Sistema de identificação de dispositivos elétricos baseado na classificação de corrente alternada com aprendizado de máquina

Abstract

Given the advancement of technologies applied to energy management, there is a growing demand for solutions capable of accurately identifying the individual consumption of appliances in residential environments. However, a significant portion of recent research relies strictly on synthetic or pre-existing datasets, inherent complexities within the physical implementation during the acquisition phase. This work developed and validated a complete non-intrusive monitoring system, ranging from the construction of instrumentation hardware to classification via Machine Learning algorithms. The proposed architecture integrated a microcontroller, a noninvasive current sensor, and a high-resolution analog-to-digital converter, ensuring fidelity in the collection of current signatures. The experimental data obtained were used to train and compare classical classifiers and deep neural networks, evaluating the robustness of the models against different electrical grid topologies. The results evidenced that the quality of instrumentation allows models with lower computational cost to achieve high precision, validating the viability of autonomous monitoring systems. Thus, the approach presents a verticalized and scalable solution, contributing to bridging the gap between theoretical modeling and practical application in energy efficiency.