Desenvolvimento de um sistema de gerenciamento de energia em edifícios prediais baseado em Edge Computing e Machine Learning

Abstract

This work presents the development and experimental validation of an energy management and optimization system for buildings, based on Edge Computing and Machine Learning techniques, using the ESP32-S3 microcontroller. The increasing demand for electrical energy, combined with the need for sustainable solutions, has driven the adoption of intelligent technologies capable of monitoring and controlling energy consumption in real time. In this context, the proposed solution integrates concepts from the Internet of Things, edge computing, and artificial intelligence, enabling local data processing while reducing latency and bandwidth consumption. The adopted methodology follows the Design Science Research (DSR) approach, encompassing literature review, hardware and software prototyping, and laboratory testing. The system architecture consists of a central unit and distributed measurement and control modules, interconnected through a wireless communication protocol based on ESP-NOW, eliminating the need for external infrastructure. The system employs current and voltage sensors based on the ATM90E36 integrated circuit, combined with signal conditioning circuits. Experimental results demonstrate high measurement accuracy, with a coefficient of determination above 0.9998 and mean errors below ±0.30%. The communication achieved a success rate of 99.5% and an average latency of 5.8 ms, even in environments with physical obstacles. The Random Forest Regressor model, trained on data collected over a 30-day period, achieved a mean absolute error of 70.71 W. The model was successfully deployed directly on the ESP32-S3 firmware, with an inference time of 15 ms, without requiring external connectivity. The results indicate that the proposed system is a low-cost, efficient, and scalable solution for energy management in buildings, demonstrating the feasibility of applying machine learning techniques in embedded systems to support energy efficiency applications.