Review on Deep Neural Networks Applied to Low-Frequency NILM

Abstract

This paper reviews non-intrusive load monitoring (NILM) approaches that employ deep neural networks to disaggregate appliances from low frequency data, i.e. data with sampling rates lower than the AC base frequency. We first review the many degrees of freedom of these approaches, what has already been done in literature, and compile the main characteristics of the reviewed publications in an extensive overview table. The second part of the paper discusses selected aspects of the literature and corresponding research gaps. In particular, we do a performance comparison with respect to reported MAE and F$_1$-scores and observe different recurring elements in the best performing approaches, namely data sampling intervals below 10\,s, a large field of view, the usage of GAN losses, multi-task learning, and post-processing. Subsequently, multiple input features, multi-task learning and related research gaps are discussed, the need for comparative studies is highlighted, and finally, missing elements for a successful deployment of NILM approaches based on deep neural networks are pointed out. We conclude the review with an outlook on possible future scenarios.

Towards Trustworthy Energy Disaggregation: A Review of Challenges, Methods, and Perspectives for Non-Intrusive Load Monitoring

“…Regarding data sampling period, a coarse division of 1s usually enables feature separation between macroscopic (low-frequency) and microscopic (high-frequency) components. Even though, low-frequency data sets greatly reduce the ability to distinguish among different types of appliances, compared to high-frequency data sets, the first are gaining more and more ground in NILM deep learning algorithms [127].…”

“…The existing surveys for NILM, highlight the importance of selecting the right dataset. Huber et al [127] summarize the main characteristics of the open access data sets. In addition, in [52] a comparison between the different open access datasets is performed.…”

Quantification of Dairy Farm Energy Consumption to Support the Transition to Sustainable Farming

“…The state of the art in terms of methodology and datasets for low-frequency NILM based on deep neural networks is summarised in a recent review [6], and shows that there has been significant progress in designing NILM methods for residential buildings and that data sampling intervals below 10 seconds, large field of view, usage of GAN losses, and post-processing, provide the most accurate results (in terms of classification and estimation accuracy). However, a recent review for NILM in commercial buildings [7], indicates that progress in commercial NILM, at any sampling rate, is slow because of the following challenges arising in non-residential buildings.…”

“…We leverage on one of the best performing low frequency deep learning based regression networks as highlighted in Fig. 1 [6], namely the WaveNet-based algorithm of [9]. Good performance was demonstrated in [9] on household appliances, which are very distinct to those observed in milk production loads.…”

Appliance Level Energy Characterization of Residential Electricity Demand: Prospects, Challenges and Recommendations

“…Some recent DNN models used for NILM include recurrent convolutional denoising autoencoder [62], Convolutional Neural Networks (CNN) [63,64], adaptive weighted recurrent graphs [65], Long Short Term Memory (LSTM) networks [62,66,67], Generative Adversial Networks (GANs) [68], sequence to sequence learning [69], Sequence To Point (seq2pt) learning [70], deep pair supervising hash [71], attention-based DNN [51], etc. Authors of [72] have presented a comprehensive review of DL-based NILM approaches applied to low-frequency data. Many researchers extended the concept of transfer learning to solve the NILM problem.…”