Real-Time Non-Intrusive Electrical Load Classification Over IoT Using Machine Learning

Abstract

In this era of technological advancement, the flow of an enormous amount of information has become such an inevitable phenomenon that makes a path for the takeover of the internet of things (IoT) based smart grid from the currently available grid system. In a smart grid, demand-side management plays a crucial role in reducing the generation capacity by shifting the user energy consumption from peak period to off-peak period, which requires detailed knowledge of the user consumption at the individual appliance level. Non-intrusive load monitoring (NILM) provides an exceptionally low-cost solution for determining individual appliance levels using a single-point measurement. This paper proposed an IoT-based real-time non-intrusive load classification (RT-NILC) system considering the variability of supply voltage using lowfrequency data. Due to the unavailability of smart meters at the household level in Bangladesh, a dataacquisition system (DAS) is developed. The DAS is capable of measuring and storing rms voltage, rms current, active power, and power factor data at a sampling rate of 1 Hz. These data are processed to train different multilabel classification models. The best-performed classification model has been selected and utilized for the implementation of RT-NILC over IoT. The Firebase real-time online database is considered for data storage to flow the data in two-way between end-user and service provider (energy distributor). The GPRS module is used for wireless data transmission as a Wi-Fi network may not be available everywhere. Windows and web application are developed for data visualization. The proposed system has been validated in real-time, using rms voltage, rms current, and active power measurements at a real house. Even under supply voltage variability, the performance evaluation of the RT-NILC system has shown an average classification accuracy of more than 94%. Good classification accuracy and the overall operation of the IoTbased information exchange systems ensure the proposed system's applicability for efficient energy management.INDEX TERMS Non-intrusive load monitoring, Real-time load classification, IoT framework, Machine learning, Variation of supply voltage.

A survey of intelligent load monitoring in IoT-enabled distributed smart grids

“…Data acquisition plan is implemented every minute to collect overall data of battery management System BMS IoT to the ECS. Ahammed et al (2021) proposed a RT-NILC (real-time non-invasive load classification) system based on the IoT, which has developed a DAS (data acquisition system) and measures and stores RMS voltage, current, active power and power factor data at a sampling rate of 1 Hz. Balakrishna et al (2018) proposed the IoT sensor data acquisition and analysis framework for data analysis and visualization.…”

Application of Internet of Things in Residential Distribution Systems

“…Efficient control is studied in [33] for monitoring temperature using computational intelligence so as to provide an IoT convection system based on data from several installed sensors to improve living quality. Non-intrusive load monitoring in an IoT environment is considered for low-cost smart home applications [34] by installing a single-entry point sensor to determine the required power for each home appliance, leading to improved appliance safety. In [35], the feasibility of integrating energy storage and energy harvesting applications to produce smart windows technology is reviewed along with its potential for net-zero buildings.…”

Analysis of Using Machine Learning Techniques for Estimating Solar Panel Performance in Edge Sensor Devices

“…Further, the examples of the implementation of AI on Arduino-based development boards are presented in the following papers. Paper [11] proposes a real-time non-intrusive load classification (RT-NILC) IoT-based system with an Arduino-based data acquisition system. In [12], a small-scale two-wheel system connected to a control unit is developed using an ARDUINO Uno Rev3 microcontroller and a Support Vector Regression (SVR) model.…”