Low-cost IoT-based smart meter for real-time power quality monitoring and disturbance detection using embedded 1D CNN

Abstract

Power quality disturbances (PQDs) such as voltage sags, swells, interruptions, and harmonics significantly impact the performance and stability of electrical distribution networks. With the growing demand for intelligent, embedded monitoring systems in smart grids, edge computing offers a practical approach to real-time disturbance detection. This paper presents a compact and efficient PQD classification system based on a quantized 1D Convolutional Neural Network (CNN), deployed on an ESP32 microcontroller. The model, developed using Edge Impulse, was compressed to approximately 2 KB, allowing deployment on low-power hardware. Key signal features such as root mean square voltage (VRMS) and total harmonic distortion (THD) were extracted to aid in classification. Real-time validation was performed using both live signals and real-world data obtained from the National Company for Electricity and Gas distribution areas, covering various single and combined PQD events. The ESP32 was connected to the Blynk IoT platform for live monitoring, where detected disturbances triggered mobile alerts, virtual LED indicators, and sound notifications. Results confirm the system’s high classification accuracy and responsiveness, demonstrating the potential of lightweight neural models for edge-based PQ monitoring.