SAED: self-attentive energy disaggregation

Abstract

The field of energy disaggregation deals with the approximation of appliance electric consumption using only the aggregate consumption measurement of a mains meter. Recent research developments have used deep neural networks and outperformed previous methods based on Hidden Markov Models. On the other hand, deep learning models are computationally heavy and require huge amounts of data. The main objective of the current paper is to incorporate the attention mechanism into neural networks in order to reduce their computational complexity. For the attention mechanism two different versions are utilized, named Additive and Dot Attention. The experiments show that they perform on par, while the Dot mechanism is slightly faster. The two versions of self-attentive neural networks are compared against two state-of-the-art energy disaggregation deep learning models. The experimental results show that the proposed architecture achieves faster or equal training and inference time and with minor performance drop depending on the device or the dataset.

Neural Fourier Energy Disaggregation

“…Models based on the attention mechanism demonstrate promising results in terms of generalization to unseen data. The attention mechanism is incorporated using the self-attention method [11,20,21] or the transformer architecture [22]. Recently, generative models have been proposed for the problem of NILM by using GANs [23] or variational approaches [24][25][26].…”

“…To overcome the aforementioned reproducibility issues, the baseline models are selected based on how easy it is to replicate past experimental results, their wider acceptance by other NILM researchers and the existence of implementations in open source projects such as NILMTK [6,37]. The baseline models are: a convolutional neural network named "sequence-to-point" (S2P) [10], a recurrent neural network named "online GRU" or "window GRU" (WGRU) [9] and a neural network based on the self-attention mechanism named "self-attentive energy dissaggragator" (SAED) Virtsionis-Gkalinikis et al [11]. The first two models have been used either as baselines or as a basis to develop new architectures.…”

“…Figure 3 shows the details of the architecture. Self-attentive energy disaggregator (SAED) is based on the mechanism of attention and is developed by Virtsionis-Gkalinikis et al [11]. It is a computationally efficient neural network.…”

“…The third contribution is a comprehensive comparative analysis that aims to find the best model per appliance through an extensive tuning methodology that takes into account both experimental and architectural hyper-parameters. The models that are compared are NFED, window-GRU (WGRU) [9], sequence-to-point (S2P) [10] and selfattentive energy disaggregator (SAED) [11]. For a fair comparison the best environmental setup is found for each of the four models that are compared.…”

Efficient Supervised Machine Learning Network for Non-Intrusive Load Monitoring

“…The active and reactive power signatures were then matched to the right appliance using a best likelihood method, and similar "steady-state" elements of the power signal were grouped together. Certain two-state (on/off) appliances have been identified with good accuracy using such clustering approaches [10,11]. These methods, on the other hand, have major trouble detecting more complicated appliances with numerous states (e.g., washing machines) and have a tendency to fail in situations when multiple appliances are operating and switching at the same time [12].…”