A non-intrusive load identification method based on convolution neural network

Lan, Zehua; Yin, Bo; Wang, Tao; Zuo, Geng-ren

doi:10.1109/ei2.2017.8245612

Cited by 21 publications

(11 citation statements)

References 3 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Although the proposed algorithm is applicable in the general smart grid context, this paper focuses on the NILM problem for demonstration purposes. In the literature, ML has been leveraged to perform NILM, and Wang et al [1], Lan et al [39], Mauch and Yang [40] are examples of some recent work in this area. This paper focuses on [1] for the Oracle.…”

Section: A the Threat Modelmentioning

confidence: 99%

“…1) Substitute Model Selection: First, the architecture of the substitute model constructed by the adversary is presented. As highlighted in recent literature, various internal architectures that include convolutional neural networks [39] [41] [42], recurrent neural networks [40] [43] [44], and auto-encoders [45] have been leveraged to construct ML models for NILM applications. One important objective is that the substitute model must be able to imitate any Oracle operating on smart meter data.…”

Section: B Stage 1: Substitute Model Constructionmentioning

confidence: 99%

See 1 more Smart Citation

Stealthy Black-Box Attacks on Deep Learning Non-Intrusive Load Monitoring Models

Wang

Srikantha

2021

IEEE Trans. Smart Grid

View full text Add to dashboard Cite

With the advent of the advanced metering infrastructure, electricity usage data is being continuously generated at large volumes by smart meters vastly deployed across the modern power grid. Electric power utility companies and third party entities such as smart home management solution providers gain significant insights into these datasets via machine learning (ML) models. These are then utilized to perform active/passive power demand management that fosters economical and sustainable electricity usage. Although ML models are powerful, these remain vulnerable to adversarial attacks. A novel stealthy black-box attack construction model is proposed that targets deep learning models utilized to perform non-intrusive load monitoring based on smart meter data. These attacks are practical as there is no assumption of the knowledge of training data, internal parameters, and architecture of the targeted ML model. The profound impact of the proposed stealthy attack constructions on energy analytics and decision-making processes is shown through comprehensive theoretical, practical, and comparative analysis. This work sheds light on vulnerabilities of ML models in the smart grid context and provides valuable insights for securely accommodating increasing prevalence of artificial intelligence in the modern power grid.

show abstract

Section: A the Threat Modelmentioning

confidence: 99%

Section: B Stage 1: Substitute Model Constructionmentioning

confidence: 99%

Stealthy Black-Box Attacks on Deep Learning Non-Intrusive Load Monitoring Models

Wang

Srikantha

2021

IEEE Trans. Smart Grid

View full text Add to dashboard Cite

show abstract

“…However, many studies propose applying CNNs to achieve NILM. Lan proposes a method to convert electric current data into a grayscale image with a resolution of 64×64, and then designs a CNN model to identify the load [24]. Yang proposes an imaging rule to convert current waveforms to greyscale image and constructs a CNN architecture premised upon VGG-16 to tackle the issue of NILM [25].…”

Section: Related Workmentioning

confidence: 99%

Non-Intrusive Load Monitoring Using a CNN-LSTM-RF Model Considering Label Correlation and Class-Imbalance

Zhang

Liu

et al. 2021

IEEE Access

View full text Add to dashboard Cite

Non-Intrusive Load Monitoring (NILM) is particularly important for demand response. This paper proposes an innovative method based on a deep learning model to recognize the working state of electrical appliances using low frequency load data. The approach includes a data processing step, a deep learning model and a new accuracy calculation method. The data processing step consists of a multi-feature and high-dimensional method (MFHDM) and a pre-training process. The deep learning model consists of a convolutional neural network (CNN), a long-term short-term memory network (LSTM) and a randomforest (RF) algorithm. The proposed method addresses the label correlation problem and the classimbalance problem. To test the proposed method, the Reference Energy Disaggregation Dataset (REDD) and the Pecan Street dataset (PSD) are used. A comparative analysis with several models shows that the proposed method can effectively improve electrical appliance recognition accuracy and realize NILM.

show abstract

“…Nosúltimos anos, as CNNs conseguiram solucionar diversos problemas de visão computacional, como classificação de imagens, detecção de objetos e alteração de imagens. Tudo isso foi possível devidoà sua capacidade de selecionar características em diferentes níveis (Lan et al, 2017). Apesar da rede 5 ter apresentado a melhor acurácia, ela também apresentou um grande número de parâmetros, um total de 2.928.800.…”

Section: Caracterização E Classificação Autônomaunclassified

“…Já no segundo caso, o monitoramentoé feito através de umúnico dispositivo conectadoà rede, o qual envia dados para uma central queé utilizada para reconhecer as cargas ativas (Rayudu et al, 2011). O método não intrusivo apresenta vantagens frente ao intrusivo, tais como: custo baixo, facilidade na instalação e na manutenção, mesmo com a entrada de novos equipamentos na rede (Lan et al, 2017). Tais vantagens garantem a viabilidade de implantação do sistema de identificação, tornando o método NILM o mais investigado.…”

Section: Introductionunclassified

Identificação não-intrusiva de cargas similares em Smart Grid usando rede neural convolucional

Firmes

Celeste

Silvestre

2019

Anais Do 14º Simpósio Brasileiro De Automação Inteligente

View full text Add to dashboard Cite

This work deals with the problem of identifying technically identical equipments, defined here as highly similar equipments, monitored from a point of common coupling (PCC). An experimental approach is given here, where four highly similar computers are used, and none or even all computers can be in simultaneous operation. Each possible configuration of computers in simultaneous operation is defined here as a load. Samples of the electrical power demand for each load are acquired. A process of automatic extraction of load features is performed using convolutional neural networks (CNN). These load features form the so-called load signature, which is the necessary and fundamental information for the training of the identification system, which proved to be very efficient when achieving a success rate of 98.96%, i. e., considerably higher than those achieved by systems trained from manually obtained signatures. Resumo: Este trabalho trata do problema de identificação de equipamentos monitorados a partir de um ponto comum de acoplamento, sendo tais equipamentos tecnicamente idênticos sob o ponto de vista elétrico, definidos aqui como equipamentos altamente similares. O problemá e tratado neste artigo de forma experimental, onde são usados quatro computadores altamente similares, sendo que nenhum ou até todos os computadores podem estar em funcionamento simultâneo. Cada configuração possível de computadores em funcionamento simultâneoé aqui definida como carga. Amostras da potência elétrica demandada por cada carga são adquiridas. Um processo de extração automática de características das cargasé realizado utilizandose redes neurais convolucionais (CNN). Tais características formam a chamada assinatura de carga, queé a informação necessária e fundamental para o treinamento do sistema de identificação, o qual se mostrou bastante eficiente ao se alcançar uma taxa de 98,96% de precisão, apresentando, portanto, um desempenho muito superior a sistemas treinados a partir de assinaturas manualmente obtidas.

show abstract

A non-intrusive load identification method based on convolution neural network

Cited by 21 publications

References 3 publications

Stealthy Black-Box Attacks on Deep Learning Non-Intrusive Load Monitoring Models

Stealthy Black-Box Attacks on Deep Learning Non-Intrusive Load Monitoring Models

Non-Intrusive Load Monitoring Using a CNN-LSTM-RF Model Considering Label Correlation and Class-Imbalance

Identificação não-intrusiva de cargas similares em Smart Grid usando rede neural convolucional

Contact Info

Product

Resources

About