Design and Implementation of a TinyML-Based Power Quality Disturbance Classification System on the STM32H743VIT6
Loading...
Date
Journal Title
Journal ISSN
Volume Title
Publisher
Higher School In Applied Sciences -Tlemcen-
DOI
Abstract
This work presents the development of an intelligent embedded system for power quality disturbance
monitoring and classification using artificial intelligence. A synthetic dataset containing 17,000 signals
across 17 classes was used to train and compare two deep learning models: CNN-1D and BiLSTM.
The models achieved test accuracies of 95.5% and 93%, respectively. The CNN-1D model was
successfully deployed and validated on the STM32H743 microcontroller using TinyML, achieving an
inference time of 24.873 ms. Noise robustness experiments revealed a significant performance degradation,
highlighting the need for further improvements and additional validation using real electrical
signals before practical deployment.
