Transfer learning for non-intrusive load monitoring & appliance identification in a smart home M. Hashim Shahab * , Hasan Mujtaba Buttar * , Ahsan Mehmood * , Waqas Aman † , M. Mahboob Ur Rahman * , M. Wasim Nawaz ‡ , Qammer H. Abbasi § * Electrical engineering department, Information Technology University, Lahore 54000, Pakistan † College of Science and Engineering, Hamad Bin Khalifa University (HBKU), Doha, Qatar ‡ Department of Computer Engineering, The University of Lahore, Lahore, 54000, Pakistan § Department of Electronics and Nano Engineering, University of Glasgow, Glasgow, G12 8QQ, UK * {mscs18011, mahboob.rahman}@itu.edu.pk, † waman@hbku.edu.qa Abstract—Non-intrusive load monitoring (NILM) or energy disaggregation is an inverse problem whereby the goal is to extract the load profiles of individual appliances, given an aggregate load profile of the mains of a home. NILM could help identify the power usage patterns of individual appliances in a home, and thus, could help realize novel energy conservation schemes for smart homes. In this backdrop, this work proposes a novel deep-learning approach to solve the NILM problem and a few related problems as follows. 1) We build upon the reputed seq2-point convolutional neural network (CNN) model to come up with the proposed seq2-[3]-point CNN model to solve the (home) NILM problem and site-NILM problem (basically, NILM at a smaller scale). 2) We solve the related problem of appliance identification by building upon the state-of-the-art (pre- trained) 2D-CNN models, i.e., AlexNet, ResNet-18, and DenseNet- 121, which are trained upon two custom datasets that consist of Wavelets and short-time Fourier transform (STFT)-based 2D electrical signatures of the appliances. 3) Finally, we do some basic qualitative inference about an individual appliance’s health by comparing the power consumption of the same appliance across multiple homes. Low-frequency REDD dataset is used to train and test the proposed deep learning models for all problems, except site-NILM where REFIT dataset has been used. As for the results, we achieve a maximum accuracy of 94.6% for home-NILM, 81% for site-NILM, and 88.9% for appliance identification (with Resnet-based model). I. I NTRODUCTION Climate change is for real, has led to increased temperatures and abnormal weather patterns worldwide lately, and is now globally accepted a real threat to the existence of life on planet earth. One major culprit to trigger climate change is ever- increasing CO 2 emissions. Thus, serious efforts have begun lately to cut down CO 2 emissions across the globe. One di- rection to pursue to partly achieve this goal is to reduce energy consumption worldwide. Thus, smart energy management or energy conservation systems are now considered as essential ingredients of smart green cities of future [1]. Energy disaggregation or non-intrusive load monitoring (NILM) is an inverse problem whereby the goal is to estimate the power consumption profile of multiple individual consumer appliances, given the aggregate power consumption profile of the mains of a home. It has been estimated that NILM could help reduce energy consumption up to 15% in a household setting [2]. Thus, NILM is being boasted as one key enabler to realize smart homes and smart cities of the future. The idea of NILM at the appliance level was first conceived by G. W. Hart [3]. He modeled each appliance as a finite state machine, and his proposed NILM algorithm consisted of four major steps: data collection, event detection, feature extraction, and load identification. He argued that both steady-state analy- sis (i.e., fundamental frequency, harmonics, direct current) and transients analysis (shape, size, duration of transients) provide valuable information for NILM purpose. NILM framework has spawned many other interesting prob- lems. For example, NILM finds its application in future smart grid systems where power utility companies will charge a customized tariff to their clients, based upon the appliance- level power usage data collected from their homes via smart meters [4]. NILM could also help consumers to cut down their energy bills by providing them a detailed picture of the power usage report of appliances in their homes. Some other examples include: energy audit of buildings, estimating the remaining useful life of an appliance, inferring the build quality of an appliance, demand response prediction, inferring the lifestyle of consumers in a home (based upon power consumption behavior of appliances in a household) etc 1 . Let us further highlight the significance of NILM, this time taking an application scenario from the domain of smart health. Smart homes of the future will have a home energy management system, which will in turn have an important component called ambient assisted living (AAL). One promise of AAL is that it could be beneficial for the elderly by determining the on and off states of the appliances in the home. Let us explain by assuming some elderly person in a home uses an oxygen machine, which is hooked up in the AAL context. If we do some basic analytics on the collected data, we may be able to infer something about the well-being of the elderly and/or the machine, and the AAL system could send an SOS alert to the local healthcare provider, if the need arises. Sleep disorder detection is another example where NILM-based AAL could help. 1 The interested reader is referred to the survey/review articles [13], [14], [15], [18] for a more systematic and detailed review of the works on NILM as well as the existing NILM datasets. 1 arXiv:2301.03018v1 [eess.SP] 8 Jan 2023