Convolutional Neural Network Based Real-Time Power Quality Disturbance Detection and Classification

Increased instalment of power electronic devices and alternative energy sources to grids leads to increase in power quality problems representing a threat to power-sensitive loads and electronically controlled equipment. Dealing with power quality problems is an important concern viewed from life-cycle of equipment and facilitating a stable manufacturing process in industrial plants. Accurate examination of anomalies in power quality in real-time is required to decide which preventive measure to take. There are numerous studies employing machine learning and optimization methods to detect and classify the power quality events in literature. However, only a few dwell on real-time performance of proposed approaches. This study presents a closed-loop approach based on one-dimensional convolutional neural network model to capture power quality events in time domain requiring no human intervention from training to inference. An application, accompanied by a data acquisition tool, is developed to demonstrate the performance of the proposed approach using the Tensorflow C++ library. The application is ported to Raspberry Pi 5 and Jetson Orin Nano single board computers to reveal the embedded system performance of the presented system.