Applications and Comparison of Deep Learning in Low Frequency Non-intrusive Load Disaggregation

Subjects: Computer Science >> Other Disciplines of Computer Science submitted time 2021-10-01

Li, Yiming Ju, Yuntao Qu, Liao

Abstract： Non-intrusive load disaggregation can comprehensively analyze user power consumption data and has been a crucial technology to evaluate users' flexible control potential. Since deep learning is currently the prevailing method of load disaggregation, the working mechanism of the three mainstream network structures, which includes denoising autoencoder, recurrent neural network and convolutional neural network, is discussed in detail when applying to load disaggregation, as well as their outlooks for applying in load disaggregation field. Afterwards, this paper proposes an evaluation framework for load disaggregation algorithms based on different dimensions and regularizes the selection of test data during the evaluation process accordingly. Finally, we make an empirical comparison of some mainstream deep learning disaggregation models by utilizing the proposed framework. And we also open source the code of the evaluation model. The evaluation result proves that the proposed framework can more comprehensively evaluate the deep learning disaggregation algorithms under specific hyper-parameters settings, and reveal the sensitivity of model performance to distinct network architectures and input features." "

Peer Review Status:Awaiting Review

 Hits 8528  Downloads 689  Comment