Machine-Learning-Based Classification of Electronic Devices Using an IoT Smart Meter

Machine-Learning-Based Classification of Electronic Devices Using an IoT Smart Meter

This study investigates the implementation of artificial intelligence (AI) algorithms on resource-constrained edge devices, such as ESP32 and Raspberry Pi, within the context of smart grid (SG) applications. Specifically, it proposes a smart-meter-based system capable of classifying and detecting th...

Full description

Saved in:
Bibliographic Details
Main Authors: Paulo Eugênio da Costa Filho, Leonardo Augusto de Aquino Marques, Israel da S. Felix de Lima, Ewerton Leandro de Sousa, Márcio Eduardo Kreutz, Augusto V. Neto, Eduardo Nogueira Cunha, Dario Vieira
Format: Article
Language:English
Published: MDPI AG 2025-05-01
Series:Informatics
Subjects:
artificial intelligence
extreme edge
IoT
smart grid
Online Access:https://www.mdpi.com/2227-9709/12/2/48
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This study investigates the implementation of artificial intelligence (AI) algorithms on resource-constrained edge devices, such as ESP32 and Raspberry Pi, within the context of smart grid (SG) applications. Specifically, it proposes a smart-meter-based system capable of classifying and detecting the Internet of Things (IoT) electronic devices at the extreme edge. The smart meter developed in this work acquires real-time voltage and current signals from connected devices, which are used to train and deploy lightweight machine learning models—Multi-Layer Perceptron (MLP) and K-Nearest Neighbor (KNN)—directly on edge hardware. The proposed system is integrated into the Artificial Intelligence in the Internet of Things for Smart Grids IAIoSGT architecture, which supports edge–cloud processing and real-time decision-making. A literature review highlights the key gaps in the existing approaches, particularly the lack of embedded intelligence for load identification at the edge. The experimental results emphasize the importance of data preprocessing—especially normalization—in optimizing model performance, revealing distinct behavior between MLP and KNN models depending on the platform. The findings confirm the feasibility of performing accurate low-latency classification directly on smart meters, reinforcing the potential of scalable AI-powered energy monitoring systems in SG.
ISSN:2227-9709

Similar Items