With the low integration costs and quick development cycle of all-IP-based 5G+ technologies, it is not surprising that the proliferation of IP devices for residential or industrial purposes is ubiquitous. Energy scheduling/management and automated device recognition are popular research areas in the engineering community, and much time and work have been invested in producing the systems required for smart city networks. However, most proposed approaches involve expensive and invasive equipment that produces huge volumes of data (high-frequency complexity) for analysis by supervised learning algorithms. In contrast to other studies in the literature, we propose an approach based on encoding consumption data into vehicular mobility and imaging systems to apply a simple convolutional neural network to recognize certain scenarios (devices powered on) in real-time and based on the Non-Intrusive Load Monitoring (NILM) paradigm. Our idea is based on a very cheap device and can be adapted at a very low cost for any real scenario. We have also created our own dataset, taken from a real domestic environment, contrary to most existing works based on synthetic data. The results of the study’s simulation demonstrate the effectiveness of this innovative and low-cost approach and its scalability in function of the number of considered appliances.
Load Monitoring and Appliance Recognition Using an Inexpensive, Low Frequency, Data-to-Image, Neural Network and Network Mobility Approach for Domestic IoT Systems
Fazio, Peppino
;
2023-01-01
Abstract
With the low integration costs and quick development cycle of all-IP-based 5G+ technologies, it is not surprising that the proliferation of IP devices for residential or industrial purposes is ubiquitous. Energy scheduling/management and automated device recognition are popular research areas in the engineering community, and much time and work have been invested in producing the systems required for smart city networks. However, most proposed approaches involve expensive and invasive equipment that produces huge volumes of data (high-frequency complexity) for analysis by supervised learning algorithms. In contrast to other studies in the literature, we propose an approach based on encoding consumption data into vehicular mobility and imaging systems to apply a simple convolutional neural network to recognize certain scenarios (devices powered on) in real-time and based on the Non-Intrusive Load Monitoring (NILM) paradigm. Our idea is based on a very cheap device and can be adapted at a very low cost for any real scenario. We have also created our own dataset, taken from a real domestic environment, contrary to most existing works based on synthetic data. The results of the study’s simulation demonstrate the effectiveness of this innovative and low-cost approach and its scalability in function of the number of considered appliances.I documenti in ARCA sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.