Abstract
Non-Intrusive Load Monitoring (NILM) makes it possible for users and energy providers to track the fine-grained energy consumption information of residential and commercial buildings. The load identification methods in NILM usually require labeling many samples for training and evaluation, which is always expensive and time-consuming. In order to reduce the labeling cost, this paper proposed a load identification method based on Active Deep Learning (ADL). In this method, Discrete Wavelet Transform (DWT) was applied to extract high-dimensional appliance features from original current signals. Then a pool-based or stream-based active deep learning model was built to learn the features and select high-value samples that worthy of labeling. A mixed dataset based on three public datasets was formed to evaluate the proposed method and three sampling approaches of active learning. The results showed that the proposed method could significantly reduce labeling cost on large datasets, and the number of samples required is 33% lower than the state-of-the-art method when the F1 score is equal. Compared with pool-based sampling approaches, the stream-based approach's benefits are that the classifier improved and the query frequency decreased with continuous input of samples.
Highlights
Residential and commercial buildings account for more than 40% of global energy consumption and produce more than one-third of the total carbon dioxide emissions [1]
According to the above analysis, this paper proposed two highdimensional Discrete Wavelet Transform (DWT)-based features for active learning and an active deep learning model that apply to high-dimensional features
Compared with pool-based approaches, its benefits are that the classifier improved and the query frequency decreased significantly with the continuous input of samples
Summary
Residential and commercial buildings account for more than 40% of global energy consumption and produce more than one-third of the total carbon dioxide emissions [1]. Only ref [9]–[11] have discussed the active learning method for event-based NILM. All these studies were based on low-dimensional power variation features and the training dataset, named BLUED, only contained a small number of appliances. Reducing the labeling cost in large datasets is the primary purpose of active learning, and performance across datasets is an important criterion for method evaluation [13]. In the field of event-based NILM, no study has discussed the feature application problem and active deep learning methods. 1) A load identification method based on active deep learning and discrete wavelet analysis was proposed.
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
