Convolutional Neural Network Applied to the Identification of Residential Equipment in Nonintrusive Load Monitoring Systems

Abstract

This paper presents the proposal of A new methodology for the identification of residential equipment in non-intrusive load monitoring systems that is based on a Convolutional Neural Network to classify equipment. The transient power signal data obtained at the time an equipment is connected in a residence is used as inputs to the system. The methodology was developed using data from a public database (REED) that presents data collected at a low frequency (1 Hz). The results obtained in the test database indicate that the proposed system is able to carry out the identification task, and presented satisfactory results when compared with the results already presented in the literature for the problem in question.

Appliance Classification by Power Signal Analysis Based on Multi-Feature Combination Multi-Layer LSTM

“…Various studies have been conducted to overcome the limitations of existing classification algorithms for NILM and mainly have been developed based on the use of deep neural networks (DNNs), so-called deep learning, which can analyze the power data more accurately in general [19][20][21][22][23][24]. It has best-in-class performance on problems that significantly outperforms other solutions in multiple different domains including speech recognition, speech synthesis, natural language processing, computer vision, computer games, and so forth, with a significant margin.…”

“…In order to overcome this, the use of deep neural networks (DNNs) has recently been proposed for appliance classification [19]. Such approaches are found to overcome the limitations of existing algorithms and to achieve higher accuracy regardless of the number of appliances [20][21][22][23][24].…”

Deep Neural Networks with Application to Transformer Failure Diagnosis

“…Esses métodos vêm sendo aplicados com sucesso em muitos campos como, reconhecimento de fala, reconhecimento de objetos visuais, detecção de objetos, tradução de idiomas, entre outros. Além dos resultados promissores para problemas envolvendo a aplicação de dados de imagem 2-D, alguns autores vêm desenvolvendo pesquisas na área de aplicação das redes neurais profundas em problemas com dados 1-D [Penha e Castro 2017], tais como dados de séries temporais. Dentre as redes neurais profundas destacam-se as Unidades Long Short Term (LSTM), as Redes Neurais Autocodificadoras, as Redes Autocodificadoras Empilhadas (Stacked Autoencoders) e as Redes Neurais Convolucionais (Convolutional Neural Network -CNN).…”

Home Appliance Identification for Nilm Systems Based on Deep Neural Networks

“…Different from the work presented in [29], the database for development of the identification system has the same amount of standards for each equipment, reaching 600 standards, considering all 6 equipments. Another difference is the fact that each pattern presents 12 transient samples of a given equipment (improving the representation with larger transients), thus forming a 12x600 two-dimensional array.…”

“…For this, it was necessary to only resize the training input matrix to 4D, assuming the dimension 1x12x1x360, and in this way CNN interprets the data as a 4-D numerical matrix (a cluster of colored images), where the first three dimensions refer to the height, width and channels and the last dimension should index the individual images, that is, index the transients. [29] For this approach, which is focused on the classification of equipment through the behavior of its power transients, an architecture based on three layers of convolution followed by pooling was used. Between each convolution and pooling layer normalization is applied in the filter sets (batches), which serves to accelerate network formation and reduce sensitivity for initialization.…”