Hybrid Attack Optimization Supported Enhanced Deep Learning to Facilitate Power System Event Detection using PMU Data

Abstract

Accurate event detection is crucial for initiating control and protection measures in power systems to ensure enhanced and reliable operation. Phasor measurement units (PMUs) play a vital role in various functional aspects of power systems, including state estimation and intelligent protection algorithms. However, the authenticity of real-time data from PMUs must be verified before feeding it into applicable real-time algorithms to prevent undesirable or erroneous operations. This paper aims to present an efficient preprocessing methodology for identifying unwanted, incorrect, missing, or noisy PMU data to facilitate robust event detection algorithms. The proposed methodology leverages real-time data-driven deep learning techniques for authenticating incoming data. Given the high sampling rate of PMUs, the presence of extraneous data can lead to false event detection, necessitating reliable data preprocessing. Challenges identified in existing literature, such as the limitations of Steady State (SS)-Local Outlier Factor (LOF) in event detection and classification, issues with detecting line tripping and inter-area oscillations, computational and bandwidth requirements for micro-PMU installations, and false alarms resulting from inaccuracies in frequency ramp rate determination, are addressed. To overcome these challenges, this research proposes a deep learning approach that utilizes modified Deep Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) classifiers to classify and extract features from PMU data, enabling highly efficient detection of disturbances in transmission lines. Additionally, a hybrid attack optimization (HAO) technique is employed to enhance convergence rates, accuracy, and efficiency. Performance evaluation of the proposed system is conducted by calculating and assessing disturbances generated by power lines using metrics such as accuracy, precision, recall, System Average Interruption Duration Index (SAIDI), and System Average Interruption Frequency Index (SAIFI).