They combine KNN and Gaussian Mixture Models (GMM) techniques. The main disadvantage is the limitations to scalability, as they store the row data in xml files [21]. ...
Article
Full-text available
The efficient use of resources is a matter of great concern in today's society, especially in the energy sector. Although the main strategy to decrease energy use has long been focused on supply, over the last few years, there has been a shift to the demand side. Under this new line of action, demand-side management networks have emerged and extended from the household level to larger installations, with the appearance of the concepts of Smart Grids and even Smart Cities. The extended use of Smart Meters for measuring residential electricity consumption facilitates the creation of such intelligent environments. In this context, this article proposes a system which extracts value from the collected consumer information to identify the appliances belonging to that smart environment by means of machine learning techniques. Considering the large amount of information that would be handled when millions of homes were sending data, big data technology has been used. An experiment to evaluate the classification method was carried out with seven devices and three different configurations. The results are also reported, achieving promising results, with recognition rates of 75 % after 1 h of training and 100 % after 4 h.
... Accurate load identification has been achieved using highfrequency smart meters [22], but this granularity of data is not typically available from off-the-shelf smart outlets. For lower frequency data, e.g., 1 Hz, the use of classifiers for learning device labels has been proposed, but not extensively evaluated [23], [24], particularly for previously unseen devices. Other approaches to device classification have focused on explicit per-device training to generate 'signatures' that can be used to detect devices in the future [25]. ...
Article
Full-text available
An increasing interest in energy-efficiency combined with the decreasing cost of embedded networked sensors is lowering the cost of outlet-level metering. If these trends continue, new buildings in the near future will be able to install 'smart' outlets, which monitor and transmit an outlets power usage in real time, for nearly the same cost as conventional outlets. One problem with the pervasive deployment of smart outlets is that users must currently identify the specific device plugged into each meter, and then manually update the outlets meta-data in software whenever a new device is plugged into the outlet. Correct meta-data is important in both interpreting historical outlet energy data and using the data for building management. To address this problem, we propose Non-Intrusive Load Identification (NILI), which automatically identifies the device attached to a smart outlet without any human intervention. In particular, in our approach to NILI, we identify an intuitive and simple-to-compute set of features from time-series energy data and then employ well-known classifiers. Our results achieve accuracy of over 90% across 15 device types on outlet-level energy traces collected from multiple real homes.
... These two parameters are also called respectively delta and delta-delta coefficients. In a work on the previous database version we demonstrated the usefulness of dynamic coefficients [13]. We computed the velocity coefficients as: ...
Article
We present ACS-F2, a new electric consumption signature database acquired from domestic appliances. The scenario of use is appliance identification with emerging applications such as domestic electricity consumption understanding, load shedding management and indirect human activity moni-toring. The novelty of our work is to use low-end electricity consumption sensors typically located at the plug. Our approach consists in acquiring signatures at a low frequency, which contrast with high frequency transient analysis approaches that are costlier and have been well studied in former research works. Electrical consumption signatures comprise real power, reactive power, RMS current, RMS voltage, frequency and phase of voltage relative to current. A total of 225 appliances were recorded over two sessions of one hour. The database is balanced with 15 different brands/models spread into 15 categories. Two realistic appliance recognition protocols are proposed and the database is made freely available to the scientific community for the experiment reproducibility. We also report on recognition results following these protocols and using baseline recognition algorithms like k-NN and GMM.
... Two evaluation protocols are proposed with the database distribution. The first intersession protocol consists in attempting to recognize signatures from appliances already seen in the training phase [29]. The second protocol, called unseen instance, aims at recognizing signatures coming from appliances never seen in the training phase. ...
Article
Full-text available
The emerging concept of Smart Building relies on an intensive use of sensors and actuators and therefore appears, at first glance, to be a domain of predilection for the IoT. However, technology providers of building automation systems have been functioning, for a long time, with dedicated networks, communication protocols and APIs. Eventually, a mix of different technologies can even be present in a given building. IoT principles are now appearing in buildings as a way to simplify and standardise application development. Nevertheless, many issues remain due to this heterogeneity between existing installations and native IP devices that induces complexity and maintenance efforts of building management systems. A key success factor for the IoT adoption in Smart Buildings is to provide a loosely-coupled Web protocol stack allowing interoperation between all devices present in a building. We review in this Chapter different strategies that are going in this direction. More specifically, we emphasise on several aspects issued from pervasive and ubiquitous computing like service discovery. Finally, making the assumption of seamless access to sensor data through IoT paradigms, we provide an overview of some of the most exciting enabling applications that rely on intelligent data analysis and machine learning for energy saving in buildings.
Chapter
According to the European Union (EU), about 72% of buildings and their equipment are not adapted to be energy efficient. This fact drives the EU countries to implement a new strategy to increase energy efficiency and promote decarbonization. Wrong estimations of real-life energy consumption and energy-consuming equipment are common factors that are degrading energy efficiency. Real-life energy consumption measurements of buildings compared with the predictions may be exceeded by more than 100%. To improve the forecasts and guide the users to configure their appliances right, a system prototype that monitors and acquires data from various appliances and provides fine-grained information about each appliance is proposed. It is designed to understand the real-life consumption of each device and compare it with the laboratory measurements observed and used by the EU energy efficiency labelling system. This work is a part of the “Self Assessment Towards Optimization” (SATO) project, where new energy assessments services are made available. These new services are data-capable and allow to express real-life utilization and life-cycle evolution.KeywordsMicrocontrollersSensorsAppliance monitoringAppliance classification
Article
Energy load monitoring via smart plugs or smart sockets has become more and more popular. Various studies have been undertaken to monitor energy consumption of household appliances and analyze the collected power data to obtain useful insights on consumers’ behaviors. The main challenge in load monitoring is to automatically recognize appliances in real time since the existing energy disaggregation process is time-consuming and labour-intensive. Although several deep learning models can achieve high accuracy on appliance classification, they usually consume large memory, hence not suitable for resources-constrained IoT devices. To resolve the issue, we demonstrate in this paper, for the first time, a novel framework named Smart Intrusive Load Monitoring based on a compact network (CompactNet), which is able to determine appliance types in real time. Specifically, our method distills the knowledge of an ensemble of large deep networks to a much more compact network. Our CompactNet accurately classifies various types of appliances, but its size is reduced by approximately eight times, making it possible to be deployed on edge IoT sensors for appliance recognition.
Conference Paper
Full-text available
Channel modeling using the correlated double ring concept on Vehicular Network communications systems has been developed previously. However, no such model has been incorporated into the OFDM multi-carrier system to simulate the effect of the transmitter and receiver velocity on the received signal quality. User velocity on the transmitter and receiver side produces a Doppler effect that damages the signal orthogonality on OFDM. In addition, the speed of the transmitter and receiver also affect the Power Spectral Density of the received signal. This paper used the Correlated Double Ring channel modeling to simulate the transmitter and receiver movement against the Power Spectral Density parameter with non-moving scatterers. The IFFT output of the OFDM transmitter is consonant with the Channel Impulse Response (CIR) of the channel modeling output and coupled with the AWGN noise. Simulation is done by dividing the movement of sender and receiver into 3-speed regions, i.e. low speed, medium and high speed. Simulation results show the faster movement of the sender and receiver cause Doppler Shift larger and make the value of power spectral density parameters become getting muffled.
Article
Electrical plug-loads and appliances have evolved over time to provide sophisticated functions to occupants in a building, and they account for a big share of building's energy consumption. Despite their evolution, they are still unmanageable in commercial and industrial buildings; unlike lighting or air-conditioning that are fixed infrastructures managed by an intelligent software layer such as a building management system. The qualities of heterogeneity, numerosity, spatial irregularity and displaceability have impeded efficient management of plug-loads, unless they are appropriately identified in real-time. Decades of efforts in both non-intrusive and nonintrusive load monitoring have not yielded any sufficiently reliable universal mechanism to identify plug-loads. In this work, a practical method of uniquely identifying plug-loads using a novel socket cyber-physical system is proposed. This socket/receptacle can access the metadata and operational-data of plug-loads in an automated fashion, thus yielding a live digital profile of plug-loads connected to sockets. Based on acquired intelligence, a supervisory control system can contextually identify the plug-loads to provide several digitized services. Various benefits and use-cases of the proposed smart cyber-physical system are explained, inclusive of improvement in operational energy efficiency of plug-loads.
Conference Paper
Home Energy Management System (HEMS) is acknowledged as a promised approach to explore household appliances dynamic energy usage. The availability of an appropriate dataset is indispensable to evaluate the performance of HEMS operations in the designing phase. In this paper, we develop a tool capable of generating long-term semi-synthetic data to avoid deficiency of available datasets particularly, the lack of the major appliances measurements and non-electric information. Accordingly, a simple household with primary appliances located in two-main zones is simulated. The paper utilizes a statistical analysis of real-world data to create probabilistic models of appliances and consequently, produce time-extended stochastic power profiles. Afterward, a simulation structure is developed to generate the power consumption profiles of major appliances consisting of Electrical Space Heaters (ESH) and Electrical Water Heaters (EWH). In order to achieve its ambition, this study executes a post-processing practice to create on/off power profiles of these appliances using their models. The results show that the proposed tool can be exploited for different HEMS scenarios.
Article
With the progress of science and technology, more and more machines are adpot to help human life better and more convenient. When the machines have been used for a longer period of time so that the machine components are getting old, the amount of power comsumption will increase and easily cause the machine to overheat. This also causes a waste of invisible resources. If the Internet of Everything (IoE) technologies are able to be applied into the enterprise information systems for monitoring the machines use time, it can not only make energy can be effectively used, but aslo create a safer living environment. To solve the above problem, the correlation predict model is established to collect the data of power consumption converted into power eigenvalues. This study takes the power eigenvalue as the independent variable and use time as the dependent variable in order to establish the decline curve. Ultimately, the scoring and estimation modules are employed to seek the best power eigenvalue as the independent variable. To predict use time, the correlation is discussed between the use time and the decline curve to improve the entire behavioural analysis of the facilitate recognition of the use time of machines.
Conference Paper
Low-cost network-connected smart outlets are now available for monitoring, controlling, and scheduling the energy usage of electrical devices. As a result, such smart outlets are being integrated into automated home management systems, which remotely control them by analyzing and interpreting their data. However, to effectively interpret data and control devices, the system must know the type of device that is plugged into each smart outlet. Existing systems require users to manually input and maintain the outlet metadata that associates a device type with a smart outlet. Such manual operation is time-consuming and error-prone: users must initially inventory all outlet-to-device mappings, enter them into the management system, and then update this metadata every time a new device is plugged in or moves to a new outlet. Inaccurate metadata may cause systems to misinterpret data or issue incorrect control actions. To address the problem, we propose AutoPlug, a system that automatically identifies and tracks the devices plugged into smart outlets in real time without user intervention. AutoPlug combines machine learning techniques with time-series analysis of device energy data in real time to accurately identify and track devices on startup, and as they move from outlet-to-outlet. We show that AutoPlug achieves ∼90% identification accuracy on real data collected from 13 distinct device types, while also detecting when a device changes outlets with an accuracy >90%. We implement an AutoPlug prototype on a Raspberry Pi and deploy it live in a real home for a period of 20 days. We show that its performance enables it to monitor up to 25 outlets, while detecting new devices or changes in devices with <50s latency.
Article
Energy monitoring is one of the important aspects of the energy management, as such there is a need to monitor the power consumption of a premises before planning some of the technical measures to minimize the energy consumption. This paper presents the current state of art of appliances’ energy management through Intrusive Load Monitoring (ILM) and Non-intrusive Load Monitoring (NILM), ILM is referred to as distributive sensing and NILM is based on single point sensing. In an effort to explore the recent trend in Energy Management for researchers in the area, various techniques of Home Energy Management (HEM) using ILM and NILM have been analyzed and categorized. Different contributions made by the researchers have been highlighted together with some techniques of reducing power consumption of a building for cost saving and greener environment. This study reveal that there are issues concerning load monitoring and load management that need the attention; issues such as more accurate recognition and the need for monitoring system that can recognize as many kind of loads as possible. More efforts to apply NILM in appliances’ energy management is also required. Finally there is need for promoting the energy management culture among the electricity consumers whether in homes, offices or industries. The study will help the researchers in the area to have a clear vision of the recent trends in the area.
Conference Paper
Full-text available
Dynamic load management, i.e., allowing electricity utilities to remotely turn electric appliances in households on or off, represents a key element of the smart grid. Appliances should however only be disconnected from mains when no negative side effects, e.g., loss of data or thawing food, are incurred thereby. This motivates the use of appliance identification techniques, which determine the type of an attached appliance based on the continuous sampling of its power consumption. While various implementations based on different sampling resolutions have been presented in existing literature, the achievable classification accuracies have rarely been analyzed. We address this shortcoming and evaluate the accuracy of appliance identification based on the characteristic features of traces collected during the 24 hours of a day. We evaluate our algorithm using more than 1,000 traces of different electrical appliances' power consumptions. The results show that our approach can identify most of the appliances at high accuracy.
Article
Full-text available
We report on the development of an innovative system which can automatically recognize home appliances based on their electric consumption profiles. The purpose of our system is to apply adequate rules to control electric appliance in order to save energy and money. The novelty of our approach is in the use of plug-based low-end sensors that measure the electric consumption at low frequency, typically every 10 seconds. Another novelty is the use of machine learning approaches to perform the classification of the appliances. In this paper, we present the system architecture, the data acquisition protocol and the evaluation framework. More details are also given on the feature extraction and classification models being used. The evaluation showed promising results with a correct rate of identification of 85\%.
Article
Full-text available
Activity sensing in the home has a variety of important applications, including healthcare, entertainment, home automation, energy monitoring and post-occupancy research studies. Many existing systems for detecting occupant activity require large numbers of sensors, invasive vision systems, or extensive installation procedures. We present an approach that uses a single plug-in sensor to detect a variety of electrical events throughout the home. This sensor detects the electrical noise on residential power lines created by the abrupt switching of electrical devices and the noise created by certain devices while in operation. We use machine learning techniques to recognize electrically noisy events such as turning on or off a particular light switch, a television set, or an electric stove. We tested our system in one home for several weeks and in five homes for one week each to evaluate the system performance over time and in different types of houses. Results indicate that we can learn and classify various electrical events with accuracies ranging from 85-90%.
Conference Paper
Full-text available
Individual appliances' electricity consumption is automatically disaggregated from a single custom metering system on the main feed to an occupied residential building. A data acquisition system samples voltage and current at 100 kHz, then calculates real and reactive power, harmonics, and other features at 20Hz. A probabilistic event-detector using the generalized likelihood ratio (GLR) matches human-labeled events to the time-series of features. Machine-learning classification was most successful with the 1-nearest-neighbor algorithm, correctly identifying 90% of the laboratory-generated training events and 79% of validation examples. The challenge of obtaining adequate training data for the real-world home leads to the development of the Wire Spy, a wirelessly-networked event detector with an inductive sensor which clamps to the cable of an appliance.
Conference Paper
Full-text available
Microgrids are well-suited for electrification of remote off-grid areas. This paper sketches the concept of a plug-and-play microgrid with a minimum of configuration effort needed for setup. When the load of such an off-grid microgrid grows over the generation capacity and energy storage is not sufficient, demand has to be reduced to prevent a blackout. In order to decide which loads are inessential and can be shedded, automated load recognition on the basis of measured power consumption profiles is needed. Two promising approaches from the area of speech recognition, Dynamic Time Warping and Hidden Markov Models, are compared for this application. It is found that a key feature to achieve good recognition efficiency is a careful selection of the features extracted from the measured power data.
Thesis
Full-text available
In this thesis, we are concerned with the two fields of automatic speech recognition (ASR) and automatic speaker recognition (ASkR) in telephony. More precisely, we are interested in systems based on hidden Markov models (HMMs) in which artificial neural networks (ANNs) are used in place of more classical tools. This work is dedicated to the analysis of three approaches. The first one, mainly original, concerns the use of Self-Organizing Maps in discrete HMMs for isolated word speech recognition. The second approach concerns continuous hybrid HMM/ANN systems, extensively studied in previous research work. The system is not original in its form but its analysis permitted to bring a new theoretical framework and to introduce some extensions regarding the way the system is trained. The last part concerns the implementation of a new ANN segmental approach for text-independent speaker verification.
Article
Full-text available
Nous présentons une méthodologie basée sur les chaînes de Markov cachées pour la modélisation et l'analyse statistique de courbes de consommation électrique. Suite à une analyse de variance qui permet d'estimer l'effet sur la log-consommation de facteurs contrôlés (mois, jour, heure, type de contrat et puissance maximale souscrite), nous modélisons les résidus par une chaîne de Markov cachée. Ensuite, les états cachés sont restaurés puis interprétés grâce à un tableau de contingence les mettant en relation avec la consommation de différents appareils électriques (ou usages) lorsque celle-ci est disponible. Nous montrons comment cette approche, par l'usage conjoint du tableau de contingence et des états cachés restaurés, permet une estimation de la consommation de chaque usage dans le cas où elle est inconnue. La prise en compte d'informations a priori sur la consommation rend cette estimation plus réaliste. Enfin, nous abordons le problème de la sélection de modèles. Les critères classiques conduisent à des modèles complexes qui rendent très difficile l'interprétation des états cachés par leur mise en correspondance avec les usages. Nous présentons et commentons des méthodes de sélection alternatives qui tiennent compte de l'objectif de mise en relation des états cachés et des classes de consommation.
Article
This paper considers additive factorial hidden Markov models, an extension to HMMs where the state factors into multiple independent chains, and the output is an additive function of all the hidden states. Although such models are very powerful, accurate inference is unfortunately difficult: exact inference is not computationally tractable, and existing approximate inference techniques are highly susceptible to local optima. In this paper we propose an alternative inference method for such models, which exploits their additive structure by 1) looking at the observed difference signal of the observation, 2) incorporating a “robust” mixture component that can account for unmodeled observations, and 3) constraining the posterior to allow at most one hidden state to change at a time. Combining these elements we develop a convex formulation of approximate inference that is computationally efficient, has no issues of local optima, and which performs much better than existing approaches in practice. The method is motivated by the problem of energy disaggregation, the task of taking a whole home electricity signal and decomposing it into its component appliances; applied to this task, our algorithm achieves state-of-the-art performance, and is able to separate many appliances almost perfectly using just the total aggregate signal.
Article
Energy and sustainability issues raise a large number of problems that can be tackled using approaches from data mining and machine learning, but traction of such problems has been slow due to the lack of publicly available data. In this paper we present the Reference Energy Disaggregation Data Set (REDD), a freely available data set containing de-tailed power usage information from several homes, which is aimed at furthering research on energy disaggregation (the task of determining the component appliance contributions from an aggregated electricity signal). We discuss past ap-proaches to disaggregation and how they have influenced our design choices in collecting data, we describe the hardware and software setups for the data collection, and we present initial benchmark disaggregation results using a well-known Factorial Hidden Markov Model (FHMM) technique.
Conference Paper
Today's solutions to inform residents about their electricity consumption are mostly confined to displaying aggregate readings collected at meter level. A reliable identification of appliances that require disproportionate amounts of energy for their operation is generally unsupported by these systems, or at least requires significant manual configuration efforts. We address this challenge by placing low-cost measurement and actuation units into the mains connection of appliances. The distributed sensors capture the current flow of individual appliances at a sampling rate of 1.6kHz and apply local signal processing to the readings in order to extract characteristic fingerprints. These fingerprints are communicated wirelessly to the evaluation server, thus keeping the required airtime and energy demand of the transmission low. The evaluation server employs machine learning techniques and caters for the actual classification of attached electric appliances based on their fingerprints, enabling the correlation of consumption data and the appliance identity. Our evaluation is based on more than 3,000 current consumption fingerprints, which we have captured for a range of household appliances. The results indicate that a high accuracy is achieved when locally extracted current consumption fingerprints are used to classify appliances.
Conference Paper
We report on the creation of a database of appliance consumption signatures and two test protocols to be used for appliance recognition tasks. By means of plug-based low-end sensors measuring the electrical consumption at low frequency, typically every 10 seconds, we made two acquisition sessions of one hour on about 100 home appliances divided into 10 categories: mobile phones (via chargers), coffee machines, computer stations (including monitor), fridges and freezers, Hi-Fi systems (CD players), lamp (CFL), laptops (via chargers), microwave oven, printers, and televisions (LCD or LED). We measured their consumption in terms of real power (W), reactive power (var), RMS current (A) and phase of voltage relative to current (φ). We now give free access to this ACS-Fl database. The proposed test protocols will help the scientific community to objectively compare new algorithms.
Article
Load signatures embedded in common electricity consumption patterns, in fact, could render much information pertaining to the nature of the appliances and their usage patterns. Based on the proposed disaggregation framework, we use three advanced disaggregation algorithms, called committee decision mechanisms (CDMs), to perform load disaggregation at the metering level. Three random switching simulators are also developed to investigate the performance of different CDMs under a variety of scenarios. Through Monte Carlo simulations, we demonstrate that all CDMs outperform any single-feature, single-algorithm-based disaggregation methods. With sensitivity analysis, we also show that the CDMs are less sensitive to any load dynamics and noise. We finally demonstrate some applications of this technology in terms of appliance usage tacking and estimated energy consumption of each appliance.
Article
Load signature is the unique consumption pattern intrinsic to each individual electrical appliance/piece of equipment. This paper focus on building a universal platform to better understand and explore the nature of electricity consumption patterns using load signatures and advanced technology, such as feature extraction and intelligent computing. Through this knowledge, we can explore and develop innovative applications to achieve better utilization of resources and develop more intelligent ways of operation. This paper depicts the basic concept, features of load signatures, structure and methodology of applying mathematical programming techniques, pattern recognition tools, and committee decision mechanism to perform load disaggregation. New indices are also introduced to aid our understanding of the nature of load signatures and different disaggregation algorithms.
Article
S ummary A broadly applicable algorithm for computing maximum likelihood estimates from incomplete data is presented at various levels of generality. Theory showing the monotone behaviour of the likelihood and convergence of the algorithm is derived. Many examples are sketched, including missing value situations, applications to grouped, censored or truncated data, finite mixture models, variance component estimation, hyperparameter estimation, iteratively reweighted least squares and factor analysis.
Article
A broadly applicable algorithm for computing maximum likelihood estimates from incomplete data is presented at various levels of generality. Theory showing the monotone behaviour of the likelihood and convergence of the algorithm is derived. Many examples are sketched, including missing value situations, applications to grouped, censored or truncated data, finite mixture models, variance component estimation, hyperparameter estimation, iteratively reweighted least squares and factor analysis.
Article
A nonintrusive appliance load monitor that determines the energy consumption of individual appliances turning on and off in an electric load, based on detailed analysis of the current and voltage of the total load, as measured at the interface to the power source is described. The theory and current practice of nonintrusive appliance load monitoring are discussed, including goals, applications, load models, appliance signatures, algorithms, prototypes field-test results, current research directions, and the advantages and disadvantages of this approach relative to intrusive monitoring
Em demystified: An expectation- maximization tutorial
  • Y Chen
  • M R Gupta
A non-intrusive appliance load monitoring system for identifying kitchen activities
  • M N V Perez