Methodology for online detection and classification of power quality disturbances based on FPGA

Abstract

The transition from conventional energy systems to decentralized generation based on renewable energy sources presents significant challenges. Sophisticated devices are required to monitor and manage the real-time flow and quality of energy. These tools require efficient algorithms that minimize computational complexity, particularly for real-time applications. This work proposes a novel, computationally efficient methodology for the real-time detection and classification of seven types of power quality disturbances (PQDs) based on Multiresolution Analysis of the Discrete Wavelet Transform (MRA-DWT) and feature extraction methods such as RMS and Logarithmic Energy Entropy. The extracted distinctive feature vector, consisting of seven elements, serves as input to a classifier based on a Feed Forward Neural Network (FFNN). The classifier identifies the type of disturbance in 8.30 microseconds, achieving classification accuracies of 97.7% with synthetic data and 98.57% with real data obtained from an arbitrary waveform generator. The proposed algorithm was implemented on the Pynq-Z1 board from Xilinx using Vitis IDE and enables online acquisition and feature extraction from approximation and detail coefficients across five levels of DWT decomposition. The system processes data within times shorter than the sampling period, remaining within 10% of the maximum processing speed required for a 10 kHz sampling rate. Its fully sequential operation avoids storing input signals or DWT coefficients. A detailed system performance analysis was also conducted, evaluating each input sample’s acquisition and processing times. The study considered 2000 samples obtained from the laboratory, demonstrating the system’s effectiveness for online and real-time applications.